Fix undefined behavior in PPC load.

VSX vec_xl is Causing a bunch of test failures and failing `-fsanitize=undefined` with g++.
Removing the instruction allows tests to pass and eliminates the warning.

.gitignore

@@ -35,7 +35,3 @@ lapack/reference
 .*project
 .settings
 Makefile
-!ci/build.gitlab-ci.yml
-!scripts/buildtests.in
-!Eigen/Core
-!Eigen/src/Core

.gitlab/issue_templates/Bug Report.md

@@ -1,69 +0,0 @@
-<!--
-Please read this!
-
-Before opening a new issue, make sure to search for keywords in the issues
-filtered by "bug::confirmed" or "bug::unconfirmed" and "bugzilla" label:
-
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bug%3A%3Aconfirmed
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bug%3A%3Aunconfirmed
-- https://gitlab.com/libeigen/eigen/-/issues?scope=all&utf8=%E2%9C%93&state=opened&label_name[]=bugzilla
-
-and verify the issue you're about to submit isn't a duplicate. -->
-
-### Summary
-<!-- Summarize the bug encountered concisely. -->
-
-### Environment
-<!-- Please provide your development environment here -->
-- **Operating System** : Windows/Linux
-- **Architecture** : x64/Arm64/PowerPC ...
-- **Eigen Version** : 3.3.9
-- **Compiler Version** : Gcc7.0
-- **Compile Flags** : -O3 -march=native
-- **Vector Extension** : SSE/AVX/NEON ...
-
-### Minimal Example
-<!-- If possible, please create a minimal example here that exhibits the problematic behavior.
-You can also link to [godbolt](https://godbolt.org). But please note that you need to click 
-the "Share" button in the top right-hand corner of the godbolt page where you reproduce the sample 
-code to get the share link instead of in your browser address bar. 
-
-You can read [the guidelines on stackoverflow](https://stackoverflow.com/help/minimal-reproducible-example)
-on how to create a good minimal example. -->
-
-```cpp
-//show your code here
-```
-
-### Steps to reproduce
-<!-- Describe how one can reproduce the issue - this is very important. Please use an ordered list. -->
-
-1. first step
-2. second step
-3. ... 
-
-### What is the current *bug* behavior?
-<!-- Describe what actually happens. -->
-
-### What is the expected *correct* behavior?
-<!-- Describe what you should see instead. -->
-
-### Relevant logs
-<!-- Add relevant code snippets or program output within blocks marked by " ``` " -->
-
-<!-- OPTIONAL: remove this section if you are not reporting a compilation warning issue.-->
-### Warning Messages
-<!-- Show us the warning messages you got! -->
-
-<!-- OPTIONAL: remove this section if you are not reporting a performance issue. -->
-### Benchmark scripts and results
-<!-- Please share any benchmark scripts - either standalone, or using [Google Benchmark](https://github.com/google/benchmark). -->
-
-### Anything else that might help
-<!-- It will be better to provide us more information to help narrow down the cause. 
-Including but not limited to the following: 
-- lines of code that might help us diagnose the problem. 
-- potential ways to address the issue.
-- last known working/first broken version (release number or commit hash). --> 
-
-- [ ] Have a plan to fix this issue.

.gitlab/issue_templates/Feature Request.md

@@ -1,7 +0,0 @@
-### Describe the feature you would like to be implemented.
-
-### Would such a feature be useful for other users? Why?
-
-### Any hints on how to implement the requested feature?
-
-### Additional resources

.gitlab/merge_request_templates/Merge Request Template.md

@@ -1,26 +0,0 @@
-<!-- 
-Thanks for contributing a merge request! Please name and fully describe your MR as you would for a commit message.
-If the MR fixes an issue, please include "Fixes #issue" in the commit message and the MR description.
-
-In addition, we recommend that first-time contributors read our [contribution guidelines](https://eigen.tuxfamily.org/index.php?title=Contributing_to_Eigen) and [git page](https://eigen.tuxfamily.org/index.php?title=Git), which will help you submit a more standardized MR.
-
-Before submitting the MR, you also need to complete the following checks:
-- Make one PR per feature/bugfix (don't mix multiple changes into one PR). Avoid committing unrelated changes.
-- Rebase before committing
-- For code changes, run the test suite (at least the tests that are likely affected by the change).
-  See our [test guidelines](https://eigen.tuxfamily.org/index.php?title=Tests).
-- If possible, add a test (both for bug-fixes as well as new features)
-- Make sure new features are documented
-
-Note that we are a team of volunteers; we appreciate your patience during the review process.
-
-Again, thanks for contributing! -->
-
-### Reference issue
-<!-- You can link to a specific issue using the gitlab syntax #<issue number>  -->
-
-### What does this implement/fix?
-<!--Please explain your changes.-->
-
-### Additional information
-<!--Any additional information you think is important.-->

COPYING.APACHE

@@ -1,203 +0,0 @@
-/*
-                                 Apache License
-                           Version 2.0, January 2004
-                        http://www.apache.org/licenses/
-
-   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
-
-   1. Definitions.
-
-      "License" shall mean the terms and conditions for use, reproduction,
-      and distribution as defined by Sections 1 through 9 of this document.
-
-      "Licensor" shall mean the copyright owner or entity authorized by
-      the copyright owner that is granting the License.
-
-      "Legal Entity" shall mean the union of the acting entity and all
-      other entities that control, are controlled by, or are under common
-      control with that entity. For the purposes of this definition,
-      "control" means (i) the power, direct or indirect, to cause the
-      direction or management of such entity, whether by contract or
-      otherwise, or (ii) ownership of fifty percent (50%) or more of the
-      outstanding shares, or (iii) beneficial ownership of such entity.
-
-      "You" (or "Your") shall mean an individual or Legal Entity
-      exercising permissions granted by this License.
-
-      "Source" form shall mean the preferred form for making modifications,
-      including but not limited to software source code, documentation
-      source, and configuration files.
-
-      "Object" form shall mean any form resulting from mechanical
-      transformation or translation of a Source form, including but
-      not limited to compiled object code, generated documentation,
-      and conversions to other media types.
-
-      "Work" shall mean the work of authorship, whether in Source or
-      Object form, made available under the License, as indicated by a
-      copyright notice that is included in or attached to the work
-      (an example is provided in the Appendix below).
-
-      "Derivative Works" shall mean any work, whether in Source or Object
-      form, that is based on (or derived from) the Work and for which the
-      editorial revisions, annotations, elaborations, or other modifications
-      represent, as a whole, an original work of authorship. For the purposes
-      of this License, Derivative Works shall not include works that remain
-      separable from, or merely link (or bind by name) to the interfaces of,
-      the Work and Derivative Works thereof.
-
-      "Contribution" shall mean any work of authorship, including
-      the original version of the Work and any modifications or additions
-      to that Work or Derivative Works thereof, that is intentionally
-      submitted to Licensor for inclusion in the Work by the copyright owner
-      or by an individual or Legal Entity authorized to submit on behalf of
-      the copyright owner. For the purposes of this definition, "submitted"
-      means any form of electronic, verbal, or written communication sent
-      to the Licensor or its representatives, including but not limited to
-      communication on electronic mailing lists, source code control systems,
-      and issue tracking systems that are managed by, or on behalf of, the
-      Licensor for the purpose of discussing and improving the Work, but
-      excluding communication that is conspicuously marked or otherwise
-      designated in writing by the copyright owner as "Not a Contribution."
-
-      "Contributor" shall mean Licensor and any individual or Legal Entity
-      on behalf of whom a Contribution has been received by Licensor and
-      subsequently incorporated within the Work.
-
-   2. Grant of Copyright License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      copyright license to reproduce, prepare Derivative Works of,
-      publicly display, publicly perform, sublicense, and distribute the
-      Work and such Derivative Works in Source or Object form.
-
-   3. Grant of Patent License. Subject to the terms and conditions of
-      this License, each Contributor hereby grants to You a perpetual,
-      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
-      (except as stated in this section) patent license to make, have made,
-      use, offer to sell, sell, import, and otherwise transfer the Work,
-      where such license applies only to those patent claims licensable
-      by such Contributor that are necessarily infringed by their
-      Contribution(s) alone or by combination of their Contribution(s)
-      with the Work to which such Contribution(s) was submitted. If You
-      institute patent litigation against any entity (including a
-      cross-claim or counterclaim in a lawsuit) alleging that the Work
-      or a Contribution incorporated within the Work constitutes direct
-      or contributory patent infringement, then any patent licenses
-      granted to You under this License for that Work shall terminate
-      as of the date such litigation is filed.
-
-   4. Redistribution. You may reproduce and distribute copies of the
-      Work or Derivative Works thereof in any medium, with or without
-      modifications, and in Source or Object form, provided that You
-      meet the following conditions:
-
-      (a) You must give any other recipients of the Work or
-          Derivative Works a copy of this License; and
-
-      (b) You must cause any modified files to carry prominent notices
-          stating that You changed the files; and
-
-      (c) You must retain, in the Source form of any Derivative Works
-          that You distribute, all copyright, patent, trademark, and
-          attribution notices from the Source form of the Work,
-          excluding those notices that do not pertain to any part of
-          the Derivative Works; and
-
-      (d) If the Work includes a "NOTICE" text file as part of its
-          distribution, then any Derivative Works that You distribute must
-          include a readable copy of the attribution notices contained
-          within such NOTICE file, excluding those notices that do not
-          pertain to any part of the Derivative Works, in at least one
-          of the following places: within a NOTICE text file distributed
-          as part of the Derivative Works; within the Source form or
-          documentation, if provided along with the Derivative Works; or,
-          within a display generated by the Derivative Works, if and
-          wherever such third-party notices normally appear. The contents
-          of the NOTICE file are for informational purposes only and
-          do not modify the License. You may add Your own attribution
-          notices within Derivative Works that You distribute, alongside
-          or as an addendum to the NOTICE text from the Work, provided
-          that such additional attribution notices cannot be construed
-          as modifying the License.
-
-      You may add Your own copyright statement to Your modifications and
-      may provide additional or different license terms and conditions
-      for use, reproduction, or distribution of Your modifications, or
-      for any such Derivative Works as a whole, provided Your use,
-      reproduction, and distribution of the Work otherwise complies with
-      the conditions stated in this License.
-
-   5. Submission of Contributions. Unless You explicitly state otherwise,
-      any Contribution intentionally submitted for inclusion in the Work
-      by You to the Licensor shall be under the terms and conditions of
-      this License, without any additional terms or conditions.
-      Notwithstanding the above, nothing herein shall supersede or modify
-      the terms of any separate license agreement you may have executed
-      with Licensor regarding such Contributions.
-
-   6. Trademarks. This License does not grant permission to use the trade
-      names, trademarks, service marks, or product names of the Licensor,
-      except as required for reasonable and customary use in describing the
-      origin of the Work and reproducing the content of the NOTICE file.
-
-   7. Disclaimer of Warranty. Unless required by applicable law or
-      agreed to in writing, Licensor provides the Work (and each
-      Contributor provides its Contributions) on an "AS IS" BASIS,
-      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
-      implied, including, without limitation, any warranties or conditions
-      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
-      PARTICULAR PURPOSE. You are solely responsible for determining the
-      appropriateness of using or redistributing the Work and assume any
-      risks associated with Your exercise of permissions under this License.
-
-   8. Limitation of Liability. In no event and under no legal theory,
-      whether in tort (including negligence), contract, or otherwise,
-      unless required by applicable law (such as deliberate and grossly
-      negligent acts) or agreed to in writing, shall any Contributor be
-      liable to You for damages, including any direct, indirect, special,
-      incidental, or consequential damages of any character arising as a
-      result of this License or out of the use or inability to use the
-      Work (including but not limited to damages for loss of goodwill,
-      work stoppage, computer failure or malfunction, or any and all
-      other commercial damages or losses), even if such Contributor
-      has been advised of the possibility of such damages.
-
-   9. Accepting Warranty or Additional Liability. While redistributing
-      the Work or Derivative Works thereof, You may choose to offer,
-      and charge a fee for, acceptance of support, warranty, indemnity,
-      or other liability obligations and/or rights consistent with this
-      License. However, in accepting such obligations, You may act only
-      on Your own behalf and on Your sole responsibility, not on behalf
-      of any other Contributor, and only if You agree to indemnify,
-      defend, and hold each Contributor harmless for any liability
-      incurred by, or claims asserted against, such Contributor by reason
-      of your accepting any such warranty or additional liability.
-
-   END OF TERMS AND CONDITIONS
-
-   APPENDIX: How to apply the Apache License to your work.
-
-      To apply the Apache License to your work, attach the following
-      boilerplate notice, with the fields enclosed by brackets "[]"
-      replaced with your own identifying information. (Don't include
-      the brackets!)  The text should be enclosed in the appropriate
-      comment syntax for the file format. We also recommend that a
-      file or class name and description of purpose be included on the
-      same "printed page" as the copyright notice for easier
-      identification within third-party archives.
-
-   Copyright [yyyy] [name of copyright owner]
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
-*/

COPYING.MINPACK

@@ -49,3 +49,4 @@ SUCH LIABILITY IS ASSERTED ON THE BASIS OF CONTRACT, TORT
 (INCLUDING NEGLIGENCE OR STRICT LIABILITY), OR OTHERWISE,
 EVEN IF ANY OF SAID PARTIES HAS BEEN WARNED OF THE
 POSSIBILITY OF SUCH LOSS OR DAMAGES.
+

CTestConfig.cmake

@@ -2,8 +2,8 @@
 ## Then modify the CMakeLists.txt file in the root directory of your
 ## project to incorporate the testing dashboard.
 ## # The following are required to uses Dart and the Cdash dashboard
-##   enable_testing()
+##   ENABLE_TESTING()
-##   include(CTest)
+##   INCLUDE(CTest)
 set(CTEST_PROJECT_NAME "Eigen")
 set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC")
 
@@ -11,7 +11,3 @@ set(CTEST_DROP_METHOD "http")
 set(CTEST_DROP_SITE "my.cdash.org")
 set(CTEST_DROP_LOCATION "/submit.php?project=Eigen")
 set(CTEST_DROP_SITE_CDASH TRUE)
-#set(CTEST_PROJECT_SUBPROJECTS
-#Official
-#Unsupported
-#)

Eigen/CMakeLists.txt

@@ -0,0 +1,19 @@
+include(RegexUtils)
+test_escape_string_as_regex()
+
+file(GLOB Eigen_directory_files "*")
+
+escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
+
+foreach(f ${Eigen_directory_files})
+  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src")
+    list(APPEND Eigen_directory_files_to_install ${f})
+  endif()
+endforeach(f ${Eigen_directory_files})
+
+install(FILES
+  ${Eigen_directory_files_to_install}
+  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
+  )
+
+install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h")

Eigen/Cholesky

@@ -43,3 +43,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLESKY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/CholmodSupport

@@ -22,7 +22,7 @@ extern "C" {
   * This module provides an interface to the Cholmod library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   * It provides the two following main factorization classes:
   * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization.
-  * - class CholmodDecomposition: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial).
+  * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial).
   *
   * For the sake of completeness, this module also propose the two following classes:
   * - class CholmodSimplicialLLT

Eigen/Core

@@ -11,55 +11,264 @@
 #ifndef EIGEN_CORE_H
 #define EIGEN_CORE_H
 
-// first thing Eigen does: stop the compiler from reporting useless warnings.
+// first thing Eigen does: stop the compiler from committing suicide
 #include "src/Core/util/DisableStupidWarnings.h"
 
-// then include this file where all our macros are defined. It's really important to do it first because
+#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
-// it's where we do all the compiler/OS/arch detections and define most defaults.
+  #define EIGEN_CUDACC __CUDACC__
-#include "src/Core/util/Macros.h"
-
-// This detects SSE/AVX/NEON/etc. and configure alignment settings
-#include "src/Core/util/ConfigureVectorization.h"
-
-// We need cuda_runtime.h/hip_runtime.h to ensure that
-// the EIGEN_USING_STD macro works properly on the device side
-#if defined(EIGEN_CUDACC)
-  #include <cuda_runtime.h>
-#elif defined(EIGEN_HIPCC)
-  #include <hip/hip_runtime.h>
 #endif
 
+#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDA_ARCH __CUDA_ARCH__
+#endif
+
+#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
+#define EIGEN_CUDACC_VER  ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
+#elif defined(__CUDACC_VER__)
+#define EIGEN_CUDACC_VER __CUDACC_VER__
+#else
+#define EIGEN_CUDACC_VER 0
+#endif
+
+// Handle NVCC/CUDA/SYCL
+#if defined(EIGEN_CUDACC) || defined(__SYCL_DEVICE_ONLY__)
+  // Do not try asserts on CUDA and SYCL!
+  #ifndef EIGEN_NO_DEBUG
+  #define EIGEN_NO_DEBUG
+  #endif
+
+  #ifdef EIGEN_INTERNAL_DEBUGGING
+  #undef EIGEN_INTERNAL_DEBUGGING
+  #endif
+
+  #ifdef EIGEN_EXCEPTIONS
+  #undef EIGEN_EXCEPTIONS
+  #endif
+
+  // All functions callable from CUDA code must be qualified with __device__
+  #ifdef EIGEN_CUDACC
+    // Do not try to vectorize on CUDA and SYCL!
+    #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+    #endif
+
+    #define EIGEN_DEVICE_FUNC __host__ __device__
+    // We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
+    // works properly on the device side
+    #include <cuda_runtime.h>
+  #else
+    #define EIGEN_DEVICE_FUNC
+  #endif
+
+#else
+  #define EIGEN_DEVICE_FUNC
+#endif
+
+#if defined(EIGEN_CUDACC)
+  #include <cuda.h>
+  #define EIGEN_CUDA_SDK_VER (CUDA_VERSION * 10)
+#else
+  #define EIGEN_CUDA_SDK_VER 0
+#endif
+
+
+// When compiling CUDA device code with NVCC, pull in math functions from the
+// global namespace.  In host mode, and when device doee with clang, use the
+// std versions.
+#if defined(__CUDA_ARCH__) && defined(__NVCC__)
+  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
+#else
+  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
+#endif
+
+#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
+  #define EIGEN_EXCEPTIONS
+#endif
 
 #ifdef EIGEN_EXCEPTIONS
   #include <new>
 #endif
 
+// then include this file where all our macros are defined. It's really important to do it first because
+// it's where we do all the alignment settings (platform detection and honoring the user's will if he
+// defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
+#include "src/Core/util/Macros.h"
+
 // Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3)
 // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details.
-#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6) && EIGEN_GNUC_AT_MOST(5,5)
+#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6)
   #pragma GCC optimize ("-fno-ipa-cp-clone")
 #endif
 
-// Prevent ICC from specializing std::complex operators that silently fail
-// on device. This allows us to use our own device-compatible specializations
-// instead.
-#if defined(EIGEN_COMP_ICC) && defined(EIGEN_GPU_COMPILE_PHASE) \
-    && !defined(_OVERRIDE_COMPLEX_SPECIALIZATION_)
-#define _OVERRIDE_COMPLEX_SPECIALIZATION_ 1
-#endif
 #include <complex>
 
 // this include file manages BLAS and MKL related macros
 // and inclusion of their respective header files
 #include "src/Core/util/MKL_support.h"
 
-
+// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
-#if defined(EIGEN_HAS_CUDA_FP16) || defined(EIGEN_HAS_HIP_FP16)
+// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
-  #define EIGEN_HAS_GPU_FP16
+#if EIGEN_MAX_ALIGN_BYTES==0
+  #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+  #endif
 #endif
 
-#if defined(EIGEN_HAS_CUDA_BF16) || defined(EIGEN_HAS_HIP_BF16)
+#if EIGEN_COMP_MSVC
-  #define EIGEN_HAS_GPU_BF16
+  #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
+  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
+    // Remember that usage of defined() in a #define is undefined by the standard.
+    // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
+      #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
+    #endif
+  #endif
+#else
+  // Remember that usage of defined() in a #define is undefined by the standard
+  #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
+    #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC
+  #endif
+#endif
+
+#if !defined(EIGEN_DONT_VECTORIZE) && !defined(EIGEN_CUDACC)
+
+  #if defined (EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
+
+    // Defines symbols for compile-time detection of which instructions are
+    // used.
+    // EIGEN_VECTORIZE_YY is defined if and only if the instruction set YY is used
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_SSE
+    #define EIGEN_VECTORIZE_SSE2
+
+    // Detect sse3/ssse3/sse4:
+    // gcc and icc defines __SSE3__, ...
+    // there is no way to know about this on msvc. You can define EIGEN_VECTORIZE_SSE* if you
+    // want to force the use of those instructions with msvc.
+    #ifdef __SSE3__
+      #define EIGEN_VECTORIZE_SSE3
+    #endif
+    #ifdef __SSSE3__
+      #define EIGEN_VECTORIZE_SSSE3
+    #endif
+    #ifdef __SSE4_1__
+      #define EIGEN_VECTORIZE_SSE4_1
+    #endif
+    #ifdef __SSE4_2__
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX__
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_SSE3
+      #define EIGEN_VECTORIZE_SSSE3
+      #define EIGEN_VECTORIZE_SSE4_1
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX2__
+      #define EIGEN_VECTORIZE_AVX2
+    #endif
+    #ifdef __FMA__
+      #define EIGEN_VECTORIZE_FMA
+    #endif
+    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
+      #define EIGEN_VECTORIZE_AVX512
+      #define EIGEN_VECTORIZE_AVX2
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_FMA
+      #ifdef __AVX512DQ__
+        #define EIGEN_VECTORIZE_AVX512DQ
+      #endif
+      #ifdef __AVX512ER__
+        #define EIGEN_VECTORIZE_AVX512ER
+      #endif
+    #endif
+
+    // include files
+
+    // This extern "C" works around a MINGW-w64 compilation issue
+    // https://sourceforge.net/tracker/index.php?func=detail&aid=3018394&group_id=202880&atid=983354
+    // In essence, intrin.h is included by windows.h and also declares intrinsics (just as emmintrin.h etc. below do).
+    // However, intrin.h uses an extern "C" declaration, and g++ thus complains of duplicate declarations
+    // with conflicting linkage.  The linkage for intrinsics doesn't matter, but at that stage the compiler doesn't know;
+    // so, to avoid compile errors when windows.h is included after Eigen/Core, ensure intrinsics are extern "C" here too.
+    // notice that since these are C headers, the extern "C" is theoretically needed anyways.
+    extern "C" {
+      // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
+      // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
+      #if EIGEN_COMP_ICC >= 1110
+        #include <immintrin.h>
+      #else
+        #include <mmintrin.h>
+        #include <emmintrin.h>
+        #include <xmmintrin.h>
+        #ifdef  EIGEN_VECTORIZE_SSE3
+        #include <pmmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSSE3
+        #include <tmmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSE4_1
+        #include <smmintrin.h>
+        #endif
+        #ifdef EIGEN_VECTORIZE_SSE4_2
+        #include <nmmintrin.h>
+        #endif
+        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
+        #include <immintrin.h>
+        #endif
+      #endif
+    } // end extern "C"
+  #elif defined __VSX__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_FMA
+    #define EIGEN_VECTORIZE_VSX
+    #include <altivec.h>
+    // We need to #undef all these ugly tokens defined in <altivec.h>
+    // => use __vector instead of vector
+    #undef bool
+    #undef vector
+    #undef pixel
+  #elif defined __ALTIVEC__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_FMA
+    #define EIGEN_VECTORIZE_ALTIVEC
+    #include <altivec.h>
+    // We need to #undef all these ugly tokens defined in <altivec.h>
+    // => use __vector instead of vector
+    #undef bool
+    #undef vector
+    #undef pixel
+  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_NEON
+    #include <arm_neon.h>
+    // Enable FMA for ARM.
+    #if defined(__ARM_FEATURE_FMA)
+      #define EIGEN_VECTORIZE_FMA
+    #endif
+  #elif (defined __s390x__ && defined __VEC__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_ZVECTOR
+    #include <vecintrin.h>
+  #endif
+#endif
+
+#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
+  // We can use the optimized fp16 to float and float to fp16 conversion routines
+  #define EIGEN_HAS_FP16_C
+#endif
+
+#if defined EIGEN_CUDACC
+  #define EIGEN_VECTORIZE_CUDA
+  #include <vector_types.h>
+  #if EIGEN_CUDA_SDK_VER >= 70500
+    #define EIGEN_HAS_CUDA_FP16
+  #endif
+#endif
+
+#if defined EIGEN_HAS_CUDA_FP16
+  #include <cuda_runtime_api.h>
+  #include <cuda_fp16.h>
 #endif
 
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
@@ -83,8 +292,8 @@
 #include <cmath>
 #include <cassert>
 #include <functional>
+#include <sstream>
 #ifndef EIGEN_NO_IO
-  #include <sstream>
   #include <iosfwd>
 #endif
 #include <cstring>
@@ -94,10 +303,6 @@
 // for min/max:
 #include <algorithm>
 
-#if EIGEN_HAS_CXX11
-#include <array>
-#endif
-
 // for std::is_nothrow_move_assignable
 #ifdef EIGEN_INCLUDE_TYPE_TRAITS
 #include <type_traits>
@@ -109,30 +314,42 @@
 #endif
 
 // required for __cpuid, needs to be included after cmath
-// also required for _BitScanReverse on Windows on ARM
+#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
-#if EIGEN_COMP_MSVC && (EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM64) && !EIGEN_OS_WINCE
   #include <intrin.h>
 #endif
 
-#if defined(EIGEN_USE_SYCL)
+/** \brief Namespace containing all symbols from the %Eigen library. */
-  #undef min
+namespace Eigen {
-  #undef max
-  #undef isnan
-  #undef isinf
-  #undef isfinite
-  #include <CL/sycl.hpp>
-  #include <map>
-  #include <memory>
-  #include <utility>
-  #include <thread>
-  #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM0
-  #define EIGEN_SYCL_LOCAL_THREAD_DIM0 16
-  #endif
-  #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM1
-  #define EIGEN_SYCL_LOCAL_THREAD_DIM1 16
-  #endif
-#endif
 
+inline static const char *SimdInstructionSetsInUse(void) {
+#if defined(EIGEN_VECTORIZE_AVX512)
+  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_AVX)
+  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_2)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_1)
+  return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
+#elif defined(EIGEN_VECTORIZE_SSSE3)
+  return "SSE, SSE2, SSE3, SSSE3";
+#elif defined(EIGEN_VECTORIZE_SSE3)
+  return "SSE, SSE2, SSE3";
+#elif defined(EIGEN_VECTORIZE_SSE2)
+  return "SSE, SSE2";
+#elif defined(EIGEN_VECTORIZE_ALTIVEC)
+  return "AltiVec";
+#elif defined(EIGEN_VECTORIZE_VSX)
+  return "VSX";
+#elif defined(EIGEN_VECTORIZE_NEON)
+  return "ARM NEON";
+#elif defined(EIGEN_VECTORIZE_ZVECTOR)
+  return "S390X ZVECTOR";
+#else
+  return "None";
+#endif
+}
+
+} // end namespace Eigen
 
 #if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
 // This will generate an error message:
@@ -141,7 +358,7 @@
 
 namespace Eigen {
 
-// we use size_t frequently and we'll never remember to prepend it with std:: every time just to
+// we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
 // ensure QNX/QCC support
 using std::size_t;
 // gcc 4.6.0 wants std:: for ptrdiff_t
@@ -165,90 +382,60 @@ using std::ptrdiff_t;
 #include "src/Core/util/StaticAssert.h"
 #include "src/Core/util/XprHelper.h"
 #include "src/Core/util/Memory.h"
-#include "src/Core/util/IntegralConstant.h"
-#include "src/Core/util/SymbolicIndex.h"
 
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
 #include "src/Core/MathFunctionsImpl.h"
 #include "src/Core/arch/Default/ConjHelper.h"
-// Generic half float support
-#include "src/Core/arch/Default/Half.h"
-#include "src/Core/arch/Default/BFloat16.h"
-#include "src/Core/arch/Default/TypeCasting.h"
-#include "src/Core/arch/Default/GenericPacketMathFunctionsFwd.h"
 
 #if defined EIGEN_VECTORIZE_AVX512
   #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
-  #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX/TypeCasting.h"
-  #include "src/Core/arch/AVX/Complex.h"
-  #include "src/Core/arch/AVX512/PacketMath.h"
-  #include "src/Core/arch/AVX512/TypeCasting.h"
-  #include "src/Core/arch/AVX512/Complex.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
   #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX512/PacketMath.h"
   #include "src/Core/arch/AVX512/MathFunctions.h"
 #elif defined EIGEN_VECTORIZE_AVX
   // Use AVX for floats and doubles, SSE for integers
   #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
   #include "src/Core/arch/SSE/Complex.h"
-  #include "src/Core/arch/AVX/PacketMath.h"
-  #include "src/Core/arch/AVX/TypeCasting.h"
-  #include "src/Core/arch/AVX/Complex.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
   #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX/Complex.h"
+  #include "src/Core/arch/AVX/TypeCasting.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
 #elif defined EIGEN_VECTORIZE_SSE
   #include "src/Core/arch/SSE/PacketMath.h"
-  #include "src/Core/arch/SSE/TypeCasting.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
   #include "src/Core/arch/SSE/Complex.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
 #elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
   #include "src/Core/arch/AltiVec/PacketMath.h"
   #include "src/Core/arch/AltiVec/MathFunctions.h"
   #include "src/Core/arch/AltiVec/Complex.h"
 #elif defined EIGEN_VECTORIZE_NEON
   #include "src/Core/arch/NEON/PacketMath.h"
-  #include "src/Core/arch/NEON/TypeCasting.h"
   #include "src/Core/arch/NEON/MathFunctions.h"
   #include "src/Core/arch/NEON/Complex.h"
-#elif defined EIGEN_VECTORIZE_SVE
-  #include "src/Core/arch/SVE/PacketMath.h"
-  #include "src/Core/arch/SVE/TypeCasting.h"
-  #include "src/Core/arch/SVE/MathFunctions.h"
 #elif defined EIGEN_VECTORIZE_ZVECTOR
   #include "src/Core/arch/ZVector/PacketMath.h"
   #include "src/Core/arch/ZVector/MathFunctions.h"
   #include "src/Core/arch/ZVector/Complex.h"
-#elif defined EIGEN_VECTORIZE_MSA
-  #include "src/Core/arch/MSA/PacketMath.h"
-  #include "src/Core/arch/MSA/MathFunctions.h"
-  #include "src/Core/arch/MSA/Complex.h"
 #endif
 
-#if defined EIGEN_VECTORIZE_GPU
+// Half float support
-  #include "src/Core/arch/GPU/PacketMath.h"
+#include "src/Core/arch/CUDA/Half.h"
-  #include "src/Core/arch/GPU/MathFunctions.h"
+#include "src/Core/arch/CUDA/PacketMathHalf.h"
-  #include "src/Core/arch/GPU/TypeCasting.h"
+#include "src/Core/arch/CUDA/TypeCasting.h"
-#endif
 
-#if defined(EIGEN_USE_SYCL)
+#if defined EIGEN_VECTORIZE_CUDA
-  #include "src/Core/arch/SYCL/SyclMemoryModel.h"
+  #include "src/Core/arch/CUDA/PacketMath.h"
-  #include "src/Core/arch/SYCL/InteropHeaders.h"
+  #include "src/Core/arch/CUDA/MathFunctions.h"
-#if !defined(EIGEN_DONT_VECTORIZE_SYCL)
-  #include "src/Core/arch/SYCL/PacketMath.h"
-  #include "src/Core/arch/SYCL/MathFunctions.h"
-  #include "src/Core/arch/SYCL/TypeCasting.h"
-#endif
 #endif
 
 #include "src/Core/arch/Default/Settings.h"
-// This file provides generic implementations valid for scalar as well
-#include "src/Core/arch/Default/GenericPacketMathFunctions.h"
 
 #include "src/Core/functors/TernaryFunctors.h"
 #include "src/Core/functors/BinaryFunctors.h"
@@ -259,16 +446,9 @@ using std::ptrdiff_t;
 
 // Specialized functors to enable the processing of complex numbers
 // on CUDA devices
-#ifdef EIGEN_CUDACC
 #include "src/Core/arch/CUDA/Complex.h"
-#endif
 
-#include "src/Core/util/IndexedViewHelper.h"
+#include "src/Core/IO.h"
-#include "src/Core/util/ReshapedHelper.h"
-#include "src/Core/ArithmeticSequence.h"
-#ifndef EIGEN_NO_IO
-  #include "src/Core/IO.h"
-#endif
 #include "src/Core/DenseCoeffsBase.h"
 #include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.h"
@@ -309,8 +489,6 @@ using std::ptrdiff_t;
 #include "src/Core/Ref.h"
 #include "src/Core/Block.h"
 #include "src/Core/VectorBlock.h"
-#include "src/Core/IndexedView.h"
-#include "src/Core/Reshaped.h"
 #include "src/Core/Transpose.h"
 #include "src/Core/DiagonalMatrix.h"
 #include "src/Core/Diagonal.h"
@@ -347,21 +525,13 @@ using std::ptrdiff_t;
 #include "src/Core/CoreIterators.h"
 #include "src/Core/ConditionEstimator.h"
 
-#if defined(EIGEN_VECTORIZE_VSX)
-  #include "src/Core/arch/AltiVec/MatrixProduct.h"
-#elif defined EIGEN_VECTORIZE_NEON
-  #include "src/Core/arch/NEON/GeneralBlockPanelKernel.h"
-#endif
-
 #include "src/Core/BooleanRedux.h"
 #include "src/Core/Select.h"
 #include "src/Core/VectorwiseOp.h"
-#include "src/Core/PartialReduxEvaluator.h"
 #include "src/Core/Random.h"
 #include "src/Core/Replicate.h"
 #include "src/Core/Reverse.h"
 #include "src/Core/ArrayWrapper.h"
-#include "src/Core/StlIterators.h"
 
 #ifdef EIGEN_USE_BLAS
 #include "src/Core/products/GeneralMatrixMatrix_BLAS.h"

Eigen/Eigenvalues

@@ -58,3 +58,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_EIGENVALUES_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Geometry

@@ -49,11 +49,14 @@
 #include "src/Geometry/AlignedBox.h"
 #include "src/Geometry/Umeyama.h"
 
-// Use the SSE optimized version whenever possible.
+// Use the SSE optimized version whenever possible. At the moment the
-#if (defined EIGEN_VECTORIZE_SSE) || (defined EIGEN_VECTORIZE_NEON)
+// SSE version doesn't compile when AVX is enabled
-#include "src/Geometry/arch/Geometry_SIMD.h"
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+#include "src/Geometry/arch/Geometry_SSE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_GEOMETRY_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
+

Eigen/Householder

@@ -27,3 +27,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_HOUSEHOLDER_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Jacobi

@@ -29,4 +29,5 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_JACOBI_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */
 

Eigen/KLUSupport

@@ -1,41 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_KLUSUPPORT_MODULE_H
-#define EIGEN_KLUSUPPORT_MODULE_H
-
-#include <Eigen/SparseCore>
-
-#include <Eigen/src/Core/util/DisableStupidWarnings.h>
-
-extern "C" {
-#include <btf.h>
-#include <klu.h>
-   }
-
-/** \ingroup Support_modules
-  * \defgroup KLUSupport_Module KLUSupport module
-  *
-  * This module provides an interface to the KLU library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
-  * It provides the following factorization class:
-  * - class KLU: a sparse LU factorization, well-suited for circuit simulation.
-  *
-  * \code
-  * #include <Eigen/KLUSupport>
-  * \endcode
-  *
-  * In order to use this module, the klu and btf headers must be accessible from the include paths, and your binary must be linked to the klu library and its dependencies.
-  * The dependencies depend on how umfpack has been compiled.
-  * For a cmake based project, you can use our FindKLU.cmake module to help you in this task.
-  *
-  */
-
-#include "src/KLUSupport/KLUSupport.h"
-
-#include <Eigen/src/Core/util/ReenableStupidWarnings.h>
-
-#endif // EIGEN_KLUSUPPORT_MODULE_H

Eigen/LU

@@ -38,10 +38,13 @@
 #include "src/LU/Determinant.h"
 #include "src/LU/InverseImpl.h"
 
-#if defined EIGEN_VECTORIZE_SSE || defined EIGEN_VECTORIZE_NEON
+// Use the SSE optimized version whenever possible. At the moment the
-  #include "src/LU/arch/InverseSize4.h"
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+  #include "src/LU/arch/Inverse_SSE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_LU_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/OrderingMethods

@@ -63,7 +63,10 @@
   * \endcode
   */
 
+#ifndef EIGEN_MPL2_ONLY
 #include "src/OrderingMethods/Amd.h"
+#endif
+
 #include "src/OrderingMethods/Ordering.h"
 #include "src/Core/util/ReenableStupidWarnings.h"
 

Eigen/PaStiXSupport

@@ -36,7 +36,6 @@ extern "C" {
   * \endcode
   *
   * In order to use this module, the PaSTiX headers must be accessible from the include paths, and your binary must be linked to the PaSTiX library and its dependencies.
-  * This wrapper resuires PaStiX version 5.x compiled without MPI support.
   * The dependencies depend on how PaSTiX has been compiled.
   * For a cmake based project, you can use our FindPaSTiX.cmake module to help you in this task.
   *

Eigen/QR

@@ -48,3 +48,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_QR_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/QtAlignedMalloc

@@ -37,3 +37,4 @@ void *qRealloc(void *ptr, std::size_t size)
 #endif
 
 #endif // EIGEN_QTMALLOC_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/SVD

@@ -48,3 +48,4 @@
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SVD_MODULE_H
+/* vim: set filetype=cpp et sw=2 ts=2 ai: */

Eigen/Sparse

@@ -25,7 +25,9 @@
 
 #include "SparseCore"
 #include "OrderingMethods"
+#ifndef EIGEN_MPL2_ONLY
 #include "SparseCholesky"
+#endif
 #include "SparseLU"
 #include "SparseQR"
 #include "IterativeLinearSolvers"

Eigen/SparseCholesky

@@ -30,8 +30,16 @@
   * \endcode
   */
 
+#ifdef EIGEN_MPL2_ONLY
+#error The SparseCholesky module has nothing to offer in MPL2 only mode
+#endif
+
 #include "src/SparseCholesky/SimplicialCholesky.h"
+
+#ifndef EIGEN_MPL2_ONLY
 #include "src/SparseCholesky/SimplicialCholesky_impl.h"
+#endif
+
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SPARSECHOLESKY_MODULE_H

Eigen/SparseLU

@@ -23,7 +23,7 @@
 // Ordering interface
 #include "OrderingMethods"
 
-#include "src/Core/util/DisableStupidWarnings.h"
+#include "src/SparseLU/SparseLU_gemm_kernel.h"
 
 #include "src/SparseLU/SparseLU_Structs.h"
 #include "src/SparseLU/SparseLU_SupernodalMatrix.h"
@@ -43,6 +43,4 @@
 #include "src/SparseLU/SparseLU_Utils.h"
 #include "src/SparseLU/SparseLU.h"
 
-#include "src/Core/util/ReenableStupidWarnings.h"
-
 #endif // EIGEN_SPARSELU_MODULE_H

Eigen/src/Cholesky/LDLT.h

@@ -16,15 +16,6 @@
 namespace Eigen {
 
 namespace internal {
-  template<typename _MatrixType, int _UpLo> struct traits<LDLT<_MatrixType, _UpLo> >
-   : traits<_MatrixType>
-  {
-    typedef MatrixXpr XprKind;
-    typedef SolverStorage StorageKind;
-    typedef int StorageIndex;
-    enum { Flags = 0 };
-  };
-
   template<typename MatrixType, int UpLo> struct LDLT_Traits;
 
   // PositiveSemiDef means positive semi-definite and non-zero; same for NegativeSemiDef
@@ -45,7 +36,7 @@ namespace internal {
   * matrix \f$ A \f$ such that \f$ A =  P^TLDL^*P \f$, where P is a permutation matrix, L
   * is lower triangular with a unit diagonal and D is a diagonal matrix.
   *
-  * The decomposition uses pivoting to ensure stability, so that D will have
+  * The decomposition uses pivoting to ensure stability, so that L will have
   * zeros in the bottom right rank(A) - n submatrix. Avoiding the square root
   * on D also stabilizes the computation.
   *
@@ -57,19 +48,20 @@ namespace internal {
   * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
   */
 template<typename _MatrixType, int _UpLo> class LDLT
-        : public SolverBase<LDLT<_MatrixType, _UpLo> >
 {
   public:
     typedef _MatrixType MatrixType;
-    typedef SolverBase<LDLT> Base;
-    friend class SolverBase<LDLT>;
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(LDLT)
     enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
       UpLo = _UpLo
     };
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
@@ -188,7 +180,6 @@ template<typename _MatrixType, int _UpLo> class LDLT
       return m_sign == internal::NegativeSemiDef || m_sign == internal::ZeroSign;
     }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \returns a solution x of \f$ A x = b \f$ using the current decomposition of A.
       *
       * This function also supports in-place solves using the syntax <tt>x = decompositionObject.solve(x)</tt> .
@@ -200,14 +191,19 @@ template<typename _MatrixType, int _UpLo> class LDLT
       * \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then
       * \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the
       * least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function
-      * computes the least-square solution of \f$ A x = b \f$ if \f$ A \f$ is singular.
+      * computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular.
       *
       * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
       */
     template<typename Rhs>
     inline const Solve<LDLT, Rhs>
-    solve(const MatrixBase<Rhs>& b) const;
+    solve(const MatrixBase<Rhs>& b) const
-    #endif
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LDLT::solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<LDLT, Rhs>(*this, b.derived());
+    }
 
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
@@ -246,12 +242,12 @@ template<typename _MatrixType, int _UpLo> class LDLT
       */
     const LDLT& adjoint() const { return *this; };
 
-    EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    inline Index rows() const { return m_matrix.rows(); }
-    EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    inline Index cols() const { return m_matrix.cols(); }
 
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was successful,
+      * \returns \c Success if computation was succesful,
       *          \c NumericalIssue if the factorization failed because of a zero pivot.
       */
     ComputationInfo info() const
@@ -262,10 +258,8 @@ template<typename _MatrixType, int _UpLo> class LDLT
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
     void _solve_impl(const RhsType &rhs, DstType &dst) const;
-
-    template<bool Conjugate, typename RhsType, typename DstType>
-    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
     #endif
 
   protected:
@@ -564,24 +558,16 @@ LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Deri
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename _MatrixType, int _UpLo>
 template<typename RhsType, typename DstType>
-void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+EIGEN_DEVICE_FUNC void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
-{
-  _solve_impl_transposed<true>(rhs, dst);
-}
-
-template<typename _MatrixType,int _UpLo>
-template<bool Conjugate, typename RhsType, typename DstType>
-void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
 {
+  eigen_assert(rhs.rows() == rows());
   // dst = P b
   dst = m_transpositions * rhs;
 
   // dst = L^-1 (P b)
-  // dst = L^-*T (P b)
+  matrixL().solveInPlace(dst);
-  matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
 
-  // dst = D^-* (L^-1 P b)
+  // dst = D^-1 (L^-1 P b)
-  // dst = D^-1 (L^-*T P b)
   // more precisely, use pseudo-inverse of D (see bug 241)
   using std::abs;
   const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
@@ -593,6 +579,7 @@ void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType
   // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
   // Using numeric_limits::min() gives us more robustness to denormals.
   RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
+
   for (Index i = 0; i < vecD.size(); ++i)
   {
     if(abs(vecD(i)) > tolerance)
@@ -601,12 +588,10 @@ void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType
       dst.row(i).setZero();
   }
 
-  // dst = L^-* (D^-* L^-1 P b)
+  // dst = L^-T (D^-1 L^-1 P b)
-  // dst = L^-T (D^-1 L^-*T P b)
+  matrixU().solveInPlace(dst);
-  matrixL().transpose().template conjugateIf<Conjugate>().solveInPlace(dst);
 
-  // dst = P^T (L^-* D^-* L^-1 P b) = A^-1 b
+  // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
-  // dst = P^-T (L^-T D^-1 L^-*T P b) = A^-1 b
   dst = m_transpositions.transpose() * dst;
 }
 #endif

Eigen/src/Cholesky/LLT.h

@@ -13,16 +13,6 @@
 namespace Eigen {
 
 namespace internal{
-
-template<typename _MatrixType, int _UpLo> struct traits<LLT<_MatrixType, _UpLo> >
- : traits<_MatrixType>
-{
-  typedef MatrixXpr XprKind;
-  typedef SolverStorage StorageKind;
-  typedef int StorageIndex;
-  enum { Flags = 0 };
-};
-
 template<typename MatrixType, int UpLo> struct LLT_Traits;
 }
 
@@ -64,17 +54,18 @@ template<typename MatrixType, int UpLo> struct LLT_Traits;
   * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
   */
 template<typename _MatrixType, int _UpLo> class LLT
-        : public SolverBase<LLT<_MatrixType, _UpLo> >
 {
   public:
     typedef _MatrixType MatrixType;
-    typedef SolverBase<LLT> Base;
-    friend class SolverBase<LLT>;
-
-    EIGEN_GENERIC_PUBLIC_INTERFACE(LLT)
     enum {
+      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
 
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -109,7 +100,7 @@ template<typename _MatrixType, int _UpLo> class LLT
       compute(matrix.derived());
     }
 
-    /** \brief Constructs a LLT factorization from a given matrix
+    /** \brief Constructs a LDLT factorization from a given matrix
       *
       * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
       * \c MatrixType is a Eigen::Ref.
@@ -138,7 +129,6 @@ template<typename _MatrixType, int _UpLo> class LLT
       return Traits::getL(m_matrix);
     }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
       *
       * Since this LLT class assumes anyway that the matrix A is invertible, the solution
@@ -151,8 +141,13 @@ template<typename _MatrixType, int _UpLo> class LLT
       */
     template<typename Rhs>
     inline const Solve<LLT, Rhs>
-    solve(const MatrixBase<Rhs>& b) const;
+    solve(const MatrixBase<Rhs>& b) const
-    #endif
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_matrix.rows()==b.rows()
+                && "LLT::solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<LLT, Rhs>(*this, b.derived());
+    }
 
     template<typename Derived>
     void solveInPlace(const MatrixBase<Derived> &bAndX) const;
@@ -185,7 +180,7 @@ template<typename _MatrixType, int _UpLo> class LLT
 
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was successful,
+      * \returns \c Success if computation was succesful,
       *          \c NumericalIssue if the matrix.appears not to be positive definite.
       */
     ComputationInfo info() const
@@ -199,20 +194,18 @@ template<typename _MatrixType, int _UpLo> class LLT
       * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
       * \code x = decomposition.adjoint().solve(b) \endcode
       */
-    const LLT& adjoint() const EIGEN_NOEXCEPT { return *this; };
+    const LLT& adjoint() const { return *this; };
 
-    inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    inline Index rows() const { return m_matrix.rows(); }
-    inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    inline Index cols() const { return m_matrix.cols(); }
 
     template<typename VectorType>
-    LLT & rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
+    LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
     void _solve_impl(const RhsType &rhs, DstType &dst) const;
-
-    template<bool Conjugate, typename RhsType, typename DstType>
-    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
     #endif
 
   protected:
@@ -466,7 +459,7 @@ LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>
   */
 template<typename _MatrixType, int _UpLo>
 template<typename VectorType>
-LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
+LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType);
   eigen_assert(v.size()==m_matrix.cols());
@@ -482,19 +475,10 @@ LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v,
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename _MatrixType,int _UpLo>
 template<typename RhsType, typename DstType>
-void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+EIGEN_DEVICE_FUNC void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  _solve_impl_transposed<true>(rhs, dst);
+  dst = rhs;
-}
+  solveInPlace(dst);
-
-template<typename _MatrixType,int _UpLo>
-template<bool Conjugate, typename RhsType, typename DstType>
-void LLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
-{
-    dst = rhs;
-
-    matrixL().template conjugateIf<!Conjugate>().solveInPlace(dst);
-    matrixU().template conjugateIf<!Conjugate>().solveInPlace(dst);
 }
 #endif
 

Eigen/src/CholmodSupport/CholmodSupport.h

@@ -32,7 +32,7 @@ template<> struct cholmod_configure_matrix<std::complex<double> > {
   }
 };
 
-// Other scalar types are not yet supported by Cholmod
+// Other scalar types are not yet suppotred by Cholmod
 // template<> struct cholmod_configure_matrix<float> {
 //   template<typename CholmodType>
 //   static void run(CholmodType& mat) {
@@ -84,7 +84,7 @@ cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> >
   {
     res.itype = CHOLMOD_INT;
   }
-  else if (internal::is_same<_StorageIndex,SuiteSparse_long>::value)
+  else if (internal::is_same<_StorageIndex,long>::value)
   {
     res.itype = CHOLMOD_LONG;
   }
@@ -124,9 +124,6 @@ cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Sca
   
   if(UpLo==Upper) res.stype =  1;
   if(UpLo==Lower) res.stype = -1;
-  // swap stype for rowmajor matrices (only works for real matrices)
-  EIGEN_STATIC_ASSERT((_Options & RowMajorBit) == 0 || NumTraits<_Scalar>::IsComplex == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
-  if(_Options & RowMajorBit) res.stype *=-1;
 
   return res;
 }
@@ -162,44 +159,6 @@ MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
           static_cast<StorageIndex*>(cm.p), static_cast<StorageIndex*>(cm.i),static_cast<Scalar*>(cm.x) );
 }
 
-namespace internal {
-
-// template specializations for int and long that call the correct cholmod method
-
-#define EIGEN_CHOLMOD_SPECIALIZE0(ret, name) \
-    template<typename _StorageIndex> inline ret cm_ ## name       (cholmod_common &Common) { return cholmod_ ## name   (&Common); } \
-    template<>                       inline ret cm_ ## name<SuiteSparse_long> (cholmod_common &Common) { return cholmod_l_ ## name (&Common); }
-
-#define EIGEN_CHOLMOD_SPECIALIZE1(ret, name, t1, a1) \
-    template<typename _StorageIndex> inline ret cm_ ## name       (t1& a1, cholmod_common &Common) { return cholmod_ ## name   (&a1, &Common); } \
-    template<>                       inline ret cm_ ## name<SuiteSparse_long> (t1& a1, cholmod_common &Common) { return cholmod_l_ ## name (&a1, &Common); }
-
-EIGEN_CHOLMOD_SPECIALIZE0(int, start)
-EIGEN_CHOLMOD_SPECIALIZE0(int, finish)
-
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_factor, cholmod_factor*, L)
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_dense,  cholmod_dense*,  X)
-EIGEN_CHOLMOD_SPECIALIZE1(int, free_sparse, cholmod_sparse*, A)
-
-EIGEN_CHOLMOD_SPECIALIZE1(cholmod_factor*, analyze, cholmod_sparse, A)
-
-template<typename _StorageIndex> inline cholmod_dense*  cm_solve         (int sys, cholmod_factor& L, cholmod_dense&  B, cholmod_common &Common) { return cholmod_solve     (sys, &L, &B, &Common); }
-template<>                       inline cholmod_dense*  cm_solve<SuiteSparse_long>   (int sys, cholmod_factor& L, cholmod_dense&  B, cholmod_common &Common) { return cholmod_l_solve   (sys, &L, &B, &Common); }
-
-template<typename _StorageIndex> inline cholmod_sparse* cm_spsolve       (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_spsolve   (sys, &L, &B, &Common); }
-template<>                       inline cholmod_sparse* cm_spsolve<SuiteSparse_long> (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_l_spsolve (sys, &L, &B, &Common); }
-
-template<typename _StorageIndex>
-inline int  cm_factorize_p       (cholmod_sparse*  A, double beta[2], _StorageIndex* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_factorize_p   (A, beta, fset, fsize, L, &Common); }
-template<>
-inline int  cm_factorize_p<SuiteSparse_long> (cholmod_sparse*  A, double beta[2], SuiteSparse_long* fset,          std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_l_factorize_p (A, beta, fset, fsize, L, &Common); }
-
-#undef EIGEN_CHOLMOD_SPECIALIZE0
-#undef EIGEN_CHOLMOD_SPECIALIZE1
-
-}  // namespace internal
-
-
 enum CholmodMode {
   CholmodAuto, CholmodSimplicialLLt, CholmodSupernodalLLt, CholmodLDLt
 };
@@ -236,7 +195,7 @@ class CholmodBase : public SparseSolverBase<Derived>
     {
       EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
       m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
-      internal::cm_start<StorageIndex>(m_cholmod);
+      cholmod_start(&m_cholmod);
     }
 
     explicit CholmodBase(const MatrixType& matrix)
@@ -244,15 +203,15 @@ class CholmodBase : public SparseSolverBase<Derived>
     {
       EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
       m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
-      internal::cm_start<StorageIndex>(m_cholmod);
+      cholmod_start(&m_cholmod);
       compute(matrix);
     }
 
     ~CholmodBase()
     {
       if(m_cholmodFactor)
-        internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
+        cholmod_free_factor(&m_cholmodFactor, &m_cholmod);
-      internal::cm_finish<StorageIndex>(m_cholmod);
+      cholmod_finish(&m_cholmod);
     }
     
     inline StorageIndex cols() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
@@ -260,7 +219,7 @@ class CholmodBase : public SparseSolverBase<Derived>
     
     /** \brief Reports whether previous computation was successful.
       *
-      * \returns \c Success if computation was successful,
+      * \returns \c Success if computation was succesful,
       *          \c NumericalIssue if the matrix.appears to be negative.
       */
     ComputationInfo info() const
@@ -287,11 +246,11 @@ class CholmodBase : public SparseSolverBase<Derived>
     {
       if(m_cholmodFactor)
       {
-        internal::cm_free_factor<StorageIndex>(m_cholmodFactor, m_cholmod);
+        cholmod_free_factor(&m_cholmodFactor, &m_cholmod);
         m_cholmodFactor = 0;
       }
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
-      m_cholmodFactor = internal::cm_analyze<StorageIndex>(A, m_cholmod);
+      m_cholmodFactor = cholmod_analyze(&A, &m_cholmod);
       
       this->m_isInitialized = true;
       this->m_info = Success;
@@ -309,7 +268,7 @@ class CholmodBase : public SparseSolverBase<Derived>
     {
       eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
-      internal::cm_factorize_p<StorageIndex>(&A, m_shiftOffset, 0, 0, m_cholmodFactor, m_cholmod);
+      cholmod_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, &m_cholmod);
 
       // If the factorization failed, minor is the column at which it did. On success minor == n.
       this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue);
@@ -330,20 +289,19 @@ class CholmodBase : public SparseSolverBase<Derived>
       EIGEN_UNUSED_VARIABLE(size);
       eigen_assert(size==b.rows());
       
-      // Cholmod needs column-major storage without inner-stride, which corresponds to the default behavior of Ref.
+      // Cholmod needs column-major stoarge without inner-stride, which corresponds to the default behavior of Ref.
       Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
 
       cholmod_dense b_cd = viewAsCholmod(b_ref);
-      cholmod_dense* x_cd = internal::cm_solve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cd, m_cholmod);
+      cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod);
       if(!x_cd)
       {
         this->m_info = NumericalIssue;
         return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
-      // NOTE Actually, the copy can be avoided by calling cholmod_solve2 instead of cholmod_solve
       dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
-      internal::cm_free_dense<StorageIndex>(x_cd, m_cholmod);
+      cholmod_free_dense(&x_cd, &m_cholmod);
     }
     
     /** \internal */
@@ -358,16 +316,15 @@ class CholmodBase : public SparseSolverBase<Derived>
       // note: cs stands for Cholmod Sparse
       Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
       cholmod_sparse b_cs = viewAsCholmod(b_ref);
-      cholmod_sparse* x_cs = internal::cm_spsolve<StorageIndex>(CHOLMOD_A, *m_cholmodFactor, b_cs, m_cholmod);
+      cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod);
       if(!x_cs)
       {
         this->m_info = NumericalIssue;
         return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
-      // NOTE cholmod_spsolve in fact just calls the dense solver for blocks of 4 columns at a time (similar to Eigen's sparse solver)
       dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
-      internal::cm_free_sparse<StorageIndex>(x_cs, m_cholmod);
+      cholmod_free_sparse(&x_cs, &m_cholmod);
     }
     #endif // EIGEN_PARSED_BY_DOXYGEN
     

Eigen/src/Core/ArithmeticSequence.h

@@ -1,406 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_ARITHMETIC_SEQUENCE_H
-#define EIGEN_ARITHMETIC_SEQUENCE_H
-
-namespace Eigen {
-
-namespace internal {
-
-#if (!EIGEN_HAS_CXX11) || !((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
-template<typename T> struct aseq_negate {};
-
-template<> struct aseq_negate<Index> {
-  typedef Index type;
-};
-
-template<int N> struct aseq_negate<FixedInt<N> > {
-  typedef FixedInt<-N> type;
-};
-
-// Compilation error in the following case:
-template<> struct aseq_negate<FixedInt<DynamicIndex> > {};
-
-template<typename FirstType,typename SizeType,typename IncrType,
-         bool FirstIsSymbolic=symbolic::is_symbolic<FirstType>::value,
-         bool SizeIsSymbolic =symbolic::is_symbolic<SizeType>::value>
-struct aseq_reverse_first_type {
-  typedef Index type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,true> {
-  typedef symbolic::AddExpr<FirstType,
-                            symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
-                                                  symbolic::ValueExpr<IncrType> >
-                           > type;
-};
-
-template<typename SizeType,typename IncrType,typename EnableIf = void>
-struct aseq_reverse_first_type_aux {
-  typedef Index type;
-};
-
-template<typename SizeType,typename IncrType>
-struct aseq_reverse_first_type_aux<SizeType,IncrType,typename internal::enable_if<bool((SizeType::value+IncrType::value)|0x1)>::type> {
-  typedef FixedInt<(SizeType::value-1)*IncrType::value> type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,true,false> {
-  typedef typename aseq_reverse_first_type_aux<SizeType,IncrType>::type Aux;
-  typedef symbolic::AddExpr<FirstType,symbolic::ValueExpr<Aux> > type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct aseq_reverse_first_type<FirstType,SizeType,IncrType,false,true> {
-  typedef symbolic::AddExpr<symbolic::ProductExpr<symbolic::AddExpr<SizeType,symbolic::ValueExpr<FixedInt<-1> > >,
-                                                  symbolic::ValueExpr<IncrType> >,
-                            symbolic::ValueExpr<> > type;
-};
-#endif
-
-// Helper to cleanup the type of the increment:
-template<typename T> struct cleanup_seq_incr {
-  typedef typename cleanup_index_type<T,DynamicIndex>::type type;
-};
-
-}
-
-//--------------------------------------------------------------------------------
-// seq(first,last,incr) and seqN(first,size,incr)
-//--------------------------------------------------------------------------------
-
-template<typename FirstType=Index,typename SizeType=Index,typename IncrType=internal::FixedInt<1> >
-class ArithmeticSequence;
-
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                   typename internal::cleanup_index_type<SizeType>::type,
-                   typename internal::cleanup_seq_incr<IncrType>::type >
-seqN(FirstType first, SizeType size, IncrType incr);
-
-/** \class ArithmeticSequence
-  * \ingroup Core_Module
-  *
-  * This class represents an arithmetic progression \f$ a_0, a_1, a_2, ..., a_{n-1}\f$ defined by
-  * its \em first value \f$ a_0 \f$, its \em size (aka length) \em n, and the \em increment (aka stride)
-  * that is equal to \f$ a_{i+1}-a_{i}\f$ for any \em i.
-  *
-  * It is internally used as the return type of the Eigen::seq and Eigen::seqN functions, and as the input arguments
-  * of DenseBase::operator()(const RowIndices&, const ColIndices&), and most of the time this is the
-  * only way it is used.
-  *
-  * \tparam FirstType type of the first element, usually an Index,
-  *                   but internally it can be a symbolic expression
-  * \tparam SizeType type representing the size of the sequence, usually an Index
-  *                  or a compile time integral constant. Internally, it can also be a symbolic expression
-  * \tparam IncrType type of the increment, can be a runtime Index, or a compile time integral constant (default is compile-time 1)
-  *
-  * \sa Eigen::seq, Eigen::seqN, DenseBase::operator()(const RowIndices&, const ColIndices&), class IndexedView
-  */
-template<typename FirstType,typename SizeType,typename IncrType>
-class ArithmeticSequence
-{
-public:
-  ArithmeticSequence(FirstType first, SizeType size) : m_first(first), m_size(size) {}
-  ArithmeticSequence(FirstType first, SizeType size, IncrType incr) : m_first(first), m_size(size), m_incr(incr) {}
-
-  enum {
-    SizeAtCompileTime = internal::get_fixed_value<SizeType>::value,
-    IncrAtCompileTime = internal::get_fixed_value<IncrType,DynamicIndex>::value
-  };
-
-  /** \returns the size, i.e., number of elements, of the sequence */
-  Index size()  const { return m_size; }
-
-  /** \returns the first element \f$ a_0 \f$ in the sequence */
-  Index first()  const { return m_first; }
-
-  /** \returns the value \f$ a_i \f$ at index \a i in the sequence. */
-  Index operator[](Index i) const { return m_first + i * m_incr; }
-
-  const FirstType& firstObject() const { return m_first; }
-  const SizeType&  sizeObject()  const { return m_size; }
-  const IncrType&  incrObject()  const { return m_incr; }
-
-protected:
-  FirstType m_first;
-  SizeType  m_size;
-  IncrType  m_incr;
-
-public:
-
-#if EIGEN_HAS_CXX11 && ((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48)
-  auto reverse() const -> decltype(Eigen::seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr)) {
-    return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
-  }
-#else
-protected:
-  typedef typename internal::aseq_negate<IncrType>::type ReverseIncrType;
-  typedef typename internal::aseq_reverse_first_type<FirstType,SizeType,IncrType>::type ReverseFirstType;
-public:
-  ArithmeticSequence<ReverseFirstType,SizeType,ReverseIncrType>
-  reverse() const {
-    return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr);
-  }
-#endif
-};
-
-/** \returns an ArithmeticSequence starting at \a first, of length \a size, and increment \a incr
-  *
-  * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type >
-seqN(FirstType first, SizeType size, IncrType incr)  {
-  return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type,typename internal::cleanup_seq_incr<IncrType>::type>(first,size,incr);
-}
-
-/** \returns an ArithmeticSequence starting at \a first, of length \a size, and unit increment
-  *
-  * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType) */
-template<typename FirstType,typename SizeType>
-ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type >
-seqN(FirstType first, SizeType size)  {
-  return ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,typename internal::cleanup_index_type<SizeType>::type>(first,size);
-}
-
-
-#if EIGEN_HAS_CXX11
-
-/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and with positive (or negative) increment \a incr
-  *
-  * It is essentially an alias to:
-  * \code
-  * seqN(f, (l-f+incr)/incr, incr);
-  * \endcode
-  *
-  * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType)
-  */
-template<typename FirstType,typename LastType>
-auto seq(FirstType f, LastType l) -> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-                                                   (  typename internal::cleanup_index_type<LastType>::type(l)
-                                                    - typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())))
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              (typename internal::cleanup_index_type<LastType>::type(l)
-               -typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
-}
-
-/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and unit increment
-  *
-  * It is essentially an alias to:
-  * \code
-  * seqN(f,l-f+1);
-  * \endcode
-  *
-  * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType)
-  */
-template<typename FirstType,typename LastType, typename IncrType>
-auto seq(FirstType f, LastType l, IncrType incr)
-  -> decltype(seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-                   (   typename internal::cleanup_index_type<LastType>::type(l)
-                     - typename internal::cleanup_index_type<FirstType>::type(f)+typename internal::cleanup_seq_incr<IncrType>::type(incr)
-                   ) / typename internal::cleanup_seq_incr<IncrType>::type(incr),
-                   typename internal::cleanup_seq_incr<IncrType>::type(incr)))
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              ( typename internal::cleanup_index_type<LastType>::type(l)
-               -typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr)) / CleanedIncrType(incr),
-              CleanedIncrType(incr));
-}
-
-#else // EIGEN_HAS_CXX11
-
-template<typename FirstType,typename LastType>
-typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
-                             ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index> >::type
-seq(FirstType f, LastType l)
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>())));
-}
-
-template<typename FirstTypeDerived,typename LastType>
-typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
-    ArithmeticSequence<FirstTypeDerived, symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,symbolic::ValueExpr<> >,
-                                                            symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l)
-{
-  return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+fix<1>()));
-}
-
-template<typename FirstType,typename LastTypeDerived>
-typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                        symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
-                                          symbolic::ValueExpr<internal::FixedInt<1> > > > >::type
-seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l)
-{
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),(l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+fix<1>()));
-}
-
-template<typename FirstTypeDerived,typename LastTypeDerived>
-ArithmeticSequence<FirstTypeDerived,
-                    symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::NegateExpr<FirstTypeDerived> >,symbolic::ValueExpr<internal::FixedInt<1> > > >
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l)
-{
-  return seqN(f.derived(),(l.derived()-f.derived()+fix<1>()));
-}
-
-
-template<typename FirstType,typename LastType, typename IncrType>
-typename internal::enable_if<!(symbolic::is_symbolic<FirstType>::value || symbolic::is_symbolic<LastType>::value),
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,Index,typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(FirstType f, LastType l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              Index((typename internal::cleanup_index_type<LastType>::type(l)-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr)), incr);
-}
-
-template<typename FirstTypeDerived,typename LastType, typename IncrType>
-typename internal::enable_if<!symbolic::is_symbolic<LastType>::value,
-    ArithmeticSequence<FirstTypeDerived,
-                        symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<symbolic::NegateExpr<FirstTypeDerived>,
-                                                                                   symbolic::ValueExpr<> >,
-                                                                 symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                              symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                        typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, LastType l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(f.derived(),(typename internal::cleanup_index_type<LastType>::type(l)-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-
-template<typename FirstType,typename LastTypeDerived, typename IncrType>
-typename internal::enable_if<!symbolic::is_symbolic<FirstType>::value,
-    ArithmeticSequence<typename internal::cleanup_index_type<FirstType>::type,
-                        symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,symbolic::ValueExpr<> >,
-                                                                 symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                               symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                        typename internal::cleanup_seq_incr<IncrType>::type> >::type
-seq(FirstType f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(typename internal::cleanup_index_type<FirstType>::type(f),
-              (l.derived()-typename internal::cleanup_index_type<FirstType>::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-
-template<typename FirstTypeDerived,typename LastTypeDerived, typename IncrType>
-ArithmeticSequence<FirstTypeDerived,
-                    symbolic::QuotientExpr<symbolic::AddExpr<symbolic::AddExpr<LastTypeDerived,
-                                                                               symbolic::NegateExpr<FirstTypeDerived> >,
-                                                             symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                                          symbolic::ValueExpr<typename internal::cleanup_seq_incr<IncrType>::type> >,
-                    typename internal::cleanup_seq_incr<IncrType>::type>
-seq(const symbolic::BaseExpr<FirstTypeDerived> &f, const symbolic::BaseExpr<LastTypeDerived> &l, IncrType incr)
-{
-  typedef typename internal::cleanup_seq_incr<IncrType>::type CleanedIncrType;
-  return seqN(f.derived(),(l.derived()-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr);
-}
-#endif // EIGEN_HAS_CXX11
-
-#if EIGEN_HAS_CXX11
-/** \cpp11
-  * \returns a symbolic ArithmeticSequence representing the last \a size elements with a unit increment.
-  *
-  * \anchor indexing_lastN
-  *
-  *  It is a shortcut for: \code seq(last+fix<1>-size, last) \endcode
-  * 
-  * \sa lastN(SizeType,IncrType, seqN(FirstType,SizeType), seq(FirstType,LastType) */
-template<typename SizeType>
-auto lastN(SizeType size)
--> decltype(seqN(Eigen::last+fix<1>()-size, size))
-{
-  return seqN(Eigen::last+fix<1>()-size, size);
-}
-
-/** \cpp11
-  * \returns a symbolic ArithmeticSequence representing the last \a size elements with increment \a incr.
-  *
-  * \anchor indexing_lastN_with_incr
-  *
-  * It is a shortcut for: \code seqN(last-(size-fix<1>)*incr, size, incr) \endcode
-  * 
-  * \sa lastN(SizeType), seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */
-template<typename SizeType,typename IncrType>
-auto lastN(SizeType size, IncrType incr)
--> decltype(seqN(Eigen::last-(size-fix<1>())*incr, size, incr))
-{
-  return seqN(Eigen::last-(size-fix<1>())*incr, size, incr);
-}
-#endif
-
-namespace internal {
-
-// Convert a symbolic span into a usable one (i.e., remove last/end "keywords")
-template<typename T>
-struct make_size_type {
-  typedef typename internal::conditional<symbolic::is_symbolic<T>::value, Index, T>::type type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType,int XprSize>
-struct IndexedViewCompatibleType<ArithmeticSequence<FirstType,SizeType,IncrType>, XprSize> {
-  typedef ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType> type;
-};
-
-template<typename FirstType,typename SizeType,typename IncrType>
-ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>
-makeIndexedViewCompatible(const ArithmeticSequence<FirstType,SizeType,IncrType>& ids, Index size,SpecializedType) {
-  return ArithmeticSequence<Index,typename make_size_type<SizeType>::type,IncrType>(
-            eval_expr_given_size(ids.firstObject(),size),eval_expr_given_size(ids.sizeObject(),size),ids.incrObject());
-}
-
-template<typename FirstType,typename SizeType,typename IncrType>
-struct get_compile_time_incr<ArithmeticSequence<FirstType,SizeType,IncrType> > {
-  enum { value = get_fixed_value<IncrType,DynamicIndex>::value };
-};
-
-} // end namespace internal
-
-/** \namespace Eigen::indexing
-  * \ingroup Core_Module
-  * 
-  * The sole purpose of this namespace is to be able to import all functions
-  * and symbols that are expected to be used within operator() for indexing
-  * and slicing. If you already imported the whole Eigen namespace:
-  * \code using namespace Eigen; \endcode
-  * then you are already all set. Otherwise, if you don't want/cannot import
-  * the whole Eigen namespace, the following line:
-  * \code using namespace Eigen::indexing; \endcode
-  * is equivalent to:
-  * \code
-  using Eigen::all;
-  using Eigen::seq;
-  using Eigen::seqN;
-  using Eigen::lastN; // c++11 only
-  using Eigen::last;
-  using Eigen::lastp1;
-  using Eigen::fix;
-  \endcode
-  */
-namespace indexing {
-  using Eigen::all;
-  using Eigen::seq;
-  using Eigen::seqN;
-  #if EIGEN_HAS_CXX11
-  using Eigen::lastN;
-  #endif
-  using Eigen::last;
-  using Eigen::lastp1;
-  using Eigen::fix;
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_ARITHMETIC_SEQUENCE_H

Eigen/src/Core/Array.h

@@ -157,58 +157,11 @@ class Array
     EIGEN_DEVICE_FUNC
     Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
     {
-      Base::operator=(std::move(other));
+      other.swap(*this);
       return *this;
     }
 #endif
 
-    #if EIGEN_HAS_CXX11
-    /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11
-     *
-     * \only_for_vectors
-     *
-     * This constructor is for 1D array or vectors with more than 4 coefficients.
-     * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients.
-     *
-     * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this
-     * constructor must match the the fixed number of rows (resp. columns) of \c *this.
-     *
-     * Example: \include Array_variadic_ctor_cxx11.cpp
-     * Output: \verbinclude Array_variadic_ctor_cxx11.out
-     *
-     * \sa Array(const std::initializer_list<std::initializer_list<Scalar>>&)
-     * \sa Array(const Scalar&), Array(const Scalar&,const Scalar&)
-     */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      : Base(a0, a1, a2, a3, args...) {}
-
-    /** \brief Constructs an array and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
-      *
-      * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
-      *
-      * Example: \include Array_initializer_list_23_cxx11.cpp
-      * Output: \verbinclude Array_initializer_list_23_cxx11.out
-      *
-      * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
-      *
-      * In the case of a compile-time column 1D array, implicit transposition from a single row is allowed.
-      * Therefore <code> Array<int,Dynamic,1>{{1,2,3,4,5}}</code> is legal and the more verbose syntax
-      * <code>Array<int,Dynamic,1>{{1},{2},{3},{4},{5}}</code> can be avoided:
-      *
-      * Example: \include Array_initializer_list_vector_cxx11.cpp
-      * Output: \verbinclude Array_initializer_list_vector_cxx11.out
-      *
-      * In the case of fixed-sized arrays, the initializer list sizes must exactly match the array sizes,
-      * and implicit transposition is allowed for compile-time 1D arrays only.
-      *
-      * \sa  Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
-    EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE Array(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
-    #endif // end EIGEN_HAS_CXX11
-
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename T>
     EIGEN_DEVICE_FUNC
@@ -225,7 +178,6 @@ class Array
       Base::_check_template_params();
       this->template _init2<T0,T1>(val0, val1);
     }
-
     #else
     /** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */
     EIGEN_DEVICE_FUNC explicit Array(const Scalar *data);
@@ -237,8 +189,7 @@ class Array
       */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE explicit Array(Index dim);
-    /** constructs an initialized 1x1 Array with the given coefficient
+    /** constructs an initialized 1x1 Array with the given coefficient */
-      * \sa const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args */
     Array(const Scalar& value);
     /** constructs an uninitialized array with \a rows rows and \a cols columns.
       *
@@ -246,14 +197,11 @@ class Array
       * it is redundant to pass these parameters, so one should use the default constructor
       * Array() instead. */
     Array(Index rows, Index cols);
-    /** constructs an initialized 2D vector with given coefficients
+    /** constructs an initialized 2D vector with given coefficients */
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */
     Array(const Scalar& val0, const Scalar& val1);
-    #endif  // end EIGEN_PARSED_BY_DOXYGEN
+    #endif
 
-    /** constructs an initialized 3D vector with given coefficients
+    /** constructs an initialized 3D vector with given coefficients */
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
     {
@@ -263,9 +211,7 @@ class Array
       m_storage.data()[1] = val1;
       m_storage.data()[2] = val2;
     }
-    /** constructs an initialized 4D vector with given coefficients
+    /** constructs an initialized 4D vector with given coefficients */
-      * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args)
-      */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
     {
@@ -296,10 +242,8 @@ class Array
       : Base(other.derived())
     { }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
-    inline Index innerStride() const EIGEN_NOEXCEPT{ return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
 
     #ifdef EIGEN_ARRAY_PLUGIN
     #include EIGEN_ARRAY_PLUGIN
@@ -314,7 +258,7 @@ class Array
 /** \defgroup arraytypedefs Global array typedefs
   * \ingroup Core_Module
   *
-  * %Eigen defines several typedef shortcuts for most common 1D and 2D array types.
+  * Eigen defines several typedef shortcuts for most common 1D and 2D array types.
   *
   * The general patterns are the following:
   *
@@ -327,12 +271,6 @@ class Array
   * There are also \c ArraySizeType which are self-explanatory. For example, \c Array4cf is
   * a fixed-size 1D array of 4 complex floats.
   *
-  * With \cpp11, template alias are also defined for common sizes.
-  * They follow the same pattern as above except that the scalar type suffix is replaced by a
-  * template parameter, i.e.:
-  *   - `ArrayRowsCols<Type>` where `Rows` and `Cols` can be \c 2,\c 3,\c 4, or \c X for fixed or dynamic size.
-  *   - `ArraySize<Type>` where `Size` can be \c 2,\c 3,\c 4 or \c X for fixed or dynamic size 1D arrays.
-  *
   * \sa class Array
   */
 
@@ -365,42 +303,8 @@ EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
 
 #undef EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES
 #undef EIGEN_MAKE_ARRAY_TYPEDEFS
-#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
 
-#if EIGEN_HAS_CXX11
+#undef EIGEN_MAKE_ARRAY_TYPEDEFS_LARGE
-
-#define EIGEN_MAKE_ARRAY_TYPEDEFS(Size, SizeSuffix)               \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##SizeSuffix##SizeSuffix = Array<Type, Size, Size>;    \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##SizeSuffix = Array<Type, Size, 1>;
-
-#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Size)                     \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##Size##X = Array<Type, Size, Dynamic>;                \
-/** \ingroup arraytypedefs */                                     \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Array##X##Size = Array<Type, Dynamic, Size>;
-
-EIGEN_MAKE_ARRAY_TYPEDEFS(2, 2)
-EIGEN_MAKE_ARRAY_TYPEDEFS(3, 3)
-EIGEN_MAKE_ARRAY_TYPEDEFS(4, 4)
-EIGEN_MAKE_ARRAY_TYPEDEFS(Dynamic, X)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(2)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(3)
-EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(4)
-
-#undef EIGEN_MAKE_ARRAY_TYPEDEFS
-#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS
-
-#endif // EIGEN_HAS_CXX11
 
 #define EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
 using Eigen::Matrix##SizeSuffix##TypeSuffix; \

Eigen/src/Core/ArrayBase.h

@@ -69,7 +69,6 @@ template<typename Derived> class ArrayBase
     using Base::coeff;
     using Base::coeffRef;
     using Base::lazyAssign;
-    using Base::operator-;
     using Base::operator=;
     using Base::operator+=;
     using Base::operator-=;
@@ -89,6 +88,7 @@ template<typename Derived> class ArrayBase
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
 #define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/ArrayCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"

Eigen/src/Core/ArrayWrapper.h

@@ -60,14 +60,14 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     EIGEN_DEVICE_FUNC
     explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
+    inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
+    inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
+    inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
+    inline Index innerStride() const { return m_expression.innerStride(); }
 
     EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
@@ -90,8 +90,8 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
     EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const { dst = m_expression; }
 
-    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<NestedExpressionType>::type& 
+    EIGEN_DEVICE_FUNC
     nestedExpression() const 
     {
       return m_expression;
@@ -158,14 +158,14 @@ class MatrixWrapper : public MatrixBase<MatrixWrapper<ExpressionType> >
     EIGEN_DEVICE_FUNC
     explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
+    inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
+    inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
+    inline Index outerStride() const { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
+    inline Index innerStride() const { return m_expression.innerStride(); }
 
     EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }

Eigen/src/Core/AssignEvaluator.h

@@ -24,7 +24,7 @@ namespace internal {
 
 // copy_using_evaluator_traits is based on assign_traits
 
-template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc, int MaxPacketSize = -1>
+template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
 struct copy_using_evaluator_traits
 {
   typedef typename DstEvaluator::XprType Dst;
@@ -51,15 +51,13 @@ private:
     InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
               : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
               : int(Dst::MaxRowsAtCompileTime),
-    RestrictedInnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(InnerSize,MaxPacketSize),
-    RestrictedLinearSize = EIGEN_SIZE_MIN_PREFER_FIXED(Dst::SizeAtCompileTime,MaxPacketSize),
     OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
     MaxSizeAtCompileTime = Dst::SizeAtCompileTime
   };
 
   // TODO distinguish between linear traversal and inner-traversals
-  typedef typename find_best_packet<DstScalar,RestrictedLinearSize>::type LinearPacketType;
+  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
-  typedef typename find_best_packet<DstScalar,RestrictedInnerSize>::type InnerPacketType;
+  typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
 
   enum {
     LinearPacketSize = unpacket_traits<LinearPacketType>::size,
@@ -99,8 +97,7 @@ private:
 
 public:
   enum {
-    Traversal =  int(Dst::SizeAtCompileTime) == 0 ? int(AllAtOnceTraversal) // If compile-size is zero, traversing will fail at compile-time.
+    Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
-              : (int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize)) ? int(LinearVectorizedTraversal)
               : int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
               : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
               : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
@@ -175,8 +172,6 @@ public:
     EIGEN_DEBUG_VAR(MaySliceVectorize)
     std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
     EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
-    EIGEN_DEBUG_VAR(DstEvaluator::CoeffReadCost)
-    EIGEN_DEBUG_VAR(Dst::SizeAtCompileTime)
     EIGEN_DEBUG_VAR(UnrollingLimit)
     EIGEN_DEBUG_VAR(MayUnrollCompletely)
     EIGEN_DEBUG_VAR(MayUnrollInner)
@@ -317,22 +312,6 @@ template<typename Kernel,
          int Unrolling = Kernel::AssignmentTraits::Unrolling>
 struct dense_assignment_loop;
 
-/************************
-***** Special Cases *****
-************************/
-
-// Zero-sized assignment is a no-op.
-template<typename Kernel, int Unrolling>
-struct dense_assignment_loop<Kernel, AllAtOnceTraversal, Unrolling>
-{
-  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel& /*kernel*/)
-  {
-    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
-    EIGEN_STATIC_ASSERT(int(DstXprType::SizeAtCompileTime) == 0,
-      EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT)
-  }
-};
-
 /************************
 *** Default traversal ***
 ************************/
@@ -450,7 +429,7 @@ struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrollin
 
     enum { size = DstXprType::SizeAtCompileTime,
            packetSize =unpacket_traits<PacketType>::size,
-           alignedSize = (int(size)/packetSize)*packetSize };
+           alignedSize = (size/packetSize)*packetSize };
 
     copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
     copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
@@ -551,7 +530,7 @@ struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
     const Scalar *dst_ptr = kernel.dstDataPtr();
     if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
     {
-      // the pointer is not aligned-on scalar, so alignment is not possible
+      // the pointer is not aligend-on scalar, so alignment is not possible
       return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
     }
     const Index packetAlignedMask = packetSize - 1;
@@ -589,15 +568,14 @@ struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
     typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
     typedef typename Kernel::PacketType PacketType;
 
-    enum { innerSize = DstXprType::InnerSizeAtCompileTime,
+    enum { size = DstXprType::InnerSizeAtCompileTime,
            packetSize =unpacket_traits<PacketType>::size,
-           vectorizableSize = (int(innerSize) / int(packetSize)) * int(packetSize),
+           vectorizableSize = (size/packetSize)*packetSize };
-           size = DstXprType::SizeAtCompileTime };
 
     for(Index outer = 0; outer < kernel.outerSize(); ++outer)
     {
       copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
-      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, innerSize>::run(kernel, outer);
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
     }
   }
 };
@@ -629,8 +607,7 @@ public:
   typedef typename AssignmentTraits::PacketType PacketType;
 
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
-  generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
     : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
   {
     #ifdef EIGEN_DEBUG_ASSIGN
@@ -638,15 +615,15 @@ public:
     #endif
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_dstExpr.size(); }
+  EIGEN_DEVICE_FUNC Index size() const        { return m_dstExpr.size(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index innerSize() const EIGEN_NOEXCEPT { return m_dstExpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const   { return m_dstExpr.innerSize(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerSize() const EIGEN_NOEXCEPT { return m_dstExpr.outerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const   { return m_dstExpr.outerSize(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_dstExpr.rows(); }
+  EIGEN_DEVICE_FUNC Index rows() const        { return m_dstExpr.rows(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_dstExpr.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const        { return m_dstExpr.cols(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerStride() const EIGEN_NOEXCEPT { return m_dstExpr.outerStride(); }
+  EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
 
-  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() EIGEN_NOEXCEPT { return m_dst; }
+  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
-  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const EIGEN_NOEXCEPT { return m_src; }
+  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
 
   /// Assign src(row,col) to dst(row,col) through the assignment functor.
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
@@ -720,27 +697,6 @@ protected:
   DstXprType& m_dstExpr;
 };
 
-// Special kernel used when computing small products whose operands have dynamic dimensions.  It ensures that the
-// PacketSize used is no larger than 4, thereby increasing the chance that vectorized instructions will be used
-// when computing the product.
-
-template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor>
-class restricted_packet_dense_assignment_kernel : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, BuiltIn>
-{
-protected:
-  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, BuiltIn> Base;
- public:
-    typedef typename Base::Scalar Scalar;
-    typedef typename Base::DstXprType DstXprType;
-    typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, 4> AssignmentTraits;
-    typedef typename AssignmentTraits::PacketType PacketType;
-
-    EIGEN_DEVICE_FUNC restricted_packet_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
-    : Base(dst, src, func, dstExpr)
-  {
-  }
- };
-
 /***************************************************************************
 * Part 5 : Entry point for dense rectangular assignment
 ***************************************************************************/
@@ -785,16 +741,6 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType
   dense_assignment_loop<Kernel>::run(kernel);
 }
 
-// Specialization for filling the destination with a constant value.
-#ifndef EIGEN_GPU_COMPILE_PHASE
-template<typename DstXprType>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const Eigen::CwiseNullaryOp<Eigen::internal::scalar_constant_op<typename DstXprType::Scalar>, DstXprType>& src, const internal::assign_op<typename DstXprType::Scalar,typename DstXprType::Scalar>& func)
-{
-  resize_if_allowed(dst, src, func);
-  std::fill_n(dst.data(), dst.size(), src.functor()());
-}
-#endif
-
 template<typename DstXprType, typename SrcXprType>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
 {
@@ -810,7 +756,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType
 // AssignmentKind must define a Kind typedef.
 template<typename DstShape, typename SrcShape> struct AssignmentKind;
 
-// Assignment kind defined in this file:
+// Assignement kind defined in this file:
 struct Dense2Dense {};
 struct EigenBase2EigenBase {};
 
@@ -889,27 +835,6 @@ void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
 
   Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
 }
-
-template<typename Dst, typename Src, typename Func>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-void call_restricted_packet_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
-{
-    typedef evaluator<Dst> DstEvaluatorType;
-    typedef evaluator<Src> SrcEvaluatorType;
-    typedef restricted_packet_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Func> Kernel;
-
-    EIGEN_STATIC_ASSERT_LVALUE(Dst)
-    EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
-
-    SrcEvaluatorType srcEvaluator(src);
-    resize_if_allowed(dst, src, func);
-
-    DstEvaluatorType dstEvaluator(dst);
-    Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
-
-    dense_assignment_loop<Kernel>::run(kernel);
-}
-
 template<typename Dst, typename Src>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
 void call_assignment_no_alias(Dst& dst, const Src& src)
@@ -944,8 +869,7 @@ template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, con
 template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
 struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
 {
-  EIGEN_DEVICE_FUNC
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
   {
 #ifndef EIGEN_NO_DEBUG
     internal::check_for_aliasing(dst, src);
@@ -962,8 +886,7 @@ struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
 template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
 struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
 {
-  EIGEN_DEVICE_FUNC
+  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
-  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();
@@ -974,7 +897,7 @@ struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
     src.evalTo(dst);
   }
 
-  // NOTE The following two functions are templated to avoid their instantiation if not needed
+  // NOTE The following two functions are templated to avoid their instanciation if not needed
   //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
   template<typename SrcScalarType>
   EIGEN_DEVICE_FUNC

Eigen/src/Core/Assign_MKL.h

@@ -68,16 +68,16 @@ class vml_assign_traits
 
 #define EIGEN_PP_EXPAND(ARG) ARG
 #if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
-#define EIGEN_VMLMODE_EXPAND_xLA , VML_HA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_HA
 #else
-#define EIGEN_VMLMODE_EXPAND_xLA , VML_LA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
 #endif
 
-#define EIGEN_VMLMODE_EXPAND_x_
+#define EIGEN_VMLMODE_EXPAND__ 
 
-#define EIGEN_VMLMODE_PREFIX_xLA vm
+#define EIGEN_VMLMODE_PREFIX_LA vm
-#define EIGEN_VMLMODE_PREFIX_x_  v
+#define EIGEN_VMLMODE_PREFIX__  v
-#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_x,VMLMODE)
+#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,VMLMODE)
 
 #define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
   template< typename DstXprType, typename SrcXprNested>                                                                         \
@@ -89,7 +89,7 @@ class vml_assign_traits
       eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
       if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
         VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) );                                           \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                           \
       } else {                                                                                                                  \
         const Index outerSize = dst.outerSize();                                                                                \
         for(Index outer = 0; outer < outerSize; ++outer) {                                                                      \
@@ -97,7 +97,7 @@ class vml_assign_traits
                                                       &(src.nestedExpression().coeffRef(0, outer));                             \
           EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                           \
           VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr,                                                                      \
-                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE));                                             \
+                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                             \
         }                                                                                                                       \
       }                                                                                                                         \
     }                                                                                                                           \
@@ -152,7 +152,7 @@ EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
       if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
       {                                                                                                                       \
         VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
-              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) );                                         \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                         \
       } else {                                                                                                                \
         const Index outerSize = dst.outerSize();                                                                              \
         for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
@@ -160,7 +160,7 @@ EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
                                                       &(src.lhs().coeffRef(0, outer));                                        \
           EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
           VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
-                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE));                                          \
+                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                          \
         }                                                                                                                     \
       }                                                                                                                       \
     }                                                                                                                         \

Eigen/src/Core/BandMatrix.h

@@ -67,7 +67,7 @@ class BandMatrixBase : public EigenBase<Derived>
       * \warning the internal storage must be column major. */
     inline Block<CoefficientsType,Dynamic,1> col(Index i)
     {
-      EIGEN_STATIC_ASSERT((int(Options) & int(RowMajor)) == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
+      EIGEN_STATIC_ASSERT((Options&RowMajor)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
       Index start = 0;
       Index len = coeffs().rows();
       if (i<=supers())
@@ -90,7 +90,7 @@ class BandMatrixBase : public EigenBase<Derived>
 
     template<int Index> struct DiagonalIntReturnType {
       enum {
-        ReturnOpposite = (int(Options) & int(SelfAdjoint)) && (((Index) > 0 && Supers == 0) || ((Index) < 0 && Subs == 0)),
+        ReturnOpposite = (Options&SelfAdjoint) && (((Index)>0 && Supers==0) || ((Index)<0 && Subs==0)),
         Conjugate = ReturnOpposite && NumTraits<Scalar>::IsComplex,
         ActualIndex = ReturnOpposite ? -Index : Index,
         DiagonalSize = (RowsAtCompileTime==Dynamic || ColsAtCompileTime==Dynamic)
@@ -192,7 +192,7 @@ struct traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
     Options = _Options,
     DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic
   };
-  typedef Matrix<Scalar, DataRowsAtCompileTime, ColsAtCompileTime, int(Options) & int(RowMajor) ? RowMajor : ColMajor> CoefficientsType;
+  typedef Matrix<Scalar,DataRowsAtCompileTime,ColsAtCompileTime,Options&RowMajor?RowMajor:ColMajor> CoefficientsType;
 };
 
 template<typename _Scalar, int Rows, int Cols, int Supers, int Subs, int Options>
@@ -211,16 +211,16 @@ class BandMatrix : public BandMatrixBase<BandMatrix<_Scalar,Rows,Cols,Supers,Sub
     }
 
     /** \returns the number of columns */
-    inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
+    inline Index rows() const { return m_rows.value(); }
 
     /** \returns the number of rows */
-    inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
+    inline Index cols() const { return m_coeffs.cols(); }
 
     /** \returns the number of super diagonals */
-    inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
+    inline Index supers() const { return m_supers.value(); }
 
     /** \returns the number of sub diagonals */
-    inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
+    inline Index subs() const { return m_subs.value(); }
 
     inline const CoefficientsType& coeffs() const { return m_coeffs; }
     inline CoefficientsType& coeffs() { return m_coeffs; }
@@ -275,16 +275,16 @@ class BandMatrixWrapper : public BandMatrixBase<BandMatrixWrapper<_CoefficientsT
     }
 
     /** \returns the number of columns */
-    inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); }
+    inline Index rows() const { return m_rows.value(); }
 
     /** \returns the number of rows */
-    inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); }
+    inline Index cols() const { return m_coeffs.cols(); }
 
     /** \returns the number of super diagonals */
-    inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); }
+    inline Index supers() const { return m_supers.value(); }
 
     /** \returns the number of sub diagonals */
-    inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); }
+    inline Index subs() const { return m_subs.value(); }
 
     inline const CoefficientsType& coeffs() const { return m_coeffs; }
 

Eigen/src/Core/Block.h

@@ -114,8 +114,8 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
   
     /** Column or Row constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    Block(XprType& xpr, Index i) : Impl(xpr,i)
+    inline Block(XprType& xpr, Index i) : Impl(xpr,i)
     {
       eigen_assert( (i>=0) && (
           ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i<xpr.rows())
@@ -124,8 +124,8 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    Block(XprType& xpr, Index startRow, Index startCol)
+    inline Block(XprType& xpr, Index startRow, Index startCol)
       : Impl(xpr, startRow, startCol)
     {
       EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
@@ -135,8 +135,8 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    Block(XprType& xpr,
+    inline Block(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
       : Impl(xpr, startRow, startCol, blockRows, blockCols)
@@ -159,10 +159,10 @@ class BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, Dense>
   public:
     typedef Impl Base;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
     EIGEN_DEVICE_FUNC
-    EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
       : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
 };
 
@@ -260,19 +260,19 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     }
 
     template<int LoadMode>
-    EIGEN_DEVICE_FUNC inline PacketScalar packet(Index rowId, Index colId) const
+    inline PacketScalar packet(Index rowId, Index colId) const
     {
       return m_xpr.template packet<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
     template<int LoadMode>
-    EIGEN_DEVICE_FUNC inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
+    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
     {
       m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
     }
 
     template<int LoadMode>
-    EIGEN_DEVICE_FUNC inline PacketScalar packet(Index index) const
+    inline PacketScalar packet(Index index) const
     {
       return m_xpr.template packet<Unaligned>
               (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
@@ -280,7 +280,7 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     }
 
     template<int LoadMode>
-    EIGEN_DEVICE_FUNC inline void writePacket(Index index, const PacketScalar& val)
+    inline void writePacket(Index index, const PacketScalar& val)
     {
       m_xpr.template writePacket<Unaligned>
          (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
@@ -294,23 +294,23 @@ template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, bool H
     EIGEN_DEVICE_FUNC inline Index outerStride() const;
     #endif
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     XprType& nestedExpression() { return m_xpr; }
       
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    StorageIndex startRow() const EIGEN_NOEXCEPT
+    StorageIndex startRow() const
     { 
       return m_startRow.value(); 
     }
       
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    StorageIndex startCol() const EIGEN_NOEXCEPT
+    StorageIndex startCol() const
     { 
       return m_startCol.value(); 
     }
@@ -334,17 +334,6 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     enum {
       XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
     };
-
-    /** \internal Returns base+offset (unless base is null, in which case returns null).
-      * Adding an offset to nullptr is undefined behavior, so we must avoid it.
-      */
-    template <typename Scalar>
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE
-    static Scalar* add_to_nullable_pointer(Scalar* base, Index offset)
-    {
-      return base != NULL ? base+offset : NULL;
-    }
-
   public:
 
     typedef MapBase<BlockType> Base;
@@ -353,11 +342,10 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Column or Row constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    BlockImpl_dense(XprType& xpr, Index i)
+    inline BlockImpl_dense(XprType& xpr, Index i)
-      : Base((BlockRows == 0 || BlockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(),
+      : Base(xpr.data() + i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) 
-                 i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor))
+                                || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
-                       || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride())),
              BlockRows==1 ? 1 : xpr.rows(),
              BlockCols==1 ? 1 : xpr.cols()),
         m_xpr(xpr),
@@ -369,10 +357,9 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
-      : Base((BlockRows == 0 || BlockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(),
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
-                 xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol))),
         m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
@@ -380,30 +367,28 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    BlockImpl_dense(XprType& xpr,
+    inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Base((blockRows == 0 || blockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(),
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
-                 xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
-             blockRows, blockCols),
         m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const EIGEN_NOEXCEPT
+    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     XprType& nestedExpression() { return m_xpr; }
       
     /** \sa MapBase::innerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index innerStride() const EIGEN_NOEXCEPT
+    inline Index innerStride() const
     {
       return internal::traits<BlockType>::HasSameStorageOrderAsXprType
              ? m_xpr.innerStride()
@@ -411,19 +396,23 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     }
 
     /** \sa MapBase::outerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index outerStride() const EIGEN_NOEXCEPT
+    inline Index outerStride() const
     {
-      return internal::traits<BlockType>::HasSameStorageOrderAsXprType
+      return m_outerStride;
-                    ? m_xpr.outerStride()
-                    : m_xpr.innerStride();
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    StorageIndex startRow() const EIGEN_NOEXCEPT { return m_startRow.value(); }
+    StorageIndex startRow() const
+    {
+      return m_startRow.value();
+    }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    StorageIndex startCol() const EIGEN_NOEXCEPT { return m_startCol.value(); }
+    StorageIndex startCol() const
+    {
+      return m_startCol.value();
+    }
 
   #ifndef __SUNPRO_CC
   // FIXME sunstudio is not friendly with the above friend...
@@ -433,8 +422,8 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
+    inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
       init();
@@ -442,7 +431,7 @@ class BlockImpl_dense<XprType,BlockRows,BlockCols, InnerPanel,true>
     #endif
 
   protected:
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     void init()
     {
       m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType

Eigen/src/Core/BooleanRedux.h

@@ -14,56 +14,56 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Derived, int UnrollCount, int InnerSize>
+template<typename Derived, int UnrollCount>
 struct all_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    IsRowMajor = (int(Derived::Flags) & int(RowMajor)),
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
-    i = (UnrollCount-1) / InnerSize,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
-    j = (UnrollCount-1) % InnerSize
   };
 
   EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
   {
-    return all_unroller<Derived, UnrollCount-1, InnerSize>::run(mat) && mat.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i);
+    return all_unroller<Derived, UnrollCount-1>::run(mat) && mat.coeff(row, col);
   }
 };
 
-template<typename Derived, int InnerSize>
+template<typename Derived>
-struct all_unroller<Derived, 0, InnerSize>
+struct all_unroller<Derived, 0>
 {
   EIGEN_DEVICE_FUNC static inline bool run(const Derived &/*mat*/) { return true; }
 };
 
-template<typename Derived, int InnerSize>
+template<typename Derived>
-struct all_unroller<Derived, Dynamic, InnerSize>
+struct all_unroller<Derived, Dynamic>
 {
   EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
 };
 
-template<typename Derived, int UnrollCount, int InnerSize>
+template<typename Derived, int UnrollCount>
 struct any_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    IsRowMajor = (int(Derived::Flags) & int(RowMajor)),
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
-    i = (UnrollCount-1) / InnerSize,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
-    j = (UnrollCount-1) % InnerSize
   };
   
   EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat)
   {
-    return any_unroller<Derived, UnrollCount-1, InnerSize>::run(mat) || mat.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i);
+    return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
   }
 };
 
-template<typename Derived, int InnerSize>
+template<typename Derived>
-struct any_unroller<Derived, 0, InnerSize>
+struct any_unroller<Derived, 0>
 {
   EIGEN_DEVICE_FUNC static inline bool run(const Derived & /*mat*/) { return false; }
 };
 
-template<typename Derived, int InnerSize>
+template<typename Derived>
-struct any_unroller<Derived, Dynamic, InnerSize>
+struct any_unroller<Derived, Dynamic>
 {
   EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; }
 };
@@ -83,16 +83,16 @@ EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::all() const
   typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits<Scalar>::AddCost)) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
   Evaluator evaluator(derived());
   if(unroll)
-    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, InnerSizeAtCompileTime>::run(evaluator);
+    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
-    for(Index i = 0; i < derived().outerSize(); ++i)
+    for(Index j = 0; j < cols(); ++j)
-      for(Index j = 0; j < derived().innerSize(); ++j)
+      for(Index i = 0; i < rows(); ++i)
-        if (!evaluator.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i)) return false;
+        if (!evaluator.coeff(i, j)) return false;
     return true;
   }
 }
@@ -107,16 +107,16 @@ EIGEN_DEVICE_FUNC inline bool DenseBase<Derived>::any() const
   typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits<Scalar>::AddCost)) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
   Evaluator evaluator(derived());
   if(unroll)
-    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic, InnerSizeAtCompileTime>::run(evaluator);
+    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
-    for(Index i = 0; i < derived().outerSize(); ++i)
+    for(Index j = 0; j < cols(); ++j)
-      for(Index j = 0; j < derived().innerSize(); ++j)
+      for(Index i = 0; i < rows(); ++i)
-        if (evaluator.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i)) return true;
+        if (evaluator.coeff(i, j)) return true;
     return false;
   }
 }

Eigen/src/Core/CommaInitializer.h

@@ -33,8 +33,6 @@ struct CommaInitializer
   inline CommaInitializer(XprType& xpr, const Scalar& s)
     : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
   {
-    eigen_assert(m_xpr.rows() > 0 && m_xpr.cols() > 0
-      && "Cannot comma-initialize a 0x0 matrix (operator<<)");
     m_xpr.coeffRef(0,0) = s;
   }
 
@@ -43,8 +41,6 @@ struct CommaInitializer
   inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
     : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
   {
-    eigen_assert(m_xpr.rows() >= other.rows() && m_xpr.cols() >= other.cols()
-      && "Cannot comma-initialize a 0x0 matrix (operator<<)");
     m_xpr.block(0, 0, other.rows(), other.cols()) = other;
   }
 
@@ -107,7 +103,7 @@ struct CommaInitializer
   EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
 #endif
   {
-    finished();
+      finished();
   }
 
   /** \returns the built matrix once all its coefficients have been set.

Eigen/src/Core/CoreEvaluators.h

@@ -90,8 +90,7 @@ template<typename T>
 struct evaluator : public unary_evaluator<T>
 {
   typedef unary_evaluator<T> Base;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const T& xpr) : Base(xpr) {}
-  explicit evaluator(const T& xpr) : Base(xpr) {}
 };
 
 
@@ -100,14 +99,14 @@ template<typename T>
 struct evaluator<const T>
   : evaluator<T>
 {
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC
   explicit evaluator(const T& xpr) : evaluator<T>(xpr) {}
 };
 
 // ---------- base class for all evaluators ----------
 
 template<typename ExpressionType>
-struct evaluator_base
+struct evaluator_base : public noncopyable
 {
   // TODO that's not very nice to have to propagate all these traits. They are currently only needed to handle outer,inner indices.
   typedef traits<ExpressionType> ExpressionTraits;
@@ -115,14 +114,6 @@ struct evaluator_base
   enum {
     Alignment = 0
   };
-  // noncopyable:
-  // Don't make this class inherit noncopyable as this kills EBO (Empty Base Optimization)
-  // and make complex evaluator much larger than then should do.
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE evaluator_base() {}
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ~evaluator_base() {}
-private:
-  EIGEN_DEVICE_FUNC evaluator_base(const evaluator_base&);
-  EIGEN_DEVICE_FUNC const evaluator_base& operator=(const evaluator_base&);
 };
 
 // -------------------- Matrix and Array --------------------
@@ -132,33 +123,6 @@ private:
 // Here we directly specialize evaluator. This is not really a unary expression, and it is, by definition, dense,
 // so no need for more sophisticated dispatching.
 
-// this helper permits to completely eliminate m_outerStride if it is known at compiletime.
-template<typename Scalar,int OuterStride> class plainobjectbase_evaluator_data {
-public:
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  plainobjectbase_evaluator_data(const Scalar* ptr, Index outerStride) : data(ptr)
-  {
-#ifndef EIGEN_INTERNAL_DEBUGGING
-    EIGEN_UNUSED_VARIABLE(outerStride);
-#endif
-    eigen_internal_assert(outerStride==OuterStride);
-  }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-  Index outerStride() const EIGEN_NOEXCEPT { return OuterStride; }
-  const Scalar *data;
-};
-
-template<typename Scalar> class plainobjectbase_evaluator_data<Scalar,Dynamic> {
-public:
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  plainobjectbase_evaluator_data(const Scalar* ptr, Index outerStride) : data(ptr), m_outerStride(outerStride) {}
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Index outerStride() const { return m_outerStride; }
-  const Scalar *data;
-protected:
-  Index m_outerStride;
-};
-
 template<typename Derived>
 struct evaluator<PlainObjectBase<Derived> >
   : evaluator_base<Derived>
@@ -177,23 +141,18 @@ struct evaluator<PlainObjectBase<Derived> >
     Flags = traits<Derived>::EvaluatorFlags,
     Alignment = traits<Derived>::Alignment
   };
-  enum {
-    // We do not need to know the outer stride for vectors
-    OuterStrideAtCompileTime = IsVectorAtCompileTime  ? 0
-                                                      : int(IsRowMajor) ? ColsAtCompileTime
-                                                                        : RowsAtCompileTime
-  };
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC evaluator()
-  evaluator()
+    : m_data(0),
-    : m_d(0,OuterStrideAtCompileTime)
+      m_outerStride(IsVectorAtCompileTime  ? 0 
+                                           : int(IsRowMajor) ? ColsAtCompileTime 
+                                           : RowsAtCompileTime)
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const PlainObjectType& m)
-  explicit evaluator(const PlainObjectType& m)
+    : m_data(m.data()), m_outerStride(IsVectorAtCompileTime ? 0 : m.outerStride()) 
-    : m_d(m.data(),IsVectorAtCompileTime ? 0 : m.outerStride())
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
@@ -202,30 +161,30 @@ struct evaluator<PlainObjectBase<Derived> >
   CoeffReturnType coeff(Index row, Index col) const
   {
     if (IsRowMajor)
-      return m_d.data[row * m_d.outerStride() + col];
+      return m_data[row * m_outerStride.value() + col];
     else
-      return m_d.data[row + col * m_d.outerStride()];
+      return m_data[row + col * m_outerStride.value()];
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   {
-    return m_d.data[index];
+    return m_data[index];
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   Scalar& coeffRef(Index row, Index col)
   {
     if (IsRowMajor)
-      return const_cast<Scalar*>(m_d.data)[row * m_d.outerStride() + col];
+      return const_cast<Scalar*>(m_data)[row * m_outerStride.value() + col];
     else
-      return const_cast<Scalar*>(m_d.data)[row + col * m_d.outerStride()];
+      return const_cast<Scalar*>(m_data)[row + col * m_outerStride.value()];
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   Scalar& coeffRef(Index index)
   {
-    return const_cast<Scalar*>(m_d.data)[index];
+    return const_cast<Scalar*>(m_data)[index];
   }
 
   template<int LoadMode, typename PacketType>
@@ -233,16 +192,16 @@ struct evaluator<PlainObjectBase<Derived> >
   PacketType packet(Index row, Index col) const
   {
     if (IsRowMajor)
-      return ploadt<PacketType, LoadMode>(m_d.data + row * m_d.outerStride() + col);
+      return ploadt<PacketType, LoadMode>(m_data + row * m_outerStride.value() + col);
     else
-      return ploadt<PacketType, LoadMode>(m_d.data + row + col * m_d.outerStride());
+      return ploadt<PacketType, LoadMode>(m_data + row + col * m_outerStride.value());
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
-    return ploadt<PacketType, LoadMode>(m_d.data + index);
+    return ploadt<PacketType, LoadMode>(m_data + index);
   }
 
   template<int StoreMode,typename PacketType>
@@ -251,22 +210,26 @@ struct evaluator<PlainObjectBase<Derived> >
   {
     if (IsRowMajor)
       return pstoret<Scalar, PacketType, StoreMode>
-	            (const_cast<Scalar*>(m_d.data) + row * m_d.outerStride() + col, x);
+	            (const_cast<Scalar*>(m_data) + row * m_outerStride.value() + col, x);
     else
       return pstoret<Scalar, PacketType, StoreMode>
-                    (const_cast<Scalar*>(m_d.data) + row + col * m_d.outerStride(), x);
+                    (const_cast<Scalar*>(m_data) + row + col * m_outerStride.value(), x);
   }
 
   template<int StoreMode, typename PacketType>
   EIGEN_STRONG_INLINE
   void writePacket(Index index, const PacketType& x)
   {
-    return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_d.data) + index, x);
+    return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_data) + index, x);
   }
 
 protected:
+  const Scalar *m_data;
 
-  plainobjectbase_evaluator_data<Scalar,OuterStrideAtCompileTime> m_d;
+  // We do not need to know the outer stride for vectors
+  variable_if_dynamic<Index, IsVectorAtCompileTime  ? 0 
+                                                    : int(IsRowMajor) ? ColsAtCompileTime 
+                                                    : RowsAtCompileTime> m_outerStride;
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -275,11 +238,9 @@ struct evaluator<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
 {
   typedef Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC evaluator() {}
-  evaluator() {}
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
-  explicit evaluator(const XprType& m)
     : evaluator<PlainObjectBase<XprType> >(m) 
   { }
 };
@@ -290,11 +251,9 @@ struct evaluator<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
 {
   typedef Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC evaluator() {}
-  evaluator() {}
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
-  explicit evaluator(const XprType& m)
     : evaluator<PlainObjectBase<XprType> >(m) 
   { }
 };
@@ -313,8 +272,7 @@ struct unary_evaluator<Transpose<ArgType>, IndexBased>
     Alignment = evaluator<ArgType>::Alignment
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {}
-  explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {}
 
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
@@ -561,7 +519,7 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
   typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
   
   enum {
-    CoeffReadCost = int(evaluator<ArgType>::CoeffReadCost) + int(functor_traits<UnaryOp>::Cost),
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
     
     Flags = evaluator<ArgType>::Flags
           & (HereditaryBits | LinearAccessBit | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
@@ -569,7 +527,9 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
   };
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  explicit unary_evaluator(const XprType& op) : m_d(op)
+  explicit unary_evaluator(const XprType& op)
+    : m_functor(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
@@ -580,43 +540,32 @@ struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index row, Index col) const
   {
-    return m_d.func()(m_d.argImpl.coeff(row, col));
+    return m_functor(m_argImpl.coeff(row, col));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   {
-    return m_d.func()(m_d.argImpl.coeff(index));
+    return m_functor(m_argImpl.coeff(index));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
-    return m_d.func().packetOp(m_d.argImpl.template packet<LoadMode, PacketType>(row, col));
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(row, col));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
-    return m_d.func().packetOp(m_d.argImpl.template packet<LoadMode, PacketType>(index));
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(index));
   }
 
 protected:
-
+  const UnaryOp m_functor;
-  // this helper permits to completely eliminate the functor if it is empty
+  evaluator<ArgType> m_argImpl;
-  struct Data
-  {
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Data(const XprType& xpr) : op(xpr.functor()), argImpl(xpr.nestedExpression()) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const UnaryOp& func() const { return op; }
-    UnaryOp op;
-    evaluator<ArgType> argImpl;
-  };
-
-  Data m_d;
 };
 
 // -------------------- CwiseTernaryOp --------------------
@@ -639,7 +588,7 @@ struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased
   typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
   
   enum {
-    CoeffReadCost = int(evaluator<Arg1>::CoeffReadCost) + int(evaluator<Arg2>::CoeffReadCost) + int(evaluator<Arg3>::CoeffReadCost) + int(functor_traits<TernaryOp>::Cost),
+    CoeffReadCost = evaluator<Arg1>::CoeffReadCost + evaluator<Arg2>::CoeffReadCost + evaluator<Arg3>::CoeffReadCost + functor_traits<TernaryOp>::Cost,
     
     Arg1Flags = evaluator<Arg1>::Flags,
     Arg2Flags = evaluator<Arg2>::Flags,
@@ -660,7 +609,11 @@ struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased
         evaluator<Arg3>::Alignment)
   };
 
-  EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr) : m_d(xpr)
+  EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_arg1Impl(xpr.arg1()), 
+      m_arg2Impl(xpr.arg2()), 
+      m_arg3Impl(xpr.arg3())  
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<TernaryOp>::Cost);
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
@@ -671,48 +624,38 @@ struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index row, Index col) const
   {
-    return m_d.func()(m_d.arg1Impl.coeff(row, col), m_d.arg2Impl.coeff(row, col), m_d.arg3Impl.coeff(row, col));
+    return m_functor(m_arg1Impl.coeff(row, col), m_arg2Impl.coeff(row, col), m_arg3Impl.coeff(row, col));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   {
-    return m_d.func()(m_d.arg1Impl.coeff(index), m_d.arg2Impl.coeff(index), m_d.arg3Impl.coeff(index));
+    return m_functor(m_arg1Impl.coeff(index), m_arg2Impl.coeff(index), m_arg3Impl.coeff(index));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
-    return m_d.func().packetOp(m_d.arg1Impl.template packet<LoadMode,PacketType>(row, col),
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(row, col),
-                               m_d.arg2Impl.template packet<LoadMode,PacketType>(row, col),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(row, col),
-                               m_d.arg3Impl.template packet<LoadMode,PacketType>(row, col));
+                              m_arg3Impl.template packet<LoadMode,PacketType>(row, col));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
-    return m_d.func().packetOp(m_d.arg1Impl.template packet<LoadMode,PacketType>(index),
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(index),
-                               m_d.arg2Impl.template packet<LoadMode,PacketType>(index),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(index),
-                               m_d.arg3Impl.template packet<LoadMode,PacketType>(index));
+                              m_arg3Impl.template packet<LoadMode,PacketType>(index));
   }
 
 protected:
-  // this helper permits to completely eliminate the functor if it is empty
+  const TernaryOp m_functor;
-  struct Data
+  evaluator<Arg1> m_arg1Impl;
-  {
+  evaluator<Arg2> m_arg2Impl;
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  evaluator<Arg3> m_arg3Impl;
-    Data(const XprType& xpr) : op(xpr.functor()), arg1Impl(xpr.arg1()), arg2Impl(xpr.arg2()), arg3Impl(xpr.arg3()) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const TernaryOp& func() const { return op; }
-    TernaryOp op;
-    evaluator<Arg1> arg1Impl;
-    evaluator<Arg2> arg2Impl;
-    evaluator<Arg3> arg3Impl;
-  };
-
-  Data m_d;
 };
 
 // -------------------- CwiseBinaryOp --------------------
@@ -725,8 +668,7 @@ struct evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
   typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
   typedef binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> > Base;
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
-  explicit evaluator(const XprType& xpr) : Base(xpr) {}
 };
 
 template<typename BinaryOp, typename Lhs, typename Rhs>
@@ -736,7 +678,7 @@ struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBase
   typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
   
   enum {
-    CoeffReadCost = int(evaluator<Lhs>::CoeffReadCost) + int(evaluator<Rhs>::CoeffReadCost) + int(functor_traits<BinaryOp>::Cost),
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
     
     LhsFlags = evaluator<Lhs>::Flags,
     RhsFlags = evaluator<Rhs>::Flags,
@@ -754,8 +696,10 @@ struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBase
     Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<Lhs>::Alignment,evaluator<Rhs>::Alignment)
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit binary_evaluator(const XprType& xpr)
-  explicit binary_evaluator(const XprType& xpr) : m_d(xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
@@ -766,46 +710,35 @@ struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBase
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index row, Index col) const
   {
-    return m_d.func()(m_d.lhsImpl.coeff(row, col), m_d.rhsImpl.coeff(row, col));
+    return m_functor(m_lhsImpl.coeff(row, col), m_rhsImpl.coeff(row, col));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   {
-    return m_d.func()(m_d.lhsImpl.coeff(index), m_d.rhsImpl.coeff(index));
+    return m_functor(m_lhsImpl.coeff(index), m_rhsImpl.coeff(index));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index row, Index col) const
   {
-    return m_d.func().packetOp(m_d.lhsImpl.template packet<LoadMode,PacketType>(row, col),
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(row, col),
-                               m_d.rhsImpl.template packet<LoadMode,PacketType>(row, col));
+                              m_rhsImpl.template packet<LoadMode,PacketType>(row, col));
   }
 
   template<int LoadMode, typename PacketType>
   EIGEN_STRONG_INLINE
   PacketType packet(Index index) const
   {
-    return m_d.func().packetOp(m_d.lhsImpl.template packet<LoadMode,PacketType>(index),
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(index),
-                               m_d.rhsImpl.template packet<LoadMode,PacketType>(index));
+                              m_rhsImpl.template packet<LoadMode,PacketType>(index));
   }
 
 protected:
-
+  const BinaryOp m_functor;
-  // this helper permits to completely eliminate the functor if it is empty
+  evaluator<Lhs> m_lhsImpl;
-  struct Data
+  evaluator<Rhs> m_rhsImpl;
-  {
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Data(const XprType& xpr) : op(xpr.functor()), lhsImpl(xpr.lhs()), rhsImpl(xpr.rhs()) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const BinaryOp& func() const { return op; }
-    BinaryOp op;
-    evaluator<Lhs> lhsImpl;
-    evaluator<Rhs> rhsImpl;
-  };
-
-  Data m_d;
 };
 
 // -------------------- CwiseUnaryView --------------------
@@ -817,14 +750,16 @@ struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
   typedef CwiseUnaryView<UnaryOp, ArgType> XprType;
   
   enum {
-    CoeffReadCost = int(evaluator<ArgType>::CoeffReadCost) + int(functor_traits<UnaryOp>::Cost),
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
     
     Flags = (evaluator<ArgType>::Flags & (HereditaryBits | LinearAccessBit | DirectAccessBit)),
     
     Alignment = 0 // FIXME it is not very clear why alignment is necessarily lost...
   };
 
-  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op) : m_d(op)
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+    : m_unaryOp(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
@@ -836,41 +771,30 @@ struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index row, Index col) const
   {
-    return m_d.func()(m_d.argImpl.coeff(row, col));
+    return m_unaryOp(m_argImpl.coeff(row, col));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   {
-    return m_d.func()(m_d.argImpl.coeff(index));
+    return m_unaryOp(m_argImpl.coeff(index));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   Scalar& coeffRef(Index row, Index col)
   {
-    return m_d.func()(m_d.argImpl.coeffRef(row, col));
+    return m_unaryOp(m_argImpl.coeffRef(row, col));
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   Scalar& coeffRef(Index index)
   {
-    return m_d.func()(m_d.argImpl.coeffRef(index));
+    return m_unaryOp(m_argImpl.coeffRef(index));
   }
 
 protected:
-
+  const UnaryOp m_unaryOp;
-  // this helper permits to completely eliminate the functor if it is empty
+  evaluator<ArgType> m_argImpl;
-  struct Data
-  {
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Data(const XprType& xpr) : op(xpr.functor()), argImpl(xpr.nestedExpression()) {}
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const UnaryOp& func() const { return op; }
-    UnaryOp op;
-    evaluator<ArgType> argImpl;
-  };
-
-  Data m_d;
 };
 
 // -------------------- Map --------------------
@@ -894,8 +818,7 @@ struct mapbase_evaluator : evaluator_base<Derived>
     CoeffReadCost = NumTraits<Scalar>::ReadCost
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map)
-  explicit mapbase_evaluator(const XprType& map)
     : m_data(const_cast<PointerType>(map.data())),
       m_innerStride(map.innerStride()),
       m_outerStride(map.outerStride())
@@ -959,14 +882,10 @@ struct mapbase_evaluator : evaluator_base<Derived>
     internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_innerStride.value(), x);
   }
 protected:
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
-  Index rowStride() const EIGEN_NOEXCEPT {
+  inline Index rowStride() const { return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); }
-    return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value();
+  EIGEN_DEVICE_FUNC
-  }
+  inline Index colStride() const { return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-  Index colStride() const EIGEN_NOEXCEPT {
-     return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value();
-  }
 
   PointerType m_data;
   const internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_innerStride;
@@ -1019,8 +938,7 @@ struct evaluator<Ref<PlainObjectType, RefOptions, StrideType> >
     Alignment = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Alignment
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& ref)
-  explicit evaluator(const XprType& ref)
     : mapbase_evaluator<XprType, PlainObjectType>(ref) 
   { }
 };
@@ -1075,8 +993,7 @@ struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
     Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ArgType>::Alignment, Alignment0)
   };
   typedef block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel> block_evaluator_type;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& block) : block_evaluator_type(block)
-  explicit evaluator(const XprType& block) : block_evaluator_type(block)
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
   }
@@ -1089,8 +1006,7 @@ struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /*HasDirectAcc
 {
   typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
-  explicit block_evaluator(const XprType& block)
     : unary_evaluator<XprType>(block) 
   {}
 };
@@ -1101,12 +1017,11 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
 {
   typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& block)
-  explicit unary_evaluator(const XprType& block)
     : m_argImpl(block.nestedExpression()), 
       m_startRow(block.startRow()), 
       m_startCol(block.startCol()),
-      m_linear_offset(ForwardLinearAccess?(ArgType::IsRowMajor ? block.startRow()*block.nestedExpression().cols() + block.startCol() : block.startCol()*block.nestedExpression().rows() + block.startRow()):0)
+      m_linear_offset(InnerPanel?(XprType::IsRowMajor ? block.startRow()*block.cols() : block.startCol()*block.rows()):0)
   { }
  
   typedef typename XprType::Scalar Scalar;
@@ -1114,7 +1029,7 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
 
   enum {
     RowsAtCompileTime = XprType::RowsAtCompileTime,
-    ForwardLinearAccess = (InnerPanel || int(XprType::IsRowMajor)==int(ArgType::IsRowMajor)) && bool(evaluator<ArgType>::Flags&LinearAccessBit)
+    ForwardLinearAccess = InnerPanel && bool(evaluator<ArgType>::Flags&LinearAccessBit)
   };
  
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
@@ -1126,7 +1041,10 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   CoeffReturnType coeff(Index index) const
   { 
-    return linear_coeff_impl(index, bool_constant<ForwardLinearAccess>());
+    if (ForwardLinearAccess)
+      return m_argImpl.coeff(m_linear_offset.value() + index); 
+    else
+      return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
   }
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
@@ -1138,7 +1056,10 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   Scalar& coeffRef(Index index)
   { 
-    return linear_coeffRef_impl(index, bool_constant<ForwardLinearAccess>());
+    if (ForwardLinearAccess)
+      return m_argImpl.coeffRef(m_linear_offset.value() + index); 
+    else
+      return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
   }
  
   template<int LoadMode, typename PacketType>
@@ -1179,32 +1100,10 @@ struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBa
   }
  
 protected:
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  CoeffReturnType linear_coeff_impl(Index index, internal::true_type /* ForwardLinearAccess */) const
-  {
-    return m_argImpl.coeff(m_linear_offset.value() + index);
-  }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  CoeffReturnType linear_coeff_impl(Index index, internal::false_type /* not ForwardLinearAccess */) const
-  {
-    return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& linear_coeffRef_impl(Index index, internal::true_type /* ForwardLinearAccess */)
-  {
-    return m_argImpl.coeffRef(m_linear_offset.value() + index);
-  }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& linear_coeffRef_impl(Index index, internal::false_type /* not ForwardLinearAccess */)
-  {
-    return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
-  }
-
   evaluator<ArgType> m_argImpl;
   const variable_if_dynamic<Index, (ArgType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
   const variable_if_dynamic<Index, (ArgType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
-  const variable_if_dynamic<Index, ForwardLinearAccess ? Dynamic : 0> m_linear_offset;
+  const variable_if_dynamic<Index, InnerPanel ? Dynamic : 0> m_linear_offset;
 };
 
 // TODO: This evaluator does not actually use the child evaluator; 
@@ -1218,8 +1117,7 @@ struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /* HasDirectAc
   typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
   typedef typename XprType::Scalar Scalar;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
-  explicit block_evaluator(const XprType& block)
     : mapbase_evaluator<XprType, typename XprType::PlainObject>(block) 
   {
     // TODO: for the 3.3 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
@@ -1247,8 +1145,7 @@ struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
     Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ThenMatrixType>::Alignment, evaluator<ElseMatrixType>::Alignment)
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& select)
-  explicit evaluator(const XprType& select)
     : m_conditionImpl(select.conditionMatrix()),
       m_thenImpl(select.thenMatrix()),
       m_elseImpl(select.elseMatrix())
@@ -1305,8 +1202,7 @@ struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
     Alignment = evaluator<ArgTypeNestedCleaned>::Alignment
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& replicate)
-  explicit unary_evaluator(const XprType& replicate)
     : m_arg(replicate.nestedExpression()),
       m_argImpl(m_arg),
       m_rows(replicate.nestedExpression().rows()),
@@ -1370,6 +1266,64 @@ protected:
   const variable_if_dynamic<Index, ArgType::ColsAtCompileTime> m_cols;
 };
 
+
+// -------------------- PartialReduxExpr --------------------
+
+template< typename ArgType, typename MemberOp, int Direction>
+struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
+  : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
+{
+  typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
+  typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
+  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+  typedef typename ArgType::Scalar InputScalar;
+  typedef typename XprType::Scalar Scalar;
+  enum {
+    TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) :  int(ArgType::ColsAtCompileTime)
+  };
+  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
+  enum {
+    CoeffReadCost = TraversalSize==Dynamic ? HugeCost
+                  : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
+    
+    Flags = (traits<XprType>::Flags&RowMajorBit) | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit))) | LinearAccessBit,
+    
+    Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
+    : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : int(CostOpType::value));
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index i, Index j) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(j));
+    else
+      return m_functor(m_arg.row(i));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index index) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(index));
+    else
+      return m_functor(m_arg.row(index));
+  }
+
+protected:
+  typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
+  const MemberOp m_functor;
+};
+
+
 // -------------------- MatrixWrapper and ArrayWrapper --------------------
 //
 // evaluator_wrapper_base<T> is a common base class for the
@@ -1386,8 +1340,7 @@ struct evaluator_wrapper_base
     Alignment = evaluator<ArgType>::Alignment
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {}
-  explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {}
 
   typedef typename ArgType::Scalar Scalar;
   typedef typename ArgType::CoeffReturnType CoeffReturnType;
@@ -1454,8 +1407,7 @@ struct unary_evaluator<MatrixWrapper<TArgType> >
 {
   typedef MatrixWrapper<TArgType> XprType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
-  explicit unary_evaluator(const XprType& wrapper)
     : evaluator_wrapper_base<MatrixWrapper<TArgType> >(wrapper.nestedExpression())
   { }
 };
@@ -1466,8 +1418,7 @@ struct unary_evaluator<ArrayWrapper<TArgType> >
 {
   typedef ArrayWrapper<TArgType> XprType;
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
-  explicit unary_evaluator(const XprType& wrapper)
     : evaluator_wrapper_base<ArrayWrapper<TArgType> >(wrapper.nestedExpression())
   { }
 };
@@ -1509,8 +1460,7 @@ struct unary_evaluator<Reverse<ArgType, Direction> >
     Alignment = 0 // FIXME in some rare cases, Alignment could be preserved, like a Vector4f.
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& reverse)
-  explicit unary_evaluator(const XprType& reverse)
     : m_argImpl(reverse.nestedExpression()),
       m_rows(ReverseRow ? reverse.nestedExpression().rows() : 1),
       m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1)
@@ -1617,8 +1567,7 @@ struct evaluator<Diagonal<ArgType, DiagIndex> >
     Alignment = 0
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& diagonal)
-  explicit evaluator(const XprType& diagonal)
     : m_argImpl(diagonal.nestedExpression()),
       m_index(diagonal.index())
   { }
@@ -1655,10 +1604,8 @@ protected:
   const internal::variable_if_dynamicindex<Index, XprType::DiagIndex> m_index;
 
 private:
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); }
-  Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-  Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; }
 };
 
 
@@ -1695,12 +1642,12 @@ class EvalToTemp
     return m_arg;
   }
 
-  EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT
+  Index rows() const 
   {
     return m_arg.rows();
   }
 
-  EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT
+  Index cols() const 
   {
     return m_arg.cols();
   }

Eigen/src/Core/CoreIterators.h

@@ -48,11 +48,6 @@ public:
     * Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView
     */
   EIGEN_STRONG_INLINE InnerIterator& operator++()   { m_iter.operator++(); return *this; }
-  EIGEN_STRONG_INLINE InnerIterator& operator+=(Index i) { m_iter.operator+=(i); return *this; }
-  EIGEN_STRONG_INLINE InnerIterator operator+(Index i) 
-  { InnerIterator result(*this); result+=i; return result; }
-    
-
   /// \returns the column or row index of the current coefficient.
   EIGEN_STRONG_INLINE Index index() const           { return m_iter.index(); }
   /// \returns the row index of the current coefficient.

Eigen/src/Core/CwiseBinaryOp.h

@@ -100,14 +100,8 @@ class CwiseBinaryOp :
     typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
     typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
 
-#if EIGEN_COMP_MSVC && EIGEN_HAS_CXX11
+    EIGEN_DEVICE_FUNC
-    //Required for Visual Studio or the Copy constructor will probably not get inlined!
+    EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
-    EIGEN_STRONG_INLINE
-    CwiseBinaryOp(const CwiseBinaryOp<BinaryOp,LhsType,RhsType>&) = default;
-#endif
-
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
       : m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
     {
       EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename Rhs::Scalar);
@@ -116,25 +110,31 @@ class CwiseBinaryOp :
       eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index rows() const EIGEN_NOEXCEPT {
+    EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
-      return internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic ? m_rhs.rows() : m_lhs.rows();
+      if (internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic)
+        return m_rhs.rows();
+      else
+        return m_lhs.rows();
     }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index cols() const EIGEN_NOEXCEPT {
+    EIGEN_STRONG_INLINE Index cols() const {
       // return the fixed size type if available to enable compile time optimizations
-      return internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic ? m_rhs.cols() : m_lhs.cols();
+      if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
+        return m_rhs.cols();
+      else
+        return m_lhs.cols();
     }
 
     /** \returns the left hand side nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     const _LhsNested& lhs() const { return m_lhs; }
     /** \returns the right hand side nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     const _RhsNested& rhs() const { return m_rhs; }
     /** \returns the functor representing the binary operation */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     const BinaryOp& functor() const { return m_functor; }
 
   protected:
@@ -181,3 +181,4 @@ MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_BINARY_OP_H
+

Eigen/src/Core/CwiseNullaryOp.h

@@ -74,10 +74,10 @@ class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp
             && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index rows() const { return m_rows.value(); }
+    EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index cols() const { return m_cols.value(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
 
     /** \returns the functor representing the nullary operation */
     EIGEN_DEVICE_FUNC
@@ -105,12 +105,7 @@ class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp,typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp,PlainObject>
-#endif
 DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
   return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
@@ -136,12 +131,7 @@ DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& f
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp, PlainObject>
-#endif
 DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -160,12 +150,7 @@ DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
-#else
-const CwiseNullaryOp<CustomNullaryOp, PlainObject>
-#endif
 DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
   return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
@@ -232,32 +217,27 @@ DenseBase<Derived>::Constant(const Scalar& value)
 
 /** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
   *
-  * \only_for_vectors
+  * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
-  *
-  * Example: \include DenseBase_LinSpaced_seq_deprecated.cpp
-  * Output: \verbinclude DenseBase_LinSpaced_seq_deprecated.out
-  *
-  * \sa LinSpaced(Index,const Scalar&, const Scalar&), setLinSpaced(Index,const Scalar&,const Scalar&)
   */
 template<typename Derived>
-EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
 /** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
   *
-  * \sa LinSpaced(const Scalar&, const Scalar&)
+  * \sa LinSpaced(Scalar,Scalar)
   */
 template<typename Derived>
-EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /**
@@ -288,7 +268,7 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomA
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
 /**
@@ -301,7 +281,7 @@ DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
@@ -383,33 +363,6 @@ PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
   return setConstant(val);
 }
 
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to the given value \a val. For the parameter
-  * of type NoChange_t, just pass the special value \c NoChange.
-  *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(NoChange_t, Index cols, const Scalar& val)
-{
-  return setConstant(rows(), cols, val);
-}
-
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to the given value \a val. For the parameter
-  * of type NoChange_t, just pass the special value \c NoChange.
-  *
-  * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index rows, NoChange_t, const Scalar& val)
-{
-  return setConstant(rows, cols(), val);
-}
-
-
 /**
   * \brief Sets a linearly spaced vector.
   *
@@ -430,7 +383,7 @@ template<typename Derived>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar>(low,high,newSize));
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
 }
 
 /**
@@ -583,32 +536,6 @@ PlainObjectBase<Derived>::setZero(Index rows, Index cols)
   return setConstant(Scalar(0));
 }
 
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to zero. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
-  * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Zero()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(NoChange_t, Index cols)
-{
-  return setZero(rows(), cols);
-}
-
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to zero. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
-  * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Zero()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index rows, NoChange_t)
-{
-  return setZero(rows, cols());
-}
-
 // ones:
 
 /** \returns an expression of a matrix where all coefficients equal one.
@@ -735,32 +662,6 @@ PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
   return setConstant(Scalar(1));
 }
 
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to one. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
- * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(NoChange_t, Index), class CwiseNullaryOp, MatrixBase::Ones()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index rows, NoChange_t)
-{
-  return setOnes(rows, cols());
-}
-
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to one. For the parameter of type NoChange_t,
-  * just pass the special value \c NoChange.
-  *
- * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(Index, NoChange_t) class CwiseNullaryOp, MatrixBase::Ones()
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(NoChange_t, Index cols)
-{
-  return setOnes(rows(), cols);
-}
-
 // Identity:
 
 /** \returns an expression of the identity matrix (not necessarily square).
@@ -960,42 +861,6 @@ template<typename Derived>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
-/** \brief Set the coefficients of \c *this to the i-th unit (basis) vector
-  *
-  * \param i index of the unique coefficient to be set to 1
-  *
-  * \only_for_vectors
-  *
-  * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Unit(Index,Index)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setUnit(Index i)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  eigen_assert(i<size());
-  derived().setZero();
-  derived().coeffRef(i) = Scalar(1);
-  return derived();
-}
-
-/** \brief Resizes to the given \a newSize, and writes the i-th unit (basis) vector into *this.
-  *
-  * \param newSize the new size of the vector
-  * \param i index of the unique coefficient to be set to 1
-  *
-  * \only_for_vectors
-  *
-  * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Unit(Index,Index)
-  */
-template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setUnit(Index newSize, Index i)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  eigen_assert(i<newSize);
-  derived().resize(newSize);
-  return setUnit(i);
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_NULLARY_OP_H

Eigen/src/Core/CwiseUnaryOp.h

@@ -65,10 +65,10 @@ class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal
     explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
+    Index rows() const { return m_xpr.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
+    Index cols() const { return m_xpr.cols(); }
 
     /** \returns the functor representing the unary operation */
     EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE

Eigen/src/Core/CwiseUnaryView.h

@@ -64,26 +64,24 @@ class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename in
     typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
     typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    explicit EIGEN_DEVICE_FUNC inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
+    explicit inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
-    Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
 
     /** \returns the functor representing unary operation */
-    EIGEN_DEVICE_FUNC const ViewOp& functor() const { return m_functor; }
+    const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC const typename internal::remove_all<MatrixTypeNested>::type&
+    const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC typename internal::remove_reference<MatrixTypeNested>::type&
+    typename internal::remove_reference<MatrixTypeNested>::type&
-    nestedExpression() { return m_matrix; }
+    nestedExpression() { return m_matrix.const_cast_derived(); }
 
   protected:
     MatrixTypeNested m_matrix;
@@ -114,12 +112,12 @@ class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
     EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
     EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
       return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const
+    EIGEN_DEVICE_FUNC inline Index outerStride() const
     {
       return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }

Eigen/src/Core/DenseBase.h

@@ -18,7 +18,7 @@ namespace internal {
 // The index type defined by EIGEN_DEFAULT_DENSE_INDEX_TYPE must be a signed type.
 // This dummy function simply aims at checking that at compile time.
 static inline void check_DenseIndex_is_signed() {
-  EIGEN_STATIC_ASSERT(NumTraits<DenseIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE)
+  EIGEN_STATIC_ASSERT(NumTraits<DenseIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE); 
 }
 
 } // end namespace internal
@@ -150,18 +150,13 @@ template<typename Derived> class DenseBase
           * \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime
           */
 
-      IsVectorAtCompileTime = internal::traits<Derived>::RowsAtCompileTime == 1
+      IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
-                           || internal::traits<Derived>::ColsAtCompileTime == 1,
+                           || internal::traits<Derived>::MaxColsAtCompileTime == 1,
         /**< This is set to true if either the number of rows or the number of
           * columns is known at compile-time to be equal to 1. Indeed, in that case,
           * we are dealing with a column-vector (if there is only one column) or with
           * a row-vector (if there is only one row). */
 
-      NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2,
-        /**< This value is equal to Tensor::NumDimensions, i.e. 0 for scalars, 1 for vectors,
-         * and 2 for matrices.
-         */
-
       Flags = internal::traits<Derived>::Flags,
         /**< This stores expression \ref flags flags which may or may not be inherited by new expressions
           * constructed from this one. See the \ref flags "list of flags".
@@ -211,7 +206,7 @@ template<typename Derived> class DenseBase
 
     /** \returns the number of nonzero coefficients which is in practice the number
       * of stored coefficients. */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index nonZeros() const { return size(); }
 
     /** \returns the outer size.
@@ -219,7 +214,7 @@ template<typename Derived> class DenseBase
       * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
       * column-major matrix, and the number of rows for a row-major matrix. */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     Index outerSize() const
     {
       return IsVectorAtCompileTime ? 1
@@ -231,7 +226,7 @@ template<typename Derived> class DenseBase
       * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a 
       * column-major matrix, and the number of columns for a row-major matrix. */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     Index innerSize() const
     {
       return IsVectorAtCompileTime ? this->size()
@@ -266,9 +261,9 @@ template<typename Derived> class DenseBase
     /** \internal Represents a matrix with all coefficients equal to one another*/
     typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
     /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
-    EIGEN_DEPRECATED typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> SequentialLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> SequentialLinSpacedReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar>,PlainObject> RandomAccessLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> RandomAccessLinSpacedReturnType;
     /** \internal the return type of MatrixBase::eigenvalues() */
     typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
@@ -302,17 +297,17 @@ template<typename Derived> class DenseBase
     Derived& operator=(const ReturnByValue<OtherDerived>& func);
 
     /** \internal
-      * Copies \a other into *this without evaluating other. \returns a reference to *this. */
+      * Copies \a other into *this without evaluating other. \returns a reference to *this.
+      * \deprecated */
     template<typename OtherDerived>
-    /** \deprecated */
+    EIGEN_DEVICE_FUNC
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
     Derived& lazyAssign(const DenseBase<OtherDerived>& other);
 
     EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
-    template<unsigned int Added,unsigned int Removed>
     /** \deprecated it now returns \c *this */
+    template<unsigned int Added,unsigned int Removed>
     EIGEN_DEPRECATED
     const Derived& flagged() const
     { return derived(); }
@@ -324,9 +319,9 @@ template<typename Derived> class DenseBase
     typedef Transpose<Derived> TransposeReturnType;
     EIGEN_DEVICE_FUNC
     TransposeReturnType transpose();
-    typedef Transpose<const Derived> ConstTransposeReturnType;
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
     EIGEN_DEVICE_FUNC
-    const ConstTransposeReturnType transpose() const;
+    ConstTransposeReturnType transpose() const;
     EIGEN_DEVICE_FUNC
     void transposeInPlace();
 
@@ -337,13 +332,12 @@ template<typename Derived> class DenseBase
     EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(const Scalar& value);
 
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
-    LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-
     EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
+    LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
     EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(const Scalar& low, const Scalar& high);
 
@@ -401,7 +395,7 @@ template<typename Derived> class DenseBase
       * Notice that in the case of a plain matrix or vector (not an expression) this function just returns
       * a const reference, in order to avoid a useless copy.
       * 
-      * \warning Be careful with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
+      * \warning Be carefull with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
       */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE EvalReturnType eval() const
@@ -416,7 +410,7 @@ template<typename Derived> class DenseBase
       *
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     void swap(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
@@ -428,7 +422,7 @@ template<typename Derived> class DenseBase
       *
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     void swap(PlainObjectBase<OtherDerived>& other)
     {
       eigen_assert(rows()==other.rows() && cols()==other.cols());
@@ -449,58 +443,18 @@ template<typename Derived> class DenseBase
 
     EIGEN_DEVICE_FUNC Scalar prod() const;
 
-    template<int NaNPropagation>
     EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
-    template<int NaNPropagation>
     EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
 
-
+    template<typename IndexType> EIGEN_DEVICE_FUNC
-    // By default, the fastest version with undefined NaN propagation semantics is
-    // used.
-    // TODO(rmlarsen): Replace with default template argument when we move to
-    // c++11 or beyond.
-    EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar minCoeff() const {
-      return minCoeff<PropagateFast>();
-    }
-    EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar maxCoeff() const {
-      return maxCoeff<PropagateFast>();
-    }
-
-    template<int NaNPropagation, typename IndexType>
-    EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
-    template<int NaNPropagation, typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
-    EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
-    template<int NaNPropagation, typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
-    EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
-    template<int NaNPropagation, typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
-    EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
 
-    // TODO(rmlarsen): Replace these methods with a default template argument.
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const {
-      return minCoeff<PropagateFast>(row, col);
-    }
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const {
-      return maxCoeff<PropagateFast>(row, col);
-    }
-    template<typename IndexType>
-     EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const {
-      return minCoeff<PropagateFast>(index);
-    }
-    template<typename IndexType>
-    EIGEN_DEVICE_FUNC inline
-    typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const {
-      return maxCoeff<PropagateFast>(index);
-    }
-  
     template<typename BinaryOp>
     EIGEN_DEVICE_FUNC
     Scalar redux(const BinaryOp& func) const;
@@ -539,7 +493,7 @@ template<typename Derived> class DenseBase
     typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
     typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    /** \returns a VectorwiseOp wrapper of *this for broadcasting and partial reductions
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
     *
     * Example: \include MatrixBase_rowwise.cpp
     * Output: \verbinclude MatrixBase_rowwise.out
@@ -552,7 +506,7 @@ template<typename Derived> class DenseBase
     }
     EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
 
-    /** \returns a VectorwiseOp wrapper of *this broadcasting and partial reductions
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
     *
     * Example: \include MatrixBase_colwise.cpp
     * Output: \verbinclude MatrixBase_colwise.out
@@ -570,16 +524,16 @@ template<typename Derived> class DenseBase
     static const RandomReturnType Random();
 
     template<typename ThenDerived,typename ElseDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived,ElseDerived>
+    const Select<Derived,ThenDerived,ElseDerived>
     select(const DenseBase<ThenDerived>& thenMatrix,
            const DenseBase<ElseDerived>& elseMatrix) const;
 
     template<typename ThenDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
+    inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
     select(const DenseBase<ThenDerived>& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const;
 
     template<typename ElseDerived>
-    inline EIGEN_DEVICE_FUNC const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
+    inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
     select(const typename ElseDerived::Scalar& thenScalar, const DenseBase<ElseDerived>& elseMatrix) const;
 
     template<int p> RealScalar lpNorm() const;
@@ -613,59 +567,16 @@ template<typename Derived> class DenseBase
     }
     EIGEN_DEVICE_FUNC void reverseInPlace();
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    /** STL-like <a href="https://en.cppreference.com/w/cpp/named_req/RandomAccessIterator">RandomAccessIterator</a>
-      * iterator type as returned by the begin() and end() methods.
-      */
-    typedef random_access_iterator_type iterator;
-    /** This is the const version of iterator (aka read-only) */
-    typedef random_access_iterator_type const_iterator;
-    #else
-    typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
-                                            internal::pointer_based_stl_iterator<Derived>,
-                                            internal::generic_randaccess_stl_iterator<Derived>
-                                          >::type iterator_type;
-
-    typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit,
-                                            internal::pointer_based_stl_iterator<const Derived>,
-                                            internal::generic_randaccess_stl_iterator<const Derived>
-                                          >::type const_iterator_type;
-
-    // Stl-style iterators are supported only for vectors.
-
-    typedef typename internal::conditional< IsVectorAtCompileTime,
-                                            iterator_type,
-                                            void
-                                          >::type iterator;
-
-    typedef typename internal::conditional< IsVectorAtCompileTime,
-                                            const_iterator_type,
-                                            void
-                                          >::type const_iterator;
-    #endif
-
-    inline iterator begin();
-    inline const_iterator begin() const;
-    inline const_iterator cbegin() const;
-    inline iterator end();
-    inline const_iterator end() const;
-    inline const_iterator cend() const;
-
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
 #define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
 #define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
-#define EIGEN_DOC_UNARY_ADDONS(X,Y)
-#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/BlockMethods.h"
-#   include "../plugins/IndexedViewMethods.h"
-#   include "../plugins/ReshapedMethods.h"
 #   ifdef EIGEN_DENSEBASE_PLUGIN
 #     include EIGEN_DENSEBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
 #undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
 #undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
-#undef EIGEN_DOC_UNARY_ADDONS
 
     // disable the use of evalTo for dense objects with a nice compilation error
     template<typename Dest>

Eigen/src/Core/DenseCoeffsBase.h

@@ -22,8 +22,7 @@ template<typename T> struct add_const_on_value_type_if_arithmetic
 /** \brief Base class providing read-only coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
+  * \tparam #ReadOnlyAccessors Constant indicating read-only access
-  * \note #ReadOnlyAccessors Constant indicating read-only access
   *
   * This class defines the \c operator() \c const function and friends, which can be used to read specific
   * entries of a matrix or array.
@@ -289,8 +288,7 @@ class DenseCoeffsBase<Derived,ReadOnlyAccessors> : public EigenBase<Derived>
 /** \brief Base class providing read/write coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
+  * \tparam #WriteAccessors Constant indicating read/write access
-  * \note #WriteAccessors Constant indicating read/write access
   *
   * This class defines the non-const \c operator() function and friends, which can be used to write specific
   * entries of a matrix or array. This class inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which
@@ -468,8 +466,7 @@ class DenseCoeffsBase<Derived, WriteAccessors> : public DenseCoeffsBase<Derived,
 /** \brief Base class providing direct read-only coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
+  * \tparam #DirectAccessors Constant indicating direct access
-  * \note #DirectAccessors Constant indicating direct access
   *
   * This class defines functions to work with strides which can be used to access entries directly. This class
   * inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
@@ -495,7 +492,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -506,14 +503,14 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
     }
 
     // FIXME shall we remove it ?
-    EIGEN_CONSTEXPR inline Index stride() const
+    inline Index stride() const
     {
       return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
     }
@@ -522,7 +519,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -532,7 +529,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -542,8 +539,7 @@ class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived
 /** \brief Base class providing direct read/write coefficient access to matrices and arrays.
   * \ingroup Core_Module
   * \tparam Derived Type of the derived class
-  *
+  * \tparam #DirectWriteAccessors Constant indicating direct access
-  * \note #DirectWriteAccessors Constant indicating direct access
   *
   * This class defines functions to work with strides which can be used to access entries directly. This class
   * inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
@@ -570,8 +566,8 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa outerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT
+    inline Index innerStride() const
     {
       return derived().innerStride();
     }
@@ -581,14 +577,14 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), rowStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT
+    inline Index outerStride() const
     {
       return derived().outerStride();
     }
 
     // FIXME shall we remove it ?
-    EIGEN_CONSTEXPR inline Index stride() const EIGEN_NOEXCEPT
+    inline Index stride() const
     {
       return Derived::IsVectorAtCompileTime ? innerStride() : outerStride();
     }
@@ -597,8 +593,8 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), colStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rowStride() const EIGEN_NOEXCEPT
+    inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
     }
@@ -607,8 +603,8 @@ class DenseCoeffsBase<Derived, DirectWriteAccessors>
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index colStride() const EIGEN_NOEXCEPT
+    inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
     }
@@ -619,7 +615,7 @@ namespace internal {
 template<int Alignment, typename Derived, bool JustReturnZero>
 struct first_aligned_impl
 {
-  static EIGEN_CONSTEXPR inline Index run(const Derived&) EIGEN_NOEXCEPT
+  static inline Index run(const Derived&)
   { return 0; }
 };
 

Eigen/src/Core/DenseStorage.h

@@ -61,7 +61,7 @@ struct plain_array
 #if defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT)
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask)
 #elif EIGEN_GNUC_AT_LEAST(4,7) 
-  // GCC 4.7 is too aggressive in its optimizations and remove the alignment test based on the fact the array is declared to be aligned.
+  // GCC 4.7 is too aggressive in its optimizations and remove the alignement test based on the fact the array is declared to be aligned.
   // See this bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53900
   // Hiding the origin of the array pointer behind a function argument seems to do the trick even if the function is inlined:
   template<typename PtrType>
@@ -163,30 +163,6 @@ struct plain_array<T, 0, MatrixOrArrayOptions, Alignment>
   EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {}
 };
 
-struct plain_array_helper {
-  template<typename T, int Size, int MatrixOrArrayOptions, int Alignment>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  static void copy(const plain_array<T, Size, MatrixOrArrayOptions, Alignment>& src, const Eigen::Index size,
-                         plain_array<T, Size, MatrixOrArrayOptions, Alignment>& dst) {
-    smart_copy(src.array, src.array + size, dst.array);
-  }
-  
-  template<typename T, int Size, int MatrixOrArrayOptions, int Alignment>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  static void swap(plain_array<T, Size, MatrixOrArrayOptions, Alignment>& a, const Eigen::Index a_size,
-                   plain_array<T, Size, MatrixOrArrayOptions, Alignment>& b, const Eigen::Index b_size) {
-    if (a_size < b_size) {
-      std::swap_ranges(b.array, b.array + a_size, a.array);
-      smart_move(b.array + a_size, b.array + b_size, a.array + a_size);
-    } else if (a_size > b_size) {
-      std::swap_ranges(a.array, a.array + b_size, b.array);
-      smart_move(a.array + b_size, a.array + a_size, b.array + b_size);
-    } else {
-      std::swap_ranges(a.array, a.array + a_size, b.array);
-    }
-  }
-};
-
 } // end namespace internal
 
 /** \internal
@@ -214,41 +190,16 @@ template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseSt
     EIGEN_DEVICE_FUNC
     explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
-#if !EIGEN_HAS_CXX11 || defined(EIGEN_DENSE_STORAGE_CTOR_PLUGIN)
     EIGEN_DEVICE_FUNC 
     DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
       EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
     }
-#else
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) = default;
-#endif
-#if !EIGEN_HAS_CXX11
     EIGEN_DEVICE_FUNC 
     DenseStorage& operator=(const DenseStorage& other)
     { 
       if (this != &other) m_data = other.m_data;
       return *this; 
     }
-#else
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) = default;
-#endif
-#if EIGEN_HAS_RVALUE_REFERENCES
-#if !EIGEN_HAS_CXX11
-    EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
-      : m_data(std::move(other.m_data))
-    {
-    }
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
-    {
-      if (this != &other)
-        m_data = std::move(other.m_data);
-      return *this;
-    }
-#else
-    EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&&) = default;
-    EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&&) = default;
-#endif
-#endif
     EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
       EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
@@ -256,11 +207,9 @@ template<typename T, int Size, int _Rows, int _Cols, int _Options> class DenseSt
       EIGEN_UNUSED_VARIABLE(rows);
       EIGEN_UNUSED_VARIABLE(cols);
     }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
-      numext::swap(m_data, other.m_data);
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-    }
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;}
     EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
     EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
     EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
@@ -277,8 +226,8 @@ template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
     EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
     EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
     EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
     EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
     EIGEN_DEVICE_FUNC const T *data() const { return 0; }
@@ -305,28 +254,20 @@ template<typename T, int Size, int _Options> class DenseStorage<T, Size, Dynamic
     EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
     EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
-      : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows), m_cols(other.m_cols)
-    {
-      internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data);
-    }
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
     { 
       if (this != &other)
       {
+        m_data = other.m_data;
         m_rows = other.m_rows;
         m_cols = other.m_cols;
-        internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data);
       }
       return *this; 
     }
     EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
     EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
-    {
+    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-      internal::plain_array_helper::swap(m_data, m_rows * m_cols, other.m_data, other.m_rows * other.m_cols);
-      numext::swap(m_rows,other.m_rows);
-      numext::swap(m_cols,other.m_cols);
-    }
     EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
     EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
     EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
@@ -344,29 +285,20 @@ template<typename T, int Size, int _Cols, int _Options> class DenseStorage<T, Si
     EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
     EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
-      : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows)
-    {
-      internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data);
-    }
-    
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
     {
       if (this != &other)
       {
+        m_data = other.m_data;
         m_rows = other.m_rows;
-        internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data);
       }
       return *this; 
     }
     EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    { 
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-      internal::plain_array_helper::swap(m_data, m_rows * _Cols, other.m_data, other.m_rows * _Cols);
+    EIGEN_DEVICE_FUNC Index cols(void) const {return _Cols;}
-      numext::swap(m_rows, other.m_rows);
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols(void) const EIGEN_NOEXCEPT {return _Cols;}
     EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
     EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
     EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
@@ -382,29 +314,22 @@ template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Si
     EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
     EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {}
-    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) 
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
-      : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(other.m_cols)
-    {
-      internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data);
-    }
     EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
     {
       if (this != &other)
       {
+        m_data = other.m_data;
         m_cols = other.m_cols;
-        internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data);
       }
       return *this;
     }
     EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-      internal::plain_array_helper::swap(m_data, _Rows * m_cols, other.m_data, _Rows * other.m_cols);
+    EIGEN_DEVICE_FUNC Index rows(void) const {return _Rows;}
-      numext::swap(m_cols, other.m_cols);
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-    }
+    void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows(void) const EIGEN_NOEXCEPT {return _Rows;}
+    void resize(Index, Index, Index cols) { m_cols = cols; }
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
-    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
-    EIGEN_DEVICE_FUNC void resize(Index, Index, Index cols) { m_cols = cols; }
     EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
     EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
@@ -456,21 +381,18 @@ template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynam
     EIGEN_DEVICE_FUNC
     DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
-      numext::swap(m_data, other.m_data);
+      using std::swap;
-      numext::swap(m_rows, other.m_rows);
+      swap(m_data, other.m_data);
-      numext::swap(m_cols, other.m_cols);
+      swap(m_rows, other.m_rows);
+      swap(m_cols, other.m_cols);
       return *this;
     }
 #endif
     EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
     EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
-    {
+    { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-      numext::swap(m_data,other.m_data);
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-      numext::swap(m_rows,other.m_rows);
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-      numext::swap(m_cols,other.m_cols);
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
     void conservativeResize(Index size, Index rows, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
@@ -537,18 +459,16 @@ template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Ro
     EIGEN_DEVICE_FUNC
     DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
-      numext::swap(m_data, other.m_data);
+      using std::swap;
-      numext::swap(m_cols, other.m_cols);
+      swap(m_data, other.m_data);
+      swap(m_cols, other.m_cols);
       return *this;
     }
 #endif
     EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-      numext::swap(m_data,other.m_data);
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
-      numext::swap(m_cols,other.m_cols);
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
-    }
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;}
-    EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;}
     EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
@@ -613,18 +533,16 @@ template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dyn
     EIGEN_DEVICE_FUNC
     DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
     {
-      numext::swap(m_data, other.m_data);
+      using std::swap;
-      numext::swap(m_rows, other.m_rows);
+      swap(m_data, other.m_data);
+      swap(m_rows, other.m_rows);
       return *this;
     }
 #endif
     EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
-    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) {
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-      numext::swap(m_data,other.m_data);
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
-      numext::swap(m_rows,other.m_rows);
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
-    }
-    EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;}
-    EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) {return _Cols;}
     void conservativeResize(Index size, Index rows, Index)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);

Eigen/src/Core/Diagonal.h

@@ -84,16 +84,20 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
                                : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return 1; }
+    inline Index cols() const { return 1; }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT {
+    inline Index innerStride() const
+    {
       return m_matrix.outerStride() + 1;
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT { return 0; }
+    inline Index outerStride() const
+    {
+      return 0;
+    }
 
     typedef typename internal::conditional<
                        internal::is_lvalue<MatrixType>::value,
@@ -163,12 +167,12 @@ template<typename MatrixType, int _DiagIndex> class Diagonal
 
   private:
     // some compilers may fail to optimize std::max etc in case of compile-time constants...
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index absDiagIndex() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
+    EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index rowOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? 0 : -m_index.value(); }
+    EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index colOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : 0; }
+    EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; }
     // trigger a compile-time error if someone try to call packet
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
@@ -191,8 +195,7 @@ MatrixBase<Derived>::diagonal()
 
 /** This is the const version of diagonal(). */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalReturnType
-const typename MatrixBase<Derived>::ConstDiagonalReturnType
 MatrixBase<Derived>::diagonal() const
 {
   return ConstDiagonalReturnType(derived());
@@ -210,18 +213,18 @@ MatrixBase<Derived>::diagonal() const
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline Diagonal<Derived, DynamicIndex>
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::DiagonalDynamicIndexReturnType
 MatrixBase<Derived>::diagonal(Index index)
 {
-  return Diagonal<Derived, DynamicIndex>(derived(), index);
+  return DiagonalDynamicIndexReturnType(derived(), index);
 }
 
 /** This is the const version of diagonal(Index). */
 template<typename Derived>
-EIGEN_DEVICE_FUNC inline const Diagonal<const Derived, DynamicIndex>
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::ConstDiagonalDynamicIndexReturnType
 MatrixBase<Derived>::diagonal(Index index) const
 {
-  return Diagonal<const Derived, DynamicIndex>(derived(), index);
+  return ConstDiagonalDynamicIndexReturnType(derived(), index);
 }
 
 /** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this
@@ -237,21 +240,19 @@ MatrixBase<Derived>::diagonal(Index index) const
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
 template<int Index_>
-EIGEN_DEVICE_FUNC
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
-inline Diagonal<Derived, Index_>
 MatrixBase<Derived>::diagonal()
 {
-  return Diagonal<Derived, Index_>(derived());
+  return typename DiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 /** This is the const version of diagonal<int>(). */
 template<typename Derived>
 template<int Index_>
-EIGEN_DEVICE_FUNC
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
-inline const Diagonal<const Derived, Index_>
 MatrixBase<Derived>::diagonal() const
 {
-  return  Diagonal<const Derived, Index_>(derived());
+  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 } // end namespace Eigen

Eigen/src/Core/DiagonalMatrix.h

@@ -83,30 +83,6 @@ class DiagonalBase : public EigenBase<Derived>
     {
       return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
     }
-
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    inline unspecified_expression_type
-    #else
-    inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,sum) >
-    #endif
-    operator+(const DiagonalBase<OtherDerived>& other) const
-    {
-      return (diagonal() + other.diagonal()).asDiagonal();
-    }
-
-    template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    inline unspecified_expression_type
-    #else
-    inline const DiagonalWrapper<const EIGEN_CWISE_BINARY_RETURN_TYPE(DiagonalVectorType,typename OtherDerived::DiagonalVectorType,difference) >
-    #endif
-    operator-(const DiagonalBase<OtherDerived>& other) const
-    {
-      return (diagonal() - other.diagonal()).asDiagonal();
-    }
 };
 
 #endif
@@ -178,30 +154,6 @@ class DiagonalMatrix
     EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {}
 
-    #if EIGEN_HAS_CXX11
-    /** \brief Construct a diagonal matrix with fixed size from an arbitrary number of coefficients. \cpp11
-      * 
-      * There exists C++98 anologue constructors for fixed-size diagonal matrices having 2 or 3 coefficients.
-      * 
-      * \warning To construct a diagonal matrix of fixed size, the number of values passed to this 
-      * constructor must match the fixed dimension of \c *this.
-      * 
-      * \sa DiagonalMatrix(const Scalar&, const Scalar&)
-      * \sa DiagonalMatrix(const Scalar&, const Scalar&, const Scalar&)
-      */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    DiagonalMatrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const ArgTypes&... args)
-      : m_diagonal(a0, a1, a2, args...) {}
-
-    /** \brief Constructs a DiagonalMatrix and initializes it by elements given by an initializer list of initializer
-      * lists \cpp11
-      */
-    EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE DiagonalMatrix(const std::initializer_list<std::initializer_list<Scalar>>& list)
-      : m_diagonal(list) {}
-    #endif  // EIGEN_HAS_CXX11
-
     /** Copy constructor. */
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC
@@ -366,7 +318,7 @@ template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2De
 template< typename DstXprType, typename SrcXprType, typename Functor>
 struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
 {
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();

Eigen/src/Core/Dot.h

@@ -18,9 +18,14 @@ namespace internal {
 // with mismatched types, the compiler emits errors about failing to instantiate cwiseProduct BEFORE
 // looking at the static assertions. Thus this is a trick to get better compile errors.
 template<typename T, typename U,
-         bool NeedToTranspose = T::IsVectorAtCompileTime && U::IsVectorAtCompileTime &&
+// the NeedToTranspose condition here is taken straight from Assign.h
-                ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1) ||
+         bool NeedToTranspose = T::IsVectorAtCompileTime
-                 (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))>
+                && U::IsVectorAtCompileTime
+                && ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1)
+                      |  // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
+                         // revert to || as soon as not needed anymore.
+                    (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))
+>
 struct dot_nocheck
 {
   typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
@@ -202,7 +207,7 @@ struct lpNorm_selector
   EIGEN_DEVICE_FUNC
   static inline RealScalar run(const MatrixBase<Derived>& m)
   {
-    EIGEN_USING_STD(pow)
+    EIGEN_USING_STD_MATH(pow)
     return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };

Eigen/src/Core/EigenBase.h

@@ -32,9 +32,8 @@ template<typename Derived> struct EigenBase
   
   /** \brief The interface type of indices
     * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+    * \deprecated Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
     * \sa StorageIndex, \ref TopicPreprocessorDirectives.
-    * DEPRECATED: Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
-    * Deprecation is not marked with a doxygen comment because there are too many existing usages to add the deprecation attribute.
     */
   typedef Eigen::Index Index;
 
@@ -56,15 +55,15 @@ template<typename Derived> struct EigenBase
   { return *static_cast<const Derived*>(this); }
 
   /** \returns the number of rows. \sa cols(), RowsAtCompileTime */
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
-  inline Index rows() const EIGEN_NOEXCEPT { return derived().rows(); }
+  inline Index rows() const { return derived().rows(); }
   /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
-  inline Index cols() const EIGEN_NOEXCEPT { return derived().cols(); }
+  inline Index cols() const { return derived().cols(); }
   /** \returns the number of coefficients, which is rows()*cols().
     * \sa rows(), cols(), SizeAtCompileTime. */
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
-  inline Index size() const EIGEN_NOEXCEPT { return rows() * cols(); }
+  inline Index size() const { return rows() * cols(); }
 
   /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
   template<typename Dest>

Eigen/src/Core/ForceAlignedAccess.h

@@ -41,14 +41,10 @@ template<typename ExpressionType> class ForceAlignedAccess
 
     EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
-    inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); }
 
     EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {

Eigen/src/Core/GeneralProduct.h

@@ -18,16 +18,6 @@ enum {
   Small = 3
 };
 
-// Define the threshold value to fallback from the generic matrix-matrix product
-// implementation (heavy) to the lightweight coeff-based product one.
-// See generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
-// in products/GeneralMatrixMatrix.h for more details.
-// TODO This threshold should also be used in the compile-time selector below.
-#ifndef EIGEN_GEMM_TO_COEFFBASED_THRESHOLD
-// This default value has been obtained on a Haswell architecture.
-#define EIGEN_GEMM_TO_COEFFBASED_THRESHOLD 20
-#endif
-
 namespace internal {
 
 template<int Rows, int Cols, int Depth> struct product_type_selector;
@@ -35,7 +25,7 @@ template<int Rows, int Cols, int Depth> struct product_type_selector;
 template<int Size, int MaxSize> struct product_size_category
 {
   enum {
-    #ifndef EIGEN_GPU_COMPILE_PHASE
+    #ifndef EIGEN_CUDA_ARCH
     is_large = MaxSize == Dynamic ||
                Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
                (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
@@ -163,13 +153,13 @@ template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vect
 template<typename Scalar,int Size,int MaxSize>
 struct gemv_static_vector_if<Scalar,Size,MaxSize,false>
 {
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
+  EIGEN_STRONG_INLINE  Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; }
 };
 
 template<typename Scalar,int Size>
 struct gemv_static_vector_if<Scalar,Size,Dynamic,true>
 {
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { return 0; }
+  EIGEN_STRONG_INLINE Scalar* data() { return 0; }
 };
 
 template<typename Scalar,int Size,int MaxSize>
@@ -228,7 +218,8 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
     ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
     ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
 
-    ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
 
     // make sure Dest is a compile-time vector type (bug 1166)
     typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
@@ -238,7 +229,7 @@ template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
       // on, the other hand it is good for the cache to pack the vector anyways...
       EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = ((!EvalToDestAtCompileTime) || ComplexByReal) && (ActualDest::MaxSizeAtCompileTime!=0)
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
     };
 
     typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
@@ -319,12 +310,13 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
     typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
     typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
 
-    ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs);
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1 || ActualRhsTypeCleaned::MaxSizeAtCompileTime==0
+      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
     };
 
     gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
@@ -394,8 +386,7 @@ template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC inline const Product<Derived, OtherDerived>
-const Product<Derived, OtherDerived>
 MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
 {
   // A note regarding the function declaration: In MSVC, this function will sometimes
@@ -437,8 +428,7 @@ MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC const Product<Derived,OtherDerived,LazyProduct>
-const Product<Derived,OtherDerived,LazyProduct>
 MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
 {
   enum {

Eigen/src/Core/GenericPacketMath.h

@@ -44,29 +44,23 @@ struct default_packet_traits
   enum {
     HasHalfPacket = 0,
 
-    HasAdd       = 1,
+    HasAdd    = 1,
-    HasSub       = 1,
+    HasSub    = 1,
-    HasShift     = 1,
+    HasMul    = 1,
-    HasMul       = 1,
+    HasNegate = 1,
-    HasNegate    = 1,
+    HasAbs    = 1,
-    HasAbs       = 1,
+    HasArg    = 0,
-    HasArg       = 0,
+    HasAbs2   = 1,
-    HasAbs2      = 1,
+    HasMin    = 1,
-    HasAbsDiff   = 0,
+    HasMax    = 1,
-    HasMin       = 1,
+    HasConj   = 1,
-    HasMax       = 1,
-    HasConj      = 1,
     HasSetLinear = 1,
-    HasBlend     = 0,
+    HasBlend  = 0,
-    // This flag is used to indicate whether packet comparison is supported.
-    // pcmp_eq, pcmp_lt and pcmp_le should be defined for it to be true.
-    HasCmp       = 0,
 
     HasDiv    = 0,
     HasSqrt   = 0,
     HasRsqrt  = 0,
     HasExp    = 0,
-    HasExpm1  = 0,
     HasLog    = 0,
     HasLog1p  = 0,
     HasLog10  = 0,
@@ -87,18 +81,14 @@ struct default_packet_traits
     HasPolygamma = 0,
     HasErf = 0,
     HasErfc = 0,
-    HasNdtri = 0,
-    HasBessel = 0,
     HasIGamma = 0,
-    HasIGammaDerA = 0,
-    HasGammaSampleDerAlpha = 0,
     HasIGammac = 0,
     HasBetaInc = 0,
 
     HasRound  = 0,
-    HasRint   = 0,
     HasFloor  = 0,
     HasCeil   = 0,
+
     HasSign   = 0
   };
 };
@@ -129,22 +119,6 @@ template<typename T> struct packet_traits : default_packet_traits
 
 template<typename T> struct packet_traits<const T> : packet_traits<T> { };
 
-template<typename T> struct unpacket_traits
-{
-  typedef T type;
-  typedef T half;
-  enum
-  {
-    size = 1,
-    alignment = 1,
-    vectorizable = false,
-    masked_load_available=false,
-    masked_store_available=false
-  };
-};
-
-template<typename T> struct unpacket_traits<const T> : unpacket_traits<T> { };
-
 template <typename Src, typename Tgt> struct type_casting_traits {
   enum {
     VectorizedCast = 0,
@@ -153,34 +127,6 @@ template <typename Src, typename Tgt> struct type_casting_traits {
   };
 };
 
-/** \internal Wrapper to ensure that multiple packet types can map to the same
-    same underlying vector type. */
-template<typename T, int unique_id = 0>
-struct eigen_packet_wrapper
-{
-  EIGEN_ALWAYS_INLINE operator T&() { return m_val; }
-  EIGEN_ALWAYS_INLINE operator const T&() const { return m_val; }
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper() {};
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper(const T &v) : m_val(v) {}
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper& operator=(const T &v) {
-    m_val = v;
-    return *this;
-  }
-
-  T m_val;
-};
-
-
-/** \internal A convenience utility for determining if the type is a scalar.
- * This is used to enable some generic packet implementations.
- */
-template<typename Packet>
-struct is_scalar {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  enum {
-    value = internal::is_same<Packet, Scalar>::value
-  };
-};
 
 /** \internal \returns static_cast<TgtType>(a) (coeff-wise) */
 template <typename SrcPacket, typename TgtPacket>
@@ -193,406 +139,75 @@ EIGEN_DEVICE_FUNC inline TgtPacket
 pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
   return static_cast<TgtPacket>(a);
 }
+
 template <typename SrcPacket, typename TgtPacket>
 EIGEN_DEVICE_FUNC inline TgtPacket
 pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
   return static_cast<TgtPacket>(a);
 }
-template <typename SrcPacket, typename TgtPacket>
-EIGEN_DEVICE_FUNC inline TgtPacket
-pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/,
-      const SrcPacket& /*e*/, const SrcPacket& /*f*/, const SrcPacket& /*g*/, const SrcPacket& /*h*/) {
-  return static_cast<TgtPacket>(a);
-}
-
-/** \internal \returns reinterpret_cast<Target>(a) */
-template <typename Target, typename Packet>
-EIGEN_DEVICE_FUNC inline Target
-preinterpret(const Packet& a); /* { return reinterpret_cast<const Target&>(a); } */
 
 /** \internal \returns a + b (coeff-wise) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-padd(const Packet& a, const Packet& b) { return a+b; }
+padd(const Packet& a,
-// Avoid compiler warning for boolean algebra.
+        const Packet& b) { return a+b; }
-template<> EIGEN_DEVICE_FUNC inline bool
-padd(const bool& a, const bool& b) { return a || b; }
 
 /** \internal \returns a - b (coeff-wise) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-psub(const Packet& a, const Packet& b) { return a-b; }
+psub(const Packet& a,
+        const Packet& b) { return a-b; }
 
 /** \internal \returns -a (coeff-wise) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pnegate(const Packet& a) { return -a; }
 
-template<> EIGEN_DEVICE_FUNC inline bool
-pnegate(const bool& a) { return !a; }
-
 /** \internal \returns conj(a) (coeff-wise) */
+
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pconj(const Packet& a) { return numext::conj(a); }
 
 /** \internal \returns a * b (coeff-wise) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pmul(const Packet& a, const Packet& b) { return a*b; }
+pmul(const Packet& a,
-// Avoid compiler warning for boolean algebra.
+        const Packet& b) { return a*b; }
-template<> EIGEN_DEVICE_FUNC inline bool
-pmul(const bool& a, const bool& b) { return a && b; }
 
 /** \internal \returns a / b (coeff-wise) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pdiv(const Packet& a, const Packet& b) { return a/b; }
+pdiv(const Packet& a,
+        const Packet& b) { return a/b; }
 
-// In the generic case, memset to all one bits.
+/** \internal \returns the min of \a a and \a b  (coeff-wise) */
-template<typename Packet, typename EnableIf = void>
-struct ptrue_impl {
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& /*a*/){
-    Packet b;
-    memset(static_cast<void*>(&b), 0xff, sizeof(Packet));
-    return b;
-  }
-};
-
-// For non-trivial scalars, set to Scalar(1) (i.e. a non-zero value).
-// Although this is technically not a valid bitmask, the scalar path for pselect
-// uses a comparison to zero, so this should still work in most cases. We don't
-// have another option, since the scalar type requires initialization.
-template<typename T>
-struct ptrue_impl<T,
-    typename internal::enable_if<is_scalar<T>::value && NumTraits<T>::RequireInitialization>::type > {
-  static EIGEN_DEVICE_FUNC inline T run(const T& /*a*/){
-    return T(1);
-  }
-};
-
-/** \internal \returns one bits. */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-ptrue(const Packet& a) {
+pmin(const Packet& a,
-  return ptrue_impl<Packet>::run(a);
+        const Packet& b) { return numext::mini(a, b); }
-}
 
-// In the general case, memset to zero.
+/** \internal \returns the max of \a a and \a b  (coeff-wise) */
-template<typename Packet, typename EnableIf = void>
-struct pzero_impl {
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& /*a*/) {
-    Packet b;
-    memset(static_cast<void*>(&b), 0x00, sizeof(Packet));
-    return b;
-  }
-};
-
-// For scalars, explicitly set to Scalar(0), since the underlying representation
-// for zero may not consist of all-zero bits.
-template<typename T>
-struct pzero_impl<T,
-    typename internal::enable_if<is_scalar<T>::value>::type> {
-  static EIGEN_DEVICE_FUNC inline T run(const T& /*a*/) {
-    return T(0);
-  }
-};
-
-/** \internal \returns packet of zeros */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pzero(const Packet& a) {
+pmax(const Packet& a,
-  return pzero_impl<Packet>::run(a);
+        const Packet& b) { return numext::maxi(a, b); }
-}
-
-/** \internal \returns a <= b as a bit mask */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_le(const Packet& a, const Packet& b)  { return a<=b ? ptrue(a) : pzero(a); }
-
-/** \internal \returns a < b as a bit mask */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_lt(const Packet& a, const Packet& b)  { return a<b ? ptrue(a) : pzero(a); }
-
-/** \internal \returns a == b as a bit mask */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_eq(const Packet& a, const Packet& b) { return a==b ? ptrue(a) : pzero(a); }
-
-/** \internal \returns a < b or a==NaN or b==NaN as a bit mask */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pcmp_lt_or_nan(const Packet& a, const Packet& b) { return a>=b ? pzero(a) : ptrue(a); }
-
-template<typename T>
-struct bit_and {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const {
-    return a & b;
-  }
-};
-
-template<typename T>
-struct bit_or {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const {
-    return a | b;
-  }
-};
-
-template<typename T>
-struct bit_xor {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const {
-    return a ^ b;
-  }
-};
-
-template<typename T>
-struct bit_not {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a) const {
-    return ~a;
-  }
-};
-
-// Use operators &, |, ^, ~.
-template<typename T>
-struct operator_bitwise_helper {
-  EIGEN_DEVICE_FUNC static inline T bitwise_and(const T& a, const T& b) { return bit_and<T>()(a, b); }
-  EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) { return bit_or<T>()(a, b); }
-  EIGEN_DEVICE_FUNC static inline T bitwise_xor(const T& a, const T& b) { return bit_xor<T>()(a, b); }
-  EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) { return bit_not<T>()(a); }
-};
-
-// Apply binary operations byte-by-byte
-template<typename T>
-struct bytewise_bitwise_helper {
-  EIGEN_DEVICE_FUNC static inline T bitwise_and(const T& a, const T& b) {
-    return binary(a, b, bit_and<unsigned char>());
-  }
-  EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) {
-    return binary(a, b, bit_or<unsigned char>());
-   }
-  EIGEN_DEVICE_FUNC static inline T bitwise_xor(const T& a, const T& b) {
-    return binary(a, b, bit_xor<unsigned char>());
-  }
-  EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) {
-    return unary(a,bit_not<unsigned char>());
-   }
-
- private:
-  template<typename Op>
-  EIGEN_DEVICE_FUNC static inline T unary(const T& a, Op op) {
-    const unsigned char* a_ptr = reinterpret_cast<const unsigned char*>(&a);
-    T c;
-    unsigned char* c_ptr = reinterpret_cast<unsigned char*>(&c);
-    for (size_t i = 0; i < sizeof(T); ++i) {
-      *c_ptr++ = op(*a_ptr++);
-    }
-    return c;
-  }
-
-  template<typename Op>
-  EIGEN_DEVICE_FUNC static inline T binary(const T& a, const T& b, Op op) {
-    const unsigned char* a_ptr = reinterpret_cast<const unsigned char*>(&a);
-    const unsigned char* b_ptr = reinterpret_cast<const unsigned char*>(&b);
-    T c;
-    unsigned char* c_ptr = reinterpret_cast<unsigned char*>(&c);
-    for (size_t i = 0; i < sizeof(T); ++i) {
-      *c_ptr++ = op(*a_ptr++, *b_ptr++);
-    }
-    return c;
-  }
-};
-
-// In the general case, use byte-by-byte manipulation.
-template<typename T, typename EnableIf = void>
-struct bitwise_helper : public bytewise_bitwise_helper<T> {};
-
-// For integers or non-trivial scalars, use binary operators.
-template<typename T>
-struct bitwise_helper<T,
-  typename internal::enable_if<
-    is_scalar<T>::value && (NumTraits<T>::IsInteger || NumTraits<T>::RequireInitialization)>::type
-  > : public operator_bitwise_helper<T> {};
-
-/** \internal \returns the bitwise and of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pand(const Packet& a, const Packet& b) {
-  return bitwise_helper<Packet>::bitwise_and(a, b);
-}
-
-/** \internal \returns the bitwise or of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-por(const Packet& a, const Packet& b) {
-  return bitwise_helper<Packet>::bitwise_or(a, b);
-}
-
-/** \internal \returns the bitwise xor of \a a and \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pxor(const Packet& a, const Packet& b) {
-  return bitwise_helper<Packet>::bitwise_xor(a, b);
-}
-
-/** \internal \returns the bitwise not of \a a */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pnot(const Packet& a) {
-  return bitwise_helper<Packet>::bitwise_not(a);
-}
-
-/** \internal \returns the bitwise and of \a a and not \a b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pandnot(const Packet& a, const Packet& b) { return pand(a, pnot(b)); }
-
-// In the general case, use bitwise select.
-template<typename Packet, typename EnableIf = void>
-struct pselect_impl {
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& mask, const Packet& a, const Packet& b) {
-    return por(pand(a,mask),pandnot(b,mask));
-  }
-};
-
-// For scalars, use ternary select.
-template<typename Packet>
-struct pselect_impl<Packet,
-    typename internal::enable_if<is_scalar<Packet>::value>::type > {
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& mask, const Packet& a, const Packet& b) {
-    return numext::equal_strict(mask, Packet(0)) ? b : a;
-  }
-};
-
-/** \internal \returns \a or \b for each field in packet according to \mask */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pselect(const Packet& mask, const Packet& a, const Packet& b) {
-  return pselect_impl<Packet>::run(mask, a, b);
-}
-
-template<> EIGEN_DEVICE_FUNC inline bool pselect<bool>(
-    const bool& cond, const bool& a, const bool& b) {
-  return cond ? a : b;
-}
-
-/** \internal \returns the min or of \a a and \a b (coeff-wise)
-    If either \a a or \a b are NaN, the result is implementation defined. */
-template<int NaNPropagation>
-struct pminmax_impl {
-  template <typename Packet, typename Op>
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) {
-    return op(a,b);
-  }
-};
-
-/** \internal \returns the min or max of \a a and \a b (coeff-wise)
-    If either \a a or \a b are NaN, NaN is returned. */
-template<>
-struct pminmax_impl<PropagateNaN> {
-  template <typename Packet, typename Op>
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) {
-  Packet not_nan_mask_a = pcmp_eq(a, a);
-  Packet not_nan_mask_b = pcmp_eq(b, b);
-  return pselect(not_nan_mask_a,
-                 pselect(not_nan_mask_b, op(a, b), b),
-                 a);
-  }
-};
-
-/** \internal \returns the min or max of \a a and \a b (coeff-wise)
-    If both \a a and \a b are NaN, NaN is returned.
-    Equivalent to std::fmin(a, b).  */
-template<>
-struct pminmax_impl<PropagateNumbers> {
-  template <typename Packet, typename Op>
-  static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) {
-  Packet not_nan_mask_a = pcmp_eq(a, a);
-  Packet not_nan_mask_b = pcmp_eq(b, b);
-  return pselect(not_nan_mask_a,
-                 pselect(not_nan_mask_b, op(a, b), a),
-                 b);
-  }
-};
-
-
-#ifndef SYCL_DEVICE_ONLY
-#define EIGEN_BINARY_OP_NAN_PROPAGATION(Type, Func) Func
-#else
-#define EIGEN_BINARY_OP_NAN_PROPAGATION(Type, Func) \
-[](const Type& a, const Type& b) { \
-        return Func(a, b);}
-#endif
-
-/** \internal \returns the min of \a a and \a b  (coeff-wise).
-    If \a a or \b b is NaN, the return value is implementation defined. */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pmin(const Packet& a, const Packet& b) { return numext::mini(a,b); }
-
-/** \internal \returns the min of \a a and \a b  (coeff-wise).
-    NaNPropagation determines the NaN propagation semantics. */
-template <int NaNPropagation, typename Packet>
-EIGEN_DEVICE_FUNC inline Packet pmin(const Packet& a, const Packet& b) {
-  return pminmax_impl<NaNPropagation>::run(a, b, EIGEN_BINARY_OP_NAN_PROPAGATION(Packet, (pmin<Packet>)));
-}
-
-/** \internal \returns the max of \a a and \a b  (coeff-wise)
-    If \a a or \b b is NaN, the return value is implementation defined. */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pmax(const Packet& a, const Packet& b) { return numext::maxi(a, b); }
-
-/** \internal \returns the max of \a a and \a b  (coeff-wise).
-    NaNPropagation determines the NaN propagation semantics. */
-template <int NaNPropagation, typename Packet>
-EIGEN_DEVICE_FUNC inline Packet pmax(const Packet& a, const Packet& b) {
-  return pminmax_impl<NaNPropagation>::run(a, b, EIGEN_BINARY_OP_NAN_PROPAGATION(Packet,(pmax<Packet>)));
-}
 
 /** \internal \returns the absolute value of \a a */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pabs(const Packet& a) { return numext::abs(a); }
+pabs(const Packet& a) { using std::abs; return abs(a); }
-template<> EIGEN_DEVICE_FUNC inline unsigned int
-pabs(const unsigned int& a) { return a; }
-template<> EIGEN_DEVICE_FUNC inline unsigned long
-pabs(const unsigned long& a) { return a; }
-template<> EIGEN_DEVICE_FUNC inline unsigned long long
-pabs(const unsigned long long& a) { return a; }
-
-/** \internal \returns the addsub value of \a a,b */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-paddsub(const Packet& a, const Packet& b) {
-  return pselect(peven_mask(a), padd(a, b), psub(a, b));
- }
 
 /** \internal \returns the phase angle of \a a */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 parg(const Packet& a) { using numext::arg; return arg(a); }
 
-
+/** \internal \returns the bitwise and of \a a and \a b */
-/** \internal \returns \a a logically shifted by N bits to the right */
-template<int N> EIGEN_DEVICE_FUNC inline int
-parithmetic_shift_right(const int& a) { return a >> N; }
-template<int N> EIGEN_DEVICE_FUNC inline long int
-parithmetic_shift_right(const long int& a) { return a >> N; }
-
-/** \internal \returns \a a arithmetically shifted by N bits to the right */
-template<int N> EIGEN_DEVICE_FUNC inline int
-plogical_shift_right(const int& a) { return static_cast<int>(static_cast<unsigned int>(a) >> N); }
-template<int N> EIGEN_DEVICE_FUNC inline long int
-plogical_shift_right(const long int& a) { return static_cast<long>(static_cast<unsigned long>(a) >> N); }
-
-/** \internal \returns \a a shifted by N bits to the left */
-template<int N> EIGEN_DEVICE_FUNC inline int
-plogical_shift_left(const int& a) { return a << N; }
-template<int N> EIGEN_DEVICE_FUNC inline long int
-plogical_shift_left(const long int& a) { return a << N; }
-
-/** \internal \returns the significant and exponent of the underlying floating point numbers
-  * See https://en.cppreference.com/w/cpp/numeric/math/frexp
-  */
-template <typename Packet>
-EIGEN_DEVICE_FUNC inline Packet pfrexp(const Packet& a, Packet& exponent) {
-  int exp;
-  EIGEN_USING_STD(frexp);
-  Packet result = static_cast<Packet>(frexp(a, &exp));
-  exponent = static_cast<Packet>(exp);
-  return result;
-}
-
-/** \internal \returns a * 2^((int)exponent)
-  * See https://en.cppreference.com/w/cpp/numeric/math/ldexp
-  */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pldexp(const Packet &a, const Packet &exponent) {
+pand(const Packet& a, const Packet& b) { return a & b; }
-  EIGEN_USING_STD(ldexp)
-  return static_cast<Packet>(ldexp(a, static_cast<int>(exponent)));
-}
 
-/** \internal \returns the min of \a a and \a b  (coeff-wise) */
+/** \internal \returns the bitwise or of \a a and \a b */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-pabsdiff(const Packet& a, const Packet& b) { return pselect(pcmp_lt(a, b), psub(b, a), psub(a, b)); }
+por(const Packet& a, const Packet& b) { return a | b; }
+
+/** \internal \returns the bitwise xor of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pxor(const Packet& a, const Packet& b) { return a ^ b; }
+
+/** \internal \returns the bitwise andnot of \a a and \a b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pandnot(const Packet& a, const Packet& b) { return a & (!b); }
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -602,22 +217,10 @@ pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
-/** \internal \returns a packet version of \a *from, (un-aligned masked load)
- * There is no generic implementation. We only have implementations for specialized
- * cases. Generic case should not be called.
- */
-template<typename Packet> EIGEN_DEVICE_FUNC inline
-typename enable_if<unpacket_traits<Packet>::masked_load_available, Packet>::type
-ploadu(const typename unpacket_traits<Packet>::type* from, typename unpacket_traits<Packet>::mask_t umask);
-
 /** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
 
-/** \internal \returns a packet with constant coefficients set from bits */
-template<typename Packet,typename BitsType> EIGEN_DEVICE_FUNC inline Packet
-pset1frombits(BitsType a);
-
 /** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(*a); }
@@ -678,20 +281,6 @@ inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
 template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
 plset(const typename unpacket_traits<Packet>::type& a) { return a; }
 
-/** \internal \returns a packet with constant coefficients \a a, e.g.: (x, 0, x, 0),
-     where x is the value of all 1-bits. */
-template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
-peven_mask(const Packet& /*a*/) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  const size_t n = unpacket_traits<Packet>::size;
-  EIGEN_ALIGN_TO_BOUNDARY(sizeof(Packet)) Scalar elements[n];
-  for(size_t i = 0; i < n; ++i) {
-    memset(elements+i, ((i & 1) == 0 ? 0xff : 0), sizeof(Scalar));
-  }
-  return ploadu<Packet>(elements);
-}
-
-
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
 { (*to) = from; }
@@ -700,15 +289,6 @@ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(
 template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
 {  (*to) = from; }
 
-/** \internal copy the packet \a from to \a *to, (un-aligned store with a mask)
- * There is no generic implementation. We only have implementations for specialized
- * cases. Generic case should not be called.
- */
-template<typename Scalar, typename Packet>
-EIGEN_DEVICE_FUNC inline
-typename enable_if<unpacket_traits<Packet>::masked_store_available, void>::type
-pstoreu(Scalar* to, const Packet& from, typename unpacket_traits<Packet>::mask_t umask);
-
  template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/)
  { return ploadu<Packet>(from); }
 
@@ -718,9 +298,7 @@ pstoreu(Scalar* to, const Packet& from, typename unpacket_traits<Packet>::mask_t
 /** \internal tries to do cache prefetching of \a addr */
 template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr)
 {
-#if defined(EIGEN_HIP_DEVICE_COMPILE)
+#ifdef __CUDA_ARCH__
-  // do nothing
-#elif defined(EIGEN_CUDA_ARCH)
 #if defined(__LP64__) || EIGEN_OS_WIN64
   // 64-bit pointer operand constraint for inlined asm
   asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr));
@@ -733,6 +311,39 @@ template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* a
 #endif
 }
 
+/** \internal \returns the first element of a packet */
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
+{ return a; }
+
+/** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+preduxp(const Packet* vecs) { return vecs[0]; }
+
+/** \internal \returns the sum of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
+{ return a; }
+
+/** \internal \returns the sum of the elements of \a a by block of 4 elements.
+  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
+  * For packet-size smaller or equal to 4, this boils down to a noop.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline
+typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
+predux_downto4(const Packet& a)
+{ return a; }
+
+/** \internal \returns the product of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
+{ return a; }
+
+/** \internal \returns the min of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
+{ return a; }
+
+/** \internal \returns the max of the elements of \a a*/
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
+{ return a; }
+
 /** \internal \returns the reversed elements of \a a*/
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
 { return a; }
@@ -740,7 +351,7 @@ template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet&
 /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
 template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
 {
-  return Packet(numext::imag(a),numext::real(a));
+  return Packet(a.imag(),a.real());
 }
 
 /**************************
@@ -748,202 +359,84 @@ template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet
 ***************************/
 
 /** \internal \returns the sine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet psin(const Packet& a) { EIGEN_USING_STD(sin); return sin(a); }
+Packet psin(const Packet& a) { using std::sin; return sin(a); }
 
 /** \internal \returns the cosine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pcos(const Packet& a) { EIGEN_USING_STD(cos); return cos(a); }
+Packet pcos(const Packet& a) { using std::cos; return cos(a); }
 
 /** \internal \returns the tan of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet ptan(const Packet& a) { EIGEN_USING_STD(tan); return tan(a); }
+Packet ptan(const Packet& a) { using std::tan; return tan(a); }
 
 /** \internal \returns the arc sine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pasin(const Packet& a) { EIGEN_USING_STD(asin); return asin(a); }
+Packet pasin(const Packet& a) { using std::asin; return asin(a); }
 
 /** \internal \returns the arc cosine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pacos(const Packet& a) { EIGEN_USING_STD(acos); return acos(a); }
+Packet pacos(const Packet& a) { using std::acos; return acos(a); }
 
 /** \internal \returns the arc tangent of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet patan(const Packet& a) { EIGEN_USING_STD(atan); return atan(a); }
+Packet patan(const Packet& a) { using std::atan; return atan(a); }
 
 /** \internal \returns the hyperbolic sine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet psinh(const Packet& a) { EIGEN_USING_STD(sinh); return sinh(a); }
+Packet psinh(const Packet& a) { using std::sinh; return sinh(a); }
 
 /** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pcosh(const Packet& a) { EIGEN_USING_STD(cosh); return cosh(a); }
+Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); }
 
 /** \internal \returns the hyperbolic tan of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet ptanh(const Packet& a) { EIGEN_USING_STD(tanh); return tanh(a); }
+Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); }
 
 /** \internal \returns the exp of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pexp(const Packet& a) { EIGEN_USING_STD(exp); return exp(a); }
+Packet pexp(const Packet& a) { using std::exp; return exp(a); }
-
-/** \internal \returns the expm1 of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet pexpm1(const Packet& a) { return numext::expm1(a); }
 
 /** \internal \returns the log of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet plog(const Packet& a) { EIGEN_USING_STD(log); return log(a); }
+Packet plog(const Packet& a) { using std::log; return log(a); }
 
 /** \internal \returns the log1p of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet plog1p(const Packet& a) { return numext::log1p(a); }
 
 /** \internal \returns the log10 of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet plog10(const Packet& a) { EIGEN_USING_STD(log10); return log10(a); }
+Packet plog10(const Packet& a) { using std::log10; return log10(a); }
-
-/** \internal \returns the log10 of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet plog2(const Packet& a) {
-  typedef typename internal::unpacket_traits<Packet>::type Scalar;
-  return pmul(pset1<Packet>(Scalar(EIGEN_LOG2E)), plog(a));
-}
 
 /** \internal \returns the square-root of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet psqrt(const Packet& a) { return numext::sqrt(a); }
+Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); }
 
 /** \internal \returns the reciprocal square-root of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet prsqrt(const Packet& a) {
-  typedef typename internal::unpacket_traits<Packet>::type Scalar;
+  return pdiv(pset1<Packet>(1), psqrt(a));
-  return pdiv(pset1<Packet>(Scalar(1)), psqrt(a));
 }
 
 /** \internal \returns the rounded value of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pround(const Packet& a) { using numext::round; return round(a); }
 
 /** \internal \returns the floor of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pfloor(const Packet& a) { using numext::floor; return floor(a); }
 
-/** \internal \returns the rounded value of \a a (coeff-wise) with current
- * rounding mode */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
-Packet print(const Packet& a) { using numext::rint; return rint(a); }
-
 /** \internal \returns the ceil of \a a (coeff-wise) */
-template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); }
 
-/** \internal \returns the first element of a packet */
-template<typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type
-pfirst(const Packet& a)
-{ return a; }
-
-/** \internal \returns the sum of the elements of upper and lower half of \a a if \a a is larger than 4.
-  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
-  * For packet-size smaller or equal to 4, this boils down to a noop.
-  */
-template<typename Packet>
-EIGEN_DEVICE_FUNC inline typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
-predux_half_dowto4(const Packet& a)
-{ return a; }
-
-// Slow generic implementation of Packet reduction.
-template <typename Packet, typename Op>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type
-predux_helper(const Packet& a, Op op) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  const size_t n = unpacket_traits<Packet>::size;
-  EIGEN_ALIGN_TO_BOUNDARY(sizeof(Packet)) Scalar elements[n];
-  pstoreu<Scalar>(elements, a);
-  for(size_t k = n / 2; k > 0; k /= 2)  {
-    for(size_t i = 0; i < k; ++i) {
-      elements[i] = op(elements[i], elements[i + k]);
-    }
-  }
-  return elements[0];
-}
-
-/** \internal \returns the sum of the elements of \a a*/
-template<typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type
-predux(const Packet& a)
-{
-  return a;
-}
-
-/** \internal \returns the product of the elements of \a a */
-template <typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(
-    const Packet& a) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmul<Scalar>)));
-}
-
-/** \internal \returns the min of the elements of \a a */
-template <typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(
-    const Packet &a) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin<PropagateFast, Scalar>)));
-}
-
-template <int NaNPropagation, typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(
-    const Packet& a) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin<NaNPropagation, Scalar>)));
-}
-
-/** \internal \returns the min of the elements of \a a */
-template <typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(
-    const Packet &a) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax<PropagateFast, Scalar>)));
-}
-
-template <int NaNPropagation, typename Packet>
-EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(
-    const Packet& a) {
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax<NaNPropagation, Scalar>)));
-}
-
-#undef EIGEN_BINARY_OP_NAN_PROPAGATION
-
-/** \internal \returns true if all coeffs of \a a means "true"
-  * It is supposed to be called on values returned by pcmp_*.
-  */
-// not needed yet
-// template<typename Packet> EIGEN_DEVICE_FUNC inline bool predux_all(const Packet& a)
-// { return bool(a); }
-
-/** \internal \returns true if any coeffs of \a a means "true"
-  * It is supposed to be called on values returned by pcmp_*.
-  */
-template<typename Packet> EIGEN_DEVICE_FUNC inline bool predux_any(const Packet& a)
-{
-  // Dirty but generic implementation where "true" is assumed to be non 0 and all the sames.
-  // It is expected that "true" is either:
-  //  - Scalar(1)
-  //  - bits full of ones (NaN for floats),
-  //  - or first bit equals to 1 (1 for ints, smallest denormal for floats).
-  // For all these cases, taking the sum is just fine, and this boils down to a no-op for scalars.
-  typedef typename unpacket_traits<Packet>::type Scalar;
-  return numext::not_equal_strict(predux(a), Scalar(0));
-}
-
 /***************************************************************************
 * The following functions might not have to be overwritten for vectorized types
 ***************************************************************************/
 
-/** \internal copy a packet with constant coefficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */
+/** \internal copy a packet with constant coeficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */
 // NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type)
 template<typename Packet>
 inline void pstore1(typename unpacket_traits<Packet>::type* to, const typename unpacket_traits<Packet>::type& a)
@@ -991,12 +484,41 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_t
   return ploadt<Packet, LoadMode>(from);
 }
 
+/** \internal default implementation of palign() allowing partial specialization */
+template<int Offset,typename PacketType>
+struct palign_impl
+{
+  // by default data are aligned, so there is nothing to be done :)
+  static inline void run(PacketType&, const PacketType&) {}
+};
+
+/** \internal update \a first using the concatenation of the packet_size minus \a Offset last elements
+  * of \a first and \a Offset first elements of \a second.
+  *
+  * This function is currently only used to optimize matrix-vector products on unligned matrices.
+  * It takes 2 packets that represent a contiguous memory array, and returns a packet starting
+  * at the position \a Offset. For instance, for packets of 4 elements, we have:
+  *  Input:
+  *  - first = {f0,f1,f2,f3}
+  *  - second = {s0,s1,s2,s3}
+  * Output:
+  *   - if Offset==0 then {f0,f1,f2,f3}
+  *   - if Offset==1 then {f1,f2,f3,s0}
+  *   - if Offset==2 then {f2,f3,s0,s1}
+  *   - if Offset==3 then {f3,s0,s1,s3}
+  */
+template<int Offset,typename PacketType>
+inline void palign(PacketType& first, const PacketType& second)
+{
+  palign_impl<Offset,PacketType>::run(first,second);
+}
+
 /***************************************************************************
 * Fast complex products (GCC generates a function call which is very slow)
 ***************************************************************************/
 
 // Eigen+CUDA does not support complexes.
-#if !defined(EIGEN_GPUCC)
+#ifndef EIGEN_CUDACC
 
 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
 { return std::complex<float>(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); }
@@ -1033,6 +555,34 @@ pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& th
   return ifPacket.select[0] ? thenPacket : elsePacket;
 }
 
+/** \internal \returns \a a with the first coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  mask.select[0] = true;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=1; i<unpacket_traits<Packet>::size; ++i)
+    mask.select[i] = false;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
+/** \internal \returns \a a with the last coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i)
+    mask.select[i] = false;
+  mask.select[unpacket_traits<Packet>::size-1] = true;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen

Eigen/src/Core/GlobalFunctions.h

@@ -66,31 +66,21 @@ namespace Eigen
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
-#if EIGEN_HAS_CXX11_MATH
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asinh,scalar_asinh_op,inverse hyperbolic sine,\sa ArrayBase::asinh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acosh,scalar_acosh_op,inverse hyperbolic cosine,\sa ArrayBase::acosh)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atanh,scalar_atanh_op,inverse hyperbolic tangent,\sa ArrayBase::atanh)
-#endif
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(logistic,scalar_logistic_op,logistic function,\sa ArrayBase::logistic)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ndtri,scalar_ndtri_op,inverse normal distribution function,\sa ArrayBase::ndtri)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(expm1,scalar_expm1_op,exponential of a value minus 1,\sa ArrayBase::expm1)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log10)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log2,scalar_log2_op,base 2 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log2)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg DOXCOMMA MatrixBase::cwiseArg)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rint,scalar_rint_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
   EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
@@ -112,18 +102,17 @@ namespace Eigen
   inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
   pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
 #else
-  template <typename Derived,typename ScalarExponent>
+  template<typename Derived,typename ScalarExponent>
-  EIGEN_DEVICE_FUNC inline
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,ScalarExponent>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent),
-  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
+          const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,ScalarExponent,pow) >::type
-    const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<typename Derived::Scalar
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent) {
-                                                 EIGEN_COMMA ScalarExponent EIGEN_COMMA
+    return x.derived().pow(exponent);
-                                                 EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type,pow))
+  }
-  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent)
+
-  {
+  template<typename Derived>
-    typedef typename internal::promote_scalar_arg<typename Derived::Scalar,ScalarExponent,
+  inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename Derived::Scalar,pow)
-                                                  EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent)>::type PromotedExponent;
+  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
-    return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedExponent,pow)(x.derived(),
+    return x.derived().pow(exponent);
-           typename internal::plain_constant_type<Derived,PromotedExponent>::type(x.derived().rows(), x.derived().cols(), internal::scalar_constant_op<PromotedExponent>(exponent)));
   }
 #endif
 
@@ -166,17 +155,21 @@ namespace Eigen
   inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
   pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
 #else
-  template <typename Scalar, typename Derived>
+  template<typename Scalar, typename Derived>
-  EIGEN_DEVICE_FUNC inline
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,Scalar>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar),
-  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(
+          const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow) >::type
-    const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<typename Derived::Scalar
+  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
-                                                 EIGEN_COMMA Scalar EIGEN_COMMA
+  {
-                                                 EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type,Derived,pow))
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow)(
-  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents) {
+            typename internal::plain_constant_type<Derived,Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
-    typedef typename internal::promote_scalar_arg<typename Derived::Scalar,Scalar,
+  }
-                                                  EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar)>::type PromotedScalar;
+
-    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedScalar,Derived,pow)(
+  template<typename Derived>
-           typename internal::plain_constant_type<Derived,PromotedScalar>::type(exponents.derived().rows(), exponents.derived().cols(), internal::scalar_constant_op<PromotedScalar>(x)), exponents.derived());
+  inline const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)
+  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  {
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)(
+      typename internal::plain_constant_type<Derived,typename Derived::Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
   }
 #endif
 

Eigen/src/Core/IO.h

@@ -41,7 +41,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   *  - \b rowSuffix string printed at the end of each row
   *  - \b matPrefix string printed at the beginning of the matrix
   *  - \b matSuffix string printed at the end of the matrix
-  *  - \b fill character printed to fill the empty space in aligned columns
   *
   * Example: \include IOFormat.cpp
   * Output: \verbinclude IOFormat.out
@@ -54,9 +53,9 @@ struct IOFormat
   IOFormat(int _precision = StreamPrecision, int _flags = 0,
     const std::string& _coeffSeparator = " ",
     const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="",
-    const std::string& _matPrefix="", const std::string& _matSuffix="", const char _fill=' ')
+    const std::string& _matPrefix="", const std::string& _matSuffix="")
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
-    rowSpacer(""), coeffSeparator(_coeffSeparator), fill(_fill), precision(_precision), flags(_flags)
+    rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
   {
     // TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline
     // don't add rowSpacer if columns are not to be aligned
@@ -72,7 +71,6 @@ struct IOFormat
   std::string matPrefix, matSuffix;
   std::string rowPrefix, rowSuffix, rowSeparator, rowSpacer;
   std::string coeffSeparator;
-  char fill;
   int precision;
   int flags;
 };
@@ -130,9 +128,6 @@ struct significant_decimals_impl
 template<typename Derived>
 std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt)
 {
-  using internal::is_same;
-  using internal::conditional;
-
   if(_m.size() == 0)
   {
     s << fmt.matPrefix << fmt.matSuffix;
@@ -141,22 +136,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   
   typename Derived::Nested m = _m;
   typedef typename Derived::Scalar Scalar;
-  typedef typename
-      conditional<
-          is_same<Scalar, char>::value ||
-            is_same<Scalar, unsigned char>::value ||
-            is_same<Scalar, numext::int8_t>::value ||
-            is_same<Scalar, numext::uint8_t>::value,
-          int,
-          typename conditional<
-              is_same<Scalar, std::complex<char> >::value ||
-                is_same<Scalar, std::complex<unsigned char> >::value ||
-                is_same<Scalar, std::complex<numext::int8_t> >::value ||
-                is_same<Scalar, std::complex<numext::uint8_t> >::value,
-              std::complex<int>,
-              const Scalar&
-            >::type
-        >::type PrintType;
 
   Index width = 0;
 
@@ -193,31 +172,23 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
       {
         std::stringstream sstr;
         sstr.copyfmt(s);
-        sstr << static_cast<PrintType>(m.coeff(i,j));
+        sstr << m.coeff(i,j);
         width = std::max<Index>(width, Index(sstr.str().length()));
       }
   }
-  std::streamsize old_width = s.width();
-  char old_fill_character = s.fill();
   s << fmt.matPrefix;
   for(Index i = 0; i < m.rows(); ++i)
   {
     if (i)
       s << fmt.rowSpacer;
     s << fmt.rowPrefix;
-    if(width) {
+    if(width) s.width(width);
-      s.fill(fmt.fill);
+    s << m.coeff(i, 0);
-      s.width(width);
-    }
-    s << static_cast<PrintType>(m.coeff(i, 0));
     for(Index j = 1; j < m.cols(); ++j)
     {
       s << fmt.coeffSeparator;
-      if(width) {
+      if (width) s.width(width);
-        s.fill(fmt.fill);
+      s << m.coeff(i, j);
-        s.width(width);
-      }
-      s << static_cast<PrintType>(m.coeff(i, j));
     }
     s << fmt.rowSuffix;
     if( i < m.rows() - 1)
@@ -225,10 +196,6 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat&
   }
   s << fmt.matSuffix;
   if(explicit_precision) s.precision(old_precision);
-  if(width) {
-    s.fill(old_fill_character);
-    s.width(old_width);
-  }
   return s;
 }
 

Eigen/src/Core/IndexedView.h

@@ -1,247 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_INDEXED_VIEW_H
-#define EIGEN_INDEXED_VIEW_H
-
-namespace Eigen {
-
-namespace internal {
-
-template<typename XprType, typename RowIndices, typename ColIndices>
-struct traits<IndexedView<XprType, RowIndices, ColIndices> >
- : traits<XprType>
-{
-  enum {
-    RowsAtCompileTime = int(array_size<RowIndices>::value),
-    ColsAtCompileTime = int(array_size<ColIndices>::value),
-    MaxRowsAtCompileTime = RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime) : Dynamic,
-    MaxColsAtCompileTime = ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime) : Dynamic,
-
-    XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
-    IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
-               : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
-               : XprTypeIsRowMajor,
-
-    RowIncr = int(get_compile_time_incr<RowIndices>::value),
-    ColIncr = int(get_compile_time_incr<ColIndices>::value),
-    InnerIncr = IsRowMajor ? ColIncr : RowIncr,
-    OuterIncr = IsRowMajor ? RowIncr : ColIncr,
-
-    HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
-    XprInnerStride = HasSameStorageOrderAsXprType ? int(inner_stride_at_compile_time<XprType>::ret) : int(outer_stride_at_compile_time<XprType>::ret),
-    XprOuterstride = HasSameStorageOrderAsXprType ? int(outer_stride_at_compile_time<XprType>::ret) : int(inner_stride_at_compile_time<XprType>::ret),
-
-    InnerSize = XprTypeIsRowMajor ? ColsAtCompileTime : RowsAtCompileTime,
-    IsBlockAlike = InnerIncr==1 && OuterIncr==1,
-    IsInnerPannel = HasSameStorageOrderAsXprType && is_same<AllRange<InnerSize>,typename conditional<XprTypeIsRowMajor,ColIndices,RowIndices>::type>::value,
-
-    InnerStrideAtCompileTime = InnerIncr<0 || InnerIncr==DynamicIndex || XprInnerStride==Dynamic ? Dynamic : XprInnerStride * InnerIncr,
-    OuterStrideAtCompileTime = OuterIncr<0 || OuterIncr==DynamicIndex || XprOuterstride==Dynamic ? Dynamic : XprOuterstride * OuterIncr,
-
-    ReturnAsScalar = is_same<RowIndices,SingleRange>::value && is_same<ColIndices,SingleRange>::value,
-    ReturnAsBlock = (!ReturnAsScalar) && IsBlockAlike,
-    ReturnAsIndexedView = (!ReturnAsScalar) && (!ReturnAsBlock),
-
-    // FIXME we deal with compile-time strides if and only if we have DirectAccessBit flag,
-    // but this is too strict regarding negative strides...
-    DirectAccessMask = (int(InnerIncr)!=UndefinedIncr && int(OuterIncr)!=UndefinedIncr && InnerIncr>=0 && OuterIncr>=0) ? DirectAccessBit : 0,
-    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
-    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-    Flags = (traits<XprType>::Flags & (HereditaryBits | DirectAccessMask )) | FlagsLvalueBit | FlagsRowMajorBit | FlagsLinearAccessBit
-  };
-
-  typedef Block<XprType,RowsAtCompileTime,ColsAtCompileTime,IsInnerPannel> BlockType;
-};
-
-}
-
-template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
-class IndexedViewImpl;
-
-
-/** \class IndexedView
-  * \ingroup Core_Module
-  *
-  * \brief Expression of a non-sequential sub-matrix defined by arbitrary sequences of row and column indices
-  *
-  * \tparam XprType the type of the expression in which we are taking the intersections of sub-rows and sub-columns
-  * \tparam RowIndices the type of the object defining the sequence of row indices
-  * \tparam ColIndices the type of the object defining the sequence of column indices
-  *
-  * This class represents an expression of a sub-matrix (or sub-vector) defined as the intersection
-  * of sub-sets of rows and columns, that are themself defined by generic sequences of row indices \f$ \{r_0,r_1,..r_{m-1}\} \f$
-  * and column indices \f$ \{c_0,c_1,..c_{n-1} \}\f$. Let \f$ A \f$  be the nested matrix, then the resulting matrix \f$ B \f$ has \c m
-  * rows and \c n columns, and its entries are given by: \f$ B(i,j) = A(r_i,c_j) \f$.
-  *
-  * The \c RowIndices and \c ColIndices types must be compatible with the following API:
-  * \code
-  * <integral type> operator[](Index) const;
-  * Index size() const;
-  * \endcode
-  *
-  * Typical supported types thus include:
-  *  - std::vector<int>
-  *  - std::valarray<int>
-  *  - std::array<int>
-  *  - Plain C arrays: int[N]
-  *  - Eigen::ArrayXi
-  *  - decltype(ArrayXi::LinSpaced(...))
-  *  - Any view/expressions of the previous types
-  *  - Eigen::ArithmeticSequence
-  *  - Eigen::internal::AllRange      (helper for Eigen::all)
-  *  - Eigen::internal::SingleRange  (helper for single index)
-  *  - etc.
-  *
-  * In typical usages of %Eigen, this class should never be used directly. It is the return type of
-  * DenseBase::operator()(const RowIndices&, const ColIndices&).
-  *
-  * \sa class Block
-  */
-template<typename XprType, typename RowIndices, typename ColIndices>
-class IndexedView : public IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>
-{
-public:
-  typedef typename IndexedViewImpl<XprType, RowIndices, ColIndices, typename internal::traits<XprType>::StorageKind>::Base Base;
-  EIGEN_GENERIC_PUBLIC_INTERFACE(IndexedView)
-  EIGEN_INHERIT_ASSIGNMENT_OPERATORS(IndexedView)
-
-  typedef typename internal::ref_selector<XprType>::non_const_type MatrixTypeNested;
-  typedef typename internal::remove_all<XprType>::type NestedExpression;
-
-  template<typename T0, typename T1>
-  IndexedView(XprType& xpr, const T0& rowIndices, const T1& colIndices)
-    : m_xpr(xpr), m_rowIndices(rowIndices), m_colIndices(colIndices)
-  {}
-
-  /** \returns number of rows */
-  Index rows() const { return internal::index_list_size(m_rowIndices); }
-
-  /** \returns number of columns */
-  Index cols() const { return internal::index_list_size(m_colIndices); }
-
-  /** \returns the nested expression */
-  const typename internal::remove_all<XprType>::type&
-  nestedExpression() const { return m_xpr; }
-
-  /** \returns the nested expression */
-  typename internal::remove_reference<XprType>::type&
-  nestedExpression() { return m_xpr; }
-
-  /** \returns a const reference to the object storing/generating the row indices */
-  const RowIndices& rowIndices() const { return m_rowIndices; }
-
-  /** \returns a const reference to the object storing/generating the column indices */
-  const ColIndices& colIndices() const { return m_colIndices; }
-
-protected:
-  MatrixTypeNested m_xpr;
-  RowIndices m_rowIndices;
-  ColIndices m_colIndices;
-};
-
-
-// Generic API dispatcher
-template<typename XprType, typename RowIndices, typename ColIndices, typename StorageKind>
-class IndexedViewImpl
-  : public internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type
-{
-public:
-  typedef typename internal::generic_xpr_base<IndexedView<XprType, RowIndices, ColIndices> >::type Base;
-};
-
-namespace internal {
-
-
-template<typename ArgType, typename RowIndices, typename ColIndices>
-struct unary_evaluator<IndexedView<ArgType, RowIndices, ColIndices>, IndexBased>
-  : evaluator_base<IndexedView<ArgType, RowIndices, ColIndices> >
-{
-  typedef IndexedView<ArgType, RowIndices, ColIndices> XprType;
-
-  enum {
-    CoeffReadCost = evaluator<ArgType>::CoeffReadCost /* TODO + cost of row/col index */,
-
-    FlagsLinearAccessBit = (traits<XprType>::RowsAtCompileTime == 1 || traits<XprType>::ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-
-    FlagsRowMajorBit = traits<XprType>::FlagsRowMajorBit, 
-
-    Flags = (evaluator<ArgType>::Flags & (HereditaryBits & ~RowMajorBit /*| LinearAccessBit | DirectAccessBit*/)) | FlagsLinearAccessBit | FlagsRowMajorBit,
-
-    Alignment = 0
-  };
-
-  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr)
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-
-  typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  CoeffReturnType coeff(Index row, Index col) const
-  {
-    eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows()
-                 && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols());
-    return m_argImpl.coeff(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& coeffRef(Index row, Index col)
-  {
-    eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows()
-                 && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols());
-    return m_argImpl.coeffRef(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  Scalar& coeffRef(Index index)
-  {
-    EIGEN_STATIC_ASSERT_LVALUE(XprType)
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows()
-                 && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols());
-    return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar& coeffRef(Index index) const
-  {
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows()
-                 && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols());
-    return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const CoeffReturnType coeff(Index index) const
-  {
-    Index row = XprType::RowsAtCompileTime == 1 ? 0 : index;
-    Index col = XprType::RowsAtCompileTime == 1 ? index : 0;
-    eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows()
-                 && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols());
-    return m_argImpl.coeff( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]);
-  }
-
-protected:
-
-  evaluator<ArgType> m_argImpl;
-  const XprType& m_xpr;
-
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_INDEXED_VIEW_H

Eigen/src/Core/Inverse.h

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2014-2019 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -44,6 +44,7 @@ class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::S
 {
 public:
   typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename XprType::PlainObject                       PlainObject;
   typedef typename XprType::Scalar                            Scalar;
   typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
   typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
@@ -54,8 +55,8 @@ public:
     : m_xpr(xpr)
   {}
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR  Index rows() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR  Index cols() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
 
   EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
 

Eigen/src/Core/Map.h

@@ -47,7 +47,7 @@ private:
   * \brief A matrix or vector expression mapping an existing array of data.
   *
   * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
+  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
   *                The default is \c #Unaligned.
   * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
   *                   of an ordinary, contiguous array. This can be overridden by specifying strides.
@@ -104,19 +104,19 @@ template<typename PlainObjectType, int MapOptions, typename StrideType> class Ma
     EIGEN_DEVICE_FUNC
     inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
-      return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
+      return int(StrideType::OuterStrideAtCompileTime) != 0 ? m_stride.outer()
-           : internal::traits<Map>::OuterStrideAtCompileTime != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
+           : int(internal::traits<Map>::OuterStrideAtCompileTime) != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
            : IsVectorAtCompileTime ? (this->size() * innerStride())
-           : int(Flags)&RowMajorBit ? (this->cols() * innerStride())
+           : (int(Flags)&RowMajorBit) ? (this->cols() * innerStride())
            : (this->rows() * innerStride());
     }
 

Eigen/src/Core/MapBase.h

@@ -87,11 +87,9 @@ template<typename Derived> class MapBase<Derived, ReadOnlyAccessors>
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
     /** \copydoc DenseBase::rows() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_rows.value(); }
-    inline Index rows() const EIGEN_NOEXCEPT { return m_rows.value(); }
     /** \copydoc DenseBase::cols() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_cols.value(); }
-    inline Index cols() const EIGEN_NOEXCEPT { return m_cols.value(); }
 
     /** Returns a pointer to the first coefficient of the matrix or vector.
       *

Eigen/src/Core/MathFunctionsImpl.h

@@ -17,28 +17,24 @@ namespace internal {
 
 /** \internal \returns the hyperbolic tan of \a a (coeff-wise)
     Doesn't do anything fancy, just a 13/6-degree rational interpolant which
-    is accurate up to a couple of ulps in the (approximate) range [-8, 8],
+    is accurate up to a couple of ulp in the range [-9, 9], outside of which
-    outside of which tanh(x) = +/-1 in single precision. The input is clamped
+    the tanh(x) = +/-1.
-    to the range [-c, c]. The value c is chosen as the smallest value where
-    the approximation evaluates to exactly 1. In the reange [-0.0004, 0.0004]
-    the approxmation tanh(x) ~= x is used for better accuracy as x tends to zero.
 
     This implementation works on both scalars and packets.
 */
 template<typename T>
 T generic_fast_tanh_float(const T& a_x)
 {
-  // Clamp the inputs to the range [-c, c]
+  // Clamp the inputs to the range [-9, 9] since anything outside
-#ifdef EIGEN_VECTORIZE_FMA
+  // this range is +/-1.0f in single-precision.
-  const T plus_clamp = pset1<T>(7.99881172180175781f);
+  const T plus_9 = pset1<T>(9.f);
-  const T minus_clamp = pset1<T>(-7.99881172180175781f);
+  const T minus_9 = pset1<T>(-9.f);
-#else
+  // NOTE GCC prior to 6.3 might improperly optimize this max/min
-  const T plus_clamp = pset1<T>(7.90531110763549805f);
+  //      step such that if a_x is nan, x will be either 9 or -9,
-  const T minus_clamp = pset1<T>(-7.90531110763549805f);
+  //      and tanh will return 1 or -1 instead of nan.
-#endif
+  //      This is supposed to be fixed in gcc6.3,
-  const T tiny = pset1<T>(0.0004f);
+  //      see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72867
-  const T x = pmax(pmin(a_x, plus_clamp), minus_clamp);
+  const T x = pmax(minus_9,pmin(plus_9,a_x));
-  const T tiny_mask = pcmp_lt(pabs(a_x), tiny);
   // The monomial coefficients of the numerator polynomial (odd).
   const T alpha_1 = pset1<T>(4.89352455891786e-03f);
   const T alpha_3 = pset1<T>(6.37261928875436e-04f);
@@ -66,26 +62,20 @@ T generic_fast_tanh_float(const T& a_x)
   p = pmadd(x2, p, alpha_1);
   p = pmul(x, p);
 
-  // Evaluate the denominator polynomial q.
+  // Evaluate the denominator polynomial p.
   T q = pmadd(x2, beta_6, beta_4);
   q = pmadd(x2, q, beta_2);
   q = pmadd(x2, q, beta_0);
 
   // Divide the numerator by the denominator.
-  return pselect(tiny_mask, x, pdiv(p, q));
+  return pdiv(p, q);
 }
 
 template<typename RealScalar>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_STRONG_INLINE
 RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
 {
-  // IEEE IEC 6059 special cases.
+  EIGEN_USING_STD_MATH(sqrt);
-  if ((numext::isinf)(x) || (numext::isinf)(y))
-    return NumTraits<RealScalar>::infinity();
-  if ((numext::isnan)(x) || (numext::isnan)(y))
-    return NumTraits<RealScalar>::quiet_NaN();
-    
-  EIGEN_USING_STD(sqrt);
   RealScalar p, qp;
   p = numext::maxi(x,y);
   if(p==RealScalar(0)) return RealScalar(0);
@@ -97,102 +87,13 @@ template<typename Scalar>
 struct hypot_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  static EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const Scalar& x, const Scalar& y)
-  inline RealScalar run(const Scalar& x, const Scalar& y)
   {
-    EIGEN_USING_STD(abs);
+    EIGEN_USING_STD_MATH(abs);
     return positive_real_hypot<RealScalar>(abs(x), abs(y));
   }
 };
 
-// Generic complex sqrt implementation that correctly handles corner cases
-// according to https://en.cppreference.com/w/cpp/numeric/complex/sqrt
-template<typename T>
-EIGEN_DEVICE_FUNC std::complex<T> complex_sqrt(const std::complex<T>& z) {
-  // Computes the principal sqrt of the input.
-  //
-  // For a complex square root of the number x + i*y. We want to find real
-  // numbers u and v such that
-  //    (u + i*v)^2 = x + i*y  <=>
-  //    u^2 - v^2 + i*2*u*v = x + i*v.
-  // By equating the real and imaginary parts we get:
-  //    u^2 - v^2 = x
-  //    2*u*v = y.
-  //
-  // For x >= 0, this has the numerically stable solution
-  //    u = sqrt(0.5 * (x + sqrt(x^2 + y^2)))
-  //    v = y / (2 * u)
-  // and for x < 0,
-  //    v = sign(y) * sqrt(0.5 * (-x + sqrt(x^2 + y^2)))
-  //    u = y / (2 * v)
-  //
-  // Letting w = sqrt(0.5 * (|x| + |z|)),
-  //   if x == 0: u = w, v = sign(y) * w
-  //   if x > 0:  u = w, v = y / (2 * w)
-  //   if x < 0:  u = |y| / (2 * w), v = sign(y) * w
-
-  const T x = numext::real(z);
-  const T y = numext::imag(z);
-  const T zero = T(0);
-  const T w = numext::sqrt(T(0.5) * (numext::abs(x) + numext::hypot(x, y)));
-
-  return
-    (numext::isinf)(y) ? std::complex<T>(NumTraits<T>::infinity(), y)
-      : x == zero ? std::complex<T>(w, y < zero ? -w : w)
-      : x > zero ? std::complex<T>(w, y / (2 * w))
-      : std::complex<T>(numext::abs(y) / (2 * w), y < zero ? -w : w );
-}
-
-// Generic complex rsqrt implementation.
-template<typename T>
-EIGEN_DEVICE_FUNC std::complex<T> complex_rsqrt(const std::complex<T>& z) {
-  // Computes the principal reciprocal sqrt of the input.
-  //
-  // For a complex reciprocal square root of the number z = x + i*y. We want to
-  // find real numbers u and v such that
-  //    (u + i*v)^2 = 1 / (x + i*y)  <=>
-  //    u^2 - v^2 + i*2*u*v = x/|z|^2 - i*v/|z|^2.
-  // By equating the real and imaginary parts we get:
-  //    u^2 - v^2 = x/|z|^2
-  //    2*u*v = y/|z|^2.
-  //
-  // For x >= 0, this has the numerically stable solution
-  //    u = sqrt(0.5 * (x + |z|)) / |z|
-  //    v = -y / (2 * u * |z|)
-  // and for x < 0,
-  //    v = -sign(y) * sqrt(0.5 * (-x + |z|)) / |z|
-  //    u = -y / (2 * v * |z|)
-  //
-  // Letting w = sqrt(0.5 * (|x| + |z|)),
-  //   if x == 0: u = w / |z|, v = -sign(y) * w / |z|
-  //   if x > 0:  u = w / |z|, v = -y / (2 * w * |z|)
-  //   if x < 0:  u = |y| / (2 * w * |z|), v = -sign(y) * w / |z|
-
-  const T x = numext::real(z);
-  const T y = numext::imag(z);
-  const T zero = T(0);
-
-  const T abs_z = numext::hypot(x, y);
-  const T w = numext::sqrt(T(0.5) * (numext::abs(x) + abs_z));
-  const T woz = w / abs_z;
-  // Corner cases consistent with 1/sqrt(z) on gcc/clang.
-  return
-    abs_z == zero ? std::complex<T>(NumTraits<T>::infinity(), NumTraits<T>::quiet_NaN())
-      : ((numext::isinf)(x) || (numext::isinf)(y)) ? std::complex<T>(zero, zero)
-      : x == zero ? std::complex<T>(woz, y < zero ? woz : -woz)
-      : x > zero ? std::complex<T>(woz, -y / (2 * w * abs_z))
-      : std::complex<T>(numext::abs(y) / (2 * w * abs_z), y < zero ? woz : -woz );
-}
-
-template<typename T>
-EIGEN_DEVICE_FUNC std::complex<T> complex_log(const std::complex<T>& z) {
-  // Computes complex log.
-  T a = numext::abs(z);
-  EIGEN_USING_STD(atan2);
-  T b = atan2(z.imag(), z.real());
-  return std::complex<T>(numext::log(a), b);
-}
-
 } // end namespace internal
 
 } // end namespace Eigen

Eigen/src/Core/Matrix.h

@@ -225,6 +225,8 @@ class Matrix
       return Base::_set(other);
     }
 
+    /* Here, doxygen failed to copy the brief information when using \copydoc */
+
     /**
       * \brief Copies the generic expression \a other into *this.
       * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
@@ -253,103 +255,53 @@ class Matrix
       *
       * \sa resize(Index,Index)
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    Matrix() : Base()
+    EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
     // FIXME is it still needed
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     explicit Matrix(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     { Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED }
 
 #if EIGEN_HAS_RVALUE_REFERENCES
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
       : Base(std::move(other))
     {
       Base::_check_template_params();
     }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
     {
-      Base::operator=(std::move(other));
+      other.swap(*this);
       return *this;
     }
 #endif
 
-#if EIGEN_HAS_CXX11
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11
-     *
-     * \only_for_vectors
-     *
-     * This constructor is for 1D array or vectors with more than 4 coefficients.
-     * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients.
-     *
-     * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this
-     * constructor must match the the fixed number of rows (resp. columns) of \c *this.
-     *
-     * Example: \include Matrix_variadic_ctor_cxx11.cpp
-     * Output: \verbinclude Matrix_variadic_ctor_cxx11.out
-     *
-     * \sa Matrix(const std::initializer_list<std::initializer_list<Scalar>>&)
-     */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2,  const Scalar& a3, const ArgTypes&... args)
-      : Base(a0, a1, a2, a3, args...) {}
-
-    /** \brief Constructs a Matrix and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11
-      *
-      * \anchor matrix_constructor_initializer_list
-      *
-      * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients:
-      *
-      * Example: \include Matrix_initializer_list_23_cxx11.cpp
-      * Output: \verbinclude Matrix_initializer_list_23_cxx11.out
-      *
-      * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered.
-      *
-      * In the case of a compile-time column vector, implicit transposition from a single row is allowed.
-      * Therefore <code>VectorXd{{1,2,3,4,5}}</code> is legal and the more verbose syntax
-      * <code>RowVectorXd{{1},{2},{3},{4},{5}}</code> can be avoided:
-      *
-      * Example: \include Matrix_initializer_list_vector_cxx11.cpp
-      * Output: \verbinclude Matrix_initializer_list_vector_cxx11.out
-      *
-      * In the case of fixed-sized matrices, the initializer list sizes must exactly match the matrix sizes,
-      * and implicit transposition is allowed for compile-time vectors only.
-      *
-      * \sa Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2,  const Scalar& a3, const ArgTypes&... args)
-      */
-    EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE Matrix(const std::initializer_list<std::initializer_list<Scalar>>& list) : Base(list) {}
-#endif // end EIGEN_HAS_CXX11
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
 
     // This constructor is for both 1x1 matrices and dynamic vectors
     template<typename T>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    explicit Matrix(const T& x)
+    EIGEN_STRONG_INLINE explicit Matrix(const T& x)
     {
       Base::_check_template_params();
       Base::template _init1<T>(x);
     }
 
     template<typename T0, typename T1>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    Matrix(const T0& x, const T1& y)
+    EIGEN_STRONG_INLINE Matrix(const T0& x, const T1& y)
     {
       Base::_check_template_params();
       Base::template _init2<T0,T1>(x, y);
     }
-
+    #else
-
-#else
     /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
     EIGEN_DEVICE_FUNC
     explicit Matrix(const Scalar *data);
@@ -367,8 +319,7 @@ class Matrix
       * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
       */
     EIGEN_STRONG_INLINE explicit Matrix(Index dim);
-    /** \brief Constructs an initialized 1x1 matrix with the given coefficient
+    /** \brief Constructs an initialized 1x1 matrix with the given coefficient */
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...) */
     Matrix(const Scalar& x);
     /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
@@ -385,14 +336,11 @@ class Matrix
     EIGEN_DEVICE_FUNC
     Matrix(Index rows, Index cols);
     
-    /** \brief Constructs an initialized 2D vector with given coefficients
+    /** \brief Constructs an initialized 2D vector with given coefficients */
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...) */
     Matrix(const Scalar& x, const Scalar& y);
-    #endif  // end EIGEN_PARSED_BY_DOXYGEN
+    #endif
 
-    /** \brief Constructs an initialized 3D vector with given coefficients
+    /** \brief Constructs an initialized 3D vector with given coefficients */
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...)
-      */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
@@ -402,9 +350,7 @@ class Matrix
       m_storage.data()[1] = y;
       m_storage.data()[2] = z;
     }
-    /** \brief Constructs an initialized 4D vector with given coefficients
+    /** \brief Constructs an initialized 4D vector with given coefficients */
-      * \sa Matrix(const Scalar&, const Scalar&, const Scalar&,  const Scalar&, const ArgTypes&...)
-      */
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
@@ -431,10 +377,8 @@ class Matrix
       : Base(other.derived())
     { }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
-    inline Index innerStride() const EIGEN_NOEXCEPT { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); }
 
     /////////// Geometry module ///////////
 
@@ -461,7 +405,7 @@ class Matrix
   *
   * \ingroup Core_Module
   *
-  * %Eigen defines several typedef shortcuts for most common matrix and vector types.
+  * Eigen defines several typedef shortcuts for most common matrix and vector types.
   *
   * The general patterns are the following:
   *
@@ -474,35 +418,21 @@ class Matrix
   * There are also \c VectorSizeType and \c RowVectorSizeType which are self-explanatory. For example, \c Vector4cf is
   * a fixed-size vector of 4 complex floats.
   *
-  * With \cpp11, template alias are also defined for common sizes.
-  * They follow the same pattern as above except that the scalar type suffix is replaced by a
-  * template parameter, i.e.:
-  *   - `MatrixSize<Type>` where `Size` can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size.
-  *   - `MatrixXSize<Type>` and `MatrixSizeX<Type>` where `Size` can be \c 2,\c 3,\c 4 for hybrid dynamic/fixed matrices.
-  *   - `VectorSize<Type>` and `RowVectorSize<Type>` for column and row vectors.
-  *
-  * With \cpp11, you can also use fully generic column and row vector types: `Vector<Type,Size>` and `RowVector<Type,Size>`.
-  *
   * \sa class Matrix
   */
 
 #define EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix)   \
 /** \ingroup matrixtypedefs */                                    \
-/** \brief \noop            */                                    \
 typedef Matrix<Type, Size, Size> Matrix##SizeSuffix##TypeSuffix;  \
 /** \ingroup matrixtypedefs */                                    \
-/** \brief \noop            */                                    \
 typedef Matrix<Type, Size, 1>    Vector##SizeSuffix##TypeSuffix;  \
 /** \ingroup matrixtypedefs */                                    \
-/** \brief \noop            */                                    \
 typedef Matrix<Type, 1, Size>    RowVector##SizeSuffix##TypeSuffix;
 
 #define EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, Size)         \
 /** \ingroup matrixtypedefs */                                    \
-/** \brief \noop            */                                    \
 typedef Matrix<Type, Size, Dynamic> Matrix##Size##X##TypeSuffix;  \
 /** \ingroup matrixtypedefs */                                    \
-/** \brief \noop            */                                    \
 typedef Matrix<Type, Dynamic, Size> Matrix##X##Size##TypeSuffix;
 
 #define EIGEN_MAKE_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \
@@ -524,55 +454,6 @@ EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex<double>, cd)
 #undef EIGEN_MAKE_TYPEDEFS
 #undef EIGEN_MAKE_FIXED_TYPEDEFS
 
-#if EIGEN_HAS_CXX11
-
-#define EIGEN_MAKE_TYPEDEFS(Size, SizeSuffix)                     \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##SizeSuffix = Matrix<Type, Size, Size>;              \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Vector##SizeSuffix = Matrix<Type, Size, 1>;                 \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using RowVector##SizeSuffix = Matrix<Type, 1, Size>;
-
-#define EIGEN_MAKE_FIXED_TYPEDEFS(Size)                           \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##Size##X = Matrix<Type, Size, Dynamic>;              \
-/** \ingroup matrixtypedefs */                                    \
-/** \brief \cpp11 */                                              \
-template <typename Type>                                          \
-using Matrix##X##Size = Matrix<Type, Dynamic, Size>;
-
-EIGEN_MAKE_TYPEDEFS(2, 2)
-EIGEN_MAKE_TYPEDEFS(3, 3)
-EIGEN_MAKE_TYPEDEFS(4, 4)
-EIGEN_MAKE_TYPEDEFS(Dynamic, X)
-EIGEN_MAKE_FIXED_TYPEDEFS(2)
-EIGEN_MAKE_FIXED_TYPEDEFS(3)
-EIGEN_MAKE_FIXED_TYPEDEFS(4)
-
-/** \ingroup matrixtypedefs
-  * \brief \cpp11 */
-template <typename Type, int Size>
-using Vector = Matrix<Type, Size, 1>;
-
-/** \ingroup matrixtypedefs
-  * \brief \cpp11 */
-template <typename Type, int Size>
-using RowVector = Matrix<Type, 1, Size>;
-
-#undef EIGEN_MAKE_TYPEDEFS
-#undef EIGEN_MAKE_FIXED_TYPEDEFS
-
-#endif // EIGEN_HAS_CXX11
-
 } // end namespace Eigen
 
 #endif // EIGEN_MATRIX_H

Eigen/src/Core/MatrixBase.h

@@ -76,7 +76,6 @@ template<typename Derived> class MatrixBase
     using Base::coeffRef;
     using Base::lazyAssign;
     using Base::eval;
-    using Base::operator-;
     using Base::operator+=;
     using Base::operator-=;
     using Base::operator*=;
@@ -123,6 +122,7 @@ template<typename Derived> class MatrixBase
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
 #define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseBinaryOps.h"
@@ -206,22 +206,28 @@ template<typename Derived> class MatrixBase
     EIGEN_DEVICE_FUNC
     DiagonalReturnType diagonal();
 
-    typedef Diagonal<const Derived> ConstDiagonalReturnType;
+    typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
     EIGEN_DEVICE_FUNC
-    const ConstDiagonalReturnType diagonal() const;
+    ConstDiagonalReturnType diagonal() const;
+
+    template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
+    template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
 
     template<int Index>
     EIGEN_DEVICE_FUNC
-    Diagonal<Derived, Index> diagonal();
+    typename DiagonalIndexReturnType<Index>::Type diagonal();
 
     template<int Index>
     EIGEN_DEVICE_FUNC
-    const Diagonal<const Derived, Index> diagonal() const;
+    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+
+    typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
+    typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
 
     EIGEN_DEVICE_FUNC
-    Diagonal<Derived, DynamicIndex> diagonal(Index index);
+    DiagonalDynamicIndexReturnType diagonal(Index index);
     EIGEN_DEVICE_FUNC
-    const Diagonal<const Derived, DynamicIndex> diagonal(Index index) const;
+    ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
 
     template<unsigned int Mode> struct TriangularViewReturnType { typedef TriangularView<Derived, Mode> Type; };
     template<unsigned int Mode> struct ConstTriangularViewReturnType { typedef const TriangularView<const Derived, Mode> Type; };
@@ -262,8 +268,6 @@ template<typename Derived> class MatrixBase
     Derived& setIdentity();
     EIGEN_DEVICE_FUNC
     Derived& setIdentity(Index rows, Index cols);
-    EIGEN_DEVICE_FUNC Derived& setUnit(Index i);
-    EIGEN_DEVICE_FUNC Derived& setUnit(Index newSize, Index i);
 
     bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
     bool isDiagonal(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
@@ -292,7 +296,7 @@ template<typename Derived> class MatrixBase
     EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
     { return cwiseNotEqual(other).any(); }
 
-    NoAlias<Derived,Eigen::MatrixBase > EIGEN_DEVICE_FUNC noalias();
+    NoAlias<Derived,Eigen::MatrixBase > noalias();
 
     // TODO forceAlignedAccess is temporarily disabled
     // Need to find a nicer workaround.
@@ -322,7 +326,6 @@ template<typename Derived> class MatrixBase
 
     inline const PartialPivLU<PlainObject> lu() const;
 
-    EIGEN_DEVICE_FUNC
     inline const Inverse<Derived> inverse() const;
 
     template<typename ResultType>
@@ -332,15 +335,12 @@ template<typename Derived> class MatrixBase
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
-
     template<typename ResultType>
     inline void computeInverseWithCheck(
       ResultType& inverse,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
-
-    EIGEN_DEVICE_FUNC
     Scalar determinant() const;
 
 /////////// Cholesky module ///////////
@@ -412,19 +412,15 @@ template<typename Derived> class MatrixBase
 
 ////////// Householder module ///////////
 
-    EIGEN_DEVICE_FUNC
     void makeHouseholderInPlace(Scalar& tau, RealScalar& beta);
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void makeHouseholder(EssentialPart& essential,
                          Scalar& tau, RealScalar& beta) const;
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void applyHouseholderOnTheLeft(const EssentialPart& essential,
                                    const Scalar& tau,
                                    Scalar* workspace);
     template<typename EssentialPart>
-    EIGEN_DEVICE_FUNC
     void applyHouseholderOnTheRight(const EssentialPart& essential,
                                     const Scalar& tau,
                                     Scalar* workspace);
@@ -432,10 +428,8 @@ template<typename Derived> class MatrixBase
 ///////// Jacobi module /////////
 
     template<typename OtherScalar>
-    EIGEN_DEVICE_FUNC
     void applyOnTheLeft(Index p, Index q, const JacobiRotation<OtherScalar>& j);
     template<typename OtherScalar>
-    EIGEN_DEVICE_FUNC
     void applyOnTheRight(Index p, Index q, const JacobiRotation<OtherScalar>& j);
 
 ///////// SparseCore module /////////
@@ -462,11 +456,6 @@ template<typename Derived> class MatrixBase
     const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
     EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine)
     EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine)
-#if EIGEN_HAS_CXX11_MATH
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, atanh, inverse hyperbolic cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, acosh, inverse hyperbolic cosine)
-    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, asinh, inverse hyperbolic sine)
-#endif
     EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine)
     EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine)
     EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root)

Eigen/src/Core/NestByValue.h

@@ -16,11 +16,7 @@ namespace Eigen {
 namespace internal {
 template<typename ExpressionType>
 struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
-{
+{};
-  enum {
-    Flags = traits<ExpressionType>::Flags & ~NestByRefBit
-  };
-};
 }
 
 /** \class NestByValue
@@ -45,13 +41,57 @@ template<typename ExpressionType> class NestByValue
 
     EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
+
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
+    {
+      return m_expression.coeff(row, col);
+    }
+
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
+    {
+      return m_expression.const_cast_derived().coeffRef(row, col);
+    }
+
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
+    {
+      return m_expression.coeff(index);
+    }
+
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
+    {
+      return m_expression.const_cast_derived().coeffRef(index);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index row, Index col) const
+    {
+      return m_expression.template packet<LoadMode>(row, col);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index row, Index col, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(row, col, x);
+    }
+
+    template<int LoadMode>
+    inline const PacketScalar packet(Index index) const
+    {
+      return m_expression.template packet<LoadMode>(index);
+    }
+
+    template<int LoadMode>
+    inline void writePacket(Index index, const PacketScalar& x)
+    {
+      m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
+    }
 
     EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
-    EIGEN_DEVICE_FUNC const ExpressionType& nestedExpression() const { return m_expression; }
-
   protected:
     const ExpressionType m_expression;
 };
@@ -65,21 +105,6 @@ DenseBase<Derived>::nestByValue() const
   return NestByValue<Derived>(derived());
 }
 
-namespace internal {
-
-// Evaluator of Solve -> eval into a temporary
-template<typename ArgType>
-struct evaluator<NestByValue<ArgType> >
-  : public evaluator<ArgType>
-{
-  typedef evaluator<ArgType> Base;
-
-  EIGEN_DEVICE_FUNC explicit evaluator(const NestByValue<ArgType>& xpr)
-    : Base(xpr.nestedExpression())
-  {}
-};
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_NESTBYVALUE_H

Eigen/src/Core/NoAlias.h

@@ -33,7 +33,6 @@ class NoAlias
   public:
     typedef typename ExpressionType::Scalar Scalar;
     
-    EIGEN_DEVICE_FUNC
     explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
     
     template<typename OtherDerived>
@@ -75,10 +74,10 @@ class NoAlias
   *
   * More precisely, noalias() allows to bypass the EvalBeforeAssignBit flag.
   * Currently, even though several expressions may alias, only product
-  * expressions have this flag. Therefore, noalias() is only useful when
+  * expressions have this flag. Therefore, noalias() is only usefull when
   * the source expression contains a matrix product.
   *
-  * Here are some examples where noalias is useful:
+  * Here are some examples where noalias is usefull:
   * \code
   * D.noalias()  = A * B;
   * D.noalias() += A.transpose() * B;
@@ -99,7 +98,7 @@ class NoAlias
   * \sa class NoAlias
   */
 template<typename Derived>
-NoAlias<Derived,MatrixBase> EIGEN_DEVICE_FUNC MatrixBase<Derived>::noalias()
+NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
   return NoAlias<Derived, Eigen::MatrixBase >(derived());
 }

Eigen/src/Core/NumTraits.h

@@ -21,14 +21,12 @@ template< typename T,
           bool is_integer = NumTraits<T>::IsInteger>
 struct default_digits10_impl
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static int run() { return std::numeric_limits<T>::digits10; }
 };
 
 template<typename T>
 struct default_digits10_impl<T,false,false> // Floating point
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static int run() {
     using std::log10;
     using std::ceil;
@@ -40,65 +38,11 @@ struct default_digits10_impl<T,false,false> // Floating point
 template<typename T>
 struct default_digits10_impl<T,false,true> // Integer
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return 0; }
-};
-
-
-// default implementation of digits(), based on numeric_limits if specialized,
-// 0 for integer types, and log2(epsilon()) otherwise.
-template< typename T,
-          bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
-          bool is_integer = NumTraits<T>::IsInteger>
-struct default_digits_impl
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() { return std::numeric_limits<T>::digits; }
-};
-
-template<typename T>
-struct default_digits_impl<T,false,false> // Floating point
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static int run() {
-    using std::log;
-    using std::ceil;
-    typedef typename NumTraits<T>::Real Real;
-    return int(ceil(-log(NumTraits<Real>::epsilon())/log(static_cast<Real>(2))));
-  }
-};
-
-template<typename T>
-struct default_digits_impl<T,false,true> // Integer
-{
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static int run() { return 0; }
 };
 
 } // end namespace internal
 
-namespace numext {
-/** \internal bit-wise cast without changing the underlying bit representation. */
-
-// TODO: Replace by std::bit_cast (available in C++20)
-template <typename Tgt, typename Src>
-EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) {
-#if EIGEN_HAS_TYPE_TRAITS
-  // The behaviour of memcpy is not specified for non-trivially copyable types
-  EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Src>::value, THIS_TYPE_IS_NOT_SUPPORTED);
-  EIGEN_STATIC_ASSERT(std::is_trivially_copyable<Tgt>::value && std::is_default_constructible<Tgt>::value,
-                      THIS_TYPE_IS_NOT_SUPPORTED);
-#endif
-
-  EIGEN_STATIC_ASSERT(sizeof(Src) == sizeof(Tgt), THIS_TYPE_IS_NOT_SUPPORTED);
-  Tgt tgt;
-  EIGEN_USING_STD(memcpy)
-  memcpy(&tgt, &src, sizeof(Tgt));
-  return tgt;
-}
-}  // namespace numext
-
-// clang-format off
 /** \class NumTraits
   * \ingroup Core_Module
   *
@@ -114,43 +58,32 @@ EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) {
   *     is a typedef to \a U.
   * \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values,
   *     such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives
-  *     \a T again. Note however that many Eigen functions such as `internal::sqrt` simply refuse to
+  *     \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to
   *     take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is
   *     only intended as a helper for code that needs to explicitly promote types.
-  * \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for `std::complex<U>`,
+  * \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for `std::complex<U>`, Literal is defined as \c U.
-  *     Literal is defined as \c U.
   *     Of course, this type must be fully compatible with \a T. In doubt, just use \a T here.
   * \li A typedef \c Nested giving the type to use to nest a value inside of the expression tree. If you don't know what
   *     this means, just use \a T here.
-  * \li An enum value \c IsComplex. It is equal to 1 if \a T is a \c std::complex
+  * \li An enum value \c IsComplex. It is equal to 1 if \a T is a `std::complex`
   *     type, and to 0 otherwise.
   * \li An enum value \c IsInteger. It is equal to \c 1 if \a T is an integer type such as \c int,
   *     and to \c 0 otherwise.
   * \li Enum values \c ReadCost, \c AddCost and \c MulCost representing a rough estimate of the number of CPU cycles needed
   *     to by move / add / mul instructions respectively, assuming the data is already stored in CPU registers.
-  *     Stay vague here. No need to do architecture-specific stuff. If you don't know what this means, just use \c Eigen::HugeCost.
+  *     Stay vague here. No need to do architecture-specific stuff.
   * \li An enum value \c IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned.
   * \li An enum value \c RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must
   *     be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise.
   * \li An `epsilon()` function which, unlike <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/epsilon">`std::numeric_limits::epsilon()`</a>,
-  *     it returns a \c Real instead of a \a T.
+  *     it returns a \a Real instead of a \a T.
   * \li A `dummy_precision()` function returning a weak epsilon value. It is mainly used as a default
   *     value by the fuzzy comparison operators.
   * \li `highest()` and `lowest()` functions returning the highest and lowest possible values respectively.
-  * \li `digits()` function returning the number of radix digits (non-sign digits for integers, mantissa for floating-point). This is
-  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits">std::numeric_limits<T>::digits</a>
-  *     which is used as the default implementation if specialized.
   * \li `digits10()` function returning the number of decimal digits that can be represented without change. This is
-  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">std::numeric_limits<T>::digits10</a>
+  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">`std::numeric_limits<T>::digits10`</a>
   *     which is used as the default implementation if specialized.
-  * \li `min_exponent()` and `max_exponent()` functions returning the highest and lowest possible values, respectively,
-  *     such that the radix raised to the power exponent-1 is a normalized floating-point number.  These are equivalent to
-  *     <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/min_exponent">`std::numeric_limits<T>::min_exponent`</a>/
-  *     <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/max_exponent">`std::numeric_limits<T>::max_exponent`</a>.
-  * \li `infinity()` function returning a representation of positive infinity, if available.
-  * \li `quiet_NaN` function returning a non-signaling "not-a-number", if available.
   */
-  // clang-format on
 
 template<typename T> struct GenericNumTraits
 {
@@ -173,60 +106,42 @@ template<typename T> struct GenericNumTraits
   typedef T Nested;
   typedef T Literal;
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline Real epsilon()
   {
     return numext::numeric_limits<T>::epsilon();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline int digits10()
   {
     return internal::default_digits10_impl<T>::run();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
-  static inline int digits()
-  {
-    return internal::default_digits_impl<T>::run();
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int min_exponent()
-  {
-    return numext::numeric_limits<T>::min_exponent;
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-  static inline int max_exponent()
-  {
-    return numext::numeric_limits<T>::max_exponent;
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline Real dummy_precision()
   {
     // make sure to override this for floating-point types
     return Real(0);
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+
+  EIGEN_DEVICE_FUNC
   static inline T highest() {
     return (numext::numeric_limits<T>::max)();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline T lowest()  {
-    return IsInteger ? (numext::numeric_limits<T>::min)()
+    return IsInteger ? (numext::numeric_limits<T>::min)() : (-(numext::numeric_limits<T>::max)());
-                     : static_cast<T>(-(numext::numeric_limits<T>::max)());
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline T infinity() {
     return numext::numeric_limits<T>::infinity();
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline T quiet_NaN() {
     return numext::numeric_limits<T>::quiet_NaN();
   }
@@ -238,27 +153,23 @@ template<typename T> struct NumTraits : GenericNumTraits<T>
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline double dummy_precision() { return 1e-12; }
 };
 
-// GPU devices treat `long double` as `double`.
-#ifndef EIGEN_GPU_COMPILE_PHASE
 template<> struct NumTraits<long double>
   : GenericNumTraits<long double>
 {
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline long double dummy_precision() { return static_cast<long double>(1e-15l); }
 
 #if defined(EIGEN_ARCH_PPC) && (__LDBL_MANT_DIG__ == 106)
   // PowerPC double double causes issues with some values
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
   static inline long double epsilon()
   {
     // 2^(-(__LDBL_MANT_DIG__)+1)
@@ -266,7 +177,6 @@ template<> struct NumTraits<long double>
   }
 #endif
 };
-#endif
 
 template<typename _Real> struct NumTraits<std::complex<_Real> >
   : GenericNumTraits<std::complex<_Real> >
@@ -281,11 +191,11 @@ template<typename _Real> struct NumTraits<std::complex<_Real> >
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline int digits10() { return NumTraits<Real>::digits10(); }
 };
 
@@ -305,17 +215,16 @@ struct NumTraits<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
     IsInteger = NumTraits<Scalar>::IsInteger,
     IsSigned  = NumTraits<Scalar>::IsSigned,
     RequireInitialization = 1,
-    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::ReadCost),
+    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
-    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::AddCost),
+    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
-    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits<Scalar>::MulCost)
+    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+  EIGEN_DEVICE_FUNC
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
 
-  EIGEN_CONSTEXPR
   static inline int digits10() { return NumTraits<Scalar>::digits10(); }
 };
 
@@ -329,7 +238,6 @@ template<> struct NumTraits<std::string>
     MulCost  = HugeCost
   };
 
-  EIGEN_CONSTEXPR
   static inline int digits10() { return 0; }
 
 private:
@@ -344,8 +252,6 @@ private:
 // Empty specialization for void to allow template specialization based on NumTraits<T>::Real with T==void and SFINAE.
 template<> struct NumTraits<void> {};
 
-template<> struct NumTraits<bool> : GenericNumTraits<bool> {};
-
 } // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H

Eigen/src/Core/PartialReduxEvaluator.h

@@ -1,237 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011-2018 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_PARTIALREDUX_H
-#define EIGEN_PARTIALREDUX_H
-
-namespace Eigen { 
-
-namespace internal {
-
-
-/***************************************************************************
-*
-* This file provides evaluators for partial reductions.
-* There are two modes:
-*
-*  - scalar path: simply calls the respective function on the column or row.
-*    -> nothing special here, all the tricky part is handled by the return
-*       types of VectorwiseOp's members. They embed the functor calling the
-*       respective DenseBase's member function.
-*
-*  - vectorized path: implements a packet-wise reductions followed by
-*    some (optional) processing of the outcome, e.g., division by n for mean.
-*
-* For the vectorized path let's observe that the packet-size and outer-unrolling
-* are both decided by the assignement logic. So all we have to do is to decide
-* on the inner unrolling.
-*
-* For the unrolling, we can reuse "internal::redux_vec_unroller" from Redux.h,
-* but be need to be careful to specify correct increment.
-*
-***************************************************************************/
-
-
-/* logic deciding a strategy for unrolling of vectorized paths */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_traits
-{
-  enum {
-    OuterSize = int(Evaluator::IsRowMajor) ? Evaluator::RowsAtCompileTime : Evaluator::ColsAtCompileTime,
-    Cost = OuterSize == Dynamic ? HugeCost
-         : OuterSize * Evaluator::CoeffReadCost + (OuterSize-1) * functor_traits<Func>::Cost,
-    Unrolling = Cost <= EIGEN_UNROLLING_LIMIT ? CompleteUnrolling : NoUnrolling
-  };
-
-};
-
-/* Value to be returned when size==0 , by default let's return 0 */
-template<typename PacketType,typename Func>
-EIGEN_DEVICE_FUNC
-PacketType packetwise_redux_empty_value(const Func& ) {
-  const typename unpacket_traits<PacketType>::type zero(0);
-  return pset1<PacketType>(zero);
-}
-
-/* For products the default is 1 */
-template<typename PacketType,typename Scalar>
-EIGEN_DEVICE_FUNC
-PacketType packetwise_redux_empty_value(const scalar_product_op<Scalar,Scalar>& ) {
-  return pset1<PacketType>(Scalar(1));
-}
-
-/* Perform the actual reduction */
-template<typename Func, typename Evaluator,
-         int Unrolling = packetwise_redux_traits<Func, Evaluator>::Unrolling
->
-struct packetwise_redux_impl;
-
-/* Perform the actual reduction with unrolling */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_impl<Func, Evaluator, CompleteUnrolling>
-{
-  typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
-  typedef typename Evaluator::Scalar Scalar;
-
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
-  PacketType run(const Evaluator &eval, const Func& func, Index /*size*/)
-  {
-    return redux_vec_unroller<Func, Evaluator, 0, packetwise_redux_traits<Func, Evaluator>::OuterSize>::template run<PacketType>(eval,func);
-  }
-};
-
-/* Add a specialization of redux_vec_unroller for size==0 at compiletime.
- * This specialization is not required for general reductions, which is
- * why it is defined here.
- */
-template<typename Func, typename Evaluator, int Start>
-struct redux_vec_unroller<Func, Evaluator, Start, 0>
-{
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE PacketType run(const Evaluator &, const Func& f)
-  {
-    return packetwise_redux_empty_value<PacketType>(f);
-  }
-};
-
-/* Perform the actual reduction for dynamic sizes */
-template<typename Func, typename Evaluator>
-struct packetwise_redux_impl<Func, Evaluator, NoUnrolling>
-{
-  typedef typename Evaluator::Scalar Scalar;
-  typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
-
-  template<typename PacketType>
-  EIGEN_DEVICE_FUNC
-  static PacketType run(const Evaluator &eval, const Func& func, Index size)
-  {
-    if(size==0)
-      return packetwise_redux_empty_value<PacketType>(func);
-    
-    const Index size4 = (size-1)&(~3);
-    PacketType p = eval.template packetByOuterInner<Unaligned,PacketType>(0,0);
-    Index i = 1;
-    // This loop is optimized for instruction pipelining:
-    // - each iteration generates two independent instructions
-    // - thanks to branch prediction and out-of-order execution we have independent instructions across loops
-    for(; i<size4; i+=4)
-      p = func.packetOp(p,
-            func.packetOp(
-              func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+0,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+1,0)),
-              func.packetOp(eval.template packetByOuterInner<Unaligned,PacketType>(i+2,0),eval.template packetByOuterInner<Unaligned,PacketType>(i+3,0))));
-    for(; i<size; ++i)
-      p = func.packetOp(p, eval.template packetByOuterInner<Unaligned,PacketType>(i,0));
-    return p;
-  }
-};
-
-template< typename ArgType, typename MemberOp, int Direction>
-struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
-  : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
-{
-  typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
-  typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
-  typedef typename internal::add_const_on_value_type<ArgTypeNested>::type ConstArgTypeNested;
-  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
-  typedef typename ArgType::Scalar InputScalar;
-  typedef typename XprType::Scalar Scalar;
-  enum {
-    TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) :  int(ArgType::ColsAtCompileTime)
-  };
-  typedef typename MemberOp::template Cost<int(TraversalSize)> CostOpType;
-  enum {
-    CoeffReadCost = TraversalSize==Dynamic ? HugeCost
-                  : TraversalSize==0 ? 1
-                  : int(TraversalSize) * int(evaluator<ArgType>::CoeffReadCost) + int(CostOpType::value),
-    
-    _ArgFlags = evaluator<ArgType>::Flags,
-
-    _Vectorizable =  bool(int(_ArgFlags)&PacketAccessBit)
-                  && bool(MemberOp::Vectorizable)
-                  && (Direction==int(Vertical) ? bool(_ArgFlags&RowMajorBit) : (_ArgFlags&RowMajorBit)==0)
-                  && (TraversalSize!=0),
-                  
-    Flags = (traits<XprType>::Flags&RowMajorBit)
-          | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit)))
-          | (_Vectorizable ? PacketAccessBit : 0)
-          | LinearAccessBit,
-    
-    Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
-  };
-
-  EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
-    : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : (TraversalSize==0 ? 1 : int(CostOpType::value)));
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar coeff(Index i, Index j) const
-  {
-    return coeff(Direction==Vertical ? j : i);
-  }
-
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  const Scalar coeff(Index index) const
-  {
-    return m_functor(m_arg.template subVector<DirectionType(Direction)>(index));
-  }
-
-  template<int LoadMode,typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  PacketType packet(Index i, Index j) const
-  {
-    return packet<LoadMode,PacketType>(Direction==Vertical ? j : i);
-  }
-  
-  template<int LoadMode,typename PacketType>
-  EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
-  PacketType packet(Index idx) const
-  {
-    enum { PacketSize = internal::unpacket_traits<PacketType>::size };
-    typedef Block<const ArgTypeNestedCleaned,
-                  Direction==Vertical ? int(ArgType::RowsAtCompileTime) : int(PacketSize),
-                  Direction==Vertical ? int(PacketSize) : int(ArgType::ColsAtCompileTime),
-                  true /* InnerPanel */> PanelType;
-    
-    PanelType panel(m_arg,
-                    Direction==Vertical ? 0 : idx,
-                    Direction==Vertical ? idx : 0,
-                    Direction==Vertical ? m_arg.rows() : Index(PacketSize),
-                    Direction==Vertical ? Index(PacketSize) : m_arg.cols());
-
-    // FIXME
-    // See bug 1612, currently if PacketSize==1 (i.e. complex<double> with 128bits registers) then the storage-order of panel get reversed
-    // and methods like packetByOuterInner do not make sense anymore in this context.
-    // So let's just by pass "vectorization" in this case:
-    if(PacketSize==1)
-      return internal::pset1<PacketType>(coeff(idx));
-    
-    typedef typename internal::redux_evaluator<PanelType> PanelEvaluator;
-    PanelEvaluator panel_eval(panel);
-    typedef typename MemberOp::BinaryOp BinaryOp;
-    PacketType p = internal::packetwise_redux_impl<BinaryOp,PanelEvaluator>::template run<PacketType>(panel_eval,m_functor.binaryFunc(),m_arg.outerSize());
-    return p;
-  }
-
-protected:
-  ConstArgTypeNested m_arg;
-  const MemberOp m_functor;
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_PARTIALREDUX_H

Eigen/src/Core/PermutationMatrix.h

@@ -88,13 +88,13 @@ class PermutationBase : public EigenBase<Derived>
     }
 
     /** \returns the number of rows */
-    inline EIGEN_DEVICE_FUNC Index rows() const { return Index(indices().size()); }
+    inline Index rows() const { return Index(indices().size()); }
 
     /** \returns the number of columns */
-    inline EIGEN_DEVICE_FUNC Index cols() const { return Index(indices().size()); }
+    inline Index cols() const { return Index(indices().size()); }
 
     /** \returns the size of a side of the respective square matrix, i.e., the number of indices */
-    inline EIGEN_DEVICE_FUNC Index size() const { return Index(indices().size()); }
+    inline Index size() const { return Index(indices().size()); }
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename DenseDerived>

Eigen/src/Core/PlainObjectBase.h

@@ -13,10 +13,10 @@
 
 #if defined(EIGEN_INITIALIZE_MATRICES_BY_ZERO)
 # define EIGEN_INITIALIZE_COEFFS
-# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(Index i=0;i<base().size();++i) coeffRef(i)=Scalar(0);
+# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(int i=0;i<base().size();++i) coeffRef(i)=Scalar(0);
 #elif defined(EIGEN_INITIALIZE_MATRICES_BY_NAN)
 # define EIGEN_INITIALIZE_COEFFS
-# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(Index i=0;i<base().size();++i) coeffRef(i)=std::numeric_limits<Scalar>::quiet_NaN();
+# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(int i=0;i<base().size();++i) coeffRef(i)=std::numeric_limits<Scalar>::quiet_NaN();
 #else
 # undef EIGEN_INITIALIZE_COEFFS
 # define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
@@ -118,8 +118,16 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
 
+    template<typename PlainObjectType, int MapOptions, typename StrideType> friend class Eigen::Map;
+    friend  class Eigen::Map<Derived, Unaligned>;
     typedef Eigen::Map<Derived, Unaligned>  MapType;
+    friend  class Eigen::Map<const Derived, Unaligned>;
     typedef const Eigen::Map<const Derived, Unaligned> ConstMapType;
+#if EIGEN_MAX_ALIGN_BYTES>0
+    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
+    friend  class Eigen::Map<Derived, AlignedMax>;
+    friend  class Eigen::Map<const Derived, AlignedMax>;
+#endif
     typedef Eigen::Map<Derived, AlignedMax> AlignedMapType;
     typedef const Eigen::Map<const Derived, AlignedMax> ConstAlignedMapType;
     template<typename StrideType> struct StridedMapType { typedef Eigen::Map<Derived, Unaligned, StrideType> type; };
@@ -139,10 +147,10 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     EIGEN_DEVICE_FUNC
     const Base& base() const { return *static_cast<const Base*>(this); }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index rows() const EIGEN_NOEXCEPT { return m_storage.rows(); }
+    EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index cols() const EIGEN_NOEXCEPT { return m_storage.cols(); }
+    EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
 
     /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index,Index) const
       * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
@@ -500,8 +508,8 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     EIGEN_DEVICE_FUNC
     PlainObjectBase& operator=(PlainObjectBase&& other) EIGEN_NOEXCEPT
     {
-      _check_template_params();
+      using std::swap;
-      m_storage = std::move(other.m_storage);
+      swap(m_storage, other.m_storage);
       return *this;
     }
 #endif
@@ -518,71 +526,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
 //       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
-    #if EIGEN_HAS_CXX11
-    /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11
-      *
-      * \only_for_vectors
-      *
-      * This constructor is for 1D array or vectors with more than 4 coefficients.
-      * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients.
-      *
-      * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this
-      * constructor must match the the fixed number of rows (resp. columns) of \c *this.
-      */
-    template <typename... ArgTypes>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    PlainObjectBase(const Scalar& a0, const Scalar& a1, const Scalar& a2,  const Scalar& a3, const ArgTypes&... args)
-      : m_storage()
-    {
-      _check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, sizeof...(args) + 4);
-      m_storage.data()[0] = a0;
-      m_storage.data()[1] = a1;
-      m_storage.data()[2] = a2;
-      m_storage.data()[3] = a3;
-      Index i = 4;
-      auto x = {(m_storage.data()[i++] = args, 0)...};
-      static_cast<void>(x);
-    }
-
-    /** \brief Constructs a Matrix or Array and initializes it by elements given by an initializer list of initializer
-      * lists \cpp11
-      */
-    EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE PlainObjectBase(const std::initializer_list<std::initializer_list<Scalar>>& list)
-      : m_storage()
-    {
-      _check_template_params();
-
-      size_t list_size = 0;
-      if (list.begin() != list.end()) {
-        list_size = list.begin()->size();
-      }
-
-      // This is to allow syntax like VectorXi {{1, 2, 3, 4}}
-      if (ColsAtCompileTime == 1 && list.size() == 1) {
-        eigen_assert(list_size == static_cast<size_t>(RowsAtCompileTime) || RowsAtCompileTime == Dynamic);
-        resize(list_size, ColsAtCompileTime);
-        std::copy(list.begin()->begin(), list.begin()->end(), m_storage.data());
-      } else {
-        eigen_assert(list.size() == static_cast<size_t>(RowsAtCompileTime) || RowsAtCompileTime == Dynamic);
-        eigen_assert(list_size == static_cast<size_t>(ColsAtCompileTime) || ColsAtCompileTime == Dynamic);
-        resize(list.size(), list_size);
-
-        Index row_index = 0;
-        for (const std::initializer_list<Scalar>& row : list) {
-          eigen_assert(list_size == row.size());
-          Index col_index = 0;
-          for (const Scalar& e : row) {
-            coeffRef(row_index, col_index) = e;
-            ++col_index;
-          }
-          ++row_index;
-        }
-      }
-    }
-    #endif  // end EIGEN_HAS_CXX11
-
     /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC
@@ -709,26 +652,18 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     using Base::setConstant;
     EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& val);
     EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& val);
-    EIGEN_DEVICE_FUNC Derived& setConstant(NoChange_t, Index cols, const Scalar& val);
-    EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, NoChange_t, const Scalar& val);
 
     using Base::setZero;
     EIGEN_DEVICE_FUNC Derived& setZero(Index size);
     EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols);
-    EIGEN_DEVICE_FUNC Derived& setZero(NoChange_t, Index cols);
-    EIGEN_DEVICE_FUNC Derived& setZero(Index rows, NoChange_t);
 
     using Base::setOnes;
     EIGEN_DEVICE_FUNC Derived& setOnes(Index size);
     EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols);
-    EIGEN_DEVICE_FUNC Derived& setOnes(NoChange_t, Index cols);
-    EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, NoChange_t);
 
     using Base::setRandom;
     Derived& setRandom(Index size);
     Derived& setRandom(Index rows, Index cols);
-    Derived& setRandom(NoChange_t, Index cols);
-    Derived& setRandom(Index rows, NoChange_t);
 
     #ifdef EIGEN_PLAINOBJECTBASE_PLUGIN
     #include EIGEN_PLAINOBJECTBASE_PLUGIN
@@ -770,7 +705,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       *
       * \internal
       */
-    // aliasing is dealt once in internal::call_assignment
+    // aliasing is dealt once in internall::call_assignment
     // so at this stage we have to assume aliasing... and resising has to be done later.
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC 
@@ -847,7 +782,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       resize(size);
     }
     
-    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitly converted)
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
     template<typename T>
     EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
@@ -949,7 +884,7 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * of same type it is enough to swap the data pointers.
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     void swap(DenseBase<OtherDerived> & other)
     {
       enum { SwapPointers = internal::is_same<Derived, OtherDerived>::value && Base::SizeAtCompileTime==Dynamic };
@@ -960,15 +895,15 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
       * \brief const version forwarded to DenseBase::swap
       */
     template<typename OtherDerived>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     void swap(DenseBase<OtherDerived> const & other)
     { Base::swap(other.derived()); }
     
     EIGEN_DEVICE_FUNC 
     static EIGEN_STRONG_INLINE void _check_template_params()
     {
-      EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (int(Options)&RowMajor)==RowMajor)
+      EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
-                        && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (int(Options)&RowMajor)==0)
+                        && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (Options&RowMajor)==0)
                         && ((RowsAtCompileTime == Dynamic) || (RowsAtCompileTime >= 0))
                         && ((ColsAtCompileTime == Dynamic) || (ColsAtCompileTime >= 0))
                         && ((MaxRowsAtCompileTime == Dynamic) || (MaxRowsAtCompileTime >= 0))
@@ -980,17 +915,6 @@ class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
     }
 
     enum { IsPlainObjectBase = 1 };
-#endif
-  public:
-    // These apparently need to be down here for nvcc+icc to prevent duplicate
-    // Map symbol.
-    template<typename PlainObjectType, int MapOptions, typename StrideType> friend class Eigen::Map;
-    friend class Eigen::Map<Derived, Unaligned>;
-    friend class Eigen::Map<const Derived, Unaligned>;
-#if EIGEN_MAX_ALIGN_BYTES>0
-    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
-    friend class Eigen::Map<Derived, AlignedMax>;
-    friend class Eigen::Map<const Derived, AlignedMax>;
 #endif
 };
 
@@ -999,19 +923,13 @@ namespace internal {
 template <typename Derived, typename OtherDerived, bool IsVector>
 struct conservative_resize_like_impl
 {
-  #if EIGEN_HAS_TYPE_TRAITS
-  static const bool IsRelocatable = std::is_trivially_copyable<typename Derived::Scalar>::value;
-  #else
-  static const bool IsRelocatable = !NumTraits<typename Derived::Scalar>::RequireInitialization;
-  #endif
   static void run(DenseBase<Derived>& _this, Index rows, Index cols)
   {
     if (_this.rows() == rows && _this.cols() == cols) return;
     EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
 
-    if ( IsRelocatable
+    if ( ( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows
-          && (( Derived::IsRowMajor && _this.cols() == cols) ||  // row-major and we change only the number of rows
+         (!Derived::IsRowMajor && _this.rows() == rows) )  // column-major and we change only the number of columns
-              (!Derived::IsRowMajor && _this.rows() == rows) ))  // column-major and we change only the number of columns
     {
       internal::check_rows_cols_for_overflow<Derived::MaxSizeAtCompileTime>::run(rows, cols);
       _this.derived().m_storage.conservativeResize(rows*cols,rows,cols);
@@ -1019,7 +937,7 @@ struct conservative_resize_like_impl
     else
     {
       // The storage order does not allow us to use reallocation.
-      Derived tmp(rows,cols);
+      typename Derived::PlainObject tmp(rows,cols);
       const Index common_rows = numext::mini(rows, _this.rows());
       const Index common_cols = numext::mini(cols, _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
@@ -1039,9 +957,8 @@ struct conservative_resize_like_impl
     EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived)
     EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(OtherDerived)
 
-    if ( IsRelocatable &&
+    if ( ( Derived::IsRowMajor && _this.cols() == other.cols()) || // row-major and we change only the number of rows
-          (( Derived::IsRowMajor && _this.cols() == other.cols()) ||  // row-major and we change only the number of rows
+         (!Derived::IsRowMajor && _this.rows() == other.rows()) )  // column-major and we change only the number of columns
-           (!Derived::IsRowMajor && _this.rows() == other.rows()) ))  // column-major and we change only the number of columns
     {
       const Index new_rows = other.rows() - _this.rows();
       const Index new_cols = other.cols() - _this.cols();
@@ -1054,7 +971,7 @@ struct conservative_resize_like_impl
     else
     {
       // The storage order does not allow us to use reallocation.
-      Derived tmp(other);
+      typename Derived::PlainObject tmp(other);
       const Index common_rows = numext::mini(tmp.rows(), _this.rows());
       const Index common_cols = numext::mini(tmp.cols(), _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
@@ -1069,18 +986,13 @@ template <typename Derived, typename OtherDerived>
 struct conservative_resize_like_impl<Derived,OtherDerived,true>
   : conservative_resize_like_impl<Derived,OtherDerived,false>
 {
-  typedef conservative_resize_like_impl<Derived,OtherDerived,false> Base;
+  using conservative_resize_like_impl<Derived,OtherDerived,false>::run;
-  using Base::run;
-  using Base::IsRelocatable;
   
   static void run(DenseBase<Derived>& _this, Index size)
   {
     const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
     const Index new_cols = Derived::RowsAtCompileTime==1 ? size : 1;
-    if(IsRelocatable)
+    _this.derived().m_storage.conservativeResize(size,new_rows,new_cols);
-      _this.derived().m_storage.conservativeResize(size,new_rows,new_cols);
-    else
-      Base::run(_this.derived(), new_rows, new_cols);
   }
 
   static void run(DenseBase<Derived>& _this, const DenseBase<OtherDerived>& other)
@@ -1091,10 +1003,7 @@ struct conservative_resize_like_impl<Derived,OtherDerived,true>
 
     const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : other.rows();
     const Index new_cols = Derived::RowsAtCompileTime==1 ? other.cols() : 1;
-    if(IsRelocatable)
+    _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols);
-      _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols);
-    else
-      Base::run(_this.derived(), new_rows, new_cols);
 
     if (num_new_elements > 0)
       _this.tail(num_new_elements) = other.tail(num_new_elements);
@@ -1105,7 +1014,7 @@ template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
 struct matrix_swap_impl
 {
   EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE void run(MatrixTypeA& a, MatrixTypeB& b)
+  static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     a.base().swap(b);
   }

Eigen/src/Core/Product.h

@@ -90,23 +90,18 @@ class Product : public ProductImpl<_Lhs,_Rhs,Option,
     typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
     typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
-    Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
     {
       eigen_assert(lhs.cols() == rhs.rows()
         && "invalid matrix product"
         && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    Index rows() const EIGEN_NOEXCEPT { return m_lhs.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
-    const LhsNestedCleaned& lhs() const { return m_lhs; }
+    EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const RhsNestedCleaned& rhs() const { return m_rhs; }
 
   protected:
 
@@ -121,7 +116,7 @@ class dense_product_base
  : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
 {};
 
-/** Conversion to scalar for inner-products */
+/** Convertion to scalar for inner-products */
 template<typename Lhs, typename Rhs, int Option>
 class dense_product_base<Lhs, Rhs, Option, InnerProduct>
  : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
@@ -132,7 +127,7 @@ public:
   using Base::derived;
   typedef typename Base::Scalar Scalar;
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE operator const Scalar() const
+  EIGEN_STRONG_INLINE operator const Scalar() const
   {
     return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
   }

Eigen/src/Core/ProductEvaluators.h

@@ -20,7 +20,7 @@ namespace internal {
 /** \internal
   * Evaluator of a product expression.
   * Since products require special treatments to handle all possible cases,
-  * we simply defer the evaluation logic to a product_evaluator class
+  * we simply deffer the evaluation logic to a product_evaluator class
   * which offers more partial specialization possibilities.
   *
   * \sa class product_evaluator
@@ -128,7 +128,7 @@ protected:
   PlainObject m_result;
 };
 
-// The following three shortcuts are enabled only if the scalar types match exactly.
+// The following three shortcuts are enabled only if the scalar types match excatly.
 // TODO: we could enable them for different scalar types when the product is not vectorized.
 
 // Dense = Product
@@ -137,7 +137,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::assign_op<Scal
   typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
 {
   typedef Product<Lhs,Rhs,Options> SrcXprType;
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
   void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
   {
     Index dstRows = src.rows();
@@ -155,7 +155,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::add_assign_op<
   typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
 {
   typedef Product<Lhs,Rhs,Options> SrcXprType;
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
   void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,Scalar> &)
   {
     eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
@@ -170,7 +170,7 @@ struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::sub_assign_op<
   typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
 {
   typedef Product<Lhs,Rhs,Options> SrcXprType;
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
   void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,Scalar> &)
   {
     eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
@@ -190,7 +190,7 @@ struct Assignment<DstXprType, CwiseBinaryOp<internal::scalar_product_op<ScalarBi
   typedef CwiseBinaryOp<internal::scalar_product_op<ScalarBis,Scalar>,
                         const CwiseNullaryOp<internal::scalar_constant_op<ScalarBis>,Plain>,
                         const Product<Lhs,Rhs,DefaultProduct> > SrcXprType;
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
   void run(DstXprType &dst, const SrcXprType &src, const AssignFunc& func)
   {
     call_assignment_no_alias(dst, (src.lhs().functor().m_other * src.rhs().lhs())*src.rhs().rhs(), func);
@@ -217,7 +217,7 @@ template<typename DstXprType, typename OtherXpr, typename ProductType, typename
 struct assignment_from_xpr_op_product
 {
   template<typename SrcXprType, typename InitialFunc>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  static EIGEN_STRONG_INLINE
   void run(DstXprType &dst, const SrcXprType &src, const InitialFunc& /*func*/)
   {
     call_assignment_no_alias(dst, src.lhs(), Func1());
@@ -246,19 +246,19 @@ template<typename Lhs, typename Rhs>
 struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,InnerProduct>
 {
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     dst.coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     dst.coeffRef(0,0) += (lhs.transpose().cwiseProduct(rhs)).sum();
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   { dst.coeffRef(0,0) -= (lhs.transpose().cwiseProduct(rhs)).sum(); }
 };
 
@@ -269,10 +269,10 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,InnerProduct>
 
 // Column major result
 template<typename Dst, typename Lhs, typename Rhs, typename Func>
-void EIGEN_DEVICE_FUNC outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&)
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&)
 {
   evaluator<Rhs> rhsEval(rhs);
-  ei_declare_local_nested_eval(Lhs,lhs,Rhs::SizeAtCompileTime,actual_lhs);
+  typename nested_eval<Lhs,Rhs::SizeAtCompileTime>::type actual_lhs(lhs);
   // FIXME if cols is large enough, then it might be useful to make sure that lhs is sequentially stored
   // FIXME not very good if rhs is real and lhs complex while alpha is real too
   const Index cols = dst.cols();
@@ -282,10 +282,10 @@ void EIGEN_DEVICE_FUNC outer_product_selector_run(Dst& dst, const Lhs &lhs, cons
 
 // Row major result
 template<typename Dst, typename Lhs, typename Rhs, typename Func>
-void EIGEN_DEVICE_FUNC outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&)
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&)
 {
   evaluator<Lhs> lhsEval(lhs);
-  ei_declare_local_nested_eval(Rhs,rhs,Lhs::SizeAtCompileTime,actual_rhs);
+  typename nested_eval<Rhs,Lhs::SizeAtCompileTime>::type actual_rhs(rhs);
   // FIXME if rows is large enough, then it might be useful to make sure that rhs is sequentially stored
   // FIXME not very good if lhs is real and rhs complex while alpha is real too
   const Index rows = dst.rows();
@@ -300,37 +300,37 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,OuterProduct>
   typedef typename Product<Lhs,Rhs>::Scalar Scalar;
 
   // TODO it would be nice to be able to exploit our *_assign_op functors for that purpose
-  struct set  { template<typename Dst, typename Src> EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
+  struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
-  struct add  { template<typename Dst, typename Src> EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
+  struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
-  struct sub  { template<typename Dst, typename Src> EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
+  struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
   struct adds {
     Scalar m_scale;
     explicit adds(const Scalar& s) : m_scale(s) {}
-    template<typename Dst, typename Src> void EIGEN_DEVICE_FUNC operator()(const Dst& dst, const Src& src) const {
+    template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
       dst.const_cast_derived() += m_scale * src;
     }
   };
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     internal::outer_product_selector_run(dst, lhs, rhs, set(), is_row_major<Dst>());
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     internal::outer_product_selector_run(dst, lhs, rhs, add(), is_row_major<Dst>());
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     internal::outer_product_selector_run(dst, lhs, rhs, sub(), is_row_major<Dst>());
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
   {
     internal::outer_product_selector_run(dst, lhs, rhs, adds(alpha), is_row_major<Dst>());
   }
@@ -345,19 +345,19 @@ struct generic_product_impl_base
   typedef typename Product<Lhs,Rhs>::Scalar Scalar;
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   { dst.setZero(); scaleAndAddTo(dst, lhs, rhs, Scalar(1)); }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   { scaleAndAddTo(dst,lhs, rhs, Scalar(1)); }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   { scaleAndAddTo(dst, lhs, rhs, Scalar(-1)); }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
   { Derived::scaleAndAddTo(dst,lhs,rhs,alpha); }
 
 };
@@ -373,13 +373,8 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct>
   typedef typename internal::remove_all<typename internal::conditional<int(Side)==OnTheRight,LhsNested,RhsNested>::type>::type MatrixType;
 
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
   {
-    // Fallback to inner product if both the lhs and rhs is a runtime vector.
-    if (lhs.rows() == 1 && rhs.cols() == 1) {
-      dst.coeffRef(0,0) += alpha * lhs.row(0).conjugate().dot(rhs.col(0));
-      return;
-    }
     LhsNested actual_lhs(lhs);
     RhsNested actual_rhs(rhs);
     internal::gemv_dense_selector<Side,
@@ -403,71 +398,48 @@ struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode>
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     // dst.noalias() += lhs.lazyProduct(rhs);
     call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::add_assign_op<typename Dst::Scalar,Scalar>());
   }
 
   template<typename Dst>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
   {
     // dst.noalias() -= lhs.lazyProduct(rhs);
     call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op<typename Dst::Scalar,Scalar>());
   }
 
-  // This is a special evaluation path called from generic_product_impl<...,GemmProduct> in file GeneralMatrixMatrix.h
+  // Catch "dst {,+,-}= (s*A)*B" and evaluate it lazily by moving out the scalar factor:
-  // This variant tries to extract scalar multiples from both the LHS and RHS and factor them out. For instance:
+  //    dst {,+,-}= s * (A.lazyProduct(B))
-  //   dst {,+,-}= (s1*A)*(B*s2)
+  // This is a huge benefit for heap-allocated matrix types as it save one costly allocation.
-  // will be rewritten as:
+  // For them, this strategy is also faster than simply by-passing the heap allocation through
-  //   dst {,+,-}= (s1*s2) * (A.lazyProduct(B))
+  // stack allocation.
-  // There are at least four benefits of doing so:
+  // For fixed sizes matrices, this is less obvious, it is sometimes x2 faster, but sometimes x3 slower,
-  //  1 - huge performance gain for heap-allocated matrix types as it save costly allocations.
+  // and the behavior depends also a lot on the compiler... so let's be conservative and enable them for dynamic-size only,
-  //  2 - it is faster than simply by-passing the heap allocation through stack allocation.
+  // that is when coming from generic_product_impl<...,GemmProduct> in file GeneralMatrixMatrix.h
-  //  3 - it makes this fallback consistent with the heavy GEMM routine.
+  template<typename Dst, typename Scalar1, typename Scalar2, typename Plain1, typename Xpr2, typename Func>
-  //  4 - it fully by-passes huge stack allocation attempts when multiplying huge fixed-size matrices.
-  //      (see https://stackoverflow.com/questions/54738495)
-  // For small fixed sizes matrices, howver, the gains are less obvious, it is sometimes x2 faster, but sometimes x3 slower,
-  // and the behavior depends also a lot on the compiler... This is why this re-writting strategy is currently
-  // enabled only when falling back from the main GEMM.
-  template<typename Dst, typename Func>
   static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  void eval_dynamic(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Func &func)
+  void eval_dynamic(Dst& dst, const CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                                           const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>, Xpr2>& lhs, const Rhs& rhs, const Func &func)
   {
-    enum {
+    call_assignment_no_alias(dst, lhs.lhs().functor().m_other * lhs.rhs().lazyProduct(rhs), func);
-      HasScalarFactor = blas_traits<Lhs>::HasScalarFactor || blas_traits<Rhs>::HasScalarFactor,
-      ConjLhs = blas_traits<Lhs>::NeedToConjugate,
-      ConjRhs = blas_traits<Rhs>::NeedToConjugate
-    };
-    // FIXME: in c++11 this should be auto, and extractScalarFactor should also return auto
-    //        this is important for real*complex_mat
-    Scalar actualAlpha = combine_scalar_factors<Scalar>(lhs, rhs);
-
-    eval_dynamic_impl(dst,
-                      blas_traits<Lhs>::extract(lhs).template conjugateIf<ConjLhs>(),
-                      blas_traits<Rhs>::extract(rhs).template conjugateIf<ConjRhs>(),
-                      func,
-                      actualAlpha,
-                      typename conditional<HasScalarFactor,true_type,false_type>::type());
   }
 
-protected:
+  // Here, we we always have LhsT==Lhs, but we need to make it a template type to make the above
-
+  // overload more specialized.
-  template<typename Dst, typename LhsT, typename RhsT, typename Func, typename Scalar>
+  template<typename Dst, typename LhsT, typename Func>
   static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-  void eval_dynamic_impl(Dst& dst, const LhsT& lhs, const RhsT& rhs, const Func &func, const Scalar&  s /* == 1 */, false_type)
+  void eval_dynamic(Dst& dst, const LhsT& lhs, const Rhs& rhs, const Func &func)
   {
-    EIGEN_UNUSED_VARIABLE(s);
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), func);
-    eigen_internal_assert(s==Scalar(1));
-    call_restricted_packet_assignment_no_alias(dst, lhs.lazyProduct(rhs), func);
   }
 
-  template<typename Dst, typename LhsT, typename RhsT, typename Func, typename Scalar>
+
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+//   template<typename Dst>
-  void eval_dynamic_impl(Dst& dst, const LhsT& lhs, const RhsT& rhs, const Func &func, const Scalar& s, true_type)
+//   static inline void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
-  {
+//   { dst.noalias() += alpha * lhs.lazyProduct(rhs); }
-    call_restricted_packet_assignment_no_alias(dst, s * lhs.lazyProduct(rhs), func);
-  }
 };
 
 // This specialization enforces the use of a coefficient-based evaluation strategy
@@ -549,7 +521,7 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
     RhsCoeffReadCost = RhsEtorType::CoeffReadCost,
     CoeffReadCost = InnerSize==0 ? NumTraits<Scalar>::ReadCost
                   : InnerSize == Dynamic ? HugeCost
-                    : InnerSize * (NumTraits<Scalar>::MulCost + int(LhsCoeffReadCost) + int(RhsCoeffReadCost))
+                  : InnerSize * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
                     + (InnerSize - 1) * NumTraits<Scalar>::AddCost,
 
     Unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT,
@@ -576,7 +548,7 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
                     : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
                     : (bool(RhsRowMajor) && !CanVectorizeLhs),
 
-    Flags = ((int(LhsFlags) | int(RhsFlags)) & HereditaryBits & ~RowMajorBit)
+    Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & ~RowMajorBit)
           | (EvalToRowMajor ? RowMajorBit : 0)
           // TODO enable vectorization for mixed types
           | (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0)
@@ -597,8 +569,8 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
     CanVectorizeInner =    SameType
                         && LhsRowMajor
                         && (!RhsRowMajor)
-                        && (int(LhsFlags) & int(RhsFlags) & ActualPacketAccessBit)
+                        && (LhsFlags & RhsFlags & ActualPacketAccessBit)
-                        && (int(InnerSize) % packet_traits<Scalar>::size == 0)
+                        && (InnerSize % packet_traits<Scalar>::size == 0)
   };
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index row, Index col) const
@@ -610,8 +582,7 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
    * which is why we don't set the LinearAccessBit.
    * TODO: this seems possible when the result is a vector
    */
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC const CoeffReturnType coeff(Index index) const
-  const CoeffReturnType coeff(Index index) const
   {
     const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
     const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0;
@@ -619,7 +590,6 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
   }
 
   template<int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   const PacketType packet(Index row, Index col) const
   {
     PacketType res;
@@ -631,7 +601,6 @@ struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape,
   }
 
   template<int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
   const PacketType packet(Index index) const
   {
     const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
@@ -660,8 +629,7 @@ struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, LazyCoeffBasedProduc
   enum {
     Flags = Base::Flags | EvalBeforeNestingBit
   };
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
-  explicit product_evaluator(const XprType& xpr)
     : Base(BaseProduct(xpr.lhs(),xpr.rhs()))
   {}
 };
@@ -673,7 +641,7 @@ struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, LazyCoeffBasedProduc
 template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
   {
     etor_product_packet_impl<RowMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
     res =  pmadd(pset1<Packet>(lhs.coeff(row, Index(UnrollingIndex-1))), rhs.template packet<LoadMode,Packet>(Index(UnrollingIndex-1), col), res);
@@ -683,7 +651,7 @@ struct etor_product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, Load
 template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
   {
     etor_product_packet_impl<ColMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
     res =  pmadd(lhs.template packet<LoadMode,Packet>(row, Index(UnrollingIndex-1)), pset1<Packet>(rhs.coeff(Index(UnrollingIndex-1), col)), res);
@@ -693,7 +661,7 @@ struct etor_product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, Load
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<RowMajor, 1, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
   {
     res = pmul(pset1<Packet>(lhs.coeff(row, Index(0))),rhs.template packet<LoadMode,Packet>(Index(0), col));
   }
@@ -702,7 +670,7 @@ struct etor_product_packet_impl<RowMajor, 1, Lhs, Rhs, Packet, LoadMode>
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<ColMajor, 1, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
   {
     res = pmul(lhs.template packet<LoadMode,Packet>(row, Index(0)), pset1<Packet>(rhs.coeff(Index(0), col)));
   }
@@ -711,7 +679,7 @@ struct etor_product_packet_impl<ColMajor, 1, Lhs, Rhs, Packet, LoadMode>
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
   {
     res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
   }
@@ -720,7 +688,7 @@ struct etor_product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
   {
     res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
   }
@@ -729,7 +697,7 @@ struct etor_product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
   {
     res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
     for(Index i = 0; i < innerDim; ++i)
@@ -740,7 +708,7 @@ struct etor_product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
 template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
 struct etor_product_packet_impl<ColMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
 {
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
   {
     res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
     for(Index i = 0; i < innerDim; ++i)
@@ -799,8 +767,7 @@ struct generic_product_impl<Lhs,Rhs,SelfAdjointShape,DenseShape,ProductTag>
   typedef typename Product<Lhs,Rhs>::Scalar Scalar;
 
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
-  void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
   {
     selfadjoint_product_impl<typename Lhs::MatrixType,Lhs::Mode,false,Rhs,0,Rhs::IsVectorAtCompileTime>::run(dst, lhs.nestedExpression(), rhs, alpha);
   }
@@ -831,25 +798,17 @@ struct diagonal_product_evaluator_base
    typedef typename ScalarBinaryOpTraits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
 public:
   enum {
-    CoeffReadCost = int(NumTraits<Scalar>::MulCost) + int(evaluator<MatrixType>::CoeffReadCost) + int(evaluator<DiagonalType>::CoeffReadCost),
+    CoeffReadCost = NumTraits<Scalar>::MulCost + evaluator<MatrixType>::CoeffReadCost + evaluator<DiagonalType>::CoeffReadCost,
 
     MatrixFlags = evaluator<MatrixType>::Flags,
     DiagFlags = evaluator<DiagonalType>::Flags,
-
+    _StorageOrder = MatrixFlags & RowMajorBit ? RowMajor : ColMajor,
-    _StorageOrder = (Derived::MaxRowsAtCompileTime==1 && Derived::MaxColsAtCompileTime!=1) ? RowMajor
-                  : (Derived::MaxColsAtCompileTime==1 && Derived::MaxRowsAtCompileTime!=1) ? ColMajor
-                  : MatrixFlags & RowMajorBit ? RowMajor : ColMajor,
-    _SameStorageOrder = _StorageOrder == (MatrixFlags & RowMajorBit ? RowMajor : ColMajor),
-
     _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
                            ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
     _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
     // FIXME currently we need same types, but in the future the next rule should be the one
     //_Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagFlags)&PacketAccessBit))),
-    _Vectorizable =   bool(int(MatrixFlags)&PacketAccessBit)
+    _Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagFlags)&PacketAccessBit))),
-                  &&  _SameTypes
-                  && (_SameStorageOrder || (MatrixFlags&LinearAccessBit)==LinearAccessBit)
-                  && (_ScalarAccessOnDiag || (bool(int(DiagFlags)&PacketAccessBit))),
     _LinearAccessMask = (MatrixType::RowsAtCompileTime==1 || MatrixType::ColsAtCompileTime==1) ? LinearAccessBit : 0,
     Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixFlags)) | (_Vectorizable ? PacketAccessBit : 0),
     Alignment = evaluator<MatrixType>::Alignment,
@@ -859,7 +818,7 @@ public:
                       ||  (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::ColsAtCompileTime==1 && ProductOrder==OnTheRight)
   };
 
-  EIGEN_DEVICE_FUNC diagonal_product_evaluator_base(const MatrixType &mat, const DiagonalType &diag)
+  diagonal_product_evaluator_base(const MatrixType &mat, const DiagonalType &diag)
     : m_diagImpl(diag), m_matImpl(mat)
   {
     EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::MulCost);
@@ -910,10 +869,10 @@ struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DiagonalSha
 
   typedef Product<Lhs, Rhs, ProductKind> XprType;
   typedef typename XprType::PlainObject PlainObject;
-  typedef typename Lhs::DiagonalVectorType DiagonalType;
 
-
+  enum {
-  enum { StorageOrder = Base::_StorageOrder };
+    StorageOrder = int(Rhs::Flags) & RowMajorBit ? RowMajor : ColMajor
+  };
 
   EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
     : Base(xpr.rhs(), xpr.lhs().diagonal())
@@ -925,7 +884,7 @@ struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DiagonalSha
     return m_diagImpl.coeff(row) * m_matImpl.coeff(row, col);
   }
 
-#ifndef EIGEN_GPUCC
+#ifndef __CUDACC__
   template<int LoadMode,typename PacketType>
   EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
   {
@@ -957,7 +916,7 @@ struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DenseShape,
   typedef Product<Lhs, Rhs, ProductKind> XprType;
   typedef typename XprType::PlainObject PlainObject;
 
-  enum { StorageOrder = Base::_StorageOrder };
+  enum { StorageOrder = int(Lhs::Flags) & RowMajorBit ? RowMajor : ColMajor };
 
   EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
     : Base(xpr.lhs(), xpr.rhs().diagonal())
@@ -969,7 +928,7 @@ struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DenseShape,
     return m_matImpl.coeff(row, col) * m_diagImpl.coeff(col);
   }
 
-#ifndef EIGEN_GPUCC
+#ifndef __CUDACC__
   template<int LoadMode,typename PacketType>
   EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
   {
@@ -1004,7 +963,7 @@ struct permutation_matrix_product<ExpressionType, Side, Transposed, DenseShape>
     typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
 
     template<typename Dest, typename PermutationType>
-    static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
+    static inline void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
     {
       MatrixType mat(xpr);
       const Index n = Side==OnTheLeft ? mat.rows() : mat.cols();
@@ -1058,7 +1017,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Rhs, PermutationShape, MatrixShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
   {
     permutation_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
   }
@@ -1068,7 +1027,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Rhs, MatrixShape, PermutationShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
   {
     permutation_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
   }
@@ -1078,7 +1037,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Inverse<Lhs>, Rhs, PermutationShape, MatrixShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Inverse<Lhs>& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Inverse<Lhs>& lhs, const Rhs& rhs)
   {
     permutation_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
   }
@@ -1088,7 +1047,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Inverse<Rhs>, MatrixShape, PermutationShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Inverse<Rhs>& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Inverse<Rhs>& rhs)
   {
     permutation_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
   }
@@ -1112,7 +1071,7 @@ struct transposition_matrix_product
   typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
 
   template<typename Dest, typename TranspositionType>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Dest& dst, const TranspositionType& tr, const ExpressionType& xpr)
+  static inline void run(Dest& dst, const TranspositionType& tr, const ExpressionType& xpr)
   {
     MatrixType mat(xpr);
     typedef typename TranspositionType::StorageIndex StorageIndex;
@@ -1135,7 +1094,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Rhs, TranspositionsShape, MatrixShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
   {
     transposition_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
   }
@@ -1145,7 +1104,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Rhs, MatrixShape, TranspositionsShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
   {
     transposition_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
   }
@@ -1156,7 +1115,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Transpose<Lhs>, Rhs, TranspositionsShape, MatrixShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Transpose<Lhs>& lhs, const Rhs& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Transpose<Lhs>& lhs, const Rhs& rhs)
   {
     transposition_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
   }
@@ -1166,7 +1125,7 @@ template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
 struct generic_product_impl<Lhs, Transpose<Rhs>, MatrixShape, TranspositionsShape, ProductTag>
 {
   template<typename Dest>
-  static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Transpose<Rhs>& rhs)
+  static EIGEN_DEVICE_FUNC void evalTo(Dest& dst, const Lhs& lhs, const Transpose<Rhs>& rhs)
   {
     transposition_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
   }

Eigen/src/Core/Random.h

@@ -177,42 +177,6 @@ PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
   return setRandom();
 }
 
-/** Resizes to the given size, changing only the number of columns, and sets all
-  * coefficients in this expression to random values. For the parameter of type
-  * NoChange_t, just pass the special value \c NoChange.
-  *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  *
-  * \not_reentrant
-  *
-  * \sa DenseBase::setRandom(), setRandom(Index), setRandom(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Random()
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setRandom(NoChange_t, Index cols)
-{
-  return setRandom(rows(), cols);
-}
-
-/** Resizes to the given size, changing only the number of rows, and sets all
-  * coefficients in this expression to random values. For the parameter of type
-  * NoChange_t, just pass the special value \c NoChange.
-  *
-  * Numbers are uniformly spread through their whole definition range for integer types,
-  * and in the [-1:1] range for floating point scalar types.
-  *
-  * \not_reentrant
-  *
-  * \sa DenseBase::setRandom(), setRandom(Index), setRandom(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Random()
-  */
-template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setRandom(Index rows, NoChange_t)
-{
-  return setRandom(rows, cols());
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_RANDOM_H

Eigen/src/Core/Redux.h

@@ -23,29 +23,23 @@ namespace internal {
 * Part 1 : the logic deciding a strategy for vectorization and unrolling
 ***************************************************************************/
 
-template<typename Func, typename Evaluator>
+template<typename Func, typename Derived>
 struct redux_traits
 {
 public:
-    typedef typename find_best_packet<typename Evaluator::Scalar,Evaluator::SizeAtCompileTime>::type PacketType;
+    typedef typename find_best_packet<typename Derived::Scalar,Derived::SizeAtCompileTime>::type PacketType;
   enum {
     PacketSize = unpacket_traits<PacketType>::size,
-    InnerMaxSize = int(Evaluator::IsRowMajor)
+    InnerMaxSize = int(Derived::IsRowMajor)
-                 ? Evaluator::MaxColsAtCompileTime
+                 ? Derived::MaxColsAtCompileTime
-                 : Evaluator::MaxRowsAtCompileTime,
+                 : Derived::MaxRowsAtCompileTime
-    OuterMaxSize = int(Evaluator::IsRowMajor)
-                 ? Evaluator::MaxRowsAtCompileTime
-                 : Evaluator::MaxColsAtCompileTime,
-    SliceVectorizedWork = int(InnerMaxSize)==Dynamic ? Dynamic
-                        : int(OuterMaxSize)==Dynamic ? (int(InnerMaxSize)>=int(PacketSize) ? Dynamic : 0)
-                        : (int(InnerMaxSize)/int(PacketSize)) * int(OuterMaxSize)
   };
 
   enum {
-    MightVectorize = (int(Evaluator::Flags)&ActualPacketAccessBit)
+    MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit)
                   && (functor_traits<Func>::PacketAccess),
-    MayLinearVectorize = bool(MightVectorize) && (int(Evaluator::Flags)&LinearAccessBit),
+    MayLinearVectorize = bool(MightVectorize) && (int(Derived::Flags)&LinearAccessBit),
-    MaySliceVectorize  = bool(MightVectorize) && (int(SliceVectorizedWork)==Dynamic || int(SliceVectorizedWork)>=3)
+    MaySliceVectorize  = bool(MightVectorize) && int(InnerMaxSize)>=3*PacketSize
   };
 
 public:
@@ -57,8 +51,8 @@ public:
 
 public:
   enum {
-    Cost = Evaluator::SizeAtCompileTime == Dynamic ? HugeCost
+    Cost = Derived::SizeAtCompileTime == Dynamic ? HugeCost
-         : int(Evaluator::SizeAtCompileTime) * int(Evaluator::CoeffReadCost) + (Evaluator::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
+         : Derived::SizeAtCompileTime * Derived::CoeffReadCost + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
     UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
   };
 
@@ -70,20 +64,18 @@ public:
 #ifdef EIGEN_DEBUG_ASSIGN
   static void debug()
   {
-    std::cerr << "Xpr: " << typeid(typename Evaluator::XprType).name() << std::endl;
+    std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
     std::cerr.setf(std::ios::hex, std::ios::basefield);
-    EIGEN_DEBUG_VAR(Evaluator::Flags)
+    EIGEN_DEBUG_VAR(Derived::Flags)
     std::cerr.unsetf(std::ios::hex);
     EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(OuterMaxSize)
-    EIGEN_DEBUG_VAR(SliceVectorizedWork)
     EIGEN_DEBUG_VAR(PacketSize)
     EIGEN_DEBUG_VAR(MightVectorize)
     EIGEN_DEBUG_VAR(MayLinearVectorize)
     EIGEN_DEBUG_VAR(MaySliceVectorize)
-    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
+    EIGEN_DEBUG_VAR(Traversal)
     EIGEN_DEBUG_VAR(UnrollingLimit)
-    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
+    EIGEN_DEBUG_VAR(Unrolling)
     std::cerr << std::endl;
   }
 #endif
@@ -95,86 +87,88 @@ public:
 
 /*** no vectorization ***/
 
-template<typename Func, typename Evaluator, int Start, int Length>
+template<typename Func, typename Derived, int Start, int Length>
 struct redux_novec_unroller
 {
   enum {
     HalfLength = Length/2
   };
 
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
 
   EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func& func)
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
-    return func(redux_novec_unroller<Func, Evaluator, Start, HalfLength>::run(eval,func),
+    return func(redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
-                redux_novec_unroller<Func, Evaluator, Start+HalfLength, Length-HalfLength>::run(eval,func));
+                redux_novec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func));
   }
 };
 
-template<typename Func, typename Evaluator, int Start>
+template<typename Func, typename Derived, int Start>
-struct redux_novec_unroller<Func, Evaluator, Start, 1>
+struct redux_novec_unroller<Func, Derived, Start, 1>
 {
   enum {
-    outer = Start / Evaluator::InnerSizeAtCompileTime,
+    outer = Start / Derived::InnerSizeAtCompileTime,
-    inner = Start % Evaluator::InnerSizeAtCompileTime
+    inner = Start % Derived::InnerSizeAtCompileTime
   };
 
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
 
   EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func&)
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&)
   {
-    return eval.coeffByOuterInner(outer, inner);
+    return mat.coeffByOuterInner(outer, inner);
   }
 };
 
 // This is actually dead code and will never be called. It is required
 // to prevent false warnings regarding failed inlining though
 // for 0 length run() will never be called at all.
-template<typename Func, typename Evaluator, int Start>
+template<typename Func, typename Derived, int Start>
-struct redux_novec_unroller<Func, Evaluator, Start, 0>
+struct redux_novec_unroller<Func, Derived, Start, 0>
 {
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
   EIGEN_DEVICE_FUNC
-  static EIGEN_STRONG_INLINE Scalar run(const Evaluator&, const Func&) { return Scalar(); }
+  static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); }
 };
 
 /*** vectorization ***/
 
-template<typename Func, typename Evaluator, int Start, int Length>
+template<typename Func, typename Derived, int Start, int Length>
 struct redux_vec_unroller
 {
-  template<typename PacketType>
+  enum {
-  EIGEN_DEVICE_FUNC
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
-  static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func& func)
+    HalfLength = Length/2
-  {
+  };
-    enum {
-      PacketSize = unpacket_traits<PacketType>::size,
-      HalfLength = Length/2
-    };
 
+  typedef typename Derived::Scalar Scalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+
+  static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func)
+  {
     return func.packetOp(
-            redux_vec_unroller<Func, Evaluator, Start, HalfLength>::template run<PacketType>(eval,func),
+            redux_vec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
-            redux_vec_unroller<Func, Evaluator, Start+HalfLength, Length-HalfLength>::template run<PacketType>(eval,func) );
+            redux_vec_unroller<Func, Derived, Start+HalfLength, Length-HalfLength>::run(mat,func) );
   }
 };
 
-template<typename Func, typename Evaluator, int Start>
+template<typename Func, typename Derived, int Start>
-struct redux_vec_unroller<Func, Evaluator, Start, 1>
+struct redux_vec_unroller<Func, Derived, Start, 1>
 {
-  template<typename PacketType>
+  enum {
-  EIGEN_DEVICE_FUNC
+    index = Start * redux_traits<Func, Derived>::PacketSize,
-  static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func&)
+    outer = index / int(Derived::InnerSizeAtCompileTime),
+    inner = index % int(Derived::InnerSizeAtCompileTime),
+    alignment = Derived::Alignment
+  };
+
+  typedef typename Derived::Scalar Scalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
+
+  static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&)
   {
-    enum {
+    return mat.template packetByOuterInner<alignment,PacketScalar>(outer, inner);
-      PacketSize = unpacket_traits<PacketType>::size,
-      index = Start * PacketSize,
-      outer = index / int(Evaluator::InnerSizeAtCompileTime),
-      inner = index % int(Evaluator::InnerSizeAtCompileTime),
-      alignment = Evaluator::Alignment
-    };
-    return eval.template packetByOuterInner<alignment,PacketType>(outer, inner);
   }
 };
 
@@ -182,65 +176,53 @@ struct redux_vec_unroller<Func, Evaluator, Start, 1>
 * Part 3 : implementation of all cases
 ***************************************************************************/
 
-template<typename Func, typename Evaluator,
+template<typename Func, typename Derived,
-         int Traversal = redux_traits<Func, Evaluator>::Traversal,
+         int Traversal = redux_traits<Func, Derived>::Traversal,
-         int Unrolling = redux_traits<Func, Evaluator>::Unrolling
+         int Unrolling = redux_traits<Func, Derived>::Unrolling
 >
 struct redux_impl;
 
-template<typename Func, typename Evaluator>
+template<typename Func, typename Derived>
-struct redux_impl<Func, Evaluator, DefaultTraversal, NoUnrolling>
+struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 {
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
-
+  EIGEN_DEVICE_FUNC
-  template<typename XprType>
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
-  Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
   {
-    eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     Scalar res;
-    res = eval.coeffByOuterInner(0, 0);
+    res = mat.coeffByOuterInner(0, 0);
-    for(Index i = 1; i < xpr.innerSize(); ++i)
+    for(Index i = 1; i < mat.innerSize(); ++i)
-      res = func(res, eval.coeffByOuterInner(0, i));
+      res = func(res, mat.coeffByOuterInner(0, i));
-    for(Index i = 1; i < xpr.outerSize(); ++i)
+    for(Index i = 1; i < mat.outerSize(); ++i)
-      for(Index j = 0; j < xpr.innerSize(); ++j)
+      for(Index j = 0; j < mat.innerSize(); ++j)
-        res = func(res, eval.coeffByOuterInner(i, j));
+        res = func(res, mat.coeffByOuterInner(i, j));
     return res;
   }
 };
 
-template<typename Func, typename Evaluator>
+template<typename Func, typename Derived>
-struct redux_impl<Func,Evaluator, DefaultTraversal, CompleteUnrolling>
+struct redux_impl<Func,Derived, DefaultTraversal, CompleteUnrolling>
-  : redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime>
+  : public redux_novec_unroller<Func,Derived, 0, Derived::SizeAtCompileTime>
+{};
+
+template<typename Func, typename Derived>
+struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
 {
-  typedef redux_novec_unroller<Func,Evaluator, 0, Evaluator::SizeAtCompileTime> Base;
+  typedef typename Derived::Scalar Scalar;
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
-  template<typename XprType>
+
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
+  static Scalar run(const Derived &mat, const Func& func)
-  Scalar run(const Evaluator &eval, const Func& func, const XprType& /*xpr*/)
   {
-    return Base::run(eval,func);
+    const Index size = mat.size();
-  }
-};
 
-template<typename Func, typename Evaluator>
+    const Index packetSize = redux_traits<Func, Derived>::PacketSize;
-struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, NoUnrolling>
-{
-  typedef typename Evaluator::Scalar Scalar;
-  typedef typename redux_traits<Func, Evaluator>::PacketType PacketScalar;
-
-  template<typename XprType>
-  static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
-  {
-    const Index size = xpr.size();
-    
-    const Index packetSize = redux_traits<Func, Evaluator>::PacketSize;
     const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
     enum {
-      alignment0 = (bool(Evaluator::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
+      alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
-      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Evaluator::Alignment)
+      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
     };
-    const Index alignedStart = internal::first_default_aligned(xpr);
+    const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
     const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
     const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
     const Index alignedEnd2 = alignedStart + alignedSize2;
@@ -248,34 +230,34 @@ struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, NoUnrolling>
     Scalar res;
     if(alignedSize)
     {
-      PacketScalar packet_res0 = eval.template packet<alignment,PacketScalar>(alignedStart);
+      PacketScalar packet_res0 = mat.template packet<alignment,PacketScalar>(alignedStart);
       if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
       {
-        PacketScalar packet_res1 = eval.template packet<alignment,PacketScalar>(alignedStart+packetSize);
+        PacketScalar packet_res1 = mat.template packet<alignment,PacketScalar>(alignedStart+packetSize);
         for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
         {
-          packet_res0 = func.packetOp(packet_res0, eval.template packet<alignment,PacketScalar>(index));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(index));
-          packet_res1 = func.packetOp(packet_res1, eval.template packet<alignment,PacketScalar>(index+packetSize));
+          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment,PacketScalar>(index+packetSize));
         }
 
         packet_res0 = func.packetOp(packet_res0,packet_res1);
         if(alignedEnd>alignedEnd2)
-          packet_res0 = func.packetOp(packet_res0, eval.template packet<alignment,PacketScalar>(alignedEnd2));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(alignedEnd2));
       }
       res = func.predux(packet_res0);
 
       for(Index index = 0; index < alignedStart; ++index)
-        res = func(res,eval.coeff(index));
+        res = func(res,mat.coeff(index));
 
       for(Index index = alignedEnd; index < size; ++index)
-        res = func(res,eval.coeff(index));
+        res = func(res,mat.coeff(index));
     }
     else // too small to vectorize anything.
          // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
     {
-      res = eval.coeff(0);
+      res = mat.coeff(0);
       for(Index index = 1; index < size; ++index)
-        res = func(res,eval.coeff(index));
+        res = func(res,mat.coeff(index));
     }
 
     return res;
@@ -283,108 +265,130 @@ struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, NoUnrolling>
 };
 
 // NOTE: for SliceVectorizedTraversal we simply bypass unrolling
-template<typename Func, typename Evaluator, int Unrolling>
+template<typename Func, typename Derived, int Unrolling>
-struct redux_impl<Func, Evaluator, SliceVectorizedTraversal, Unrolling>
+struct redux_impl<Func, Derived, SliceVectorizedTraversal, Unrolling>
 {
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
-  typedef typename redux_traits<Func, Evaluator>::PacketType PacketType;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketType;
 
-  template<typename XprType>
+  EIGEN_DEVICE_FUNC static Scalar run(const Derived &mat, const Func& func)
-  EIGEN_DEVICE_FUNC static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr)
   {
-    eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    const Index innerSize = xpr.innerSize();
+    const Index innerSize = mat.innerSize();
-    const Index outerSize = xpr.outerSize();
+    const Index outerSize = mat.outerSize();
     enum {
-      packetSize = redux_traits<Func, Evaluator>::PacketSize
+      packetSize = redux_traits<Func, Derived>::PacketSize
     };
     const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
     Scalar res;
     if(packetedInnerSize)
     {
-      PacketType packet_res = eval.template packet<Unaligned,PacketType>(0,0);
+      PacketType packet_res = mat.template packet<Unaligned,PacketType>(0,0);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
-          packet_res = func.packetOp(packet_res, eval.template packetByOuterInner<Unaligned,PacketType>(j,i));
+          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned,PacketType>(j,i));
 
       res = func.predux(packet_res);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=packetedInnerSize; i<innerSize; ++i)
-          res = func(res, eval.coeffByOuterInner(j,i));
+          res = func(res, mat.coeffByOuterInner(j,i));
     }
     else // too small to vectorize anything.
          // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize.
     {
-      res = redux_impl<Func, Evaluator, DefaultTraversal, NoUnrolling>::run(eval, func, xpr);
+      res = redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>::run(mat, func);
     }
 
     return res;
   }
 };
 
-template<typename Func, typename Evaluator>
+template<typename Func, typename Derived>
-struct redux_impl<Func, Evaluator, LinearVectorizedTraversal, CompleteUnrolling>
+struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 {
-  typedef typename Evaluator::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar;
 
-  typedef typename redux_traits<Func, Evaluator>::PacketType PacketType;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
   enum {
-    PacketSize = redux_traits<Func, Evaluator>::PacketSize,
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
-    Size = Evaluator::SizeAtCompileTime,
+    Size = Derived::SizeAtCompileTime,
-    VectorizedSize = (int(Size) / int(PacketSize)) * int(PacketSize)
+    VectorizedSize = (Size / PacketSize) * PacketSize
   };
-
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
-  template<typename XprType>
-  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE
-  Scalar run(const Evaluator &eval, const Func& func, const XprType &xpr)
   {
-    EIGEN_ONLY_USED_FOR_DEBUG(xpr)
+    eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix");
     if (VectorizedSize > 0) {
-      Scalar res = func.predux(redux_vec_unroller<Func, Evaluator, 0, Size / PacketSize>::template run<PacketType>(eval,func));
+      Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
       if (VectorizedSize != Size)
-        res = func(res,redux_novec_unroller<Func, Evaluator, VectorizedSize, Size-VectorizedSize>::run(eval,func));
+        res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
       return res;
     }
     else {
-      return redux_novec_unroller<Func, Evaluator, 0, Size>::run(eval,func);
+      return redux_novec_unroller<Func, Derived, 0, Size>::run(mat,func);
     }
   }
 };
 
 // evaluator adaptor
 template<typename _XprType>
-class redux_evaluator : public internal::evaluator<_XprType>
+class redux_evaluator
 {
-  typedef internal::evaluator<_XprType> Base;
 public:
   typedef _XprType XprType;
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC explicit redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
-  explicit redux_evaluator(const XprType &xpr) : Base(xpr) {}
 
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
   typedef typename XprType::PacketScalar PacketScalar;
+  typedef typename XprType::PacketReturnType PacketReturnType;
 
   enum {
     MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
     // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
-    Flags = Base::Flags & ~DirectAccessBit,
+    Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
     IsRowMajor = XprType::IsRowMajor,
     SizeAtCompileTime = XprType::SizeAtCompileTime,
-    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime
+    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
+    CoeffReadCost = evaluator<XprType>::CoeffReadCost,
+    Alignment = evaluator<XprType>::Alignment
   };
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
-  { return Base::coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const { return m_xpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const { return m_xpr.outerSize(); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index index) const
+  { return m_evaluator.coeff(index); }
 
   template<int LoadMode, typename PacketType>
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
-  PacketType packetByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(row, col); }
-  { return Base::template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
 
+  template<int LoadMode, typename PacketType>
+  PacketType packet(Index index) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(index); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+
+  template<int LoadMode, typename PacketType>
+  PacketType packetByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+
+  const XprType & nestedExpression() const { return m_xpr; }
+
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
 };
 
 } // end namespace internal
@@ -399,8 +403,6 @@ public:
   * The template parameter \a BinaryOp is the type of the functor \a func which must be
   * an associative operator. Both current C++98 and C++11 functor styles are handled.
   *
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
-  *
   * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
   */
 template<typename Derived>
@@ -413,39 +415,27 @@ DenseBase<Derived>::redux(const Func& func) const
   typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
   ThisEvaluator thisEval(derived());
 
-  // The initial expression is passed to the reducer as an additional argument instead of
+  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
-  // passing it as a member of redux_evaluator to help  
-  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func, derived());
 }
 
 /** \returns the minimum of all coefficients of \c *this.
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  * \warning the result is undefined if \c *this contains NaN.
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is minimum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   */
 template<typename Derived>
-template<int NaNPropagation>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff() const
 {
-  return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar, NaNPropagation>());
+  return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar>());
 }
 
 /** \returns the maximum of all coefficients of \c *this.
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
+  * \warning the result is undefined if \c *this contains NaN.
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   */
 template<typename Derived>
-template<int NaNPropagation>
 EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff() const
 {
-  return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar, NaNPropagation>());
+  return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar>());
 }
 
 /** \returns the sum of all coefficients of \c *this
@@ -489,7 +479,7 @@ DenseBase<Derived>::mean() const
   * \sa sum(), mean(), trace()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC  EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::prod() const
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
@@ -504,7 +494,7 @@ DenseBase<Derived>::prod() const
   * \sa diagonal(), sum()
   */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC  EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 MatrixBase<Derived>::trace() const
 {
   return derived().diagonal().sum();

Eigen/src/Core/Ref.h

@@ -67,12 +67,12 @@ public:
   typedef MapBase<Derived> Base;
   EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const
+  EIGEN_DEVICE_FUNC inline Index innerStride() const
   {
     return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
   }
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const
+  EIGEN_DEVICE_FUNC inline Index outerStride() const
   {
     return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
          : IsVectorAtCompileTime ? this->size()
@@ -93,115 +93,29 @@ protected:
 
   typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
 
-  // Resolves inner stride if default 0.
-  static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveInnerStride(Index inner) {
-    return inner == 0 ? 1 : inner;
-  }
-
-  // Resolves outer stride if default 0.
-  static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveOuterStride(Index inner, Index outer, Index rows, Index cols, bool isVectorAtCompileTime, bool isRowMajor) {
-    return outer == 0 ? isVectorAtCompileTime ? inner * rows * cols : isRowMajor ? inner * cols : inner * rows : outer;
-  }
-
-  // Returns true if construction is valid, false if there is a stride mismatch,
-  // and fails if there is a size mismatch.
   template<typename Expression>
-  EIGEN_DEVICE_FUNC bool construct(Expression& expr)
+  EIGEN_DEVICE_FUNC void construct(Expression& expr)
   {
-    // Check matrix sizes.  If this is a compile-time vector, we do allow
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(PlainObjectType,Expression);
-    // implicitly transposing.
-    EIGEN_STATIC_ASSERT(
-      EIGEN_PREDICATE_SAME_MATRIX_SIZE(PlainObjectType, Expression)
-      // If it is a vector, the transpose sizes might match.
-      || ( PlainObjectType::IsVectorAtCompileTime
-            && ((int(PlainObjectType::RowsAtCompileTime)==Eigen::Dynamic
-              || int(Expression::ColsAtCompileTime)==Eigen::Dynamic
-              || int(PlainObjectType::RowsAtCompileTime)==int(Expression::ColsAtCompileTime))
-            &&  (int(PlainObjectType::ColsAtCompileTime)==Eigen::Dynamic
-              || int(Expression::RowsAtCompileTime)==Eigen::Dynamic
-              || int(PlainObjectType::ColsAtCompileTime)==int(Expression::RowsAtCompileTime)))),
-      YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES
-    )
 
-    // Determine runtime rows and columns.
-    Index rows = expr.rows();
-    Index cols = expr.cols();
     if(PlainObjectType::RowsAtCompileTime==1)
     {
       eigen_assert(expr.rows()==1 || expr.cols()==1);
-      rows = 1;
+      ::new (static_cast<Base*>(this)) Base(expr.data(), 1, expr.size());
-      cols = expr.size();
     }
     else if(PlainObjectType::ColsAtCompileTime==1)
     {
       eigen_assert(expr.rows()==1 || expr.cols()==1);
-      rows = expr.size();
+      ::new (static_cast<Base*>(this)) Base(expr.data(), expr.size(), 1);
-      cols = 1;
     }
-    // Verify that the sizes are valid.
+    else
-    eigen_assert(
+      ::new (static_cast<Base*>(this)) Base(expr.data(), expr.rows(), expr.cols());
-      (PlainObjectType::RowsAtCompileTime == Dynamic) || (PlainObjectType::RowsAtCompileTime == rows));
-    eigen_assert(
-      (PlainObjectType::ColsAtCompileTime == Dynamic) || (PlainObjectType::ColsAtCompileTime == cols));
     
-
+    if(Expression::IsVectorAtCompileTime && (!PlainObjectType::IsVectorAtCompileTime) && ((Expression::Flags&RowMajorBit)!=(PlainObjectType::Flags&RowMajorBit)))
-    // If this is a vector, we might be transposing, which means that stride should swap.
+      ::new (&m_stride) StrideBase(expr.innerStride(), StrideType::InnerStrideAtCompileTime==0?0:1);
-    const bool transpose = PlainObjectType::IsVectorAtCompileTime && (rows != expr.rows());
+    else
-    // If the storage format differs, we also need to swap the stride.
+      ::new (&m_stride) StrideBase(StrideType::OuterStrideAtCompileTime==0?0:expr.outerStride(),
-    const bool row_major = ((PlainObjectType::Flags)&RowMajorBit) != 0;
+                                   StrideType::InnerStrideAtCompileTime==0?0:expr.innerStride());    
-    const bool expr_row_major = (Expression::Flags&RowMajorBit) != 0;
-    const bool storage_differs =  (row_major != expr_row_major);
-
-    const bool swap_stride = (transpose != storage_differs);
-
-    // Determine expr's actual strides, resolving any defaults if zero.
-    const Index expr_inner_actual = resolveInnerStride(expr.innerStride());
-    const Index expr_outer_actual = resolveOuterStride(expr_inner_actual,
-                                                       expr.outerStride(),
-                                                       expr.rows(),
-                                                       expr.cols(),
-                                                       Expression::IsVectorAtCompileTime != 0,
-                                                       expr_row_major);
-
-    // If this is a column-major row vector or row-major column vector, the inner-stride
-    // is arbitrary, so set it to either the compile-time inner stride or 1.
-    const bool row_vector = (rows == 1);
-    const bool col_vector = (cols == 1);
-    const Index inner_stride =
-        ( (!row_major && row_vector) || (row_major && col_vector) ) ?
-            ( StrideType::InnerStrideAtCompileTime > 0 ? Index(StrideType::InnerStrideAtCompileTime) : 1)
-            : swap_stride ? expr_outer_actual : expr_inner_actual;
-
-    // If this is a column-major column vector or row-major row vector, the outer-stride
-    // is arbitrary, so set it to either the compile-time outer stride or vector size.
-    const Index outer_stride =
-      ( (!row_major && col_vector) || (row_major && row_vector) ) ?
-          ( StrideType::OuterStrideAtCompileTime > 0 ? Index(StrideType::OuterStrideAtCompileTime) : rows * cols * inner_stride)
-          : swap_stride ? expr_inner_actual : expr_outer_actual;
-
-    // Check if given inner/outer strides are compatible with compile-time strides.
-    const bool inner_valid = (StrideType::InnerStrideAtCompileTime == Dynamic)
-        || (resolveInnerStride(Index(StrideType::InnerStrideAtCompileTime)) == inner_stride);
-    if (!inner_valid) {
-      return false;
-    }
-
-    const bool outer_valid = (StrideType::OuterStrideAtCompileTime == Dynamic)
-        || (resolveOuterStride(
-              inner_stride,
-              Index(StrideType::OuterStrideAtCompileTime),
-              rows, cols, PlainObjectType::IsVectorAtCompileTime != 0,
-              row_major)
-            == outer_stride);
-    if (!outer_valid) {
-      return false;
-    }
-
-    ::new (static_cast<Base*>(this)) Base(expr.data(), rows, cols);
-    ::new (&m_stride) StrideBase(
-      (StrideType::OuterStrideAtCompileTime == 0) ? 0 : outer_stride,
-      (StrideType::InnerStrideAtCompileTime == 0) ? 0 : inner_stride );
-    return true;
   }
 
   StrideBase m_stride;
@@ -273,8 +187,6 @@ protected:
   * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
   * \endcode
   *
-  * See also the following stackoverflow questions for further references:
-  *  - <a href="http://stackoverflow.com/questions/21132538/correct-usage-of-the-eigenref-class">Correct usage of the Eigen::Ref<> class</a>
   *
   * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
   */
@@ -298,10 +210,7 @@ template<typename PlainObjectType, int Options, typename StrideType> class Ref
                                  typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     {
       EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
-      // Construction must pass since we will not create temprary storage in the non-const case.
+      Base::construct(expr.derived());
-      const bool success = Base::construct(expr.derived());
-      EIGEN_UNUSED_VARIABLE(success)
-      eigen_assert(success);
     }
     template<typename Derived>
     EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
@@ -312,13 +221,10 @@ template<typename PlainObjectType, int Options, typename StrideType> class Ref
     inline Ref(DenseBase<Derived>& expr)
     #endif
     {
-      EIGEN_STATIC_ASSERT((static_cast<bool>(internal::is_lvalue<Derived>::value)), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
-      EIGEN_STATIC_ASSERT((static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime)), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
       EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
-      // Construction must pass since we will not create temporary storage in the non-const case.
+      Base::construct(expr.const_cast_derived());
-      const bool success = Base::construct(expr.const_cast_derived());
-      EIGEN_UNUSED_VARIABLE(success)
-      eigen_assert(success);
     }
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Ref)
@@ -359,10 +265,7 @@ template<typename TPlainObjectType, int Options, typename StrideType> class Ref<
     template<typename Expression>
     EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
     {
-      // Check if we can use the underlying expr's storage directly, otherwise call the copy version.
+      Base::construct(expr);
-      if (!Base::construct(expr)) {
-        construct(expr, internal::false_type());
-      }
     }
 
     template<typename Expression>

Eigen/src/Core/Replicate.h

@@ -88,9 +88,9 @@ template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
     EIGEN_DEVICE_FUNC

Eigen/src/Core/Reshaped.h

@@ -1,454 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2017 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2014 yoco <peter.xiau@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_RESHAPED_H
-#define EIGEN_RESHAPED_H
-
-namespace Eigen {
-
-/** \class Reshaped
-  * \ingroup Core_Module
-  *
-  * \brief Expression of a fixed-size or dynamic-size reshape
-  *
-  * \tparam XprType the type of the expression in which we are taking a reshape
-  * \tparam Rows the number of rows of the reshape we are taking at compile time (optional)
-  * \tparam Cols the number of columns of the reshape we are taking at compile time (optional)
-  * \tparam Order can be ColMajor or RowMajor, default is ColMajor.
-  *
-  * This class represents an expression of either a fixed-size or dynamic-size reshape.
-  * It is the return type of DenseBase::reshaped(NRowsType,NColsType) and
-  * most of the time this is the only way it is used.
-  *
-  * However, in C++98, if you want to directly maniputate reshaped expressions,
-  * for instance if you want to write a function returning such an expression, you
-  * will need to use this class. In C++11, it is advised to use the \em auto
-  * keyword for such use cases.
-  *
-  * Here is an example illustrating the dynamic case:
-  * \include class_Reshaped.cpp
-  * Output: \verbinclude class_Reshaped.out
-  *
-  * Here is an example illustrating the fixed-size case:
-  * \include class_FixedReshaped.cpp
-  * Output: \verbinclude class_FixedReshaped.out
-  *
-  * \sa DenseBase::reshaped(NRowsType,NColsType)
-  */
-
-namespace internal {
-
-template<typename XprType, int Rows, int Cols, int Order>
-struct traits<Reshaped<XprType, Rows, Cols, Order> > : traits<XprType>
-{
-  typedef typename traits<XprType>::Scalar Scalar;
-  typedef typename traits<XprType>::StorageKind StorageKind;
-  typedef typename traits<XprType>::XprKind XprKind;
-  enum{
-    MatrixRows = traits<XprType>::RowsAtCompileTime,
-    MatrixCols = traits<XprType>::ColsAtCompileTime,
-    RowsAtCompileTime = Rows,
-    ColsAtCompileTime = Cols,
-    MaxRowsAtCompileTime = Rows,
-    MaxColsAtCompileTime = Cols,
-    XpxStorageOrder = ((int(traits<XprType>::Flags) & RowMajorBit) == RowMajorBit) ? RowMajor : ColMajor,
-    ReshapedStorageOrder = (RowsAtCompileTime == 1 && ColsAtCompileTime != 1) ? RowMajor
-                         : (ColsAtCompileTime == 1 && RowsAtCompileTime != 1) ? ColMajor
-                         : XpxStorageOrder,
-    HasSameStorageOrderAsXprType = (ReshapedStorageOrder == XpxStorageOrder),
-    InnerSize = (ReshapedStorageOrder==int(RowMajor)) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
-    InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? int(inner_stride_at_compile_time<XprType>::ret)
-                             : Dynamic,
-    OuterStrideAtCompileTime = Dynamic,
-
-    HasDirectAccess = internal::has_direct_access<XprType>::ret
-                    && (Order==int(XpxStorageOrder))
-                    && ((evaluator<XprType>::Flags&LinearAccessBit)==LinearAccessBit),
-
-    MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
-                       && (InnerStrideAtCompileTime == 1)
-                        ? PacketAccessBit : 0,
-    //MaskAlignedBit = ((OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0,
-    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
-    FlagsRowMajorBit = (ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0,
-    FlagsDirectAccessBit = HasDirectAccess ? DirectAccessBit : 0,
-    Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) | MaskPacketAccessBit),
-
-    Flags = (Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit | FlagsDirectAccessBit)
-  };
-};
-
-template<typename XprType, int Rows, int Cols, int Order, bool HasDirectAccess> class ReshapedImpl_dense;
-
-} // end namespace internal
-
-template<typename XprType, int Rows, int Cols, int Order, typename StorageKind> class ReshapedImpl;
-
-template<typename XprType, int Rows, int Cols, int Order> class Reshaped
-  : public ReshapedImpl<XprType, Rows, Cols, Order, typename internal::traits<XprType>::StorageKind>
-{
-    typedef ReshapedImpl<XprType, Rows, Cols, Order, typename internal::traits<XprType>::StorageKind> Impl;
-  public:
-    //typedef typename Impl::Base Base;
-    typedef Impl Base;
-    EIGEN_GENERIC_PUBLIC_INTERFACE(Reshaped)
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reshaped)
-
-    /** Fixed-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline Reshaped(XprType& xpr)
-      : Impl(xpr)
-    {
-      EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
-      eigen_assert(Rows * Cols == xpr.rows() * xpr.cols());
-    }
-
-    /** Dynamic-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline Reshaped(XprType& xpr,
-          Index reshapeRows, Index reshapeCols)
-      : Impl(xpr, reshapeRows, reshapeCols)
-    {
-      eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==reshapeRows)
-          && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==reshapeCols));
-      eigen_assert(reshapeRows * reshapeCols == xpr.rows() * xpr.cols());
-    }
-};
-
-// The generic default implementation for dense reshape simply forward to the internal::ReshapedImpl_dense
-// that must be specialized for direct and non-direct access...
-template<typename XprType, int Rows, int Cols, int Order>
-class ReshapedImpl<XprType, Rows, Cols, Order, Dense>
-  : public internal::ReshapedImpl_dense<XprType, Rows, Cols, Order,internal::traits<Reshaped<XprType,Rows,Cols,Order> >::HasDirectAccess>
-{
-    typedef internal::ReshapedImpl_dense<XprType, Rows, Cols, Order,internal::traits<Reshaped<XprType,Rows,Cols,Order> >::HasDirectAccess> Impl;
-  public:
-    typedef Impl Base;
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl)
-    EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr) : Impl(xpr) {}
-    EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr, Index reshapeRows, Index reshapeCols)
-      : Impl(xpr, reshapeRows, reshapeCols) {}
-};
-
-namespace internal {
-
-/** \internal Internal implementation of dense Reshaped in the general case. */
-template<typename XprType, int Rows, int Cols, int Order>
-class ReshapedImpl_dense<XprType,Rows,Cols,Order,false>
-  : public internal::dense_xpr_base<Reshaped<XprType, Rows, Cols, Order> >::type
-{
-    typedef Reshaped<XprType, Rows, Cols, Order> ReshapedType;
-  public:
-
-    typedef typename internal::dense_xpr_base<ReshapedType>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType)
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense)
-
-    typedef typename internal::ref_selector<XprType>::non_const_type MatrixTypeNested;
-    typedef typename internal::remove_all<XprType>::type NestedExpression;
-
-    class InnerIterator;
-
-    /** Fixed-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline ReshapedImpl_dense(XprType& xpr)
-      : m_xpr(xpr), m_rows(Rows), m_cols(Cols)
-    {}
-
-    /** Dynamic-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols)
-      : m_xpr(xpr), m_rows(nRows), m_cols(nCols)
-    {}
-
-    EIGEN_DEVICE_FUNC Index rows() const { return m_rows; }
-    EIGEN_DEVICE_FUNC Index cols() const { return m_cols; }
-
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    /** \sa MapBase::data() */
-    EIGEN_DEVICE_FUNC inline const Scalar* data() const;
-    EIGEN_DEVICE_FUNC inline Index innerStride() const;
-    EIGEN_DEVICE_FUNC inline Index outerStride() const;
-    #endif
-
-    /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC
-    const typename internal::remove_all<XprType>::type&
-    nestedExpression() const { return m_xpr; }
-
-    /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC
-    typename internal::remove_reference<XprType>::type&
-    nestedExpression() { return m_xpr; }
-
-  protected:
-
-    MatrixTypeNested m_xpr;
-    const internal::variable_if_dynamic<Index, Rows> m_rows;
-    const internal::variable_if_dynamic<Index, Cols> m_cols;
-};
-
-
-/** \internal Internal implementation of dense Reshaped in the direct access case. */
-template<typename XprType, int Rows, int Cols, int Order>
-class ReshapedImpl_dense<XprType, Rows, Cols, Order, true>
-  : public MapBase<Reshaped<XprType, Rows, Cols, Order> >
-{
-    typedef Reshaped<XprType, Rows, Cols, Order> ReshapedType;
-    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
-  public:
-
-    typedef MapBase<ReshapedType> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType)
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense)
-
-    /** Fixed-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline ReshapedImpl_dense(XprType& xpr)
-      : Base(xpr.data()), m_xpr(xpr)
-    {}
-
-    /** Dynamic-size constructor
-      */
-    EIGEN_DEVICE_FUNC
-    inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols)
-      : Base(xpr.data(), nRows, nCols),
-        m_xpr(xpr)
-    {}
-
-    EIGEN_DEVICE_FUNC
-    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
-    {
-      return m_xpr;
-    }
-
-    EIGEN_DEVICE_FUNC
-    XprType& nestedExpression() { return m_xpr; }
-
-    /** \sa MapBase::innerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index innerStride() const
-    {
-      return m_xpr.innerStride();
-    }
-
-    /** \sa MapBase::outerStride() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index outerStride() const
-    {
-      return (((Flags&RowMajorBit)==RowMajorBit) ? this->cols() : this->rows()) * m_xpr.innerStride();
-    }
-
-  protected:
-
-    XprTypeNested m_xpr;
-};
-
-// Evaluators
-template<typename ArgType, int Rows, int Cols, int Order, bool HasDirectAccess> struct reshaped_evaluator;
-
-template<typename ArgType, int Rows, int Cols, int Order>
-struct evaluator<Reshaped<ArgType, Rows, Cols, Order> >
-  : reshaped_evaluator<ArgType, Rows, Cols, Order, traits<Reshaped<ArgType,Rows,Cols,Order> >::HasDirectAccess>
-{
-  typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
-  typedef typename XprType::Scalar Scalar;
-  // TODO: should check for smaller packet types
-  typedef typename packet_traits<Scalar>::type PacketScalar;
-
-  enum {
-    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
-    HasDirectAccess = traits<XprType>::HasDirectAccess,
-
-//     RowsAtCompileTime = traits<XprType>::RowsAtCompileTime,
-//     ColsAtCompileTime = traits<XprType>::ColsAtCompileTime,
-//     MaxRowsAtCompileTime = traits<XprType>::MaxRowsAtCompileTime,
-//     MaxColsAtCompileTime = traits<XprType>::MaxColsAtCompileTime,
-//
-//     InnerStrideAtCompileTime = traits<XprType>::HasSameStorageOrderAsXprType
-//                              ? int(inner_stride_at_compile_time<ArgType>::ret)
-//                              : Dynamic,
-//     OuterStrideAtCompileTime = Dynamic,
-
-    FlagsLinearAccessBit = (traits<XprType>::RowsAtCompileTime == 1 || traits<XprType>::ColsAtCompileTime == 1 || HasDirectAccess) ? LinearAccessBit : 0,
-    FlagsRowMajorBit = (traits<XprType>::ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0,
-    FlagsDirectAccessBit =  HasDirectAccess ? DirectAccessBit : 0,
-    Flags0 = evaluator<ArgType>::Flags & (HereditaryBits & ~RowMajorBit),
-    Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit | FlagsDirectAccessBit,
-
-    PacketAlignment = unpacket_traits<PacketScalar>::alignment,
-    Alignment = evaluator<ArgType>::Alignment
-  };
-  typedef reshaped_evaluator<ArgType, Rows, Cols, Order, HasDirectAccess> reshaped_evaluator_type;
-  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : reshaped_evaluator_type(xpr)
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-};
-
-template<typename ArgType, int Rows, int Cols, int Order>
-struct reshaped_evaluator<ArgType, Rows, Cols, Order, /* HasDirectAccess */ false>
-  : evaluator_base<Reshaped<ArgType, Rows, Cols, Order> >
-{
-  typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
-
-  enum {
-    CoeffReadCost = evaluator<ArgType>::CoeffReadCost /* TODO + cost of index computations */,
-
-    Flags = (evaluator<ArgType>::Flags & (HereditaryBits /*| LinearAccessBit | DirectAccessBit*/)),
-
-    Alignment = 0
-  };
-
-  EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr)
-  {
-    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
-  }
-
-  typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::CoeffReturnType CoeffReturnType;
-
-  typedef std::pair<Index, Index> RowCol;
-
-  inline RowCol index_remap(Index rowId, Index colId) const
-  {
-    if(Order==ColMajor)
-    {
-      const Index nth_elem_idx = colId * m_xpr.rows() + rowId;
-      return RowCol(nth_elem_idx % m_xpr.nestedExpression().rows(),
-                    nth_elem_idx / m_xpr.nestedExpression().rows());
-    }
-    else
-    {
-      const Index nth_elem_idx = colId + rowId * m_xpr.cols();
-      return RowCol(nth_elem_idx / m_xpr.nestedExpression().cols(),
-                    nth_elem_idx % m_xpr.nestedExpression().cols());
-    }
-  }
-
-  EIGEN_DEVICE_FUNC
-  inline Scalar& coeffRef(Index rowId, Index colId)
-  {
-    EIGEN_STATIC_ASSERT_LVALUE(XprType)
-    const RowCol row_col = index_remap(rowId, colId);
-    return m_argImpl.coeffRef(row_col.first, row_col.second);
-  }
-
-  EIGEN_DEVICE_FUNC
-  inline const Scalar& coeffRef(Index rowId, Index colId) const
-  {
-    const RowCol row_col = index_remap(rowId, colId);
-    return m_argImpl.coeffRef(row_col.first, row_col.second);
-  }
-
-  EIGEN_DEVICE_FUNC
-  EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
-  {
-    const RowCol row_col = index_remap(rowId, colId);
-    return m_argImpl.coeff(row_col.first, row_col.second);
-  }
-
-  EIGEN_DEVICE_FUNC
-  inline Scalar& coeffRef(Index index)
-  {
-    EIGEN_STATIC_ASSERT_LVALUE(XprType)
-    const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
-                                       Rows == 1 ? index : 0);
-    return m_argImpl.coeffRef(row_col.first, row_col.second);
-
-  }
-
-  EIGEN_DEVICE_FUNC
-  inline const Scalar& coeffRef(Index index) const
-  {
-    const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
-                                       Rows == 1 ? index : 0);
-    return m_argImpl.coeffRef(row_col.first, row_col.second);
-  }
-
-  EIGEN_DEVICE_FUNC
-  inline const CoeffReturnType coeff(Index index) const
-  {
-    const RowCol row_col = index_remap(Rows == 1 ? 0 : index,
-                                       Rows == 1 ? index : 0);
-    return m_argImpl.coeff(row_col.first, row_col.second);
-  }
-#if 0
-  EIGEN_DEVICE_FUNC
-  template<int LoadMode>
-  inline PacketScalar packet(Index rowId, Index colId) const
-  {
-    const RowCol row_col = index_remap(rowId, colId);
-    return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second);
-
-  }
-
-  template<int LoadMode>
-  EIGEN_DEVICE_FUNC
-  inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
-  {
-    const RowCol row_col = index_remap(rowId, colId);
-    m_argImpl.const_cast_derived().template writePacket<Unaligned>
-            (row_col.first, row_col.second, val);
-  }
-
-  template<int LoadMode>
-  EIGEN_DEVICE_FUNC
-  inline PacketScalar packet(Index index) const
-  {
-    const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index,
-                                        RowsAtCompileTime == 1 ? index : 0);
-    return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second);
-  }
-
-  template<int LoadMode>
-  EIGEN_DEVICE_FUNC
-  inline void writePacket(Index index, const PacketScalar& val)
-  {
-    const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index,
-                                        RowsAtCompileTime == 1 ? index : 0);
-    return m_argImpl.template packet<Unaligned>(row_col.first, row_col.second, val);
-  }
-#endif
-protected:
-
-  evaluator<ArgType> m_argImpl;
-  const XprType& m_xpr;
-
-};
-
-template<typename ArgType, int Rows, int Cols, int Order>
-struct reshaped_evaluator<ArgType, Rows, Cols, Order, /* HasDirectAccess */ true>
-: mapbase_evaluator<Reshaped<ArgType, Rows, Cols, Order>,
-                      typename Reshaped<ArgType, Rows, Cols, Order>::PlainObject>
-{
-  typedef Reshaped<ArgType, Rows, Cols, Order> XprType;
-  typedef typename XprType::Scalar Scalar;
-
-  EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr)
-    : mapbase_evaluator<XprType, typename XprType::PlainObject>(xpr)
-  {
-    // TODO: for the 3.4 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
-    eigen_assert(((internal::UIntPtr(xpr.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator<XprType>::Alignment)) == 0) && "data is not aligned");
-  }
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_RESHAPED_H

Eigen/src/Core/ReturnByValue.h

@@ -60,10 +60,8 @@ template<typename Derived> class ReturnByValue
     EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const
     { static_cast<const Derived*>(this)->evalTo(dst); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
-    inline Index rows() const EIGEN_NOEXCEPT { return static_cast<const Derived*>(this)->rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return static_cast<const Derived*>(this)->cols(); }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 #define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT

Eigen/src/Core/Reverse.h

@@ -89,10 +89,8 @@ template<typename MatrixType, int Direction> class Reverse
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reverse)
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows(); }
-    inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
 
     EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
@@ -173,10 +171,8 @@ struct vectorwise_reverse_inplace_impl<Vertical>
   template<typename ExpressionType>
   static void run(ExpressionType &xpr)
   {
-    const int HalfAtCompileTime = ExpressionType::RowsAtCompileTime==Dynamic?Dynamic:ExpressionType::RowsAtCompileTime/2;
     Index half = xpr.rows()/2;
-    xpr.topRows(fix<HalfAtCompileTime>(half))
+    xpr.topRows(half).swap(xpr.bottomRows(half).colwise().reverse());
-       .swap(xpr.bottomRows(fix<HalfAtCompileTime>(half)).colwise().reverse());
   }
 };
 
@@ -186,10 +182,8 @@ struct vectorwise_reverse_inplace_impl<Horizontal>
   template<typename ExpressionType>
   static void run(ExpressionType &xpr)
   {
-    const int HalfAtCompileTime = ExpressionType::ColsAtCompileTime==Dynamic?Dynamic:ExpressionType::ColsAtCompileTime/2;
     Index half = xpr.cols()/2;
-    xpr.leftCols(fix<HalfAtCompileTime>(half))
+    xpr.leftCols(half).swap(xpr.rightCols(half).rowwise().reverse());
-       .swap(xpr.rightCols(fix<HalfAtCompileTime>(half)).rowwise().reverse());
   }
 };
 
@@ -209,7 +203,7 @@ struct vectorwise_reverse_inplace_impl<Horizontal>
 template<typename ExpressionType, int Direction>
 EIGEN_DEVICE_FUNC void VectorwiseOp<ExpressionType,Direction>::reverseInPlace()
 {
-  internal::vectorwise_reverse_inplace_impl<Direction>::run(m_matrix);
+  internal::vectorwise_reverse_inplace_impl<Direction>::run(_expression().const_cast_derived());
 }
 
 } // end namespace Eigen

Eigen/src/Core/Select.h

@@ -67,10 +67,8 @@ class Select : public internal::dense_xpr_base< Select<ConditionMatrixType, Then
       eigen_assert(m_condition.cols() == m_then.cols() && m_condition.cols() == m_else.cols());
     }
 
-    inline EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    inline EIGEN_DEVICE_FUNC Index rows() const { return m_condition.rows(); }
-    Index rows() const EIGEN_NOEXCEPT { return m_condition.rows(); }
+    inline EIGEN_DEVICE_FUNC Index cols() const { return m_condition.cols(); }
-    inline EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_condition.cols(); }
 
     inline EIGEN_DEVICE_FUNC
     const Scalar coeff(Index i, Index j) const
@@ -122,7 +120,7 @@ class Select : public internal::dense_xpr_base< Select<ConditionMatrixType, Then
   */
 template<typename Derived>
 template<typename ThenDerived,typename ElseDerived>
-inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived,ElseDerived>
+inline const Select<Derived,ThenDerived,ElseDerived>
 DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
                             const DenseBase<ElseDerived>& elseMatrix) const
 {
@@ -136,7 +134,7 @@ DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
   */
 template<typename Derived>
 template<typename ThenDerived>
-inline EIGEN_DEVICE_FUNC const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
+inline const Select<Derived,ThenDerived, typename ThenDerived::ConstantReturnType>
 DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
                            const typename ThenDerived::Scalar& elseScalar) const
 {
@@ -151,7 +149,7 @@ DenseBase<Derived>::select(const DenseBase<ThenDerived>& thenMatrix,
   */
 template<typename Derived>
 template<typename ElseDerived>
-inline EIGEN_DEVICE_FUNC const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
+inline const Select<Derived, typename ElseDerived::ConstantReturnType, ElseDerived >
 DenseBase<Derived>::select(const typename ElseDerived::Scalar& thenScalar,
                            const DenseBase<ElseDerived>& elseMatrix) const
 {

Eigen/src/Core/SelfAdjointView.h

@@ -61,12 +61,11 @@ template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
     typedef typename internal::traits<SelfAdjointView>::Scalar Scalar;
     typedef typename MatrixType::StorageIndex StorageIndex;
     typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
-    typedef SelfAdjointView<typename internal::add_const<MatrixType>::type, UpLo> ConstSelfAdjointView;
 
     enum {
       Mode = internal::traits<SelfAdjointView>::Mode,
       Flags = internal::traits<SelfAdjointView>::Flags,
-      TransposeMode = ((int(Mode) & int(Upper)) ? Lower : 0) | ((int(Mode) & int(Lower)) ? Upper : 0)
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0)
     };
     typedef typename MatrixType::PlainObject PlainObject;
 
@@ -76,14 +75,14 @@ template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
       EIGEN_STATIC_ASSERT(UpLo==Lower || UpLo==Upper,SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY);
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    inline Index rows() const { return m_matrix.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    inline Index cols() const { return m_matrix.cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT { return m_matrix.outerStride(); }
+    inline Index outerStride() const { return m_matrix.outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT { return m_matrix.innerStride(); }
+    inline Index innerStride() const { return m_matrix.innerStride(); }
 
     /** \sa MatrixBase::coeff()
       * \warning the coordinates must fit into the referenced triangular part
@@ -198,18 +197,6 @@ template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
     inline const ConjugateReturnType conjugate() const
     { return ConjugateReturnType(m_matrix.conjugate()); }
 
-    /** \returns an expression of the complex conjugate of \c *this if Cond==true,
-     *           returns \c *this otherwise.
-     */
-    template<bool Cond>
-    EIGEN_DEVICE_FUNC
-    inline typename internal::conditional<Cond,ConjugateReturnType,ConstSelfAdjointView>::type
-    conjugateIf() const
-    {
-      typedef typename internal::conditional<Cond,ConjugateReturnType,ConstSelfAdjointView>::type ReturnType;
-      return ReturnType(m_matrix.template conjugateIf<Cond>());
-    }
-
     typedef SelfAdjointView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
     /** \sa MatrixBase::adjoint() const */
     EIGEN_DEVICE_FUNC

Eigen/src/Core/Solve.h

@@ -19,7 +19,7 @@ template<typename Decomposition, typename RhsType, typename StorageKind> class S
   *
   * \brief Pseudo expression representing a solving operation
   *
-  * \tparam Decomposition the type of the matrix or decomposition object
+  * \tparam Decomposition the type of the matrix or decomposion object
   * \tparam Rhstype the type of the right-hand side
   *
   * This class represents an expression of A.solve(B)
@@ -69,8 +69,8 @@ public:
     : m_dec(dec), m_rhs(rhs)
   {}
 
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_dec.cols(); }
+  EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
 
   EIGEN_DEVICE_FUNC const Decomposition& dec() const { return m_dec; }
   EIGEN_DEVICE_FUNC const RhsType&       rhs() const { return m_rhs; }
@@ -137,7 +137,7 @@ template<typename DstXprType, typename DecType, typename RhsType, typename Scala
 struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
 {
   typedef Solve<DecType,RhsType> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();
@@ -153,7 +153,7 @@ template<typename DstXprType, typename DecType, typename RhsType, typename Scala
 struct Assignment<DstXprType, Solve<Transpose<const DecType>,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
 {
   typedef Solve<Transpose<const DecType>,RhsType> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();
@@ -170,7 +170,7 @@ struct Assignment<DstXprType, Solve<CwiseUnaryOp<internal::scalar_conjugate_op<t
                   internal::assign_op<Scalar,Scalar>, Dense2Dense>
 {
   typedef Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();
@@ -181,7 +181,7 @@ struct Assignment<DstXprType, Solve<CwiseUnaryOp<internal::scalar_conjugate_op<t
   }
 };
 
-} // end namespace internal
+} // end namepsace internal
 
 } // end namespace Eigen
 

Eigen/src/Core/SolveTriangular.h

@@ -54,7 +54,7 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,1>
   typedef blas_traits<Lhs> LhsProductTraits;
   typedef typename LhsProductTraits::ExtractType ActualLhsType;
   typedef Map<Matrix<RhsScalar,Dynamic,1>, Aligned> MappedRhs;
-  static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
+  static void run(const Lhs& lhs, Rhs& rhs)
   {
     ActualLhsType actualLhs = LhsProductTraits::extract(lhs);
 
@@ -85,7 +85,7 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
   typedef blas_traits<Lhs> LhsProductTraits;
   typedef typename LhsProductTraits::DirectLinearAccessType ActualLhsType;
 
-  static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
+  static void run(const Lhs& lhs, Rhs& rhs)
   {
     typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsProductTraits::extract(lhs);
 
@@ -118,7 +118,7 @@ struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,false> {
     DiagIndex  = IsLower ? LoopIndex : Size - LoopIndex - 1,
     StartIndex = IsLower ? 0         : DiagIndex+1
   };
-  static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
+  static void run(const Lhs& lhs, Rhs& rhs)
   {
     if (LoopIndex>0)
       rhs.coeffRef(DiagIndex) -= lhs.row(DiagIndex).template segment<LoopIndex>(StartIndex).transpose()
@@ -133,18 +133,18 @@ struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,false> {
 
 template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
 struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,true> {
-  static EIGEN_DEVICE_FUNC void run(const Lhs&, Rhs&) {}
+  static void run(const Lhs&, Rhs&) {}
 };
 
 template<typename Lhs, typename Rhs, int Mode>
 struct triangular_solver_selector<Lhs,Rhs,OnTheLeft,Mode,CompleteUnrolling,1> {
-  static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
+  static void run(const Lhs& lhs, Rhs& rhs)
   { triangular_solver_unroller<Lhs,Rhs,Mode,0,Rhs::SizeAtCompileTime>::run(lhs,rhs); }
 };
 
 template<typename Lhs, typename Rhs, int Mode>
 struct triangular_solver_selector<Lhs,Rhs,OnTheRight,Mode,CompleteUnrolling,1> {
-  static EIGEN_DEVICE_FUNC void run(const Lhs& lhs, Rhs& rhs)
+  static void run(const Lhs& lhs, Rhs& rhs)
   {
     Transpose<const Lhs> trLhs(lhs);
     Transpose<Rhs> trRhs(rhs);
@@ -168,7 +168,7 @@ EIGEN_DEVICE_FUNC void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(c
 {
   OtherDerived& other = _other.const_cast_derived();
   eigen_assert( derived().cols() == derived().rows() && ((Side==OnTheLeft && derived().cols() == other.rows()) || (Side==OnTheRight && derived().cols() == other.cols())) );
-  eigen_assert((!(int(Mode) & int(ZeroDiag))) && bool(int(Mode) & (int(Upper) | int(Lower))));
+  eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
   // If solving for a 0x0 matrix, nothing to do, simply return.
   if (derived().cols() == 0)
     return;
@@ -187,7 +187,7 @@ EIGEN_DEVICE_FUNC void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(c
 
 template<typename Derived, unsigned int Mode>
 template<int Side, typename Other>
-const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
+EIGEN_DEVICE_FUNC const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
 TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) const
 {
   return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
@@ -213,8 +213,8 @@ template<int Side, typename TriangularType, typename Rhs> struct triangular_solv
     : m_triangularMatrix(tri), m_rhs(rhs)
   {}
 
-  inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_rhs.rows(); }
+  inline Index rows() const { return m_rhs.rows(); }
-  inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); }
+  inline Index cols() const { return m_rhs.cols(); }
 
   template<typename Dest> inline void evalTo(Dest& dst) const
   {

Eigen/src/Core/SolverBase.h

@@ -14,35 +14,8 @@ namespace Eigen {
 
 namespace internal {
 
-template<typename Derived>
-struct solve_assertion {
-    template<bool Transpose_, typename Rhs>
-    static void run(const Derived& solver, const Rhs& b) { solver.template _check_solve_assertion<Transpose_>(b); }
-};
 
-template<typename Derived>
-struct solve_assertion<Transpose<Derived> >
-{
-    typedef Transpose<Derived> type;
 
-    template<bool Transpose_, typename Rhs>
-    static void run(const type& transpose, const Rhs& b)
-    {
-        internal::solve_assertion<typename internal::remove_all<Derived>::type>::template run<true>(transpose.nestedExpression(), b);
-    }
-};
-
-template<typename Scalar, typename Derived>
-struct solve_assertion<CwiseUnaryOp<Eigen::internal::scalar_conjugate_op<Scalar>, const Transpose<Derived> > >
-{
-    typedef CwiseUnaryOp<Eigen::internal::scalar_conjugate_op<Scalar>, const Transpose<Derived> > type;
-
-    template<bool Transpose_, typename Rhs>
-    static void run(const type& adjoint, const Rhs& b)
-    {
-        internal::solve_assertion<typename internal::remove_all<Transpose<Derived> >::type>::template run<true>(adjoint.nestedExpression(), b);
-    }
-};
 } // end namespace internal
 
 /** \class SolverBase
@@ -62,7 +35,7 @@ struct solve_assertion<CwiseUnaryOp<Eigen::internal::scalar_conjugate_op<Scalar>
   *
   * \warning Currently, any other usage of transpose() and adjoint() are not supported and will produce compilation errors.
   *
-  * \sa class PartialPivLU, class FullPivLU, class HouseholderQR, class ColPivHouseholderQR, class FullPivHouseholderQR, class CompleteOrthogonalDecomposition, class LLT, class LDLT, class SVDBase
+  * \sa class PartialPivLU, class FullPivLU
   */
 template<typename Derived>
 class SolverBase : public EigenBase<Derived>
@@ -73,9 +46,6 @@ class SolverBase : public EigenBase<Derived>
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef Scalar CoeffReturnType;
 
-    template<typename Derived_>
-    friend struct internal::solve_assertion;
-
     enum {
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
@@ -86,8 +56,7 @@ class SolverBase : public EigenBase<Derived>
       MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
                                                              internal::traits<Derived>::MaxColsAtCompileTime>::ret),
       IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
-                           || internal::traits<Derived>::MaxColsAtCompileTime == 1,
+                           || internal::traits<Derived>::MaxColsAtCompileTime == 1
-      NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2
     };
 
     /** Default constructor */
@@ -105,12 +74,12 @@ class SolverBase : public EigenBase<Derived>
     inline const Solve<Derived, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
-      internal::solve_assertion<typename internal::remove_all<Derived>::type>::template run<false>(derived(), b);
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
       return Solve<Derived, Rhs>(derived(), b.derived());
     }
 
     /** \internal the return type of transpose() */
-    typedef Transpose<const Derived> ConstTransposeReturnType;
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
     /** \returns an expression of the transposed of the factored matrix.
       *
       * A typical usage is to solve for the transposed problem A^T x = b:
@@ -118,16 +87,16 @@ class SolverBase : public EigenBase<Derived>
       *
       * \sa adjoint(), solve()
       */
-    inline const ConstTransposeReturnType transpose() const
+    inline ConstTransposeReturnType transpose() const
     {
       return ConstTransposeReturnType(derived());
     }
 
     /** \internal the return type of adjoint() */
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
-               CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, const ConstTransposeReturnType>,
+                        CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
-               const ConstTransposeReturnType
+                        ConstTransposeReturnType
-            >::type AdjointReturnType;
+                     >::type AdjointReturnType;
     /** \returns an expression of the adjoint of the factored matrix
       *
       * A typical usage is to solve for the adjoint problem A' x = b:
@@ -137,19 +106,12 @@ class SolverBase : public EigenBase<Derived>
       *
       * \sa transpose(), solve()
       */
-    inline const AdjointReturnType adjoint() const
+    inline AdjointReturnType adjoint() const
     {
       return AdjointReturnType(derived().transpose());
     }
 
   protected:
-
-    template<bool Transpose_, typename Rhs>
-    void _check_solve_assertion(const Rhs& b) const {
-        EIGEN_ONLY_USED_FOR_DEBUG(b);
-        eigen_assert(derived().m_isInitialized && "Solver is not initialized.");
-        eigen_assert((Transpose_?derived().cols():derived().rows())==b.rows() && "SolverBase::solve(): invalid number of rows of the right hand side matrix b");
-    }
 };
 
 namespace internal {

Eigen/src/Core/StableNorm.h

@@ -50,71 +50,6 @@ inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& sc
     ssq += (bl*invScale).squaredNorm();
 }
 
-template<typename VectorType, typename RealScalar>
-void stable_norm_impl_inner_step(const VectorType &vec, RealScalar& ssq, RealScalar& scale, RealScalar& invScale)
-{
-  typedef typename VectorType::Scalar Scalar;
-  const Index blockSize = 4096;
-  
-  typedef typename internal::nested_eval<VectorType,2>::type VectorTypeCopy;
-  typedef typename internal::remove_all<VectorTypeCopy>::type VectorTypeCopyClean;
-  const VectorTypeCopy copy(vec);
-  
-  enum {
-    CanAlign = (   (int(VectorTypeCopyClean::Flags)&DirectAccessBit)
-                || (int(internal::evaluator<VectorTypeCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
-               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
-                 && (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
-  };
-  typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<VectorTypeCopyClean>::Alignment>,
-                                                   typename VectorTypeCopyClean::ConstSegmentReturnType>::type SegmentWrapper;
-  Index n = vec.size();
-  
-  Index bi = internal::first_default_aligned(copy);
-  if (bi>0)
-    internal::stable_norm_kernel(copy.head(bi), ssq, scale, invScale);
-  for (; bi<n; bi+=blockSize)
-    internal::stable_norm_kernel(SegmentWrapper(copy.segment(bi,numext::mini(blockSize, n - bi))), ssq, scale, invScale);
-}
-
-template<typename VectorType>
-typename VectorType::RealScalar
-stable_norm_impl(const VectorType &vec, typename enable_if<VectorType::IsVectorAtCompileTime>::type* = 0 )
-{
-  using std::sqrt;
-  using std::abs;
-
-  Index n = vec.size();
-
-  if(n==1)
-    return abs(vec.coeff(0));
-
-  typedef typename VectorType::RealScalar RealScalar;
-  RealScalar scale(0);
-  RealScalar invScale(1);
-  RealScalar ssq(0); // sum of squares
-
-  stable_norm_impl_inner_step(vec, ssq, scale, invScale);
-  
-  return scale * sqrt(ssq);
-}
-
-template<typename MatrixType>
-typename MatrixType::RealScalar
-stable_norm_impl(const MatrixType &mat, typename enable_if<!MatrixType::IsVectorAtCompileTime>::type* = 0 )
-{
-  using std::sqrt;
-
-  typedef typename MatrixType::RealScalar RealScalar;
-  RealScalar scale(0);
-  RealScalar invScale(1);
-  RealScalar ssq(0); // sum of squares
-
-  for(Index j=0; j<mat.outerSize(); ++j)
-    stable_norm_impl_inner_step(mat.innerVector(j), ssq, scale, invScale);
-  return scale * sqrt(ssq);
-}
-
 template<typename Derived>
 inline typename NumTraits<typename traits<Derived>::Scalar>::Real
 blueNorm_impl(const EigenBase<Derived>& _vec)
@@ -123,43 +58,52 @@ blueNorm_impl(const EigenBase<Derived>& _vec)
   using std::pow;
   using std::sqrt;
   using std::abs;
-
-  // This program calculates the machine-dependent constants
-  // bl, b2, slm, s2m, relerr overfl
-  // from the "basic" machine-dependent numbers
-  // nbig, ibeta, it, iemin, iemax, rbig.
-  // The following define the basic machine-dependent constants.
-  // For portability, the PORT subprograms "ilmaeh" and "rlmach"
-  // are used. For any specific computer, each of the assignment
-  // statements can be replaced
-  static const int ibeta = std::numeric_limits<RealScalar>::radix;  // base for floating-point numbers
-  static const int it    = NumTraits<RealScalar>::digits();  // number of base-beta digits in mantissa
-  static const int iemin = NumTraits<RealScalar>::min_exponent();  // minimum exponent
-  static const int iemax = NumTraits<RealScalar>::max_exponent();  // maximum exponent
-  static const RealScalar rbig   = NumTraits<RealScalar>::highest();  // largest floating-point number
-  static const RealScalar b1     = RealScalar(pow(RealScalar(ibeta),RealScalar(-((1-iemin)/2))));  // lower boundary of midrange
-  static const RealScalar b2     = RealScalar(pow(RealScalar(ibeta),RealScalar((iemax + 1 - it)/2)));  // upper boundary of midrange
-  static const RealScalar s1m    = RealScalar(pow(RealScalar(ibeta),RealScalar((2-iemin)/2)));  // scaling factor for lower range
-  static const RealScalar s2m    = RealScalar(pow(RealScalar(ibeta),RealScalar(- ((iemax+it)/2))));  // scaling factor for upper range
-  static const RealScalar eps    = RealScalar(pow(double(ibeta), 1-it));
-  static const RealScalar relerr = sqrt(eps);  // tolerance for neglecting asml
-
   const Derived& vec(_vec.derived());
+  static bool initialized = false;
+  static RealScalar b1, b2, s1m, s2m, rbig, relerr;
+  if(!initialized)
+  {
+    int ibeta, it, iemin, iemax, iexp;
+    RealScalar eps;
+    // This program calculates the machine-dependent constants
+    // bl, b2, slm, s2m, relerr overfl
+    // from the "basic" machine-dependent numbers
+    // nbig, ibeta, it, iemin, iemax, rbig.
+    // The following define the basic machine-dependent constants.
+    // For portability, the PORT subprograms "ilmaeh" and "rlmach"
+    // are used. For any specific computer, each of the assignment
+    // statements can be replaced
+    ibeta = std::numeric_limits<RealScalar>::radix;                 // base for floating-point numbers
+    it    = std::numeric_limits<RealScalar>::digits;                // number of base-beta digits in mantissa
+    iemin = std::numeric_limits<RealScalar>::min_exponent;          // minimum exponent
+    iemax = std::numeric_limits<RealScalar>::max_exponent;          // maximum exponent
+    rbig  = (std::numeric_limits<RealScalar>::max)();               // largest floating-point number
+
+    iexp  = -((1-iemin)/2);
+    b1    = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // lower boundary of midrange
+    iexp  = (iemax + 1 - it)/2;
+    b2    = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // upper boundary of midrange
+
+    iexp  = (2-iemin)/2;
+    s1m   = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // scaling factor for lower range
+    iexp  = - ((iemax+it)/2);
+    s2m   = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // scaling factor for upper range
+
+    eps     = RealScalar(pow(double(ibeta), 1-it));
+    relerr  = sqrt(eps);                                            // tolerance for neglecting asml
+    initialized = true;
+  }
   Index n = vec.size();
   RealScalar ab2 = b2 / RealScalar(n);
   RealScalar asml = RealScalar(0);
   RealScalar amed = RealScalar(0);
   RealScalar abig = RealScalar(0);
-
+  for(typename Derived::InnerIterator it(vec, 0); it; ++it)
-  for(Index j=0; j<vec.outerSize(); ++j)
   {
-    for(typename Derived::InnerIterator iter(vec, j); iter; ++iter)
+    RealScalar ax = abs(it.value());
-    {
+    if(ax > ab2)     abig += numext::abs2(ax*s2m);
-      RealScalar ax = abs(iter.value());
+    else if(ax < b1) asml += numext::abs2(ax*s1m);
-      if(ax > ab2)     abig += numext::abs2(ax*s2m);
+    else             amed += numext::abs2(ax);
-      else if(ax < b1) asml += numext::abs2(ax*s1m);
-      else             amed += numext::abs2(ax);
-    }
   }
   if(amed!=amed)
     return amed;  // we got a NaN
@@ -212,7 +156,36 @@ template<typename Derived>
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::stableNorm() const
 {
-  return internal::stable_norm_impl(derived());
+  using std::sqrt;
+  using std::abs;
+  const Index blockSize = 4096;
+  RealScalar scale(0);
+  RealScalar invScale(1);
+  RealScalar ssq(0); // sum of square
+  
+  typedef typename internal::nested_eval<Derived,2>::type DerivedCopy;
+  typedef typename internal::remove_all<DerivedCopy>::type DerivedCopyClean;
+  const DerivedCopy copy(derived());
+  
+  enum {
+    CanAlign = (   (int(DerivedCopyClean::Flags)&DirectAccessBit)
+                || (int(internal::evaluator<DerivedCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
+               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
+                 && (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
+  };
+  typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<DerivedCopyClean>::Alignment>,
+                                                   typename DerivedCopyClean::ConstSegmentReturnType>::type SegmentWrapper;
+  Index n = size();
+  
+  if(n==1)
+    return abs(this->coeff(0));
+  
+  Index bi = internal::first_default_aligned(copy);
+  if (bi>0)
+    internal::stable_norm_kernel(copy.head(bi), ssq, scale, invScale);
+  for (; bi<n; bi+=blockSize)
+    internal::stable_norm_kernel(SegmentWrapper(copy.segment(bi,numext::mini(blockSize, n - bi))), ssq, scale, invScale);
+  return scale * sqrt(ssq);
 }
 
 /** \returns the \em l2 norm of \c *this using the Blue's algorithm.
@@ -240,10 +213,7 @@ template<typename Derived>
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::hypotNorm() const
 {
-  if(size()==1)
+  return this->cwiseAbs().redux(internal::scalar_hypot_op<RealScalar>());
-    return numext::abs(coeff(0,0));
-  else
-    return this->cwiseAbs().redux(internal::scalar_hypot_op<RealScalar>());
 }
 
 } // end namespace Eigen

Eigen/src/Core/StlIterators.h

@@ -1,463 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2018 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_STLITERATORS_H
-#define EIGEN_STLITERATORS_H
-
-namespace Eigen {
-
-namespace internal {
-
-template<typename IteratorType>
-struct indexed_based_stl_iterator_traits;
-
-template<typename  Derived>
-class indexed_based_stl_iterator_base
-{
-protected:
-  typedef indexed_based_stl_iterator_traits<Derived> traits;
-  typedef typename traits::XprType XprType;
-  typedef indexed_based_stl_iterator_base<typename traits::non_const_iterator> non_const_iterator;
-  typedef indexed_based_stl_iterator_base<typename traits::const_iterator> const_iterator;
-  typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
-  // NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
-  friend class indexed_based_stl_iterator_base<typename traits::const_iterator>;
-  friend class indexed_based_stl_iterator_base<typename traits::non_const_iterator>;
-public:
-  typedef Index difference_type;
-  typedef std::random_access_iterator_tag iterator_category;
-
-  indexed_based_stl_iterator_base() EIGEN_NO_THROW : mp_xpr(0), m_index(0) {}
-  indexed_based_stl_iterator_base(XprType& xpr, Index index) EIGEN_NO_THROW : mp_xpr(&xpr), m_index(index) {}
-
-  indexed_based_stl_iterator_base(const non_const_iterator& other) EIGEN_NO_THROW
-    : mp_xpr(other.mp_xpr), m_index(other.m_index)
-  {}
-
-  indexed_based_stl_iterator_base& operator=(const non_const_iterator& other)
-  {
-    mp_xpr = other.mp_xpr;
-    m_index = other.m_index;
-    return *this;
-  }
-
-  Derived& operator++() { ++m_index; return derived(); }
-  Derived& operator--() { --m_index; return derived(); }
-
-  Derived operator++(int) { Derived prev(derived()); operator++(); return prev;}
-  Derived operator--(int) { Derived prev(derived()); operator--(); return prev;}
-
-  friend Derived operator+(const indexed_based_stl_iterator_base& a, Index b) { Derived ret(a.derived()); ret += b; return ret; }
-  friend Derived operator-(const indexed_based_stl_iterator_base& a, Index b) { Derived ret(a.derived()); ret -= b; return ret; }
-  friend Derived operator+(Index a, const indexed_based_stl_iterator_base& b) { Derived ret(b.derived()); ret += a; return ret; }
-  friend Derived operator-(Index a, const indexed_based_stl_iterator_base& b) { Derived ret(b.derived()); ret -= a; return ret; }
-  
-  Derived& operator+=(Index b) { m_index += b; return derived(); }
-  Derived& operator-=(Index b) { m_index -= b; return derived(); }
-
-  difference_type operator-(const indexed_based_stl_iterator_base& other) const
-  {
-    eigen_assert(mp_xpr == other.mp_xpr);
-    return m_index - other.m_index;
-  }
-
-  difference_type operator-(const other_iterator& other) const
-  {
-    eigen_assert(mp_xpr == other.mp_xpr);
-    return m_index - other.m_index;
-  }
-
-  bool operator==(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
-  bool operator!=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
-  bool operator< (const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <  other.m_index; }
-  bool operator<=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
-  bool operator> (const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >  other.m_index; }
-  bool operator>=(const indexed_based_stl_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
-
-  bool operator==(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
-  bool operator!=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
-  bool operator< (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <  other.m_index; }
-  bool operator<=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
-  bool operator> (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >  other.m_index; }
-  bool operator>=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
-
-protected:
-
-  Derived& derived() { return static_cast<Derived&>(*this); }
-  const Derived& derived() const { return static_cast<const Derived&>(*this); }
-
-  XprType *mp_xpr;
-  Index m_index;
-};
-
-template<typename  Derived>
-class indexed_based_stl_reverse_iterator_base
-{
-protected:
-  typedef indexed_based_stl_iterator_traits<Derived> traits;
-  typedef typename traits::XprType XprType;
-  typedef indexed_based_stl_reverse_iterator_base<typename traits::non_const_iterator> non_const_iterator;
-  typedef indexed_based_stl_reverse_iterator_base<typename traits::const_iterator> const_iterator;
-  typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
-  // NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
-  friend class indexed_based_stl_reverse_iterator_base<typename traits::const_iterator>;
-  friend class indexed_based_stl_reverse_iterator_base<typename traits::non_const_iterator>;
-public:
-  typedef Index difference_type;
-  typedef std::random_access_iterator_tag iterator_category;
-
-  indexed_based_stl_reverse_iterator_base() : mp_xpr(0), m_index(0) {}
-  indexed_based_stl_reverse_iterator_base(XprType& xpr, Index index) : mp_xpr(&xpr), m_index(index) {}
-
-  indexed_based_stl_reverse_iterator_base(const non_const_iterator& other)
-    : mp_xpr(other.mp_xpr), m_index(other.m_index)
-  {}
-
-  indexed_based_stl_reverse_iterator_base& operator=(const non_const_iterator& other)
-  {
-    mp_xpr = other.mp_xpr;
-    m_index = other.m_index;
-    return *this;
-  }
-
-  Derived& operator++() { --m_index; return derived(); }
-  Derived& operator--() { ++m_index; return derived(); }
-
-  Derived operator++(int) { Derived prev(derived()); operator++(); return prev;}
-  Derived operator--(int) { Derived prev(derived()); operator--(); return prev;}
-
-  friend Derived operator+(const indexed_based_stl_reverse_iterator_base& a, Index b) { Derived ret(a.derived()); ret += b; return ret; }
-  friend Derived operator-(const indexed_based_stl_reverse_iterator_base& a, Index b) { Derived ret(a.derived()); ret -= b; return ret; }
-  friend Derived operator+(Index a, const indexed_based_stl_reverse_iterator_base& b) { Derived ret(b.derived()); ret += a; return ret; }
-  friend Derived operator-(Index a, const indexed_based_stl_reverse_iterator_base& b) { Derived ret(b.derived()); ret -= a; return ret; }
-  
-  Derived& operator+=(Index b) { m_index -= b; return derived(); }
-  Derived& operator-=(Index b) { m_index += b; return derived(); }
-
-  difference_type operator-(const indexed_based_stl_reverse_iterator_base& other) const
-  {
-    eigen_assert(mp_xpr == other.mp_xpr);
-    return other.m_index - m_index;
-  }
-
-  difference_type operator-(const other_iterator& other) const
-  {
-    eigen_assert(mp_xpr == other.mp_xpr);
-    return other.m_index - m_index;
-  }
-
-  bool operator==(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
-  bool operator!=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
-  bool operator< (const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >  other.m_index; }
-  bool operator<=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
-  bool operator> (const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <  other.m_index; }
-  bool operator>=(const indexed_based_stl_reverse_iterator_base& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
-
-  bool operator==(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index == other.m_index; }
-  bool operator!=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index != other.m_index; }
-  bool operator< (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >  other.m_index; }
-  bool operator<=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index >= other.m_index; }
-  bool operator> (const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <  other.m_index; }
-  bool operator>=(const other_iterator& other) const { eigen_assert(mp_xpr == other.mp_xpr); return m_index <= other.m_index; }
-
-protected:
-
-  Derived& derived() { return static_cast<Derived&>(*this); }
-  const Derived& derived() const { return static_cast<const Derived&>(*this); }
-
-  XprType *mp_xpr;
-  Index m_index;
-};
-
-template<typename XprType>
-class pointer_based_stl_iterator
-{
-  enum { is_lvalue  = internal::is_lvalue<XprType>::value };
-  typedef pointer_based_stl_iterator<typename internal::remove_const<XprType>::type> non_const_iterator;
-  typedef pointer_based_stl_iterator<typename internal::add_const<XprType>::type> const_iterator;
-  typedef typename internal::conditional<internal::is_const<XprType>::value,non_const_iterator,const_iterator>::type other_iterator;
-  // NOTE: in C++03 we cannot declare friend classes through typedefs because we need to write friend class:
-  friend class pointer_based_stl_iterator<typename internal::add_const<XprType>::type>;
-  friend class pointer_based_stl_iterator<typename internal::remove_const<XprType>::type>;
-public:
-  typedef Index difference_type;
-  typedef typename XprType::Scalar value_type;
-  typedef std::random_access_iterator_tag iterator_category;
-  typedef typename internal::conditional<bool(is_lvalue), value_type*, const value_type*>::type pointer;
-  typedef typename internal::conditional<bool(is_lvalue), value_type&, const value_type&>::type reference;
-
-
-  pointer_based_stl_iterator() EIGEN_NO_THROW : m_ptr(0) {}
-  pointer_based_stl_iterator(XprType& xpr, Index index) EIGEN_NO_THROW : m_incr(xpr.innerStride())
-  {
-    m_ptr = xpr.data() + index * m_incr.value();
-  }
-
-  pointer_based_stl_iterator(const non_const_iterator& other) EIGEN_NO_THROW
-    : m_ptr(other.m_ptr), m_incr(other.m_incr)
-  {}
-
-  pointer_based_stl_iterator& operator=(const non_const_iterator& other) EIGEN_NO_THROW
-  {
-    m_ptr = other.m_ptr;
-    m_incr.setValue(other.m_incr);
-    return *this;
-  }
-
-  reference operator*()         const { return *m_ptr;   }
-  reference operator[](Index i) const { return *(m_ptr+i*m_incr.value()); }
-  pointer   operator->()        const { return m_ptr;    }
-
-  pointer_based_stl_iterator& operator++() { m_ptr += m_incr.value(); return *this; }
-  pointer_based_stl_iterator& operator--() { m_ptr -= m_incr.value(); return *this; }
-
-  pointer_based_stl_iterator operator++(int) { pointer_based_stl_iterator prev(*this); operator++(); return prev;}
-  pointer_based_stl_iterator operator--(int) { pointer_based_stl_iterator prev(*this); operator--(); return prev;}
-
-  friend pointer_based_stl_iterator operator+(const pointer_based_stl_iterator& a, Index b) { pointer_based_stl_iterator ret(a); ret += b; return ret; }
-  friend pointer_based_stl_iterator operator-(const pointer_based_stl_iterator& a, Index b) { pointer_based_stl_iterator ret(a); ret -= b; return ret; }
-  friend pointer_based_stl_iterator operator+(Index a, const pointer_based_stl_iterator& b) { pointer_based_stl_iterator ret(b); ret += a; return ret; }
-  friend pointer_based_stl_iterator operator-(Index a, const pointer_based_stl_iterator& b) { pointer_based_stl_iterator ret(b); ret -= a; return ret; }
-  
-  pointer_based_stl_iterator& operator+=(Index b) { m_ptr += b*m_incr.value(); return *this; }
-  pointer_based_stl_iterator& operator-=(Index b) { m_ptr -= b*m_incr.value(); return *this; }
-
-  difference_type operator-(const pointer_based_stl_iterator& other) const {
-    return (m_ptr - other.m_ptr)/m_incr.value();
-  }
-
-  difference_type operator-(const other_iterator& other) const {
-    return (m_ptr - other.m_ptr)/m_incr.value();
-  }
-
-  bool operator==(const pointer_based_stl_iterator& other) const { return m_ptr == other.m_ptr; }
-  bool operator!=(const pointer_based_stl_iterator& other) const { return m_ptr != other.m_ptr; }
-  bool operator< (const pointer_based_stl_iterator& other) const { return m_ptr <  other.m_ptr; }
-  bool operator<=(const pointer_based_stl_iterator& other) const { return m_ptr <= other.m_ptr; }
-  bool operator> (const pointer_based_stl_iterator& other) const { return m_ptr >  other.m_ptr; }
-  bool operator>=(const pointer_based_stl_iterator& other) const { return m_ptr >= other.m_ptr; }
-
-  bool operator==(const other_iterator& other) const { return m_ptr == other.m_ptr; }
-  bool operator!=(const other_iterator& other) const { return m_ptr != other.m_ptr; }
-  bool operator< (const other_iterator& other) const { return m_ptr <  other.m_ptr; }
-  bool operator<=(const other_iterator& other) const { return m_ptr <= other.m_ptr; }
-  bool operator> (const other_iterator& other) const { return m_ptr >  other.m_ptr; }
-  bool operator>=(const other_iterator& other) const { return m_ptr >= other.m_ptr; }
-
-protected:
-
-  pointer m_ptr;
-  internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_incr;
-};
-
-template<typename _XprType>
-struct indexed_based_stl_iterator_traits<generic_randaccess_stl_iterator<_XprType> >
-{
-  typedef _XprType XprType;
-  typedef generic_randaccess_stl_iterator<typename internal::remove_const<XprType>::type> non_const_iterator;
-  typedef generic_randaccess_stl_iterator<typename internal::add_const<XprType>::type> const_iterator;
-};
-
-template<typename XprType>
-class generic_randaccess_stl_iterator : public indexed_based_stl_iterator_base<generic_randaccess_stl_iterator<XprType> >
-{
-public:
-  typedef typename XprType::Scalar value_type;
-
-protected:
-
-  enum {
-    has_direct_access = (internal::traits<XprType>::Flags & DirectAccessBit) ? 1 : 0,
-    is_lvalue  = internal::is_lvalue<XprType>::value
-  };
-
-  typedef indexed_based_stl_iterator_base<generic_randaccess_stl_iterator> Base;
-  using Base::m_index;
-  using Base::mp_xpr;
-
-  // TODO currently const Transpose/Reshape expressions never returns const references,
-  // so lets return by value too.
-  //typedef typename internal::conditional<bool(has_direct_access), const value_type&, const value_type>::type read_only_ref_t;
-  typedef const value_type read_only_ref_t;
-
-public:
-  
-  typedef typename internal::conditional<bool(is_lvalue), value_type *, const value_type *>::type pointer;
-  typedef typename internal::conditional<bool(is_lvalue), value_type&, read_only_ref_t>::type reference;
-  
-  generic_randaccess_stl_iterator() : Base() {}
-  generic_randaccess_stl_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
-  generic_randaccess_stl_iterator(const typename Base::non_const_iterator& other) : Base(other) {}
-  using Base::operator=;
-
-  reference operator*()         const { return   (*mp_xpr)(m_index);   }
-  reference operator[](Index i) const { return   (*mp_xpr)(m_index+i); }
-  pointer   operator->()        const { return &((*mp_xpr)(m_index)); }
-};
-
-template<typename _XprType, DirectionType Direction>
-struct indexed_based_stl_iterator_traits<subvector_stl_iterator<_XprType,Direction> >
-{
-  typedef _XprType XprType;
-  typedef subvector_stl_iterator<typename internal::remove_const<XprType>::type, Direction> non_const_iterator;
-  typedef subvector_stl_iterator<typename internal::add_const<XprType>::type, Direction> const_iterator;
-};
-
-template<typename XprType, DirectionType Direction>
-class subvector_stl_iterator : public indexed_based_stl_iterator_base<subvector_stl_iterator<XprType,Direction> >
-{
-protected:
-
-  enum { is_lvalue  = internal::is_lvalue<XprType>::value };
-
-  typedef indexed_based_stl_iterator_base<subvector_stl_iterator> Base;
-  using Base::m_index;
-  using Base::mp_xpr;
-
-  typedef typename internal::conditional<Direction==Vertical,typename XprType::ColXpr,typename XprType::RowXpr>::type SubVectorType;
-  typedef typename internal::conditional<Direction==Vertical,typename XprType::ConstColXpr,typename XprType::ConstRowXpr>::type ConstSubVectorType;
-
-
-public:
-  typedef typename internal::conditional<bool(is_lvalue), SubVectorType, ConstSubVectorType>::type reference;
-  typedef typename reference::PlainObject value_type;
-
-private:
-  class subvector_stl_iterator_ptr
-  {
-  public:
-      subvector_stl_iterator_ptr(const reference &subvector) : m_subvector(subvector) {}
-      reference* operator->() { return &m_subvector; }
-  private:
-      reference m_subvector;
-  };
-public:
-
-  typedef subvector_stl_iterator_ptr pointer;
-  
-  subvector_stl_iterator() : Base() {}
-  subvector_stl_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
-
-  reference operator*()         const { return (*mp_xpr).template subVector<Direction>(m_index); }
-  reference operator[](Index i) const { return (*mp_xpr).template subVector<Direction>(m_index+i); }
-  pointer   operator->()        const { return (*mp_xpr).template subVector<Direction>(m_index); }
-};
-
-template<typename _XprType, DirectionType Direction>
-struct indexed_based_stl_iterator_traits<subvector_stl_reverse_iterator<_XprType,Direction> >
-{
-  typedef _XprType XprType;
-  typedef subvector_stl_reverse_iterator<typename internal::remove_const<XprType>::type, Direction> non_const_iterator;
-  typedef subvector_stl_reverse_iterator<typename internal::add_const<XprType>::type, Direction> const_iterator;
-};
-
-template<typename XprType, DirectionType Direction>
-class subvector_stl_reverse_iterator : public indexed_based_stl_reverse_iterator_base<subvector_stl_reverse_iterator<XprType,Direction> >
-{
-protected:
-
-  enum { is_lvalue  = internal::is_lvalue<XprType>::value };
-
-  typedef indexed_based_stl_reverse_iterator_base<subvector_stl_reverse_iterator> Base;
-  using Base::m_index;
-  using Base::mp_xpr;
-
-  typedef typename internal::conditional<Direction==Vertical,typename XprType::ColXpr,typename XprType::RowXpr>::type SubVectorType;
-  typedef typename internal::conditional<Direction==Vertical,typename XprType::ConstColXpr,typename XprType::ConstRowXpr>::type ConstSubVectorType;
-
-
-public:
-  typedef typename internal::conditional<bool(is_lvalue), SubVectorType, ConstSubVectorType>::type reference;
-  typedef typename reference::PlainObject value_type;
-
-private:
-  class subvector_stl_reverse_iterator_ptr
-  {
-  public:
-      subvector_stl_reverse_iterator_ptr(const reference &subvector) : m_subvector(subvector) {}
-      reference* operator->() { return &m_subvector; }
-  private:
-      reference m_subvector;
-  };
-public:
-
-  typedef subvector_stl_reverse_iterator_ptr pointer;
-  
-  subvector_stl_reverse_iterator() : Base() {}
-  subvector_stl_reverse_iterator(XprType& xpr, Index index) : Base(xpr,index) {}
-
-  reference operator*()         const { return (*mp_xpr).template subVector<Direction>(m_index); }
-  reference operator[](Index i) const { return (*mp_xpr).template subVector<Direction>(m_index+i); }
-  pointer   operator->()        const { return (*mp_xpr).template subVector<Direction>(m_index); }
-};
-
-} // namespace internal
-
-
-/** returns an iterator to the first element of the 1D vector or array
-  * \only_for_vectors
-  * \sa end(), cbegin()
-  */
-template<typename Derived>
-inline typename DenseBase<Derived>::iterator DenseBase<Derived>::begin()
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return iterator(derived(), 0);
-}
-
-/** const version of begin() */
-template<typename Derived>
-inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::begin() const
-{
-  return cbegin();
-}
-
-/** returns a read-only const_iterator to the first element of the 1D vector or array
-  * \only_for_vectors
-  * \sa cend(), begin()
-  */
-template<typename Derived>
-inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::cbegin() const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return const_iterator(derived(), 0);
-}
-
-/** returns an iterator to the element following the last element of the 1D vector or array
-  * \only_for_vectors
-  * \sa begin(), cend()
-  */
-template<typename Derived>
-inline typename DenseBase<Derived>::iterator DenseBase<Derived>::end()
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return iterator(derived(), size());
-}
-
-/** const version of end() */
-template<typename Derived>
-inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::end() const
-{
-  return cend();
-}
-
-/** returns a read-only const_iterator to the element following the last element of the 1D vector or array
-  * \only_for_vectors
-  * \sa begin(), cend()
-  */
-template<typename Derived>
-inline typename DenseBase<Derived>::const_iterator DenseBase<Derived>::cend() const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return const_iterator(derived(), size());
-}
-
-} // namespace Eigen
-
-#endif // EIGEN_STLITERATORS_H

Eigen/src/Core/Stride.h

@@ -38,14 +38,6 @@ namespace Eigen {
   * \include Map_general_stride.cpp
   * Output: \verbinclude Map_general_stride.out
   *
-  * Both strides can be negative. However, a negative stride of -1 cannot be specified at compile time
-  * because of the ambiguity with Dynamic which is defined to -1 (historically, negative strides were
-  * not allowed).
-  *
-  * Note that for compile-time vectors (ColsAtCompileTime==1 or RowsAtCompile==1),
-  * the inner stride is the pointer increment between two consecutive elements,
-  * regardless of storage layout.
-  *
   * \sa class InnerStride, class OuterStride, \ref TopicStorageOrders
   */
 template<int _OuterStrideAtCompileTime, int _InnerStrideAtCompileTime>
@@ -63,8 +55,6 @@ class Stride
     Stride()
       : m_outer(OuterStrideAtCompileTime), m_inner(InnerStrideAtCompileTime)
     {
-      // FIXME: for Eigen 4 we should use DynamicIndex instead of Dynamic.
-      // FIXME: for Eigen 4 we should also unify this API with fix<>
       eigen_assert(InnerStrideAtCompileTime != Dynamic && OuterStrideAtCompileTime != Dynamic);
     }
 
@@ -73,6 +63,7 @@ class Stride
     Stride(Index outerStride, Index innerStride)
       : m_outer(outerStride), m_inner(innerStride)
     {
+      eigen_assert(innerStride>=0 && outerStride>=0);
     }
 
     /** Copy constructor */
@@ -82,10 +73,10 @@ class Stride
     {}
 
     /** \returns the outer stride */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index outer() const { return m_outer.value(); }
     /** \returns the inner stride */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
     inline Index inner() const { return m_inner.value(); }
 
   protected:

Eigen/src/Core/Swap.h

@@ -30,13 +30,12 @@ public:
   typedef typename Base::DstXprType DstXprType;
   typedef swap_assign_op<Scalar> Functor;
   
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
-  generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
     : Base(dst, src, func, dstExpr)
   {}
   
   template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
+  void assignPacket(Index row, Index col)
   {
     PacketType tmp = m_src.template packet<LoadMode,PacketType>(row,col);
     const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(row,col, m_dst.template packet<StoreMode,PacketType>(row,col));
@@ -44,7 +43,7 @@ public:
   }
   
   template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_STRONG_INLINE void assignPacket(Index index)
+  void assignPacket(Index index)
   {
     PacketType tmp = m_src.template packet<LoadMode,PacketType>(index);
     const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(index, m_dst.template packet<StoreMode,PacketType>(index));
@@ -53,7 +52,7 @@ public:
   
   // TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
   template<int StoreMode, int LoadMode, typename PacketType>
-  EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
+  void assignPacketByOuterInner(Index outer, Index inner)
   {
     Index row = Base::rowIndexByOuterInner(outer, inner); 
     Index col = Base::colIndexByOuterInner(outer, inner);

Eigen/src/Core/Transpose.h

@@ -61,27 +61,24 @@ template<typename MatrixType> class Transpose
     typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
     EIGEN_DEVICE_FUNC
-    explicit EIGEN_STRONG_INLINE Transpose(MatrixType& matrix) : m_matrix(matrix) {}
+    explicit inline Transpose(MatrixType& matrix) : m_matrix(matrix) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Transpose)
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.cols(); }
-    Index rows() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
     typename internal::remove_reference<MatrixTypeNested>::type&
     nestedExpression() { return m_matrix; }
 
     /** \internal */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     void resize(Index nrows, Index ncols) {
       m_matrix.resize(ncols,nrows);
     }
@@ -125,10 +122,8 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
     EIGEN_DENSE_PUBLIC_INTERFACE(Transpose<MatrixType>)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TransposeImpl)
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
-    Index innerStride() const { return derived().nestedExpression().innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    Index outerStride() const { return derived().nestedExpression().outerStride(); }
 
     typedef typename internal::conditional<
                        internal::is_lvalue<MatrixType>::value,
@@ -136,20 +131,18 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
-    ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return derived().nestedExpression().data(); }
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
-    const Scalar* data() const { return derived().nestedExpression().data(); }
 
     // FIXME: shall we keep the const version of coeffRef?
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    const Scalar& coeffRef(Index rowId, Index colId) const
+    inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return derived().nestedExpression().coeffRef(colId, rowId);
     }
 
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    const Scalar& coeffRef(Index index) const
+    inline const Scalar& coeffRef(Index index) const
     {
       return derived().nestedExpression().coeffRef(index);
     }
@@ -177,8 +170,7 @@ template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
   *
   * \sa transposeInPlace(), adjoint() */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC inline Transpose<Derived>
-typename DenseBase<Derived>::TransposeReturnType
 DenseBase<Derived>::transpose()
 {
   return TransposeReturnType(derived());
@@ -190,8 +182,7 @@ DenseBase<Derived>::transpose()
   *
   * \sa transposeInPlace(), adjoint() */
 template<typename Derived>
-EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+EIGEN_DEVICE_FUNC inline typename DenseBase<Derived>::ConstTransposeReturnType
-const typename DenseBase<Derived>::ConstTransposeReturnType
 DenseBase<Derived>::transpose() const
 {
   return ConstTransposeReturnType(derived());
@@ -239,10 +230,11 @@ struct inplace_transpose_selector;
 template<typename MatrixType>
 struct inplace_transpose_selector<MatrixType,true,false> { // square matrix
   static void run(MatrixType& m) {
-    m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose().template triangularView<StrictlyUpper>());
+    m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
   }
 };
 
+// TODO: vectorized path is currently limited to LargestPacketSize x LargestPacketSize cases only.
 template<typename MatrixType>
 struct inplace_transpose_selector<MatrixType,true,true> { // PacketSize x PacketSize
   static void run(MatrixType& m) {
@@ -259,66 +251,16 @@ struct inplace_transpose_selector<MatrixType,true,true> { // PacketSize x Packet
   }
 };
 
-
-template <typename MatrixType, Index Alignment>
-void BlockedInPlaceTranspose(MatrixType& m) {
-  typedef typename MatrixType::Scalar Scalar;
-  typedef typename internal::packet_traits<typename MatrixType::Scalar>::type Packet;
-  const Index PacketSize = internal::packet_traits<Scalar>::size;
-  eigen_assert(m.rows() == m.cols());
-  int row_start = 0;
-  for (; row_start + PacketSize <= m.rows(); row_start += PacketSize) {
-    for (int col_start = row_start; col_start + PacketSize <= m.cols(); col_start += PacketSize) {
-      PacketBlock<Packet> A;
-      if (row_start == col_start) {
-        for (Index i=0; i<PacketSize; ++i)
-          A.packet[i] = m.template packetByOuterInner<Alignment>(row_start + i,col_start);
-        internal::ptranspose(A);
-        for (Index i=0; i<PacketSize; ++i)
-          m.template writePacket<Alignment>(m.rowIndexByOuterInner(row_start + i, col_start), m.colIndexByOuterInner(row_start + i,col_start), A.packet[i]);
-      } else {
-        PacketBlock<Packet> B;
-        for (Index i=0; i<PacketSize; ++i) {
-          A.packet[i] = m.template packetByOuterInner<Alignment>(row_start + i,col_start);
-          B.packet[i] = m.template packetByOuterInner<Alignment>(col_start + i, row_start);
-        }
-        internal::ptranspose(A);
-        internal::ptranspose(B);
-        for (Index i=0; i<PacketSize; ++i) {
-          m.template writePacket<Alignment>(m.rowIndexByOuterInner(row_start + i, col_start), m.colIndexByOuterInner(row_start + i,col_start), B.packet[i]);
-          m.template writePacket<Alignment>(m.rowIndexByOuterInner(col_start + i, row_start), m.colIndexByOuterInner(col_start + i,row_start), A.packet[i]);
-        }
-      }
-    }
-  }
-  for (Index row = row_start; row < m.rows(); ++row) {
-    m.matrix().row(row).head(row).swap(
-        m.matrix().col(row).head(row).transpose());
-  }
-}
-
 template<typename MatrixType,bool MatchPacketSize>
-struct inplace_transpose_selector<MatrixType,false,MatchPacketSize> { // non square or dynamic matrix
+struct inplace_transpose_selector<MatrixType,false,MatchPacketSize> { // non square matrix
   static void run(MatrixType& m) {
-    typedef typename MatrixType::Scalar Scalar;
+    if (m.rows()==m.cols())
-    if (m.rows() == m.cols()) {
+      m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
-      const Index PacketSize = internal::packet_traits<Scalar>::size;
+    else
-      if (!NumTraits<Scalar>::IsComplex && m.rows() >= PacketSize) {
-        if ((m.rows() % PacketSize) == 0)
-          BlockedInPlaceTranspose<MatrixType,internal::evaluator<MatrixType>::Alignment>(m);
-        else
-          BlockedInPlaceTranspose<MatrixType,Unaligned>(m);
-      }
-      else {
-        m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose().template triangularView<StrictlyUpper>());
-      }
-    } else {
       m = m.transpose().eval();
-    }
   }
 };
 
-
 } // end namespace internal
 
 /** This is the "in place" version of transpose(): it replaces \c *this by its own transpose.
@@ -451,8 +393,7 @@ struct checkTransposeAliasing_impl<Derived, OtherDerived, false>
 template<typename Dst, typename Src>
 void check_for_aliasing(const Dst &dst, const Src &src)
 {
-  if((!Dst::IsVectorAtCompileTime) && dst.rows()>1 && dst.cols()>1)
+  internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
-    internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
 }
 
 } // end namespace internal

Eigen/src/Core/Transpositions.h

@@ -23,9 +23,7 @@ class TranspositionsBase
     typedef typename IndicesType::Scalar StorageIndex;
     typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
-    EIGEN_DEVICE_FUNC
     Derived& derived() { return *static_cast<Derived*>(this); }
-    EIGEN_DEVICE_FUNC
     const Derived& derived() const { return *static_cast<const Derived*>(this); }
 
     /** Copies the \a other transpositions into \c *this */
@@ -37,17 +35,13 @@ class TranspositionsBase
     }
 
     /** \returns the number of transpositions */
-    EIGEN_DEVICE_FUNC
     Index size() const { return indices().size(); }
     /** \returns the number of rows of the equivalent permutation matrix */
-    EIGEN_DEVICE_FUNC
     Index rows() const { return indices().size(); }
     /** \returns the number of columns of the equivalent permutation matrix */
-    EIGEN_DEVICE_FUNC
     Index cols() const { return indices().size(); }
 
     /** Direct access to the underlying index vector */
-    EIGEN_DEVICE_FUNC
     inline const StorageIndex& coeff(Index i) const { return indices().coeff(i); }
     /** Direct access to the underlying index vector */
     inline StorageIndex& coeffRef(Index i) { return indices().coeffRef(i); }
@@ -61,10 +55,8 @@ class TranspositionsBase
     inline StorageIndex& operator[](Index i) { return indices()(i); }
 
     /** const version of indices(). */
-    EIGEN_DEVICE_FUNC
     const IndicesType& indices() const { return derived().indices(); }
     /** \returns a reference to the stored array representing the transpositions. */
-    EIGEN_DEVICE_FUNC
     IndicesType& indices() { return derived().indices(); }
 
     /** Resizes to given size. */
@@ -81,7 +73,7 @@ class TranspositionsBase
     }
 
     // FIXME: do we want such methods ?
-    // might be useful when the target matrix expression is complex, e.g.:
+    // might be usefull when the target matrix expression is complex, e.g.:
     // object.matrix().block(..,..,..,..) = trans * object.matrix().block(..,..,..,..);
     /*
     template<typename MatrixType>
@@ -186,10 +178,8 @@ class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTim
     {}
 
     /** const version of indices(). */
-    EIGEN_DEVICE_FUNC
     const IndicesType& indices() const { return m_indices; }
     /** \returns a reference to the stored array representing the transpositions. */
-    EIGEN_DEVICE_FUNC
     IndicesType& indices() { return m_indices; }
 
   protected:
@@ -247,11 +237,9 @@ class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,P
     #endif
 
     /** const version of indices(). */
-    EIGEN_DEVICE_FUNC
     const IndicesType& indices() const { return m_indices; }
     
     /** \returns a reference to the stored array representing the transpositions. */
-    EIGEN_DEVICE_FUNC
     IndicesType& indices() { return m_indices; }
 
   protected:
@@ -291,11 +279,9 @@ class TranspositionsWrapper
     }
 
     /** const version of indices(). */
-    EIGEN_DEVICE_FUNC
     const IndicesType& indices() const { return m_indices; }
 
     /** \returns a reference to the stored array representing the transpositions. */
-    EIGEN_DEVICE_FUNC
     IndicesType& indices() { return m_indices; }
 
   protected:
@@ -349,12 +335,9 @@ class Transpose<TranspositionsBase<TranspositionsDerived> >
 
     explicit Transpose(const TranspositionType& t) : m_transpositions(t) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    Index size() const { return m_transpositions.size(); }
-    Index size() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
+    Index rows() const { return m_transpositions.size(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    Index cols() const { return m_transpositions.size(); }
-    Index rows() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
-    Index cols() const EIGEN_NOEXCEPT { return m_transpositions.size(); }
 
     /** \returns the \a matrix with the inverse transpositions applied to the columns.
       */
@@ -374,7 +357,6 @@ class Transpose<TranspositionsBase<TranspositionsDerived> >
       return Product<Transpose, OtherDerived, AliasFreeProduct>(*this, matrix.derived());
     }
     
-    EIGEN_DEVICE_FUNC
     const TranspositionType& nestedExpression() const { return m_transpositions; }
 
   protected:

Eigen/src/Core/TriangularMatrix.h

@@ -53,19 +53,18 @@ template<typename Derived> class TriangularBase : public EigenBase<Derived>
     typedef Derived const& Nested;
 
     EIGEN_DEVICE_FUNC
-    inline TriangularBase() { eigen_assert(!((int(Mode) & int(UnitDiag)) && (int(Mode) & int(ZeroDiag)))); }
+    inline TriangularBase() { eigen_assert(!((Mode&UnitDiag) && (Mode&ZeroDiag))); }
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rows() const EIGEN_NOEXCEPT { return derived().rows(); }
+    inline Index rows() const { return derived().rows(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return derived().cols(); }
+    inline Index cols() const { return derived().cols(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index outerStride() const EIGEN_NOEXCEPT { return derived().outerStride(); }
+    inline Index outerStride() const { return derived().outerStride(); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index innerStride() const EIGEN_NOEXCEPT { return derived().innerStride(); }
+    inline Index innerStride() const { return derived().innerStride(); }
 
     // dummy resize function
-    EIGEN_DEVICE_FUNC
     void resize(Index rows, Index cols)
     {
       EIGEN_UNUSED_VARIABLE(rows);
@@ -100,10 +99,12 @@ template<typename Derived> class TriangularBase : public EigenBase<Derived>
       return coeffRef(row,col);
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
     EIGEN_DEVICE_FUNC
     inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
     EIGEN_DEVICE_FUNC
     inline Derived& derived() { return *static_cast<Derived*>(this); }
+    #endif // not EIGEN_PARSED_BY_DOXYGEN
 
     template<typename DenseDerived>
     EIGEN_DEVICE_FUNC
@@ -196,7 +197,6 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularView
     typedef typename internal::traits<TriangularView>::MatrixTypeNestedNonRef MatrixTypeNestedNonRef;
 
     typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
-    typedef TriangularView<typename internal::add_const<MatrixType>::type, _Mode> ConstTriangularView;
 
   public:
 
@@ -220,11 +220,11 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularView
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TriangularView)
 
     /** \copydoc EigenBase::rows() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); }
+    inline Index rows() const { return m_matrix.rows(); }
     /** \copydoc EigenBase::cols() */
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); }
+    inline Index cols() const { return m_matrix.cols(); }
 
     /** \returns a const reference to the nested expression */
     EIGEN_DEVICE_FUNC
@@ -240,18 +240,6 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularView
     inline const ConjugateReturnType conjugate() const
     { return ConjugateReturnType(m_matrix.conjugate()); }
 
-    /** \returns an expression of the complex conjugate of \c *this if Cond==true,
-     *           returns \c *this otherwise.
-     */
-    template<bool Cond>
-    EIGEN_DEVICE_FUNC
-    inline typename internal::conditional<Cond,ConjugateReturnType,ConstTriangularView>::type
-    conjugateIf() const
-    {
-      typedef typename internal::conditional<Cond,ConjugateReturnType,ConstTriangularView>::type ReturnType;
-      return ReturnType(m_matrix.template conjugateIf<Cond>());
-    }
-
     typedef TriangularView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
     /** \sa MatrixBase::adjoint() const */
     EIGEN_DEVICE_FUNC
@@ -440,19 +428,21 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
     EIGEN_DEVICE_FUNC
     TriangularViewType& operator=(const MatrixBase<OtherDerived>& other);
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
     EIGEN_DEVICE_FUNC
     TriangularViewType& operator=(const TriangularViewImpl& other)
     { return *this = other.derived().nestedExpression(); }
 
-    template<typename OtherDerived>
     /** \deprecated */
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void lazyAssign(const TriangularBase<OtherDerived>& other);
 
-    template<typename OtherDerived>
     /** \deprecated */
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void lazyAssign(const MatrixBase<OtherDerived>& other);
+#endif
 
     /** Efficient triangular matrix times vector/matrix product */
     template<typename OtherDerived>
@@ -478,7 +468,7 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
       * \a Side==OnTheRight.
       *
-      * Note that the template parameter \c Side can be omitted, in which case \c Side==OnTheLeft
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
       *
       * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
       * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
@@ -496,6 +486,7 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       * \sa TriangularView::solveInPlace()
       */
     template<int Side, typename Other>
+    EIGEN_DEVICE_FUNC
     inline const internal::triangular_solve_retval<Side,TriangularViewType, Other>
     solve(const MatrixBase<Other>& other) const;
 
@@ -504,7 +495,7 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
       * This function will const_cast it, so constness isn't honored here.
       *
-      * Note that the template parameter \c Side can be omitted, in which case \c Side==OnTheLeft
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
       *
       * See TriangularView:solve() for the details.
       */
@@ -520,16 +511,20 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
     /** Swaps the coefficients of the common triangular parts of two matrices */
     template<typename OtherDerived>
     EIGEN_DEVICE_FUNC
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+    void swap(TriangularBase<OtherDerived> &other)
+#else
     void swap(TriangularBase<OtherDerived> const & other)
+#endif
     {
       EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
       call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
     }
 
-    /** Shortcut for \code (*this).swap(other.triangularView<(*this)::Mode>()) \endcode */
+    /** \deprecated
+      * Shortcut for \code (*this).swap(other.triangularView<(*this)::Mode>()) \endcode */
     template<typename OtherDerived>
-    /** \deprecated */
+    EIGEN_DEVICE_FUNC
-    EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
     void swap(MatrixBase<OtherDerived> const & other)
     {
       EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
@@ -544,10 +539,9 @@ template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_Mat
       this->solveInPlace(dst);
     }
 
-    template <typename ProductType>
+    template<typename ProductType>
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha,
+    EIGEN_DEVICE_FUNC
-                                                                           bool beta);
+    EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha, bool beta);
-
   protected:
     EIGEN_DEFAULT_COPY_CONSTRUCTOR(TriangularViewImpl)
     EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(TriangularViewImpl)
@@ -632,8 +626,7 @@ EIGEN_DEVICE_FUNC void TriangularBase<Derived>::evalTo(MatrixBase<DenseDerived>
   */
 template<typename Derived>
 template<unsigned int Mode>
-EIGEN_DEVICE_FUNC
+EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::template TriangularViewReturnType<Mode>::Type
-typename MatrixBase<Derived>::template TriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView()
 {
   return typename TriangularViewReturnType<Mode>::Type(derived());
@@ -642,8 +635,7 @@ MatrixBase<Derived>::triangularView()
 /** This is the const version of MatrixBase::triangularView() */
 template<typename Derived>
 template<unsigned int Mode>
-EIGEN_DEVICE_FUNC
+EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::template ConstTriangularViewReturnType<Mode>::Type
-typename MatrixBase<Derived>::template ConstTriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView() const
 {
   return typename ConstTriangularViewReturnType<Mode>::Type(derived());
@@ -725,7 +717,6 @@ struct unary_evaluator<TriangularView<MatrixType,Mode>, IndexBased>
 {
   typedef TriangularView<MatrixType,Mode> XprType;
   typedef evaluator<typename internal::remove_all<MatrixType>::type> Base;
-  EIGEN_DEVICE_FUNC
   unary_evaluator(const XprType &xpr) : Base(xpr.nestedExpression()) {}
 };
 
@@ -812,7 +803,7 @@ void call_triangular_assignment_loop(DstXprType& dst, const SrcXprType& src, con
   enum {
       unroll = DstXprType::SizeAtCompileTime != Dynamic
             && SrcEvaluatorType::CoeffReadCost < HugeCost
-            && DstXprType::SizeAtCompileTime * (int(DstEvaluatorType::CoeffReadCost) + int(SrcEvaluatorType::CoeffReadCost)) / 2 <= EIGEN_UNROLLING_LIMIT
+            && DstXprType::SizeAtCompileTime * (DstEvaluatorType::CoeffReadCost+SrcEvaluatorType::CoeffReadCost) / 2 <= EIGEN_UNROLLING_LIMIT
     };
 
   triangular_assignment_loop<Kernel, Mode, unroll ? int(DstXprType::SizeAtCompileTime) : Dynamic, SetOpposite>::run(kernel);
@@ -846,7 +837,7 @@ struct Assignment<DstXprType, SrcXprType, Functor, Triangular2Dense>
 {
   EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
   {
-    call_triangular_assignment_loop<SrcXprType::Mode, (int(SrcXprType::Mode) & int(SelfAdjoint)) == 0>(dst, src, func);
+    call_triangular_assignment_loop<SrcXprType::Mode, (SrcXprType::Mode&SelfAdjoint)==0>(dst, src, func);
   }
 };
 
@@ -944,7 +935,7 @@ template<typename DenseDerived>
 EIGEN_DEVICE_FUNC void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
 {
   other.derived().resize(this->rows(), this->cols());
-  internal::call_triangular_assignment_loop<Derived::Mode, (int(Derived::Mode) & int(SelfAdjoint)) == 0 /* SetOpposite */>(other.derived(), derived().nestedExpression());
+  internal::call_triangular_assignment_loop<Derived::Mode,(Derived::Mode&SelfAdjoint)==0 /* SetOpposite */>(other.derived(), derived().nestedExpression());
 }
 
 namespace internal {
@@ -954,14 +945,14 @@ template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
 struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
 {
   typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename SrcXprType::Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename SrcXprType::Scalar> &)
   {
     Index dstRows = src.rows();
     Index dstCols = src.cols();
     if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
       dst.resize(dstRows, dstCols);
 
-    dst._assignProduct(src, Scalar(1), false);
+    dst._assignProduct(src, 1, 0);
   }
 };
 
@@ -970,9 +961,9 @@ template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
 struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::add_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
 {
   typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,typename SrcXprType::Scalar> &)
   {
-    dst._assignProduct(src, Scalar(1), true);
+    dst._assignProduct(src, 1, 1);
   }
 };
 
@@ -981,9 +972,9 @@ template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
 struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::sub_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
 {
   typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
-  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  static EIGEN_DEVICE_FUNC void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,typename SrcXprType::Scalar> &)
   {
-    dst._assignProduct(src, Scalar(-1), true);
+    dst._assignProduct(src, -1, 1);
   }
 };
 

Eigen/src/Core/VectorBlock.h

@@ -35,7 +35,7 @@ struct traits<VectorBlock<VectorType, Size> >
   * It is the return type of DenseBase::segment(Index,Index) and DenseBase::segment<int>(Index) and
   * most of the time this is the only way it is used.
   *
-  * However, if you want to directly manipulate sub-vector expressions,
+  * However, if you want to directly maniputate sub-vector expressions,
   * for instance if you want to write a function returning such an expression, you
   * will need to use this class.
   *
@@ -71,8 +71,8 @@ template<typename VectorType, int Size> class VectorBlock
 
     /** Dynamic-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    VectorBlock(VectorType& vector, Index start, Index size)
+    inline VectorBlock(VectorType& vector, Index start, Index size)
       : Base(vector,
              IsColVector ? start : 0, IsColVector ? 0 : start,
              IsColVector ? size  : 1, IsColVector ? 1 : size)
@@ -82,8 +82,8 @@ template<typename VectorType, int Size> class VectorBlock
 
     /** Fixed-size constructor
       */
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    EIGEN_DEVICE_FUNC
-    VectorBlock(VectorType& vector, Index start)
+    inline VectorBlock(VectorType& vector, Index start)
       : Base(vector, IsColVector ? start : 0, IsColVector ? 0 : start)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorBlock);

Eigen/src/Core/VectorwiseOp.h

@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2019 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -65,10 +65,10 @@ class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<Matri
     explicit PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
       : m_matrix(mat), m_functor(func) {}
 
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index rows() const EIGEN_NOEXCEPT { return (Direction==Vertical   ? 1 : m_matrix.rows()); }
+    Index rows() const { return (Direction==Vertical   ? 1 : m_matrix.rows()); }
-    EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR
+    EIGEN_DEVICE_FUNC
-    Index cols() const EIGEN_NOEXCEPT { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
+    Index cols() const { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
 
     EIGEN_DEVICE_FUNC
     typename MatrixType::Nested nestedExpression() const { return m_matrix; }
@@ -81,46 +81,39 @@ class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<Matri
     const MemberOp m_functor;
 };
 
-template<typename A,typename B> struct partial_redux_dummy_func;
+#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST)                               \
-
+  template <typename ResultType>                                        \
-#define EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,VECTORIZABLE,BINARYOP)                \
+  struct member_##MEMBER {                                              \
-  template <typename ResultType,typename Scalar>                                                            \
+    EIGEN_EMPTY_STRUCT_CTOR(member_##MEMBER)                            \
-  struct member_##MEMBER {                                                                  \
+    typedef ResultType result_type;                                     \
-    EIGEN_EMPTY_STRUCT_CTOR(member_##MEMBER)                                                \
+    template<typename Scalar, int Size> struct Cost                     \
-    typedef ResultType result_type;                                                         \
+    { enum { value = COST }; };                                         \
-    typedef BINARYOP<Scalar,Scalar> BinaryOp;   \
+    template<typename XprType>                                          \
-    template<int Size> struct Cost { enum { value = COST }; };             \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE                               \
-    enum { Vectorizable = VECTORIZABLE };                                                   \
+    ResultType operator()(const XprType& mat) const                     \
-    template<typename XprType>                                                              \
+    { return mat.MEMBER(); } \
-    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE                                                   \
-    ResultType operator()(const XprType& mat) const                                         \
-    { return mat.MEMBER(); }                                                                \
-    BinaryOp binaryFunc() const { return BinaryOp(); }                                      \
   }
 
-#define EIGEN_MEMBER_FUNCTOR(MEMBER,COST) \
-  EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(MEMBER,COST,0,partial_redux_dummy_func)
-
 namespace internal {
 
+EIGEN_MEMBER_FUNCTOR(squaredNorm, Size * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(norm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(stableNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(blueNorm, (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(hypotNorm, (Size-1) * functor_traits<scalar_hypot_op<Scalar> >::Cost );
+EIGEN_MEMBER_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost);
+EIGEN_MEMBER_FUNCTOR(mean, (Size-1)*NumTraits<Scalar>::AddCost + NumTraits<Scalar>::MulCost);
+EIGEN_MEMBER_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
+EIGEN_MEMBER_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(all, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(any, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
+EIGEN_MEMBER_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost);
 
-EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(sum, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_sum_op);
+template <int p, typename ResultType>
-EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(minCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_min_op);
-EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(maxCoeff, (Size-1)*NumTraits<Scalar>::AddCost, 1, internal::scalar_max_op);
-EIGEN_MAKE_PARTIAL_REDUX_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost, 1, internal::scalar_product_op);
-
-template <int p, typename ResultType,typename Scalar>
 struct member_lpnorm {
   typedef ResultType result_type;
-  enum { Vectorizable = 0 };
+  template<typename Scalar, int Size> struct Cost
-  template<int Size> struct Cost
   { enum { value = (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost }; };
   EIGEN_DEVICE_FUNC member_lpnorm() {}
   template<typename XprType>
@@ -128,20 +121,17 @@ struct member_lpnorm {
   { return mat.template lpNorm<p>(); }
 };
 
-template <typename BinaryOpT, typename Scalar>
+template <typename BinaryOp, typename Scalar>
 struct member_redux {
-  typedef BinaryOpT BinaryOp;
   typedef typename result_of<
                      BinaryOp(const Scalar&,const Scalar&)
                    >::type  result_type;
-
+  template<typename _Scalar, int Size> struct Cost
-  enum { Vectorizable = functor_traits<BinaryOp>::PacketAccess };
+  { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
-  template<int Size> struct Cost { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
   EIGEN_DEVICE_FUNC explicit member_redux(const BinaryOp func) : m_functor(func) {}
   template<typename Derived>
   EIGEN_DEVICE_FUNC inline result_type operator()(const DenseBase<Derived>& mat) const
   { return mat.redux(m_functor); }
-  const BinaryOp& binaryFunc() const { return m_functor; }
   const BinaryOp m_functor;
 };
 }
@@ -149,38 +139,18 @@ struct member_redux {
 /** \class VectorwiseOp
   * \ingroup Core_Module
   *
-  * \brief Pseudo expression providing broadcasting and partial reduction operations
+  * \brief Pseudo expression providing partial reduction operations
   *
   * \tparam ExpressionType the type of the object on which to do partial reductions
-  * \tparam Direction indicates whether to operate on columns (#Vertical) or rows (#Horizontal)
+  * \tparam Direction indicates the direction of the redux (#Vertical or #Horizontal)
   *
-  * This class represents a pseudo expression with broadcasting and partial reduction features.
+  * This class represents a pseudo expression with partial reduction features.
   * It is the return type of DenseBase::colwise() and DenseBase::rowwise()
-  * and most of the time this is the only way it is explicitly used.
+  * and most of the time this is the only way it is used.
-  *
-  * To understand the logic of rowwise/colwise expression, let's consider a generic case `A.colwise().foo()`
-  * where `foo` is any method of `VectorwiseOp`. This expression is equivalent to applying `foo()` to each
-  * column of `A` and then re-assemble the outputs in a matrix expression:
-  * \code [A.col(0).foo(), A.col(1).foo(), ..., A.col(A.cols()-1).foo()] \endcode
   *
   * Example: \include MatrixBase_colwise.cpp
   * Output: \verbinclude MatrixBase_colwise.out
   *
-  * The begin() and end() methods are obviously exceptions to the previous rule as they
-  * return STL-compatible begin/end iterators to the rows or columns of the nested expression.
-  * Typical use cases include for-range-loop and calls to STL algorithms:
-  *
-  * Example: \include MatrixBase_colwise_iterator_cxx11.cpp
-  * Output: \verbinclude MatrixBase_colwise_iterator_cxx11.out
-  *
-  * For a partial reduction on an empty input, some rules apply.
-  * For the sake of clarity, let's consider a vertical reduction:
-  *   - If the number of columns is zero, then a 1x0 row-major vector expression is returned.
-  *   - Otherwise, if the number of rows is zero, then
-  *       - a row vector of zeros is returned for sum-like reductions (sum, squaredNorm, norm, etc.)
-  *       - a row vector of ones is returned for a product reduction (e.g., <code>MatrixXd(n,0).colwise().prod()</code>)
-  *       - an assert is triggered for all other reductions (minCoeff,maxCoeff,redux(bin_op))
-  *
   * \sa DenseBase::colwise(), DenseBase::rowwise(), class PartialReduxExpr
   */
 template<typename ExpressionType, int Direction> class VectorwiseOp
@@ -193,11 +163,11 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     typedef typename internal::ref_selector<ExpressionType>::non_const_type ExpressionTypeNested;
     typedef typename internal::remove_all<ExpressionTypeNested>::type ExpressionTypeNestedCleaned;
 
-    template<template<typename OutScalar,typename InputScalar> class Functor,
+    template<template<typename _Scalar> class Functor,
-                      typename ReturnScalar=Scalar> struct ReturnType
+                      typename Scalar_=Scalar> struct ReturnType
     {
       typedef PartialReduxExpr<ExpressionType,
-                               Functor<ReturnScalar,Scalar>,
+                               Functor<Scalar_>,
                                Direction
                               > Type;
     };
@@ -217,6 +187,23 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
 
   protected:
 
+    typedef typename internal::conditional<isVertical,
+                               typename ExpressionType::ColXpr,
+                               typename ExpressionType::RowXpr>::type SubVector;
+    /** \internal
+      * \returns the i-th subvector according to the \c Direction */
+    EIGEN_DEVICE_FUNC
+    SubVector subVector(Index i)
+    {
+      return SubVector(m_matrix.derived(),i);
+    }
+
+    /** \internal
+      * \returns the number of subvectors in the direction \c Direction */
+    EIGEN_DEVICE_FUNC
+    Index subVectors() const
+    { return isVertical?m_matrix.cols():m_matrix.rows(); }
+
     template<typename OtherDerived> struct ExtendedType {
       typedef Replicate<OtherDerived,
                         isVertical   ? 1 : ExpressionType::RowsAtCompileTime,
@@ -271,101 +258,42 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     EIGEN_DEVICE_FUNC
     inline const ExpressionType& _expression() const { return m_matrix; }
 
-    #ifdef EIGEN_PARSED_BY_DOXYGEN
-    /** STL-like <a href="https://en.cppreference.com/w/cpp/named_req/RandomAccessIterator">RandomAccessIterator</a>
-      * iterator type over the columns or rows as returned by the begin() and end() methods.
-      */
-    random_access_iterator_type iterator;
-    /** This is the const version of iterator (aka read-only) */
-    random_access_iterator_type const_iterator;
-    #else
-    typedef internal::subvector_stl_iterator<ExpressionType,               DirectionType(Direction)> iterator;
-    typedef internal::subvector_stl_iterator<const ExpressionType,         DirectionType(Direction)> const_iterator;
-    typedef internal::subvector_stl_reverse_iterator<ExpressionType,       DirectionType(Direction)> reverse_iterator;
-    typedef internal::subvector_stl_reverse_iterator<const ExpressionType, DirectionType(Direction)> const_reverse_iterator;
-    #endif
-
-    /** returns an iterator to the first row (rowwise) or column (colwise) of the nested expression.
-      * \sa end(), cbegin()
-      */
-    iterator                 begin()       { return iterator      (m_matrix, 0); }
-    /** const version of begin() */
-    const_iterator           begin() const { return const_iterator(m_matrix, 0); }
-    /** const version of begin() */
-    const_iterator          cbegin() const { return const_iterator(m_matrix, 0); }
-
-    /** returns a reverse iterator to the last row (rowwise) or column (colwise) of the nested expression.
-      * \sa rend(), crbegin()
-      */
-    reverse_iterator        rbegin()       { return reverse_iterator       (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
-	/** const version of rbegin() */
-    const_reverse_iterator  rbegin() const { return const_reverse_iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
-	/** const version of rbegin() */
-	const_reverse_iterator crbegin() const { return const_reverse_iterator (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()-1); }
-
-    /** returns an iterator to the row (resp. column) following the last row (resp. column) of the nested expression
-      * \sa begin(), cend()
-      */
-    iterator                 end()         { return iterator      (m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
-    /** const version of end() */
-    const_iterator           end()  const  { return const_iterator(m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
-    /** const version of end() */
-    const_iterator          cend()  const  { return const_iterator(m_matrix, m_matrix.template subVectors<DirectionType(Direction)>()); }
-
-    /** returns a reverse iterator to the row (resp. column) before the first row (resp. column) of the nested expression
-      * \sa begin(), cend()
-      */
-    reverse_iterator        rend()         { return reverse_iterator       (m_matrix, -1); }
-    /** const version of rend() */
-    const_reverse_iterator  rend()  const  { return const_reverse_iterator (m_matrix, -1); }
-    /** const version of rend() */
-    const_reverse_iterator crend()  const  { return const_reverse_iterator (m_matrix, -1); }
-
     /** \returns a row or column vector expression of \c *this reduxed by \a func
       *
       * The template parameter \a BinaryOp is the type of the functor
       * of the custom redux operator. Note that func must be an associative operator.
       *
-      * \warning the size along the reduction direction must be strictly positive,
-      *          otherwise an assertion is triggered.
-      *
       * \sa class VectorwiseOp, DenseBase::colwise(), DenseBase::rowwise()
       */
     template<typename BinaryOp>
     EIGEN_DEVICE_FUNC
     const typename ReduxReturnType<BinaryOp>::Type
     redux(const BinaryOp& func = BinaryOp()) const
-    {
+    { return typename ReduxReturnType<BinaryOp>::Type(_expression(), internal::member_redux<BinaryOp,Scalar>(func)); }
-      eigen_assert(redux_length()>0 && "you are using an empty matrix");
-      return typename ReduxReturnType<BinaryOp>::Type(_expression(), internal::member_redux<BinaryOp,Scalar>(func));
-    }
 
     typedef typename ReturnType<internal::member_minCoeff>::Type MinCoeffReturnType;
     typedef typename ReturnType<internal::member_maxCoeff>::Type MaxCoeffReturnType;
-    typedef PartialReduxExpr<const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const ExpressionTypeNestedCleaned>,internal::member_sum<RealScalar,RealScalar>,Direction> SquaredNormReturnType;
+    typedef typename ReturnType<internal::member_squaredNorm,RealScalar>::Type SquaredNormReturnType;
-    typedef CwiseUnaryOp<internal::scalar_sqrt_op<RealScalar>, const SquaredNormReturnType> NormReturnType;
+    typedef typename ReturnType<internal::member_norm,RealScalar>::Type NormReturnType;
     typedef typename ReturnType<internal::member_blueNorm,RealScalar>::Type BlueNormReturnType;
     typedef typename ReturnType<internal::member_stableNorm,RealScalar>::Type StableNormReturnType;
     typedef typename ReturnType<internal::member_hypotNorm,RealScalar>::Type HypotNormReturnType;
     typedef typename ReturnType<internal::member_sum>::Type SumReturnType;
-    typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(SumReturnType,Scalar,quotient) MeanReturnType;
+    typedef typename ReturnType<internal::member_mean>::Type MeanReturnType;
     typedef typename ReturnType<internal::member_all>::Type AllReturnType;
     typedef typename ReturnType<internal::member_any>::Type AnyReturnType;
-    typedef PartialReduxExpr<ExpressionType, internal::member_count<Index,Scalar>, Direction> CountReturnType;
+    typedef PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> CountReturnType;
     typedef typename ReturnType<internal::member_prod>::Type ProdReturnType;
     typedef Reverse<const ExpressionType, Direction> ConstReverseReturnType;
     typedef Reverse<ExpressionType, Direction> ReverseReturnType;
 
     template<int p> struct LpNormReturnType {
-      typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar,Scalar>,Direction> Type;
+      typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar>,Direction> Type;
     };
 
     /** \returns a row (or column) vector expression of the smallest coefficient
       * of each column (or row) of the referenced expression.
       *
-      * \warning the size along the reduction direction must be strictly positive,
-      *          otherwise an assertion is triggered.
-      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_minCoeff.cpp
@@ -374,17 +302,11 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa DenseBase::minCoeff() */
     EIGEN_DEVICE_FUNC
     const MinCoeffReturnType minCoeff() const
-    {
+    { return MinCoeffReturnType(_expression()); }
-      eigen_assert(redux_length()>0 && "you are using an empty matrix");
-      return MinCoeffReturnType(_expression());
-    }
 
     /** \returns a row (or column) vector expression of the largest coefficient
       * of each column (or row) of the referenced expression.
       *
-      * \warning the size along the reduction direction must be strictly positive,
-      *          otherwise an assertion is triggered.
-      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_maxCoeff.cpp
@@ -393,10 +315,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa DenseBase::maxCoeff() */
     EIGEN_DEVICE_FUNC
     const MaxCoeffReturnType maxCoeff() const
-    {
+    { return MaxCoeffReturnType(_expression()); }
-      eigen_assert(redux_length()>0 && "you are using an empty matrix");
-      return MaxCoeffReturnType(_expression());
-    }
 
     /** \returns a row (or column) vector expression of the squared norm
       * of each column (or row) of the referenced expression.
@@ -408,7 +327,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa DenseBase::squaredNorm() */
     EIGEN_DEVICE_FUNC
     const SquaredNormReturnType squaredNorm() const
-    { return SquaredNormReturnType(m_matrix.cwiseAbs2()); }
+    { return SquaredNormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression.
@@ -420,7 +339,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       * \sa DenseBase::norm() */
     EIGEN_DEVICE_FUNC
     const NormReturnType norm() const
-    { return NormReturnType(squaredNorm()); }
+    { return NormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression.
@@ -485,7 +404,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     * \sa DenseBase::mean() */
     EIGEN_DEVICE_FUNC
     const MeanReturnType mean() const
-    { return sum() / Scalar(Direction==Vertical?m_matrix.rows():m_matrix.cols()); }
+    { return MeanReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * whether \b all coefficients of each respective column (or row) are \c true.
@@ -581,7 +500,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
       //eigen_assert((m_matrix.isNull()) == (other.isNull())); FIXME
-      return m_matrix = extendedTo(other.derived());
+      return const_cast<ExpressionType&>(m_matrix = extendedTo(other.derived()));
     }
 
     /** Adds the vector \a other to each subvector of \c *this */
@@ -591,7 +510,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
-      return m_matrix += extendedTo(other.derived());
+      return const_cast<ExpressionType&>(m_matrix += extendedTo(other.derived()));
     }
 
     /** Substracts the vector \a other to each subvector of \c *this */
@@ -601,7 +520,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
-      return m_matrix -= extendedTo(other.derived());
+      return const_cast<ExpressionType&>(m_matrix -= extendedTo(other.derived()));
     }
 
     /** Multiples each subvector of \c *this by the vector \a other */
@@ -613,7 +532,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
       m_matrix *= extendedTo(other.derived());
-      return m_matrix;
+      return const_cast<ExpressionType&>(m_matrix);
     }
 
     /** Divides each subvector of \c *this by the vector \a other */
@@ -625,7 +544,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
       EIGEN_STATIC_ASSERT_ARRAYXPR(ExpressionType)
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
       m_matrix /= extendedTo(other.derived());
-      return m_matrix;
+      return const_cast<ExpressionType&>(m_matrix);
     }
 
     /** Returns the expression of the sum of the vector \a other to each subvector of \c *this */
@@ -690,7 +609,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
-                  const typename OppositeExtendedType<NormReturnType>::Type>
+                  const typename OppositeExtendedType<typename ReturnType<internal::member_norm,RealScalar>::Type>::Type>
     normalized() const { return m_matrix.cwiseQuotient(extendedToOpposite(this->norm())); }
 
 
@@ -739,15 +658,7 @@ template<typename ExpressionType, int Direction> class VectorwiseOp
     EIGEN_DEVICE_FUNC
     const HNormalizedReturnType hnormalized() const;
 
-#   ifdef EIGEN_VECTORWISEOP_PLUGIN
-#     include EIGEN_VECTORWISEOP_PLUGIN
-#   endif
-
   protected:
-    Index redux_length() const
-    {
-      return Direction==Vertical ? m_matrix.rows() : m_matrix.cols();
-    }
     ExpressionTypeNested m_matrix;
 };
 

Eigen/src/Core/Visitor.h

@@ -40,14 +40,6 @@ struct visitor_impl<Visitor, Derived, 1>
   }
 };
 
-// This specialization enables visitors on empty matrices at compile-time
-template<typename Visitor, typename Derived>
-struct visitor_impl<Visitor, Derived, 0> {
-  EIGEN_DEVICE_FUNC
-  static inline void run(const Derived &/*mat*/, Visitor& /*visitor*/)
-  {}
-};
-
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, Dynamic>
 {
@@ -79,9 +71,9 @@ public:
     CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
   };
   
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
-  EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_xpr.size(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
 
   EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
   { return m_evaluator.coeff(row, col); }
@@ -107,8 +99,6 @@ protected:
   * \note compared to one or two \em for \em loops, visitors offer automatic
   * unrolling for small fixed size matrix.
   *
-  * \note if the matrix is empty, then the visitor is left unchanged.
-  *
   * \sa minCoeff(Index*,Index*), maxCoeff(Index*,Index*), DenseBase::redux()
   */
 template<typename Derived>
@@ -116,15 +106,12 @@ template<typename Visitor>
 EIGEN_DEVICE_FUNC
 void DenseBase<Derived>::visit(Visitor& visitor) const
 {
-  if(size()==0)
-    return;
-
   typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
   ThisEvaluator thisEval(derived());
   
   enum {
     unroll =  SizeAtCompileTime != Dynamic
-           && SizeAtCompileTime * int(ThisEvaluator::CoeffReadCost) + (SizeAtCompileTime-1) * int(internal::functor_traits<Visitor>::Cost) <= EIGEN_UNROLLING_LIMIT
+           && SizeAtCompileTime * ThisEvaluator::CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost <= EIGEN_UNROLLING_LIMIT
   };
   return internal::visitor_impl<Visitor, ThisEvaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(thisEval, visitor);
 }
@@ -137,9 +124,6 @@ namespace internal {
 template <typename Derived>
 struct coeff_visitor
 {
-  // default initialization to avoid countless invalid maybe-uninitialized warnings by gcc
-  EIGEN_DEVICE_FUNC
-  coeff_visitor() : row(-1), col(-1), res(0) {}
   typedef typename Derived::Scalar Scalar;
   Index row, col;
   Scalar res;
@@ -157,7 +141,7 @@ struct coeff_visitor
   *
   * \sa DenseBase::minCoeff(Index*, Index*)
   */
-template <typename Derived, int NaNPropagation>
+template <typename Derived>
 struct min_coeff_visitor : coeff_visitor<Derived>
 {
   typedef typename Derived::Scalar Scalar;
@@ -173,40 +157,8 @@ struct min_coeff_visitor : coeff_visitor<Derived>
   }
 };
 
-template <typename Derived>
+template<typename Scalar>
-struct min_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
+struct functor_traits<min_coeff_visitor<Scalar> > {
-{
-  typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC
-  void operator() (const Scalar& value, Index i, Index j)
-  {
-    if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value < this->res))
-    {
-      this->res = value;
-      this->row = i;
-      this->col = j;
-    }
-  }
-};
-
-template <typename Derived>
-struct min_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
-{
-  typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC
-  void operator() (const Scalar& value, Index i, Index j)
-  {
-    if((numext::isnan)(value) || value < this->res)
-    {
-      this->res = value;
-      this->row = i;
-      this->col = j;
-    }
-  }
-};
-
-template<typename Scalar, int NaNPropagation>
-    struct functor_traits<min_coeff_visitor<Scalar, NaNPropagation> > {
   enum {
     Cost = NumTraits<Scalar>::AddCost
   };
@@ -217,7 +169,7 @@ template<typename Scalar, int NaNPropagation>
   *
   * \sa DenseBase::maxCoeff(Index*, Index*)
   */
-template <typename Derived, int NaNPropagation>
+template <typename Derived>
 struct max_coeff_visitor : coeff_visitor<Derived>
 {
   typedef typename Derived::Scalar Scalar; 
@@ -233,40 +185,8 @@ struct max_coeff_visitor : coeff_visitor<Derived>
   }
 };
 
-template <typename Derived>
+template<typename Scalar>
-struct max_coeff_visitor<Derived, PropagateNumbers> : coeff_visitor<Derived>
+struct functor_traits<max_coeff_visitor<Scalar> > {
-{
-  typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC
-  void operator() (const Scalar& value, Index i, Index j)
-  {
-    if((numext::isnan)(this->res) || (!(numext::isnan)(value) && value > this->res))
-    {
-      this->res = value;
-      this->row = i;
-      this->col = j;
-    }
-  }
-};
-
-template <typename Derived>
-struct max_coeff_visitor<Derived, PropagateNaN> : coeff_visitor<Derived>
-{
-  typedef typename Derived::Scalar Scalar;
-  EIGEN_DEVICE_FUNC
-  void operator() (const Scalar& value, Index i, Index j)
-  {
-    if((numext::isnan)(value) || value > this->res)
-    {
-      this->res = value;
-      this->row = i;
-      this->col = j;
-    }
-  }
-};
-
-template<typename Scalar, int NaNPropagation>
-struct functor_traits<max_coeff_visitor<Scalar, NaNPropagation> > {
   enum {
     Cost = NumTraits<Scalar>::AddCost
   };
@@ -276,24 +196,17 @@ struct functor_traits<max_coeff_visitor<Scalar, NaNPropagation> > {
 
 /** \fn DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
   * \returns the minimum of all coefficients of *this and puts in *row and *col its location.
-  *
+  * \warning the result is undefined if \c *this contains NaN.
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
-template<int NaNPropagation, typename IndexType>
+template<typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 {
-  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+  internal::min_coeff_visitor<Derived> minVisitor;
-
-  internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
   this->visit(minVisitor);
   *rowId = minVisitor.row;
   if (colId) *colId = minVisitor.col;
@@ -301,25 +214,18 @@ DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 }
 
 /** \returns the minimum of all coefficients of *this and puts in *index its location.
-  *
+  * \warning the result is undefined if \c *this contains NaN. 
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
-template<int NaNPropagation, typename IndexType>
+template<typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* index) const
 {
-  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
-
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-      internal::min_coeff_visitor<Derived, NaNPropagation> minVisitor;
+  internal::min_coeff_visitor<Derived> minVisitor;
   this->visit(minVisitor);
   *index = IndexType((RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row);
   return minVisitor.res;
@@ -327,24 +233,17 @@ DenseBase<Derived>::minCoeff(IndexType* index) const
 
 /** \fn DenseBase<Derived>::maxCoeff(IndexType* rowId, IndexType* colId) const
   * \returns the maximum of all coefficients of *this and puts in *row and *col its location.
-  *
+  * \warning the result is undefined if \c *this contains NaN. 
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()
   */
 template<typename Derived>
-template<int NaNPropagation, typename IndexType>
+template<typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 {
-  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+  internal::max_coeff_visitor<Derived> maxVisitor;
-
-  internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
   this->visit(maxVisitor);
   *rowPtr = maxVisitor.row;
   if (colPtr) *colPtr = maxVisitor.col;
@@ -352,25 +251,18 @@ DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 }
 
 /** \returns the maximum of all coefficients of *this and puts in *index its location.
-  *
+  * \warning the result is undefined if \c *this contains NaN.
-  * In case \c *this contains NaN, NaNPropagation determines the behavior:
-  *   NaNPropagation == PropagateFast : undefined
-  *   NaNPropagation == PropagateNaN : result is NaN
-  *   NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN
-  * \warning the matrix must be not empty, otherwise an assertion is triggered.
   *
   * \sa DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::maxCoeff()
   */
 template<typename Derived>
-template<int NaNPropagation, typename IndexType>
+template<typename IndexType>
 EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* index) const
 {
-  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
-
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-      internal::max_coeff_visitor<Derived, NaNPropagation> maxVisitor;
+  internal::max_coeff_visitor<Derived> maxVisitor;
   this->visit(maxVisitor);
   *index = (RowsAtCompileTime==1) ? maxVisitor.col : maxVisitor.row;
   return maxVisitor.res;

Eigen/src/Core/arch/AVX/Complex.h

@@ -22,7 +22,6 @@ struct Packet4cf
   __m256  v;
 };
 
-#ifndef EIGEN_VECTORIZE_AVX512
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet4cf type;
@@ -38,7 +37,6 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasMul    = 1,
     HasDiv    = 1,
     HasNegate = 1,
-    HasSqrt   = 1,
     HasAbs    = 0,
     HasAbs2   = 0,
     HasMin    = 0,
@@ -46,20 +44,8 @@ template<> struct packet_traits<std::complex<float> >  : default_packet_traits
     HasSetLinear = 0
   };
 };
-#endif
 
-template<> struct unpacket_traits<Packet4cf> {
+template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32}; typedef Packet2cf half; };
-  typedef std::complex<float> type;
-  typedef Packet2cf half;
-  typedef Packet8f as_real;
-  enum {
-    size=4,
-    alignment=Aligned32,
-    vectorizable=true,
-    masked_load_available=false,
-    masked_store_available=false
-  };
-};
 
 template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
@@ -81,17 +67,10 @@ template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, con
   return Packet4cf(result);
 }
 
-template <>
-EIGEN_STRONG_INLINE Packet4cf pcmp_eq(const Packet4cf& a, const Packet4cf& b) {
-  __m256 eq = _mm256_cmp_ps(a.v, b.v, _CMP_EQ_OQ);
-  return Packet4cf(_mm256_and_ps(eq, _mm256_permute_ps(eq, 0xb1)));
-}
-
-template<> EIGEN_STRONG_INLINE Packet4cf ptrue<Packet4cf>(const Packet4cf& a) { return Packet4cf(ptrue(Packet8f(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(b.v,a.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(a.v,b.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
 template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
@@ -99,9 +78,7 @@ template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<fl
 
 template<> EIGEN_STRONG_INLINE Packet4cf pset1<Packet4cf>(const std::complex<float>& from)
 {
-  const float re = std::real(from);
+  return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from)));
-  const float im = std::imag(from);
-  return Packet4cf(_mm256_set_ps(im, re, im, re, im, re, im, re));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4cf ploaddup<Packet4cf>(const std::complex<float>* from)
@@ -163,18 +140,79 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packe
                      Packet2cf(_mm256_extractf128_ps(a.v,1))));
 }
 
+template<> EIGEN_STRONG_INLINE Packet4cf preduxp<Packet4cf>(const Packet4cf* vecs)
+{
+  Packet8f t0 = _mm256_shuffle_ps(vecs[0].v, vecs[0].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t1 = _mm256_shuffle_ps(vecs[1].v, vecs[1].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t0 = _mm256_hadd_ps(t0,t1);
+  Packet8f t2 = _mm256_shuffle_ps(vecs[2].v, vecs[2].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t3 = _mm256_shuffle_ps(vecs[3].v, vecs[3].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t2 = _mm256_hadd_ps(t2,t3);
+  
+  t1 = _mm256_permute2f128_ps(t0,t2, 0 + (2<<4));
+  t3 = _mm256_permute2f128_ps(t0,t2, 1 + (3<<4));
+
+  return Packet4cf(_mm256_add_ps(t1,t3));
+}
+
 template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
 {
   return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
                          Packet2cf(_mm256_extractf128_ps(a.v, 1))));
 }
 
+template<int Offset>
+struct palign_impl<Offset,Packet4cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4cf& first, const Packet4cf& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet8f>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
 
 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
 
 template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
 {
-  return pdiv_complex(a, b);
+  Packet4cf num = pmul(a, pconj(b));
+  __m256 tmp = _mm256_mul_ps(b.v, b.v);
+  __m256 tmp2    = _mm256_shuffle_ps(tmp,tmp,0xB1);
+  __m256 denom = _mm256_add_ps(tmp, tmp2);
+  return Packet4cf(_mm256_div_ps(num.v, denom));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4cf pcplxflip<Packet4cf>(const Packet4cf& x)
@@ -190,7 +228,6 @@ struct Packet2cd
   __m256d  v;
 };
 
-#ifndef EIGEN_VECTORIZE_AVX512
 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
 {
   typedef Packet2cd type;
@@ -206,7 +243,6 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
     HasMul    = 1,
     HasDiv    = 1,
     HasNegate = 1,
-    HasSqrt   = 1,
     HasAbs    = 0,
     HasAbs2   = 0,
     HasMin    = 0,
@@ -214,20 +250,8 @@ template<> struct packet_traits<std::complex<double> >  : default_packet_traits
     HasSetLinear = 0
   };
 };
-#endif
 
-template<> struct unpacket_traits<Packet2cd> {
+template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32}; typedef Packet1cd half; };
-  typedef std::complex<double> type;
-  typedef Packet1cd half;
-  typedef Packet4d as_real;
-  enum {
-    size=2,
-    alignment=Aligned32,
-    vectorizable=true,
-    masked_load_available=false,
-    masked_store_available=false
-  };
-};
 
 template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
@@ -248,17 +272,10 @@ template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, con
   return Packet2cd(_mm256_addsub_pd(even, odd));
 }
 
-template <>
-EIGEN_STRONG_INLINE Packet2cd pcmp_eq(const Packet2cd& a, const Packet2cd& b) {
-  __m256d eq = _mm256_cmp_pd(a.v, b.v, _CMP_EQ_OQ);
-  return Packet2cd(pand(eq, _mm256_permute_pd(eq, 0x5)));
-}
-
-template<> EIGEN_STRONG_INLINE Packet2cd ptrue<Packet2cd>(const Packet2cd& a) { return Packet2cd(ptrue(Packet4d(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(b.v,a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(a.v,b.v)); }
 
 template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
 { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
@@ -310,17 +327,71 @@ template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Pack
                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
 }
 
+template<> EIGEN_STRONG_INLINE Packet2cd preduxp<Packet2cd>(const Packet2cd* vecs)
+{
+  Packet4d t0 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 0 + (2<<4));
+  Packet4d t1 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 1 + (3<<4));
+
+  return Packet2cd(_mm256_add_pd(t0,t1));
+}
+
 template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
 {
   return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
                      Packet1cd(_mm256_extractf128_pd(a.v,1))));
 }
 
+template<int Offset>
+struct palign_impl<Offset,Packet2cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cd& first, const Packet2cd& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet4d>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
 EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
 
 template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
 {
-  return pdiv_complex(a, b);
+  Packet2cd num = pmul(a, pconj(b));
+  __m256d tmp = _mm256_mul_pd(b.v, b.v);
+  __m256d denom = _mm256_hadd_pd(tmp, tmp);
+  return Packet2cd(_mm256_div_pd(num.v, denom));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cd pcplxflip<Packet2cd>(const Packet2cd& x)
@@ -353,12 +424,24 @@ ptranspose(PacketBlock<Packet2cd,2>& kernel) {
  kernel.packet[0].v = tmp;
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cd psqrt<Packet2cd>(const Packet2cd& a) {
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
-  return psqrt_complex<Packet2cd>(a);
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
 }
 
-template<> EIGEN_STRONG_INLINE Packet4cf psqrt<Packet4cf>(const Packet4cf& a) {
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
-  return psqrt_complex<Packet4cf>(a);
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
 }
 
 } // end namespace internal

Eigen/src/Core/arch/AVX/MathFunctions.h

@@ -10,7 +10,7 @@
 #ifndef EIGEN_MATH_FUNCTIONS_AVX_H
 #define EIGEN_MATH_FUNCTIONS_AVX_H
 
-/* The sin and cos functions of this file are loosely derived from
+/* The sin, cos, exp, and log functions of this file are loosely derived from
  * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
  */
 
@@ -18,50 +18,187 @@ namespace Eigen {
 
 namespace internal {
 
+inline Packet8i pshiftleft(Packet8i v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_slli_epi32(v, n);
+#else
+  __m128i lo = _mm_slli_epi32(_mm256_extractf128_si256(v, 0), n);
+  __m128i hi = _mm_slli_epi32(_mm256_extractf128_si256(v, 1), n);
+  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1);
+#endif
+}
+
+inline Packet8f pshiftright(Packet8f v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_cvtepi32_ps(_mm256_srli_epi32(_mm256_castps_si256(v), n));
+#else
+  __m128i lo = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 0), n);
+  __m128i hi = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 1), n);
+  return _mm256_cvtepi32_ps(_mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1));
+#endif
+}
+
+// Sine function
+// Computes sin(x) by wrapping x to the interval [-Pi/4,3*Pi/4] and
+// evaluating interpolants in [-Pi/4,Pi/4] or [Pi/4,3*Pi/4]. The interpolants
+// are (anti-)symmetric and thus have only odd/even coefficients
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
 psin<Packet8f>(const Packet8f& _x) {
-  return psin_float(_x);
+  Packet8f x = _x;
-}
+
-
+  // Some useful values.
-template <>
+  _EIGEN_DECLARE_CONST_Packet8i(one, 1);
-EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
-pcos<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f(two, 2.0f);
-  return pcos_float(_x);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_four, 0.25f);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_pi, 3.183098861837907e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_first, -3.140625000000000e+00f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_second, -9.670257568359375e-04f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_third, -6.278329571784980e-07f);
+  _EIGEN_DECLARE_CONST_Packet8f(four_over_pi, 1.273239544735163e+00f);
+
+  // Map x from [-Pi/4,3*Pi/4] to z in [-1,3] and subtract the shifted period.
+  Packet8f z = pmul(x, p8f_one_over_pi);
+  Packet8f shift = _mm256_floor_ps(padd(z, p8f_one_over_four));
+  x = pmadd(shift, p8f_neg_pi_first, x);
+  x = pmadd(shift, p8f_neg_pi_second, x);
+  x = pmadd(shift, p8f_neg_pi_third, x);
+  z = pmul(x, p8f_four_over_pi);
+
+  // Make a mask for the entries that need flipping, i.e. wherever the shift
+  // is odd.
+  Packet8i shift_ints = _mm256_cvtps_epi32(shift);
+  Packet8i shift_isodd = _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(shift_ints), _mm256_castsi256_ps(p8i_one)));
+  Packet8i sign_flip_mask = pshiftleft(shift_isodd, 31);
+
+  // Create a mask for which interpolant to use, i.e. if z > 1, then the mask
+  // is set to ones for that entry.
+  Packet8f ival_mask = _mm256_cmp_ps(z, p8f_one, _CMP_GT_OQ);
+
+  // Evaluate the polynomial for the interval [1,3] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_0, 9.999999724233232e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_2, -3.084242535619928e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_4, 1.584991525700324e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_6, -3.188805084631342e-04f);
+  Packet8f z_minus_two = psub(z, p8f_two);
+  Packet8f z_minus_two2 = pmul(z_minus_two, z_minus_two);
+  Packet8f right = pmadd(p8f_coeff_right_6, z_minus_two2, p8f_coeff_right_4);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_2);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_0);
+
+  // Evaluate the polynomial for the interval [-1,1] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_1, 7.853981525427295e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_3, -8.074536727092352e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_5, 2.489871967827018e-03f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_7, -3.587725841214251e-05f);
+  Packet8f z2 = pmul(z, z);
+  Packet8f left = pmadd(p8f_coeff_left_7, z2, p8f_coeff_left_5);
+  left = pmadd(left, z2, p8f_coeff_left_3);
+  left = pmadd(left, z2, p8f_coeff_left_1);
+  left = pmul(left, z);
+
+  // Assemble the results, i.e. select the left and right polynomials.
+  left = _mm256_andnot_ps(ival_mask, left);
+  right = _mm256_and_ps(ival_mask, right);
+  Packet8f res = _mm256_or_ps(left, right);
+
+  // Flip the sign on the odd intervals and return the result.
+  res = _mm256_xor_ps(res, _mm256_castsi256_ps(sign_flip_mask));
+  return res;
 }
 
+// Natural logarithm
+// Computes log(x) as log(2^e * m) = C*e + log(m), where the constant C =log(2)
+// and m is in the range [sqrt(1/2),sqrt(2)). In this range, the logarithm can
+// be easily approximated by a polynomial centered on m=1 for stability.
+// TODO(gonnet): Further reduce the interval allowing for lower-degree
+//               polynomial interpolants -> ... -> profit!
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
 plog<Packet8f>(const Packet8f& _x) {
-  return plog_float(_x);
+  Packet8f x = _x;
-}
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(126f, 126.0f);
 
-template <>
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inv_mant_mask, ~0x7f800000);
-EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
-plog<Packet4d>(const Packet4d& _x) {
-  return plog_double(_x);
-}
 
-template <>
+  // The smallest non denormalized float number.
-EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(min_norm_pos, 0x00800000);
-plog2<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(minus_inf, 0xff800000);
-  return plog2_float(_x);
-}
 
-template <>
+  // Polynomial coefficients.
-EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_SQRTHF, 0.707106781186547524f);
-plog2<Packet4d>(const Packet4d& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p0, 7.0376836292E-2f);
-  return plog2_double(_x);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p1, -1.1514610310E-1f);
-}
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p2, 1.1676998740E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p3, -1.2420140846E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p4, +1.4249322787E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p5, -1.6668057665E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p6, +2.0000714765E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p7, -2.4999993993E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p8, +3.3333331174E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q1, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q2, 0.693359375f);
 
-template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+  Packet8f invalid_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_NGE_UQ); // not greater equal is true if x is NaN
-Packet8f plog1p<Packet8f>(const Packet8f& _x) {
+  Packet8f iszero_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_EQ_OQ);
-  return generic_plog1p(_x);
-}
 
-template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+  // Truncate input values to the minimum positive normal.
-Packet8f pexpm1<Packet8f>(const Packet8f& _x) {
+  x = pmax(x, p8f_min_norm_pos);
-  return generic_expm1(_x);
+
+  Packet8f emm0 = pshiftright(x,23);
+  Packet8f e = _mm256_sub_ps(emm0, p8f_126f);
+
+  // Set the exponents to -1, i.e. x are in the range [0.5,1).
+  x = _mm256_and_ps(x, p8f_inv_mant_mask);
+  x = _mm256_or_ps(x, p8f_half);
+
+  // part2: Shift the inputs from the range [0.5,1) to [sqrt(1/2),sqrt(2))
+  // and shift by -1. The values are then centered around 0, which improves
+  // the stability of the polynomial evaluation.
+  //   if( x < SQRTHF ) {
+  //     e -= 1;
+  //     x = x + x - 1.0;
+  //   } else { x = x - 1.0; }
+  Packet8f mask = _mm256_cmp_ps(x, p8f_cephes_SQRTHF, _CMP_LT_OQ);
+  Packet8f tmp = _mm256_and_ps(x, mask);
+  x = psub(x, p8f_1);
+  e = psub(e, _mm256_and_ps(p8f_1, mask));
+  x = padd(x, tmp);
+
+  Packet8f x2 = pmul(x, x);
+  Packet8f x3 = pmul(x2, x);
+
+  // Evaluate the polynomial approximant of degree 8 in three parts, probably
+  // to improve instruction-level parallelism.
+  Packet8f y, y1, y2;
+  y = pmadd(p8f_cephes_log_p0, x, p8f_cephes_log_p1);
+  y1 = pmadd(p8f_cephes_log_p3, x, p8f_cephes_log_p4);
+  y2 = pmadd(p8f_cephes_log_p6, x, p8f_cephes_log_p7);
+  y = pmadd(y, x, p8f_cephes_log_p2);
+  y1 = pmadd(y1, x, p8f_cephes_log_p5);
+  y2 = pmadd(y2, x, p8f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  // Add the logarithm of the exponent back to the result of the interpolation.
+  y1 = pmul(e, p8f_cephes_log_q1);
+  tmp = pmul(x2, p8f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p8f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+
+  // Filter out invalid inputs, i.e. negative arg will be NAN, 0 will be -INF.
+  return _mm256_or_ps(
+      _mm256_andnot_ps(iszero_mask, _mm256_or_ps(x, invalid_mask)),
+      _mm256_and_ps(iszero_mask, p8f_minus_inf));
 }
 
 // Exponential function. Works by writing "x = m*log(2) + r" where
@@ -70,21 +207,149 @@ Packet8f pexpm1<Packet8f>(const Packet8f& _x) {
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
 pexp<Packet8f>(const Packet8f& _x) {
-  return pexp_float(_x);
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(127, 127.0f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(exp_hi, 88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet8f(exp_lo, -88.3762626647949f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_LOG2EF, 1.44269504088896341f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p5, 5.0000001201E-1f);
+
+  // Clamp x.
+  Packet8f x = pmax(pmin(_x, p8f_exp_hi), p8f_exp_lo);
+
+  // Express exp(x) as exp(m*ln(2) + r), start by extracting
+  // m = floor(x/ln(2) + 0.5).
+  Packet8f m = _mm256_floor_ps(pmadd(x, p8f_cephes_LOG2EF, p8f_half));
+
+// Get r = x - m*ln(2). If no FMA instructions are available, m*ln(2) is
+// subtracted out in two parts, m*C1+m*C2 = m*ln(2), to avoid accumulating
+// truncation errors. Note that we don't use the "pmadd" function here to
+// ensure that a precision-preserving FMA instruction is used.
+#ifdef EIGEN_VECTORIZE_FMA
+  _EIGEN_DECLARE_CONST_Packet8f(nln2, -0.6931471805599453f);
+  Packet8f r = _mm256_fmadd_ps(m, p8f_nln2, x);
+#else
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C1, 0.693359375f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C2, -2.12194440e-4f);
+  Packet8f r = psub(x, pmul(m, p8f_cephes_exp_C1));
+  r = psub(r, pmul(m, p8f_cephes_exp_C2));
+#endif
+
+  Packet8f r2 = pmul(r, r);
+
+  // TODO(gonnet): Split into odd/even polynomials and try to exploit
+  //               instruction-level parallelism.
+  Packet8f y = p8f_cephes_exp_p0;
+  y = pmadd(y, r, p8f_cephes_exp_p1);
+  y = pmadd(y, r, p8f_cephes_exp_p2);
+  y = pmadd(y, r, p8f_cephes_exp_p3);
+  y = pmadd(y, r, p8f_cephes_exp_p4);
+  y = pmadd(y, r, p8f_cephes_exp_p5);
+  y = pmadd(y, r2, r);
+  y = padd(y, p8f_1);
+
+  // Build emm0 = 2^m.
+  Packet8i emm0 = _mm256_cvttps_epi32(padd(m, p8f_127));
+  emm0 = pshiftleft(emm0, 23);
+
+  // Return 2^m * exp(r).
+  return pmax(pmul(y, _mm256_castsi256_ps(emm0)), _x);
 }
 
 // Hyperbolic Tangent function.
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
-ptanh<Packet8f>(const Packet8f& _x) {
+ptanh<Packet8f>(const Packet8f& x) {
-  return internal::generic_fast_tanh_float(_x);
+  return internal::generic_fast_tanh_float(x);
 }
 
-// Exponential function for doubles.
 template <>
 EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
 pexp<Packet4d>(const Packet4d& _x) {
-  return pexp_double(_x);
+  Packet4d x = _x;
+
+  _EIGEN_DECLARE_CONST_Packet4d(1, 1.0);
+  _EIGEN_DECLARE_CONST_Packet4d(2, 2.0);
+  _EIGEN_DECLARE_CONST_Packet4d(half, 0.5);
+
+  _EIGEN_DECLARE_CONST_Packet4d(exp_hi, 709.437);
+  _EIGEN_DECLARE_CONST_Packet4d(exp_lo, -709.436139303);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_LOG2EF, 1.4426950408889634073599);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p0, 1.26177193074810590878e-4);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p1, 3.02994407707441961300e-2);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q0, 3.00198505138664455042e-6);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q1, 2.52448340349684104192e-3);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q2, 2.27265548208155028766e-1);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q3, 2.00000000000000000009e0);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C1, 0.693145751953125);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C2, 1.42860682030941723212e-6);
+  _EIGEN_DECLARE_CONST_Packet4i(1023, 1023);
+
+  Packet4d tmp, fx;
+
+  // clamp x
+  x = pmax(pmin(x, p4d_exp_hi), p4d_exp_lo);
+  // Express exp(x) as exp(g + n*log(2)).
+  fx = pmadd(p4d_cephes_LOG2EF, x, p4d_half);
+
+  // Get the integer modulus of log(2), i.e. the "n" described above.
+  fx = _mm256_floor_pd(fx);
+
+  // Get the remainder modulo log(2), i.e. the "g" described above. Subtract
+  // n*log(2) out in two steps, i.e. n*C1 + n*C2, C1+C2=log2 to get the last
+  // digits right.
+  tmp = pmul(fx, p4d_cephes_exp_C1);
+  Packet4d z = pmul(fx, p4d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet4d x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial of the rational interpolant.
+  Packet4d px = p4d_cephes_exp_p0;
+  px = pmadd(px, x2, p4d_cephes_exp_p1);
+  px = pmadd(px, x2, p4d_cephes_exp_p2);
+  px = pmul(px, x);
+
+  // Evaluate the denominator polynomial of the rational interpolant.
+  Packet4d qx = p4d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p4d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q3);
+
+  // I don't really get this bit, copied from the SSE2 routines, so...
+  // TODO(gonnet): Figure out what is going on here, perhaps find a better
+  // rational interpolant?
+  x = _mm256_div_pd(px, psub(qx, px));
+  x = pmadd(p4d_2, x, p4d_1);
+
+  // Build e=2^n by constructing the exponents in a 128-bit vector and
+  // shifting them to where they belong in double-precision values.
+  __m128i emm0 = _mm256_cvtpd_epi32(fx);
+  emm0 = _mm_add_epi32(emm0, p4i_1023);
+  emm0 = _mm_shuffle_epi32(emm0, _MM_SHUFFLE(3, 1, 2, 0));
+  __m128i lo = _mm_slli_epi64(emm0, 52);
+  __m128i hi = _mm_slli_epi64(_mm_srli_epi64(emm0, 32), 52);
+  __m256i e = _mm256_insertf128_si256(_mm256_setzero_si256(), lo, 0);
+  e = _mm256_insertf128_si256(e, hi, 1);
+
+  // Construct the result 2^n * exp(g) = e * x. The max is used to catch
+  // non-finite values in the input.
+  return pmax(pmul(x, _mm256_castsi256_pd(e)), _x);
 }
 
 // Functions for sqrt.
@@ -97,39 +362,37 @@ pexp<Packet4d>(const Packet4d& _x) {
 // For detail see here: http://www.beyond3d.com/content/articles/8/
 #if EIGEN_FAST_MATH
 template <>
-EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
-Packet8f psqrt<Packet8f>(const Packet8f& _x) {
+psqrt<Packet8f>(const Packet8f& _x) {
-  Packet8f minus_half_x = pmul(_x, pset1<Packet8f>(-0.5f));
+  Packet8f half = pmul(_x, pset1<Packet8f>(.5f));
-  Packet8f denormal_mask = pandnot(
+  Packet8f denormal_mask = _mm256_and_ps(
-      pcmp_lt(_x, pset1<Packet8f>((std::numeric_limits<float>::min)())),
+      _mm256_cmp_ps(_x, pset1<Packet8f>((std::numeric_limits<float>::min)()),
-      pcmp_lt(_x, pzero(_x)));
+                    _CMP_LT_OQ),
+      _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_GE_OQ));
 
   // Compute approximate reciprocal sqrt.
   Packet8f x = _mm256_rsqrt_ps(_x);
   // Do a single step of Newton's iteration.
-  x = pmul(x, pmadd(minus_half_x, pmul(x,x), pset1<Packet8f>(1.5f)));
+  x = pmul(x, psub(pset1<Packet8f>(1.5f), pmul(half, pmul(x,x))));
   // Flush results for denormals to zero.
-  return pandnot(pmul(_x,x), denormal_mask);
+  return _mm256_andnot_ps(denormal_mask, pmul(_x,x));
 }
-
 #else
-
 template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet8f psqrt<Packet8f>(const Packet8f& _x) {
+Packet8f psqrt<Packet8f>(const Packet8f& x) {
-  return _mm256_sqrt_ps(_x);
+  return _mm256_sqrt_ps(x);
 }
-
 #endif
-
 template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet4d psqrt<Packet4d>(const Packet4d& _x) {
+Packet4d psqrt<Packet4d>(const Packet4d& x) {
-  return _mm256_sqrt_pd(_x);
+  return _mm256_sqrt_pd(x);
 }
-
 #if EIGEN_FAST_MATH
+
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
   _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(nan, 0x7fc00000);
   _EIGEN_DECLARE_CONST_Packet8f(one_point_five, 1.5f);
   _EIGEN_DECLARE_CONST_Packet8f(minus_half, -0.5f);
   _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(flt_min, 0x00800000);
@@ -138,88 +401,36 @@ Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
 
   // select only the inverse sqrt of positive normal inputs (denormals are
   // flushed to zero and cause infs as well).
-  Packet8f lt_min_mask = _mm256_cmp_ps(_x, p8f_flt_min, _CMP_LT_OQ);
+  Packet8f le_zero_mask = _mm256_cmp_ps(_x, p8f_flt_min, _CMP_LT_OQ);
-  Packet8f inf_mask =  _mm256_cmp_ps(_x, p8f_inf, _CMP_EQ_OQ);
+  Packet8f x = _mm256_andnot_ps(le_zero_mask, _mm256_rsqrt_ps(_x));
-  Packet8f not_normal_finite_mask = _mm256_or_ps(lt_min_mask, inf_mask);
 
-  // Compute an approximate result using the rsqrt intrinsic.
+  // Fill in NaNs and Infs for the negative/zero entries.
-  Packet8f y_approx = _mm256_rsqrt_ps(_x);
+  Packet8f neg_mask = _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_LT_OQ);
+  Packet8f zero_mask = _mm256_andnot_ps(neg_mask, le_zero_mask);
+  Packet8f infs_and_nans = _mm256_or_ps(_mm256_and_ps(neg_mask, p8f_nan),
+                                        _mm256_and_ps(zero_mask, p8f_inf));
 
-  // Do a single step of Newton-Raphson iteration to improve the approximation.
+  // Do a single step of Newton's iteration.
-  // This uses the formula y_{n+1} = y_n * (1.5 - y_n * (0.5 * x) * y_n).
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8f_one_point_five));
-  // It is essential to evaluate the inner term like this because forming
-  // y_n^2 may over- or underflow.
-  Packet8f y_newton = pmul(y_approx, pmadd(y_approx, pmul(neg_half, y_approx), p8f_one_point_five));
 
-  // Select the result of the Newton-Raphson step for positive normal arguments.
+  // Insert NaNs and Infs in all the right places.
-  // For other arguments, choose the output of the intrinsic. This will
+  return _mm256_or_ps(x, infs_and_nans);
-  // return rsqrt(+inf) = 0, rsqrt(x) = NaN if x < 0, and rsqrt(x) = +inf if
-  // x is zero or a positive denormalized float (equivalent to flushing positive
-  // denormalized inputs to zero).
-  return pselect<Packet8f>(not_normal_finite_mask, y_approx, y_newton);
 }
 
 #else
 template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
+Packet8f prsqrt<Packet8f>(const Packet8f& x) {
   _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
-  return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(_x));
+  return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(x));
 }
 #endif
 
 template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
-Packet4d prsqrt<Packet4d>(const Packet4d& _x) {
+Packet4d prsqrt<Packet4d>(const Packet4d& x) {
   _EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
-  return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(_x));
+  return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(x));
 }
 
-F16_PACKET_FUNCTION(Packet8f, Packet8h, psin)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, pcos)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, plog)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, plog2)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, plog1p)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, pexpm1)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, pexp)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, ptanh)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, psqrt)
-F16_PACKET_FUNCTION(Packet8f, Packet8h, prsqrt)
-
-template <>
-EIGEN_STRONG_INLINE Packet8h pfrexp(const Packet8h& a, Packet8h& exponent) {
-  Packet8f fexponent;
-  const Packet8h out = float2half(pfrexp<Packet8f>(half2float(a), fexponent));
-  exponent = float2half(fexponent);
-  return out;
-}
-
-template <>
-EIGEN_STRONG_INLINE Packet8h pldexp(const Packet8h& a, const Packet8h& exponent) {
-  return float2half(pldexp<Packet8f>(half2float(a), half2float(exponent)));
-}
-
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psin)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pcos)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog2)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog1p)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pexpm1)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pexp)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, ptanh)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psqrt)
-BF16_PACKET_FUNCTION(Packet8f, Packet8bf, prsqrt)
-
-template <>
-EIGEN_STRONG_INLINE Packet8bf pfrexp(const Packet8bf& a, Packet8bf& exponent) {
-  Packet8f fexponent;
-  const Packet8bf out = F32ToBf16(pfrexp<Packet8f>(Bf16ToF32(a), fexponent));
-  exponent = F32ToBf16(fexponent);
-  return out;
-}
-
-template <>
-EIGEN_STRONG_INLINE Packet8bf pldexp(const Packet8bf& a, const Packet8bf& exponent) {
-  return F32ToBf16(pldexp<Packet8f>(Bf16ToF32(a), Bf16ToF32(exponent)));
-}
 
 }  // end namespace internal
 

Eigen/src/Core/arch/AVX/TypeCasting.h

@@ -35,79 +35,15 @@ struct type_casting_traits<int, float> {
 };
 
 
-#ifndef EIGEN_VECTORIZE_AVX512
-
-template <>
-struct type_casting_traits<Eigen::half, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-
-template <>
-struct type_casting_traits<float, Eigen::half> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template <>
-struct type_casting_traits<bfloat16, float> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-template <>
-struct type_casting_traits<float, bfloat16> {
-  enum {
-    VectorizedCast = 1,
-    SrcCoeffRatio = 1,
-    TgtCoeffRatio = 1
-  };
-};
-
-#endif  // EIGEN_VECTORIZE_AVX512
 
 template<> EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {
-  return _mm256_cvttps_epi32(a);
+  return _mm256_cvtps_epi32(a);
 }
 
 template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {
   return _mm256_cvtepi32_ps(a);
 }
 
-template<> EIGEN_STRONG_INLINE Packet8i preinterpret<Packet8i,Packet8f>(const Packet8f& a) {
-  return _mm256_castps_si256(a);
-}
-
-template<> EIGEN_STRONG_INLINE Packet8f preinterpret<Packet8f,Packet8i>(const Packet8i& a) {
-  return _mm256_castsi256_ps(a);
-}
-
-template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
-  return half2float(a);
-}
-
-template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8bf, Packet8f>(const Packet8bf& a) {
-  return Bf16ToF32(a);
-}
-
-template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
-  return float2half(a);
-}
-
-template<> EIGEN_STRONG_INLINE Packet8bf pcast<Packet8f, Packet8bf>(const Packet8f& a) {
-  return F32ToBf16(a);
-}
-
 } // end namespace internal
 
 } // end namespace Eigen