devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix outdated documentation across multiple .dox files

Browse Source 
libeigen/eigen!2148

Co-authored-by: Rasmus Munk Larsen <rmlarsen@gmail.com>
This commit is contained in:
Rasmus Munk Larsen
2026-02-18 12:23:41 -08:00
parent bdec88009d
commit 073190be04
5 changed files with 60 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
doc/UsingNVCC.dox
View File

@@ -3,27 +3,36 @@ namespace Eigen {
/** \page TopicCUDA Using Eigen in CUDA kernels
Staring from CUDA 5.5 and Eigen 3.3, it is possible to use Eigen's matrices, vectors, and arrays for fixed size within CUDA kernels. This is especially useful when working on numerous but small problems. By default, when Eigen's headers are included within a .cu file compiled by nvcc most Eigen's functions and methods are prefixed by the \c __device__ \c __host__ keywords making them callable from both host and device code.
This support can be disabled by defining \c EIGEN_NO_CUDA before including any Eigen's header.
This might be useful to disable some warnings when a .cu file makes use of Eigen on the host side only.
However, in both cases, host's SIMD vectorization has to be disabled in .cu files.
It is thus \b strongly \b recommended to properly move all costly host computation from your .cu files to regular .cpp files.
\section TopicCUDA_Overview Overview
Known issues:
%Eigen's fixed-size matrices, vectors, and arrays can be used inside CUDA and HIP kernels.
This is especially useful when working on numerous but small problems.
By default, when %Eigen's headers are included within a \c .cu file compiled by \c nvcc,
most of %Eigen's functions and methods are prefixed by the \c __device__ \c __host__ keywords
making them callable from both host and device code.
- \c nvcc with MS Visual Studio does not work (patch welcome)
- \c nvcc 5.5 with gcc-4.7 (or greater) has issues with the standard \c \<limits\> header file. To workaround this, you can add the following before including any other files:
\code
// workaround issue between gcc >= 4.7 and cuda 5.5
#if (defined __GNUC__) && (__GNUC__>4 || __GNUC_MINOR__>=7)
#undef _GLIBCXX_ATOMIC_BUILTINS
#undef _GLIBCXX_USE_INT128
#endif
\endcode
- On 64bits system Eigen uses \c long \c int as the default type for indexes and sizes. On CUDA device, it would make sense to default to 32 bits \c int.
However, to keep host and CUDA code compatible, this cannot be done automatically by %Eigen, and the user is thus required to define \c EIGEN_DEFAULT_DENSE_INDEX_TYPE to \c int throughout his code (or only for CUDA code if there is no interaction between host and CUDA code through %Eigen's object).
This support can be disabled by defining \c EIGEN_NO_CUDA (or \c EIGEN_NO_HIP for HIP)
before including any %Eigen header.
This might be useful to disable some warnings when a \c .cu file makes use of %Eigen on the host side only.
\warning Host SIMD vectorization must be disabled in \c .cu files. It is \b strongly
\b recommended to move all costly host-side computation from \c .cu files to regular \c .cpp files.
\section TopicCUDA_HIP HIP support
%Eigen also supports AMD's HIP compiler (\c hipcc). The same \c EIGEN_DEVICE_FUNC mechanism applies:
functions are annotated with \c __device__ \c __host__ when compiling with HIP.
Define \c EIGEN_NO_HIP to disable this. Internally, both CUDA and HIP are unified under the
\c EIGEN_GPUCC macro.
\section TopicCUDA_KnownIssues Known issues
- On 64-bit systems %Eigen uses \c long \c int as the default type for indexes and sizes.
On CUDA/HIP devices, it would make sense to default to 32-bit \c int.
However, to keep host and device code compatible, this cannot be done automatically by %Eigen,
and the user is thus required to define \c EIGEN_DEFAULT_DENSE_INDEX_TYPE to \c int throughout
their code (or only for device code if there is no interaction between host and device code
through %Eigen's objects).
*/