Benoit Steiner
8b2afe33a1
Fixes for the forced evaluation of tensor expressions
...
More tests
2014-10-02 10:39:36 -07:00
Benoit Steiner
7caaf6453b
Added support for tensor reductions and concatenations
2014-10-01 20:38:22 -07:00
Benoit Steiner
1c236f4c9a
Added tests for tensors of const values and tensors of stringswwq::
2014-10-01 20:21:42 -07:00
Benoit Steiner
efdff15749
Fixed a typo in the contraction code
2014-09-06 13:28:24 -07:00
Benoit Steiner
74db22455a
Misc fixes.
2014-09-05 07:47:43 -07:00
Benoit Steiner
d43f737b4a
Added support for evaluation of tensor shuffling operations as lvalues
2014-09-04 20:02:28 -07:00
Benoit Steiner
f50548e86a
Added missing tensor copy constructors. As a result it is now possible to declare and initialize a tensor on the same line, as in:
...
Tensor<bla> T = A + B; or
Tensor<bla> T(A.reshape(new_shape));
2014-09-04 19:50:27 -07:00
Benoit Steiner
b24fe22b1a
Improved the performance of the tensor convolution code by a factor of about 4.
2014-09-03 11:38:13 -07:00
Benoit Steiner
2959045f2f
Optimized the tensor padding code.
2014-08-26 09:47:18 -07:00
Benoit Steiner
36fffe48f7
Misc api improvements and cleanups
2014-08-23 14:35:41 -07:00
Benoit Steiner
fb5c1e9097
Optimized and cleaned up the tensor morphing code
2014-08-23 13:18:30 -07:00
Benoit Steiner
3d298da269
Added support for broadcasting
2014-08-20 17:00:50 -07:00
Benoit Steiner
9ac3c821ea
Improved the speed of convolutions when running on cuda devices
2014-08-19 16:57:10 -07:00
Benoit Steiner
33c702c79f
Added support for fast integer divisions by a constant
...
Sped up tensor slicing by a factor of 3 by using these fast integer divisions.
2014-08-14 22:13:21 -07:00
Benoit Steiner
eeb43f9e2b
Added support for padding, stridding, and shuffling
2014-08-14 00:22:47 -07:00
Benoit Steiner
16047c8d4a
Pulled in the latest changes from the Eigen trunk
2014-08-13 22:25:29 -07:00
Benoit Steiner
916ef48846
Added ability to get the nth element from an abstract array type.
2014-08-13 08:44:47 -07:00
Benoit Steiner
f1d8c13dbc
Fixed misc typos.
2014-08-13 08:40:26 -07:00
Benoit Steiner
9faad2932f
Added missing apis.
2014-08-13 08:36:33 -07:00
Benoit Steiner
f8fad09301
Updated the convolution and contraction evaluators to follow the new EvalSubExprsIfNeeded apu.
2014-08-13 08:33:18 -07:00
Benoit Steiner
72e7529708
Fixed a typo.
2014-08-13 08:29:40 -07:00
Benoit Steiner
1aa2bf8274
Support for in place evaluation of expressions containing slicing and reshaping operations
2014-08-13 08:27:58 -07:00
Benoit Steiner
b1892ab14d
Added suppor for in place evaluation to simple tensor expressions.
...
Use mempy to speedup tensor copies whenever possible.
2014-08-13 08:26:44 -07:00
Benoit Steiner
439feca139
Reworked the TensorExecutor code to support in place evaluation.
2014-08-13 08:22:05 -07:00
Benoit Steiner
647622281e
The tensor assignment code now resizes the destination tensor as needed.
2014-07-31 17:39:04 -07:00
Benoit Steiner
2116e261fb
Made sure that the data stored in fixed sized tensor is aligned.
2014-07-25 09:47:59 -07:00
Benoit Steiner
f7bb7ee3f3
Fixed the assignment operator of the Tensor and TensorMap classes.
2014-07-22 10:31:21 -07:00
Benoit Steiner
40bb98e76a
Added primitives to compare tensor dimensions
2014-07-10 11:29:51 -07:00
Benoit Steiner
ffd3654f67
Vectorized the evaluation of expressions involving tensor slices.
2014-07-10 11:09:46 -07:00
Benoit Steiner
25b2f6624d
Improved the speed of slicing operations.
2014-07-09 12:48:34 -07:00
Benoit Steiner
ea0906dfd8
Improved evaluation of tensor expressions when used as rvalues
2014-07-08 16:43:28 -07:00
Benoit Steiner
cc1bacea5b
Improved the efficiency of the tensor evaluation code on thread pools and gpus.
2014-07-08 16:39:28 -07:00
Benoit Steiner
c285fda7f4
Extended the functionality of the TensorDeviceType classes
2014-07-08 16:30:48 -07:00
Benoit Steiner
7d53633e05
Added support for tensor slicing
2014-07-07 14:10:36 -07:00
Benoit Steiner
bc072c5cba
Added support for tensor slicing
2014-07-07 14:08:45 -07:00
Benoit Steiner
47981c5925
Added support for tensor slicing
2014-07-07 14:07:57 -07:00
Benoit Steiner
f80c8e17eb
Silenced a compilation warning
2014-06-13 10:12:12 -07:00
Benoit Steiner
38ab7e6ed0
Reworked the expression evaluation mechanism in order to make it possible to efficiently compute convolutions and contractions in the future:
...
* The scheduling of computation is moved out the the assignment code and into a new TensorExecutor class
* The assignment itself is now a regular node on the expression tree
* The expression evaluators start by recursively evaluating all their subexpressions if needed
2014-06-13 09:56:51 -07:00
Benoit Steiner
aa664eabb9
Fixed a few compilation errors.
2014-06-10 10:31:29 -07:00
Benoit Steiner
4304c73542
Pulled latest updates from the Eigen main trunk.
2014-06-10 10:23:32 -07:00
Benoit Steiner
925fb6b937
TensorEval are now typed on the device: this will make it possible to use partial template specialization to optimize the strategy of each evaluator for each device type.
...
Started work on partial evaluations.
2014-06-10 09:14:44 -07:00
Benoit Steiner
a77458a8ff
Fixes compilation errors triggered when compiling the tensor contraction code with cxx11 enabled.
2014-06-09 10:06:57 -07:00
Benoit Steiner
a669052f12
Improved support for rvalues in tensor expressions.
2014-06-09 09:45:30 -07:00
Benoit Steiner
36a2b2e9dc
Prevent the generation of unlaunchable cuda kernels when compiling in debug mode.
2014-06-09 09:43:51 -07:00
Benoit Steiner
2859a31ac8
Fixed compilation error
2014-06-09 09:42:34 -07:00
Benoit Steiner
a961d72e65
Added support for convolution and reshaping of tensors.
2014-06-06 16:25:16 -07:00
Christian Seiler
96cb58fa3b
unsupported/TensorSymmetry: factor out completely from Tensor module
...
Remove the symCoeff() method of the the Tensor module and move the
functionality into a new operator() of the symmetry classes. This makes
the Tensor module now completely self-contained without symmetry
support (even though previously it was only a forward declaration and a
otherwise harmless trivial templated method) and also removes the
inconsistency with the rest of eigen w.r.t. the method's naming scheme.
2014-06-04 20:44:22 +02:00
Christian Seiler
ea99433523
unsupported/TensorSymmetry: make symgroup construction autodetect number of indices
...
When constructing a symmetry group, make the code automatically detect
the number of indices required from the indices of the group's
generators. Also, allow the symmetry group to be applied to lists of
indices that are larger than the number of indices of the symmetry
group.
Before:
SGroup<4, Symmetry<0, 1>, Symmetry<2,3>> group;
group.apply<SomeOp, int>(std::array<int,4>{{0, 1, 2, 3}}, 0);
After:
SGroup<Symmetry<0, 1>, Symmetry<2,3>> group;
group.apply<SomeOp, int>(std::array<int,4>{{0, 1, 2, 3}}, 0);
group.apply<SomeOp, int>(std::array<int,5>{{0, 1, 2, 3, 4}}, 0);
This should make the symmetry group easier to use - especially if one
wants to reuse the same symmetry group for different tensors of maybe
different rank.
static/runtime asserts remain for the case where the length of the
index list to which a symmetry group is to be applied is too small.
2014-06-04 20:27:42 +02:00
Christian Seiler
cee62018fc
unsupported/CXX11/Core: allow gen_numeric_list to have a starting point
...
Add a template parameter to gen_numeric_list that acts as a starting
point for the list, i.e. gen_numeric_list<int, 5, 4> will generate a
numeric_list<int, 4, 5, 6, 7, 8>.
2014-06-04 19:54:22 +02:00
Christian Seiler
58cfac9a12
unsupported/ C++11 workarounds: don't use hack for libc++ if not required
...
libc++ from 3.4 onwards supports constexpr std::get, but only if
compiled with -std=c++1y. Change the detection so that libc++'s
internals are only used if either -std=c++1y is not specified or the
library is too old, making the whole hack a bit more future-proof.
2014-06-04 18:47:42 +02:00
