Commit Graph

985 Commits

Author SHA1 Message Date
Rasmus Munk Larsen
5418154a45 Fix oversharding bug in parallelFor. 2018-06-20 17:51:48 -07:00
Michael Figurnov
30fa3d0454 Merge from eigen/eigen 2018-06-07 17:57:56 +01:00
Gael Guennebaud
b3fd93207b Fix typos found using codespell 2018-06-07 14:43:02 +02:00
Michael Figurnov
4bd158fa37 Derivative of the incomplete Gamma function and the sample of a Gamma random variable.
In addition to igamma(a, x), this code implements:
* igamma_der_a(a, x) = d igamma(a, x) / da -- derivative of igamma with respect to the parameter
* gamma_sample_der_alpha(alpha, sample) -- reparameterization derivative of a Gamma(alpha, 1) random variable sample with respect to the alpha parameter

The derivatives are computed by forward mode differentiation of the igamma(a, x) code. Although gamma_sample_der_alpha can be implemented via igamma_der_a, a separate function is more accurate and efficient due to analytical cancellation of some terms. All three functions are implemented by a method parameterized with "mode" that always computes the derivatives, but does not return them unless required by the mode. The compiler is expected to (and, based on benchmarks, does) skip the unnecessary computations depending on the mode.
2018-06-06 18:49:26 +01:00
Benoit Steiner
e206f8d4a4 Merged in mfigurnov/eigen (pull request PR-400)
Exponentially scaled modified Bessel functions of order zero and one.

Approved-by: Benoit Steiner <benoit.steiner.goog@gmail.com>
2018-06-05 17:05:21 +00:00
Penporn Koanantakool
e2ed0cf8ab Add a ThreadPoolInterface* getter for ThreadPoolDevice. 2018-06-02 12:07:49 -07:00
Michael Figurnov
f216854453 Exponentially scaled modified Bessel functions of order zero and one.
The functions are conventionally called i0e and i1e. The exponentially scaled version is more numerically stable. The standard Bessel functions can be obtained as i0(x) = exp(|x|) i0e(x)

The code is ported from Cephes and tested against SciPy.
2018-05-31 15:34:53 +01:00
Katrin Leinweber
ea94543190 Hyperlink DOIs against preferred resolver 2018-05-24 18:55:40 +02:00
Vamsi Sripathi
6293ad3f39 Performance improvements to tensor broadcast operation
1. Added new packet functions using SIMD for NByOne, OneByN cases
  2. Modified existing packet functions to reduce index calculations when input stride is non-SIMD
  3. Added 4 test cases to cover the new packet functions
2018-05-23 14:02:05 -07:00
Benoit Steiner
0371380d5b Merged in rmlarsen/eigen2 (pull request PR-393)
Rename scalar_clip_op to scalar_clamp_op to prevent collision with existing functor in TensorFlow.
2018-05-16 21:45:42 +00:00
Rasmus Munk Larsen
b8d36774fa Rename clip2 to clamp. 2018-05-16 14:04:48 -07:00
Rasmus Munk Larsen
812480baa3 Rename scalar_clip_op to scalar_clip2_op to prevent collision with existing functor in TensorFlow. 2018-05-16 09:49:24 -07:00
Benoit Steiner
1403c2c15b Merged in didierjansen/eigen (pull request PR-360)
Fix bugs and typos in the contraction example of the tensor README
2018-05-16 01:16:36 +00:00
Rasmus Munk Larsen
b8c8e5f436 Add vectorized clip functor for Eigen Tensors. 2018-05-14 16:07:13 -07:00
Benoit Steiner
6118c6ff4f Enable RawAccess to tensor slices whenever possinle.
Avoid 32-bit integer overflow in TensorSlicingOp
2018-04-30 11:28:12 -07:00
Weiming Zhao
b0eda3cb9f Avoid using memcpy for non-POD elements 2018-04-11 11:37:06 +02:00
Lee.Deokjae
5b3c367926 Fix typos in the contraction example of tensor README 2018-01-06 14:36:19 +09:00
Gael Guennebaud
73214c4bd0 Workaround nvcc 9.0 issue. See PR 351.
https://bitbucket.org/eigen/eigen/pull-requests/351
2017-12-15 14:10:59 +01:00
Benoit Steiner
8f55956a57 Update the padding computation for PADDING_SAME to be consistent with TensorFlow. 2018-01-30 20:22:12 +00:00
RJ Ryan
59985cfd26 Disable use of recurrence for computing twiddle factors. Fixes FFT precision issues for large FFTs. https://github.com/tensorflow/tensorflow/issues/10749#issuecomment-354557689 2017-12-31 10:44:56 -05:00
Yangzihao Wang
3122477c86 Update the padding computation for PADDING_SAME to be consistent with TensorFlow. 2017-12-12 11:15:24 -08:00
Benoit Steiner
3949615176 Merged in JonasMu/eigen (pull request PR-329)
Added an example for a contraction to a scalar value to README.md

Approved-by: Jonas Harsch <jonas.harsch@gmail.com>
2017-10-27 07:27:46 +00:00
Benoit Steiner
8eb4b9d254 Merged in benoitsteiner/opencl (pull request PR-341) 2017-10-17 16:39:28 +00:00
Rasmus Munk Larsen
f349507e02 Specialize ThreadPoolDevice::enqueueNotification for the case with no args. As an example this reduces binary size of an TensorFlow demo app for Android by about 2.5%. 2017-10-13 15:58:12 -07:00
Mehdi Goli
2062ac9958 Changes required for new ComputeCpp CE version. 2017-09-18 18:17:39 +01:00
Rasmus Munk Larsen
1b7294f6fc Fix cut-and-paste error. 2017-09-08 16:35:58 -07:00
Rasmus Munk Larsen
94e2213b38 Avoid undefined behavior in Eigen::TensorCostModel::numThreads.
If the cost is large enough then the thread count can be larger than the maximum
representable int, so just casting it to an int is undefined behavior.

Contributed by phurst@google.com.
2017-09-08 15:49:55 -07:00
Jonas Harsch
a991c80365 Added an example for a contraction to a scalar value, e.g. a double contraction of two second order tensors and how you can get the value of the result. I lost one day to get this doen so I think it will help some guys. I also added Eigen:: to the IndexPair and and array in the same example. 2017-09-01 11:30:26 +00:00
Benoit Steiner
a4089991eb Added support for CUDA 9.0. 2017-08-31 02:49:39 +00:00
Benoit Steiner
84d7be103a Fixing Argmax that was breaking upstream TensorFlow. 2017-07-22 03:19:34 +00:00
Benoit Steiner
f0b154a4b0 Code cleanup 2017-07-10 09:54:09 -07:00
Benoit Steiner
575cda76b3 Fixed syntax errors generated by xcode 2017-07-09 11:39:01 -07:00
Benoit Steiner
5ac27d5b51 Avoid relying on cxx11 features when possible. 2017-07-08 21:58:44 -07:00
Benoit Steiner
c5a241ab9b Merged in benoitsteiner/opencl (pull request PR-323)
Improved support for OpenCL
2017-07-07 16:27:33 +00:00
Benoit Steiner
b7ae4dd9ef Merged in hughperkins/eigen/add-endif-labels-TensorReductionCuda.h (pull request PR-315)
Add labels to #ifdef, in TensorReductionCuda.h
2017-07-07 04:23:52 +00:00
Benoit Steiner
9daed67952 Merged in tntnatbry/eigen (pull request PR-319)
Tensor Trace op
2017-07-07 04:18:03 +00:00
Benoit Steiner
6795512e59 Improved the randomness of the tensor random generator 2017-07-06 21:12:45 -07:00
Benoit Steiner
dc524ac716 Fixed compilation warning 2017-07-06 21:11:15 -07:00
Benoit Steiner
62b4634ebe Merged in mehdi_goli/upstr_benoit/TensorSYCLImageVolumePatchFixed (pull request PR-14)
Applying Benoit's comment for Fixing ImageVolumePatch.

* Applying Benoit's comment for Fixing ImageVolumePatch. Fixing conflict on cmake file.

* Fixing dealocation of the memory in ImagePatch test for SYCL.

* Fixing the automerge issue.
2017-07-06 05:08:13 +00:00
Benoit Steiner
53725c10b8 Merged in mehdi_goli/opencl/DataDependancy (pull request PR-10)
DataDependancy

* Wrapping data type to the pointer class for sycl in non-terminal nodes; not having that breaks Tensorflow Conv2d code.

* Applying Ronnan's Comments.

* Applying benoit's comments
2017-06-28 17:55:23 +00:00
Hugh Perkins
9341f258d4 Add labels to #ifdef, in TensorReductionCuda.h 2017-06-06 15:51:06 +01:00
Benoit Steiner
1e736b9ead Merged in mehdi_goli/opencl/SYCLAlignAllocator (pull request PR-7)
Fixing SYCL alignment issue required by TensorFlow.
2017-05-26 17:23:00 +00:00
Benoit Steiner
9dee55ec33 Merged eigen/eigen into default 2017-05-26 09:01:04 -07:00
Mehdi Goli
0370d3576e Applying Ronnan's comments. 2017-05-26 16:01:48 +01:00
Mehdi Goli
e3f964ed55 Applying Benoit's comment;removing dead code. 2017-05-25 11:17:26 +01:00
a-doumoulakis
fb853a857a Restore misplaced comment 2017-05-24 17:50:15 +01:00
a-doumoulakis
7a8ba565f8 Merge changed from upstream 2017-05-24 17:45:29 +01:00
Mehdi Goli
76c0fc1f95 Fixing SYCL alignment issue required by TensorFlow. 2017-05-22 16:49:32 +01:00
Mehdi Goli
2d17128d6f Fixing suported device list. 2017-05-22 16:40:33 +01:00
a-doumoulakis
052426b824 Add support for triSYCL
Eigen is now able to use triSYCL with EIGEN_SYCL_TRISYCL and TRISYCL_INCLUDE_DIR options

Fix contraction kernel with correct nd_item dimension
2017-05-05 19:26:27 +01:00