Daniel Trebbien
0c57be407d
Move up the specialization of std::numeric_limits
...
This fixes a compilation error seen when building TensorFlow on macOS:
https://github.com/tensorflow/tensorflow/issues/17067
2018-02-18 15:35:45 -08:00
nluehr
aefd5fd5c4
Replace __float2half_rn with __float2half
...
The latter provides a consistent definition for CUDA 8.0 and 9.0.
2017-11-28 10:15:46 -08:00
nluehr
dd6de618c3
Fix incorrect integer cast in predux<half2>().
...
Bug corrupts results on Maxwell and earlier GPU architectures.
2017-11-21 10:47:00 -08:00
Henry Schreiner
9bb26eb8f1
Restore __device__
2017-10-21 00:50:38 +00:00
Henry Schreiner
4245475d22
Fixing missing inlines on device functions for newer CUDA cards
2017-10-20 03:20:13 +00:00
Gael Guennebaud
9c353dd145
Add C++11 max_digits10 for half.
2017-09-06 10:22:47 +02:00
Benoit Steiner
a4089991eb
Added support for CUDA 9.0.
2017-08-31 02:49:39 +00:00
Gael Guennebaud
21633e585b
bug #1462 : remove all occurences of the deprecated __CUDACC_VER__ macro by introducing EIGEN_CUDACC_VER
2017-08-24 11:06:47 +02:00
Gael Guennebaud
bbd97b4095
Add a EIGEN_NO_CUDA option, and introduce EIGEN_CUDACC and EIGEN_CUDA_ARCH aliases
2017-07-17 01:02:51 +02:00
Benoit Steiner
3baef62b9a
Added missing __device__ qualifier
2017-06-13 12:56:55 -07:00
Benoit Steiner
449936828c
Added missing __device__ qualifier
2017-06-13 12:54:57 -07:00
Gael Guennebaud
26f552c18d
fix compilation of Half in C++98 (issue introduced in previous commit)
2017-06-09 13:36:58 +02:00
Gael Guennebaud
d588822779
Add missing std::numeric_limits specialization for half, and complete NumTraits<half>
2017-06-09 11:51:53 +02:00
Abhijit Kundu
9bc0a35731
Fixed nested angle barckets >> issue when compiling with cuda 8
2017-04-27 03:09:03 -04:00
Gael Guennebaud
e958c2baac
remove UTF8 symbols
2017-03-07 10:47:40 +01:00
Benoit Steiner
7b61944669
Made most of the packet math primitives usable within CUDA kernel when compiling with clang
2017-02-28 17:05:28 -08:00
Benoit Steiner
34d9fce93b
Avoid unecessary float to double conversions.
2017-02-27 16:33:33 -08:00
Srinivas Vasudevan
f7d7c33a28
Fix expm1 CUDA implementation (do not shadow exp CUDA implementation).
2016-12-05 12:19:01 -08:00
Srinivas Vasudevan
09ee7f0c80
Fix small nit where I changed name of plog1p to pexpm1.
2016-12-02 15:30:12 -08:00
Srinivas Vasudevan
218764ee1f
Added support for expm1 in Eigen.
2016-12-02 14:13:01 -08:00
Rasmus Munk Larsen
a0329f64fb
Add a default constructor for the "fake" __half class when not using the
...
__half class provided by CUDA.
2016-11-29 13:18:09 -08:00
Benoit Steiner
dff9a049c4
Optimized the computation of exp, sqrt, ceil anf floor for fp16 on Pascal GPUs
2016-11-16 09:01:51 -08:00
Benoit Steiner
c80587c92b
Merged eigen/eigen into default
2016-11-03 03:55:11 -07:00
Benoit Steiner
7a0e96b80d
Gate the code that refers to cuda fp16 primitives more thoroughly
2016-11-01 12:08:09 -07:00
Benoit Steiner
38b6048e14
Deleted redundant implementation of predux
2016-10-12 14:37:56 -07:00
Benoit Steiner
78d2926508
Merged eigen/eigen into default
2016-10-12 13:46:29 -07:00
Benoit Steiner
2e2f48e30e
Take advantage of AVX512 instructions whenever possible to speedup the processing of 16 bit floats.
2016-10-12 13:45:39 -07:00
Benoit Steiner
d485d12c51
Added missing AVX intrinsics for fp16: in particular, implemented predux which is required by the matrix-vector code.
2016-10-06 10:41:03 -07:00
Benoit Steiner
698ff69450
Properly characterize the CUDA packet primitives for fp16 as device only
2016-10-04 16:53:30 -07:00
Benoit Steiner
409e887d78
Added support for constand std::complex numbers on GPU
2016-10-03 11:06:24 -07:00
Benoit Steiner
26f9907542
Added missing typedefs
2016-09-20 12:58:03 -07:00
RJ Ryan
b2c6dc48d9
Add CUDA-specific std::complex<T> specializations for scalar_sum_op, scalar_difference_op, scalar_product_op, and scalar_quotient_op.
2016-09-20 07:18:20 -07:00
Gael Guennebaud
8f4b4ad5fb
use ::hlog if available.
2016-08-29 11:05:32 +02:00
Gael Guennebaud
35a8e94577
bug #1167 : simplify installation of header files using cmake's install(DIRECTORY ...) command.
2016-08-29 10:59:37 +02:00
Gael Guennebaud
d937a420a2
Fix compilation with MSVC by using our portable numext::log1p implementation.
2016-08-22 15:44:21 +02:00
Igor Babuschkin
59bacfe520
Fix compilation on CUDA 8 by removing call to h2log1p
2016-08-15 23:38:05 +01:00
Igor Babuschkin
aee693ac52
Add log1p support for CUDA and half floats
2016-08-08 20:24:59 +01:00
Benoit Steiner
fe778427f2
Fixed the constructors of the new half_base class.
2016-08-04 18:32:26 -07:00
Benoit Steiner
9506343349
Fixed the isnan, isfinite and isinf operations on GPU
2016-08-04 17:25:53 -07:00
Gael Guennebaud
17b9a55d98
Move Eigen::half_impl::half to Eigen::half while preserving the free functions to the Eigen::half_impl namespace together with ADL
2016-08-04 00:00:43 +02:00
Benoit Steiner
02fe89f5ef
half implementation has been moved to half_impl namespace
2016-07-29 15:09:34 -07:00
Christoph Hertzberg
c5b893f434
bug #1266 : half implementation has been moved to half_impl namespace
2016-07-29 18:36:08 +02:00
Gael Guennebaud
395c835f4b
Fix CUDA compilation
2016-07-22 15:30:24 +02:00
Gael Guennebaud
47afc9a365
More cleaning in half:
...
- put its definition and functions in its own half_impl namespace such that the free function does not polute the Eigen namespace while still making them visible for half through ADL.
- expose Eigen::half throguh a using statement
- move operator<< from std to half_float namespace
2016-07-22 14:33:28 +02:00
Gael Guennebaud
0f350a8b7e
Fix CUDA compilation
2016-07-21 18:47:07 +02:00
Gael Guennebaud
87fbda812f
Add missing log10 and random generator for half.
2016-07-21 15:46:45 +02:00
Gael Guennebaud
01d12d3e82
Some cleanup in Halh: standard functions should be defined in the namespace of the class half to make ADL work, and thus the global is* functions can be removed.
2016-07-21 15:10:48 +02:00
Gael Guennebaud
a96a7ce3f7
Move CUDA's special functions to SpecialFunctions module.
2016-07-11 18:39:11 +02:00
Gael Guennebaud
2f7e2614e7
bug #1232 : refactor special functions as a new SpecialFunctions module, currently in unsupported/.
2016-07-08 11:13:55 +02:00
Benoit Steiner
8fd57a97f2
Enable the vectorization of adds and mults of fp16
2016-06-07 18:22:18 -07:00
