mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-06 14:14:46 +08:00
Code Issues Packages Projects Releases Wiki Activity

FP16 on CUDA are only available starting with cuda 7.5. Disable them when using an older version of CUDA

Browse Source
This commit is contained in:
Benoit Steiner 2016-02-18 23:15:23 -08:00
parent f36c0c2c65
commit 0606a0a39b
3 changed files with 11 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Eigen/Core
View File

@ -200,7 +200,10 @@
#if defined __CUDACC__ #if defined __CUDACC__
#define EIGEN_VECTORIZE_CUDA #define EIGEN_VECTORIZE_CUDA
#include <vector_types.h> #include <vector_types.h>
#include <cuda_fp16.h> #if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500
#define EIGEN_HAS_CUDA_FP16
#include <cuda_fp16.h>
#endif
#endif #endif
#if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE) #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)

3
Eigen/src/Core/arch/CUDA/TypeCasting.h
View File

@ -14,6 +14,8 @@ namespace Eigen {
namespace internal { namespace internal {
#if defined(EIGEN_HAS_CUDA_FP16)
template<> template<>
struct scalar_cast_op<float, half> { struct scalar_cast_op<float, half> {
EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op) EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
@ -92,6 +94,7 @@ template<> EIGEN_STRONG_INLINE half2 pcast<float4, half2>(const float4& a) {
#endif #endif
} }
#endif
} // end namespace internal } // end namespace internal

5
unsupported/test/cxx11_tensor_of_float16_cuda.cu
View File

@ -19,6 +19,7 @@
using Eigen::Tensor; using Eigen::Tensor;
#ifdef EIGEN_HAS_CUDA_FP16
void test_cuda_conversion() { void test_cuda_conversion() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -52,9 +53,11 @@ void test_cuda_conversion() {
gpu_device.deallocate(d_half); gpu_device.deallocate(d_half);
gpu_device.deallocate(d_conv); gpu_device.deallocate(d_conv);
} }
#endif
void test_cxx11_tensor_of_float16_cuda() void test_cxx11_tensor_of_float16_cuda()
{ {
#ifdef EIGEN_HAS_CUDA_FP16
CALL_SUBTEST_1(test_cuda_conversion()); CALL_SUBTEST_1(test_cuda_conversion());
#endif
} }