mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-15 07:10:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

merge

Browse Source
This commit is contained in:
Gael Guennebaud 2016-07-22 16:43:12 +02:00
commit 32d95e86c9
3 changed files with 40 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/src/Core/arch/CUDA/PacketMathHalf.h
View File

@ -262,7 +262,7 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half predux_mul<half2>(c
#else #else
float a1 = __low2float(a); float a1 = __low2float(a);
float a2 = __high2float(a); float a2 = __high2float(a);
return Eigen::half(internal::raw_uint16_to_half(__float2half_rn(a1 * a2))); return Eigen::half(half_impl::raw_uint16_to_half(__float2half_rn(a1 * a2)));
#endif #endif
} }

12
test/boostmultiprec.cpp
View File

@ -71,6 +71,18 @@ namespace Eigen {
template<> template<>
Real test_precision<Real>() { return 1e-50; } Real test_precision<Real>() { return 1e-50; }
// needed in C++93 mode where number does not support explicit cast.
namespace internal {
template<typename NewType>
struct cast_impl<Real,NewType>
{
static inline NewType run(const Real& x)
{
return x.template convert_to<NewType>();
}
};
}
} }
namespace boost { namespace boost {

47
unsupported/test/cxx11_tensor_of_float16_cuda.cu
View File

@ -21,6 +21,7 @@ using Eigen::Tensor;
#ifdef EIGEN_HAS_CUDA_FP16 #ifdef EIGEN_HAS_CUDA_FP16
template<typename>
void test_cuda_conversion() { void test_cuda_conversion() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -55,7 +56,7 @@ void test_cuda_conversion() {
gpu_device.deallocate(d_conv); gpu_device.deallocate(d_conv);
} }
template<typename>
void test_cuda_unary() { void test_cuda_unary() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -92,7 +93,7 @@ void test_cuda_unary() {
gpu_device.deallocate(d_res_float); gpu_device.deallocate(d_res_float);
} }
template<typename>
void test_cuda_elementwise() { void test_cuda_elementwise() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -134,6 +135,7 @@ void test_cuda_elementwise() {
gpu_device.deallocate(d_res_float); gpu_device.deallocate(d_res_float);
} }
template<typename>
void test_cuda_trancendental() { void test_cuda_trancendental() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -186,7 +188,10 @@ void test_cuda_trancendental() {
} }
for (int i = 0; i < num_elem; ++i) { for (int i = 0; i < num_elem; ++i) {
std::cout << "Checking elemwise log " << i << " input = " << input2(i) << " full = " << full_prec2(i) << " half = " << half_prec2(i) << std::endl; std::cout << "Checking elemwise log " << i << " input = " << input2(i) << " full = " << full_prec2(i) << " half = " << half_prec2(i) << std::endl;
VERIFY_IS_APPROX(full_prec2(i), half_prec2(i)); if(std::abs(input2(i)-1.f)<0.05f) // log lacks accurary nearby 1
VERIFY_IS_APPROX(full_prec2(i)+Eigen::half(0.1f), half_prec2(i)+Eigen::half(0.1f));
else
VERIFY_IS_APPROX(full_prec2(i), half_prec2(i));
} }
gpu_device.deallocate(d_float1); gpu_device.deallocate(d_float1);
gpu_device.deallocate(d_float2); gpu_device.deallocate(d_float2);
@ -196,7 +201,7 @@ void test_cuda_trancendental() {
gpu_device.deallocate(d_res2_float); gpu_device.deallocate(d_res2_float);
} }
template<typename>
void test_cuda_contractions() { void test_cuda_contractions() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -247,7 +252,7 @@ void test_cuda_contractions() {
gpu_device.deallocate(d_res_float); gpu_device.deallocate(d_res_float);
} }
template<typename>
void test_cuda_reductions(int size1, int size2, int redux) { void test_cuda_reductions(int size1, int size2, int redux) {
std::cout << "Reducing " << size1 << " by " << size2 std::cout << "Reducing " << size1 << " by " << size2
@ -296,17 +301,19 @@ void test_cuda_reductions(int size1, int size2, int redux) {
gpu_device.deallocate(d_res_float); gpu_device.deallocate(d_res_float);
} }
template<typename>
void test_cuda_reductions() { void test_cuda_reductions() {
test_cuda_reductions(13, 13, 0); test_cuda_reductions<void>(13, 13, 0);
test_cuda_reductions(13, 13, 1); test_cuda_reductions<void>(13, 13, 1);
test_cuda_reductions(35, 36, 0); test_cuda_reductions<void>(35, 36, 0);
test_cuda_reductions(35, 36, 1); test_cuda_reductions<void>(35, 36, 1);
test_cuda_reductions(36, 35, 0); test_cuda_reductions<void>(36, 35, 0);
test_cuda_reductions(36, 35, 1); test_cuda_reductions<void>(36, 35, 1);
} }
template<typename>
void test_cuda_full_reductions() { void test_cuda_full_reductions() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream); Eigen::GpuDevice gpu_device(&stream);
@ -355,7 +362,7 @@ void test_cuda_full_reductions() {
gpu_device.deallocate(d_res_float); gpu_device.deallocate(d_res_float);
} }
template<typename>
void test_cuda_forced_evals() { void test_cuda_forced_evals() {
Eigen::CudaStreamDevice stream; Eigen::CudaStreamDevice stream;
@ -409,14 +416,14 @@ void test_cuda_forced_evals() {
void test_cxx11_tensor_of_float16_cuda() void test_cxx11_tensor_of_float16_cuda()
{ {
#ifdef EIGEN_HAS_CUDA_FP16 #ifdef EIGEN_HAS_CUDA_FP16
CALL_SUBTEST_1(test_cuda_conversion()); CALL_SUBTEST_1(test_cuda_conversion<void>());
CALL_SUBTEST_1(test_cuda_unary()); CALL_SUBTEST_1(test_cuda_unary<void>());
CALL_SUBTEST_1(test_cuda_elementwise()); CALL_SUBTEST_1(test_cuda_elementwise<void>());
CALL_SUBTEST_1(test_cuda_trancendental()); CALL_SUBTEST_1(test_cuda_trancendental<void>());
CALL_SUBTEST_2(test_cuda_contractions()); CALL_SUBTEST_2(test_cuda_contractions<void>());
CALL_SUBTEST_3(test_cuda_reductions()); CALL_SUBTEST_3(test_cuda_reductions<void>());
CALL_SUBTEST_4(test_cuda_full_reductions()); CALL_SUBTEST_4(test_cuda_full_reductions<void>());
CALL_SUBTEST_5(test_cuda_forced_evals()); CALL_SUBTEST_5(test_cuda_forced_evals<void>());
#else #else
std::cout << "Half floats are not supported by this version of cuda: skipping the test" << std::endl; std::cout << "Half floats are not supported by this version of cuda: skipping the test" << std::endl;
#endif #endif