devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add a EIGEN_NO_CUDA option, and introduce EIGEN_CUDACC and EIGEN_CUDA_ARCH aliases

Browse Source
This commit is contained in:
Gael Guennebaud
2017-07-17 01:02:51 +02:00
parent f0b154a4b0
commit bbd97b4095
33 changed files with 134 additions and 126 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
Eigen/src/Core/arch/CUDA/PacketMath.h
View File

@@ -17,7 +17,7 @@ namespace internal {
// Make sure this is only available when targeting a GPU: we don't want to
// introduce conflicts between these packet_traits definitions and the ones
// we'll use on the host side (SSE, AVX, ...)
#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
#if defined(EIGEN_CUDACC) && defined(EIGEN_USE_GPU)
template<> struct is_arithmetic<float4> { enum { value = true }; };
template<> struct is_arithmetic<double2> { enum { value = true }; };
@@ -196,7 +196,7 @@ template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu<double>(double* to
template<>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Aligned>(const float* from) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
return __ldg((const float4*)from);
#else
return make_float4(from[0], from[1], from[2], from[3]);
@@ -204,7 +204,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Aligned>(const fl
}
template<>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Aligned>(const double* from) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
return __ldg((const double2*)from);
#else
return make_double2(from[0], from[1]);
@@ -213,7 +213,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Aligned>(const
template<>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Unaligned>(const float* from) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
return make_float4(__ldg(from+0), __ldg(from+1), __ldg(from+2), __ldg(from+3));
#else
return make_float4(from[0], from[1], from[2], from[3]);
@@ -221,7 +221,7 @@ EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Unaligned>(const
}
template<>
EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Unaligned>(const double* from) {
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
return make_double2(__ldg(from+0), __ldg(from+1));
#else
return make_double2(from[0], from[1]);