GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-29 23:34:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

Marked the CUDA packet primitives as EIGEN_DEVICE_FUNC since they'll end up being executed on the GPU device.

Browse Source
This commit is contained in:
Benoit Steiner 2015-02-19 21:22:51 -08:00
parent 92ceb02c6d
commit 7765039f1c
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
Eigen/src/Core/arch/CUDA/MathFunctions.h
View File

@ -18,49 +18,49 @@ namespace internal {
// introduce conflicts between these packet_traits definitions and the ones // introduce conflicts between these packet_traits definitions and the ones
// we'll use on the host side (SSE, AVX, ...) // we'll use on the host side (SSE, AVX, ...)
#if defined(__CUDACC__) && defined(EIGEN_USE_GPU) #if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 plog<float4>(const float4& a) float4 plog<float4>(const float4& a)
{ {
return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w)); return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 plog<double2>(const double2& a) double2 plog<double2>(const double2& a)
{ {
return make_double2(log(a.x), log(a.y)); return make_double2(log(a.x), log(a.y));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 pexp<float4>(const float4& a) float4 pexp<float4>(const float4& a)
{ {
return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w)); return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 pexp<double2>(const double2& a) double2 pexp<double2>(const double2& a)
{ {
return make_double2(exp(a.x), exp(a.y)); return make_double2(exp(a.x), exp(a.y));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 psqrt<float4>(const float4& a) float4 psqrt<float4>(const float4& a)
{ {
return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w)); return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 psqrt<double2>(const double2& a) double2 psqrt<double2>(const double2& a)
{ {
return make_double2(sqrt(a.x), sqrt(a.y)); return make_double2(sqrt(a.x), sqrt(a.y));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
float4 prsqrt<float4>(const float4& a) float4 prsqrt<float4>(const float4& a)
{ {
return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w)); return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w));
} }
template<> EIGEN_STRONG_INLINE template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
double2 prsqrt<double2>(const double2& a) double2 prsqrt<double2>(const double2& a)
{ {
return make_double2(rsqrt(a.x), rsqrt(a.y)); return make_double2(rsqrt(a.x), rsqrt(a.y));