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

Fixing a CUDA / P100 regression introduced by PR 181

Browse Source 
PR 181 ( https://gitlab.com/libeigen/eigen/-/merge_requests/181 ) adds `__launch_bounds__(1024)` attribute to GPU kernels, that did not have that attribute explicitly specified.

That PR seems to cause regressions on the CUDA platform. This PR/commit makes the changes in PR 181, to be applicable for HIP only
This commit is contained in:
Deven Desai
2020-08-19 20:06:39 +00:00
parent c060114a25
commit 603e213d13
6 changed files with 40 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
unsupported/Eigen/CXX11/src/Tensor/TensorReduction.h
View File

@@ -15,7 +15,7 @@
// so we'll use a macro to make clang happy.
#ifndef KERNEL_FRIEND
#if defined(__clang__) && (defined(__CUDA__) || defined(__HIP__))
#define KERNEL_FRIEND friend __global__ __launch_bounds__(1024)
#define KERNEL_FRIEND friend __global__ EIGEN_HIP_LAUNCH_BOUNDS_1024
#else
#define KERNEL_FRIEND friend
#endif
@@ -427,24 +427,24 @@ struct GenericReducer {
#if defined(EIGEN_USE_GPU) && (defined(EIGEN_GPUCC))
template <int B, int N, typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void FullReductionKernel(R, const S, I_, typename S::CoeffReturnType*, unsigned int*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void FullReductionKernel(R, const S, I_, typename S::CoeffReturnType*, unsigned int*);
#if defined(EIGEN_HAS_GPU_FP16)
template <typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void ReductionInitFullReduxKernelHalfFloat(R, const S, I_, internal::packet_traits<half>::type*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void ReductionInitFullReduxKernelHalfFloat(R, const S, I_, internal::packet_traits<half>::type*);
template <int B, int N, typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void FullReductionKernelHalfFloat(R, const S, I_, half*, internal::packet_traits<half>::type*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void FullReductionKernelHalfFloat(R, const S, I_, half*, internal::packet_traits<half>::type*);
template <int NPT, typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void InnerReductionKernelHalfFloat(R, const S, I_, I_, half*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void InnerReductionKernelHalfFloat(R, const S, I_, I_, half*);
#endif
template <int NPT, typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void InnerReductionKernel(R, const S, I_, I_, typename S::CoeffReturnType*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void InnerReductionKernel(R, const S, I_, I_, typename S::CoeffReturnType*);
template <int NPT, typename S, typename R, typename I_>
__global__ __launch_bounds__(1024) void OuterReductionKernel(R, const S, I_, I_, typename S::CoeffReturnType*);
__global__ EIGEN_HIP_LAUNCH_BOUNDS_1024 void OuterReductionKernel(R, const S, I_, I_, typename S::CoeffReturnType*);
#endif
/**