portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-10-30 19:44:32 +00:00
Code Releases Activity

GPU work allocation improved

Browse Source
This commit is contained in:
Peter Boyle
2018-03-22 18:04:24 -04:00
parent 55be842d23
commit 299d119013
1 changed files with 12 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
lib/threads/Pragmas.h
View File

@@ -77,26 +77,33 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
////////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////////
#ifdef GRID_NVCC #ifdef GRID_NVCC
constexpr uint32_t gpu_threads = 8;
template<typename lambda> __global__ template<typename lambda> __global__
void LambdaApply(uint64_t Num, lambda Lambda) void LambdaApply(uint64_t Num, lambda Lambda)
{ {
uint64_t ss = blockIdx.x; uint64_t ss = blockIdx.x*blockDim.x + threadIdx.x;
if ( ss < Num ) { if ( ss < Num ) {
Lambda(ss); Lambda(ss);
} }
} }
#define accelerator __host__ __device__ #define accelerator __host__ __device__
#define accelerator_inline __host__ __device__ inline #define accelerator_inline __host__ __device__ inline
#define accelerator_loop( iterator, range, ... ) \ #define accelerator_loop( iterator, range, ... ) \
typedef decltype(range.begin()) Iterator; \ typedef decltype(range.begin()) Iterator; \
auto lambda = [=] accelerator (Iterator iterator) mutable { \ auto lambda = [=] accelerator (Iterator iterator) mutable { \
__VA_ARGS__; \ __VA_ARGS__; \
}; \ }; \
Iterator num = range.end(); \ Iterator num = range.end(); \
LambdaApply<<<num,1>>>(num,lambda); \ Iterator num_block = (num + gpu_threads - 1)/gpu_threads; \
cudaDeviceSynchronize(); LambdaApply<<<num_block,gpu_threads>>>(num,lambda); \
cudaError err = cudaGetLastError(); \
cudaDeviceSynchronize(); \
if ( cudaSuccess != err ) { \
printf("Cuda error %s\n",cudaGetErrorString( err )); \
exit(0); \
} \
#define cpu_loop( iterator, range, ... ) thread_loop( (auto iterator = range.begin();iterator<range.end();iterator++), { __VA_ARGS__ }); #define cpu_loop( iterator, range, ... ) thread_loop( (auto iterator = range.begin();iterator<range.end();iterator++), { __VA_ARGS__ });