Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix device query

Browse Source
This commit is contained in:
amdsc21 2023-04-19 19:53:26 +02:00
parent f645cf51c1
commit 65d83e288f
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/tree/gpu_hist/histogram.cu
View File

@ -306,12 +306,14 @@ void BuildGradientHistogram(CUDAContext const* ctx, EllpackDeviceAccessor const&
dh::safe_cuda(cudaDeviceGetAttribute(&n_mps, cudaDevAttrMultiProcessorCount, device)); dh::safe_cuda(cudaDeviceGetAttribute(&n_mps, cudaDevAttrMultiProcessorCount, device));
int n_blocks_per_mp = 0; int n_blocks_per_mp = 0;
dh::safe_cuda(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&n_blocks_per_mp, kernel, dh::safe_cuda(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&n_blocks_per_mp, kernel,
kBlockThreads, smem_size));
#elif defined(XGBOOST_USE_HIP) #elif defined(XGBOOST_USE_HIP)
dh::safe_cuda(hipDeviceGetAttribute(&n_mps, hipDeviceAttributeMultiprocessorCount, device)); dh::safe_cuda(hipDeviceGetAttribute(&n_mps, hipDeviceAttributeMultiprocessorCount, device));
int n_blocks_per_mp = 0; int n_blocks_per_mp = 0;
dh::safe_cuda(hipOccupancyMaxActiveBlocksPerMultiprocessor(&n_blocks_per_mp, kernel, dh::safe_cuda(hipOccupancyMaxActiveBlocksPerMultiprocessor(&n_blocks_per_mp, kernel,
#endif
kBlockThreads, smem_size)); kBlockThreads, smem_size));
#endif
// This gives the number of blocks to keep the device occupied // This gives the number of blocks to keep the device occupied
// Use this as the maximum number of blocks // Use this as the maximum number of blocks
unsigned grid_size = n_blocks_per_mp * n_mps; unsigned grid_size = n_blocks_per_mp * n_mps;