NASP/cuda-samples
50
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2025-08-23 23:35:31 +08:00
Code Issues Packages Projects Releases Wiki Activity

Bug 5412815: Fix the issue of cudaTensorCoreGemm.cu

Browse Source
This commit is contained in:
shawnz 2025-07-25 15:16:12 +08:00
parent 4a631c9fd6
commit b38ed29c95
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Samples/3_CUDA_Features/cudaTensorCoreGemm/cudaTensorCoreGemm.cu
View File

@ -224,7 +224,7 @@ __global__ void compute_gemm(const half *A, const half *B, const float *C, float
// there's no such tile, all warps in this CTA exit. // there's no such tile, all warps in this CTA exit.
for (unsigned int block_pos = blockIdx.x;; block_pos += gridDim.x) { for (unsigned int block_pos = blockIdx.x;; block_pos += gridDim.x) {
const unsigned int block_tile_i = ((block_pos * BLOCK_ROW_TILES) / N_TILES) * (BLOCK_COL_TILES); const unsigned int block_tile_i = ((block_pos * BLOCK_ROW_TILES) / N_TILES) * (BLOCK_COL_TILES);
const unsigned int block_tile_j = (block_pos * BLOCK_COL_TILES) % N_TILES; const unsigned int block_tile_j = (block_pos * BLOCK_ROW_TILES) % N_TILES;
// Stop when there are no more D matrix tiles to compute in this CTA. // Stop when there are no more D matrix tiles to compute in this CTA.
if (block_tile_i >= M_TILES) { if (block_tile_i >= M_TILES) {