DeepSeek/DeepGEMM
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepGEMM synced 2025-05-05 09:54:21 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

OOB bugs fixed

Browse Source
This commit is contained in:
Chenggang Zhao 2025-04-11 11:00:47 +08:00
parent 99eb6ec563
commit b0d64817a7
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
deep_gemm/include/deep_gemm/fp8_gemm.cuh
View File

@ -375,7 +375,8 @@ fp8_gemm_kernel(__nv_bfloat16* gmem_d, float* scales_b, int* grouped_layout,
uint64_t gmem_m_offset = scheduler.get_global_idx(shape_m, BLOCK_M, m_block_idx);
auto smem_ptr = smem_d + (m_offset + warp_idx * 16 + lane_idx) * (BLOCK_N + BLOCK_N_PADDING);
auto gmem_ptr = gmem_d + (gmem_m_offset + m_offset + warp_idx * 16 + lane_idx) * SHAPE_N + n_block_idx * BLOCK_N;
cute::SM90_BULK_COPY_S2G::copy(smem_ptr, gmem_ptr, BLOCK_N * sizeof(nv_bfloat16));
auto num_valid_cols = (n_block_idx == ceil_div(SHAPE_N, BLOCK_N) - 1) ? (SHAPE_N - n_block_idx * BLOCK_N) : BLOCK_N;
cute::SM90_BULK_COPY_S2G::copy(smem_ptr, gmem_ptr, num_valid_cols * sizeof(nv_bfloat16));
}
__syncwarp();
}