biondizzle/nvfp4-megamoe-kernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

debug: print specific indices for SF layout coordinate decomposition

Browse Source
This commit is contained in:
biondizzle
2026-05-14 15:57:55 +00:00
parent 8ee3f90e44
commit 17e6033ade
1 changed files with 8 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/nvfp4_megamoe_kernel/cutlass_nvfp4_gemm/cutlass_nvfp4_gemm.cu
View File

@@ -131,17 +131,20 @@ __global__ void remap_sf_to_cutlass_kernel(
printf("[remap] flat_rank=%d MN=%d K_sf=%d total=%d\n", R, MN, K_sf, total);
}
// Debug: print first few coordinates
if (dst_idx < 3) {
printf("[remap] idx=%d flat_rank=%d vals=", dst_idx, R);
// Debug: print specific indices to understand the coordinate decomposition
// Print idx 0, 1, 4, 16, 128, 512, 512*448, 128*448
if (dst_idx == 0 || dst_idx == 1 || dst_idx == 4 || dst_idx == 16 ||
dst_idx == 128 || dst_idx == 512 || dst_idx == 512*448 ||
dst_idx == 128*448 || dst_idx == 4 || dst_idx == 3) {
printf("[remap] idx=%d", dst_idx);
if constexpr (R >= 8) {
printf("%d,%d,%d,%d,%d,%d,%d,%d",
printf(" f=%d,%d,%d,%d,%d,%d,%d,%d",
int(cute::get<0>(flat)), int(cute::get<1>(flat)),
int(cute::get<2>(flat)), int(cute::get<3>(flat)),
int(cute::get<4>(flat)), int(cute::get<5>(flat)),
int(cute::get<6>(flat)), int(cute::get<7>(flat)));
}
printf("\n");
printf(" m=%d k=%d\n", m, k_sf);
}
int m = 0, k_sf = 0;