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

auto: pre-test commit

Browse Source
This commit is contained in:
biondizzle
2026-05-28 15:46:53 +00:00
parent 1b86860c19
commit 9583cbc67a
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/unit/test_fmha_hd64_debug.cu
View File

@@ -242,8 +242,8 @@ int main() {
printf("Cosine: %.8f\n", cs_reg);
// Compare MMA PV vs reference
printf("\n--- MMA PV vs FP32 reference ---\n");
for (int d=0;d<8;d++) printf(" MMA[%d]=%.6f ref[%d]=%.6f ratio=%.4f\n",
printf("\n--- MMA PV vs FP32 reference (all 64 elements) ---\n");
for (int d=0;d<HD;d++) printf(" MMA[%2d]=%10.6f ref[%2d]=%10.6f ratio=%.4f\n",
d, bf16_to_f32_host(h_o_mma[d]), d, h_o_ref[d],
fabsf(h_o_ref[d])>1e-6f ? bf16_to_f32_host(h_o_mma[d])/h_o_ref[d] : 0);
float cs_mma = 0, na_mma = 0, nb_mma = 0;