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

test: add max_abs_diff to FMHA test output

Browse Source
This commit is contained in:
biondizzle
2026-05-28 07:44:45 +00:00
parent 579dd061cd
commit 53c676c8a6
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/unit/test_fmha_sm100_standalone.cu
View File

@@ -101,8 +101,10 @@ int test_kernel(const char* name, int HD_val, int sk, float scale,
for (int i = 0; i < B*H*HD_val; i++) ho_gpu[i] = bf16_to_f32_cpu(hob[i]);
float cos = cosine_sim(ho_gpu, ho_ref, B*H*HD_val);
float max_diff = 0;
for(int i=0;i<B*H*HD_val;i++) max_diff = fmaxf(max_diff, fabsf(ho_gpu[i]-ho_ref[i]));
int pass = cos > 0.999f;
printf(" %s hd=%d s_k=%d: cos %.6f %s\n", name, HD_val, sk, cos, pass ? "✅" : "❌");
printf(" %s hd=%d s_k=%d: cos %.6f max_diff %.6f %s\n", name, HD_val, sk, cos, max_diff, pass ? "✅" : "❌");
if (!pass) {
printf(" GPU[:4] = %.6f %.6f %.6f %.6f\n", ho_gpu[0], ho_gpu[1], ho_gpu[2], ho_gpu[3]);