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

Fix nvcc goto-bypasses-init: move var decls before goto targets

Browse Source
This commit is contained in:
biondizzle
2026-05-28 20:04:59 +00:00
parent 2b09d4f2ef
commit bf4dfd131b
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/unit/test_fmha_6warp_multirow.cu
View File

@@ -70,6 +70,8 @@ static int test_single_T(int T, int n_h = 1, int batch = 1) {
printf("\n=== Test T=%d, n_h=%d, batch=%d, HD=%d, SK=%d (%s) ===\n", T, n_h, batch, HD, SK, mode);
const float SCALE = 1.0f / sqrtf((float)HD);
int pass = 1;
int checked = 0, failed = 0;
float min_cos = 1.0f;
int total_heads = batch * n_h;
@@ -142,10 +144,6 @@ static int test_single_T(int T, int n_h = 1, int batch = 1) {
pass = 0; goto cleanup;
}
// Verify each head (declared before goto targets)
int checked = 0, failed = 0;
float min_cos = 1.0f;
cudaMemcpy(h_o, d_o, total_heads * T * HD * sizeof(bf16_t), cudaMemcpyDeviceToHost);
cudaMemcpy(h_lse, d_lse, total_heads * T * sizeof(float), cudaMemcpyDeviceToHost);
for (int b = 0; b < batch; b++) {