clean: remove debug prints, multirow kernel complete with multi-tile KV merge

2026-05-28 23:57:31 +00:00
parent c65baabcc9
commit d1c1eaeddc
1 changed files with 0 additions and 7 deletions
--- a/tests/unit/test_fmha_6warp_multirow.cu
+++ b/tests/unit/test_fmha_6warp_multirow.cu
@@ -230,13 +230,6 @@ static int test_multitile_merge(int T) {
        free(h_o_tile);
    }

-    // Debug: print first LSE values per tile
-    for (int tile = 0; tile < N_TILES; tile++) {
-        printf("  tile %d lse[0]=%.6f", tile, lse_per_tile[tile * T]);
-        if (T > 1) printf(" lse[1]=%.6f", lse_per_tile[tile * T + 1]);
-        printf("\n");
-    }
-
    // Python KV merge with normalized O + LSE:
    // O = Σ exp(lse_i - L) * O_i_norm / Σ exp(lse_i - L)
    // where L = max(lse_i) for numerical stability