From 1dca8d8cfa672bc71fffcf630d29f8f8f099dfb7 Mon Sep 17 00:00:00 2001 From: biondizzle Date: Sat, 30 May 2026 04:46:11 +0000 Subject: [PATCH] debug: unbuffered stdout --- tests/unit/test_fmha_6warp_tma_multirow_multitile.cu | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu b/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu index ee97af41..20b6a6ec 100644 --- a/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu +++ b/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu @@ -153,6 +153,7 @@ static int test_single(int T, int s_k, int n_h = 1, int batch = 1) { dim3 grid(1, n_h, batch); printf(" Launching kernel: grid=(%d,%d,%d) smem=%zu\n", grid.x, grid.y, grid.z, smem); fflush(stdout); fmha_6warp_tma_multirow_multitile_kernel<<>>(params); + printf(" Kernel launched.\n"); cudaError_t lerr = cudaGetLastError(); if (lerr != cudaSuccess) { @@ -200,8 +201,8 @@ static int test_single(int T, int s_k, int n_h = 1, int batch = 1) { } int main() { + setbuf(stdout, NULL); printf("START: test_fmha_6warp_tma_multirow_multitile HD=%d\n", HD); - fflush(stdout); int total_fail = 0; // Just the most basic test first