diff --git a/tests/unit/test_umma_qk.cu b/tests/unit/test_umma_qk.cu index f32ed781..48ad1cba 100644 --- a/tests/unit/test_umma_qk.cu +++ b/tests/unit/test_umma_qk.cu @@ -78,9 +78,11 @@ test_umma_qk_hd16( } __syncthreads(); - // Zero TMEM — only zero 1 column to test + // Zero TMEM if (wid == 0) { - tmem_store(tmem_base, 0, 0, 0, 0); + for (int col = 0; col < 128; col++) { + tmem_store(tmem_base + col, 0, 0, 0, 0); + } tmem_fence_store(); } __syncthreads();