diff --git a/tests/unit/test_umma_qk.cu b/tests/unit/test_umma_qk.cu index b7dd55bb..cdd0175e 100644 --- a/tests/unit/test_umma_qk.cu +++ b/tests/unit/test_umma_qk.cu @@ -80,14 +80,12 @@ test_umma_qk_hd16( } __syncthreads(); - // Read S from TMEM + // Read S from TMEM — only read 1 column if (wid == 0) { - for (int col = 0; col < 32; col++) { - uint32_t u0, u1, u2, u3; - tmem_load(tmem_base + col, u0, u1, u2, u3); - if (lane == 0) { - s_out[col] = u32_to_f32(u0); - } + uint32_t u0, u1, u2, u3; + tmem_load(tmem_base, u0, u1, u2, u3); + if (lane == 0) { + s_out[0] = u32_to_f32(u0); // S[0, 0] } } __syncthreads();