From 39aef1284fc4a49e772326ade01e5b94c3f061c6 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Fri, 29 May 2026 18:37:38 +0000
Subject: [PATCH] fix: smem size in minimal QK test

---
 tests/unit/test_qk_minimal.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/unit/test_qk_minimal.cu b/tests/unit/test_qk_minimal.cu
index 17e31337..f260d802 100644
--- a/tests/unit/test_qk_minimal.cu
+++ b/tests/unit/test_qk_minimal.cu
@@ -133,7 +133,7 @@ int main() {
     cudaMemcpy(d_q, h_q, T * HD * sizeof(bf16_t), cudaMemcpyHostToDevice);
     cudaMemcpy(d_k, h_k, SK * HD * sizeof(bf16_t), cudaMemcpyHostToDevice);
 
-    int smem = 4 + 128 + TILE_SZ*2 + 4096;
+    int smem = 4 + 128 + 128*16*2 + 128*16*2 + 4096;
     test_qk_minimal_kernel<<<1, 192, smem>>>(d_out, d_q, d_k, T, SK);
     cudaError_t err = cudaDeviceSynchronize();
     if (err != cudaSuccess) { printf("CUDA ERROR: %s\n", cudaGetErrorString(err)); return 1; }