From 8059ed15ad182d692a1a4db047c61ab0070a5aa3 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 28 May 2026 10:08:35 +0000
Subject: [PATCH] test: explicitly zero padding between Q and K

---
 tests/unit/test_umma_qk.cu | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/tests/unit/test_umma_qk.cu b/tests/unit/test_umma_qk.cu
index 239ff9fe..1c59d32f 100644
--- a/tests/unit/test_umma_qk.cu
+++ b/tests/unit/test_umma_qk.cu
@@ -43,6 +43,10 @@ test_umma_qk_hd16(const bf16_t* q, const bf16_t* k,
     // Load Q and K
     write_q_to_smem<16>(sQ, q);
     write_k_to_smem<128, 16>(sK, k);
+    // Zero the padding between Q and K
+    bf16_t* sQ_pad = sQ + 128 * 16;  // After Q data
+    bf16_t* sK_end = sK + 128 * 16;  // After K data
+    for (int i = tid; i < 4096; i += NTHREADS) sQ_pad[i] = 0;  // 8KB padding
     __syncthreads();
 
     // Descriptors