From f244c4fdd232a48bd68f73a0a6c08477d391ba8e Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 28 May 2026 12:56:39 +0000
Subject: [PATCH] test: single-thread MMA (tid==0) for Layout D

---
 tests/unit/test_umma_qk_hd64.cu | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/unit/test_umma_qk_hd64.cu b/tests/unit/test_umma_qk_hd64.cu
index 8b3663e0..54fcddc2 100644
--- a/tests/unit/test_umma_qk_hd64.cu
+++ b/tests/unit/test_umma_qk_hd64.cu
@@ -107,8 +107,8 @@ test_umma_hd64(const bf16_t* __restrict__ q, const bf16_t* __restrict__ k,
         uint64_t dq = make_umma_desc_kmajor_none(q_addr, BLOCK_MN);
         uint64_t dk = make_umma_desc_kmajor_none(k_addr, BLOCK_MN);
 
-        // 4 warp leaders call MMA (same as working HD=16 test)
-        if (lane == 0) {
+        // Single thread calls MMA (gau-nernst's elect_one pattern)
+        if (tid == 0) {
             umma_ss_f16(tb, dq, dk, idesc, kt > 0);
         }
         asm volatile("tcgen05.fence::after_thread_sync;" ::: "memory");