From 20f3ccd992f025167b1f3943136dc01eb27d5be2 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Sat, 30 May 2026 07:02:41 +0000
Subject: [PATCH] D1.5 complete: HD=512 support via hd_chunk tiling with native
 TMEM columns

---
 tests/unit/test_fmha_6warp_tma_multirow_multitile.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu b/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu
index f66417af..fa19861f 100644
--- a/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu
+++ b/tests/unit/test_fmha_6warp_tma_multirow_multitile.cu
@@ -180,7 +180,7 @@ static int test_single(int T, int s_k, int n_h = 1, int batch = 1) {
             o_ref, nullptr, HD, T, s_k, SCALE);
 
         float cs = 0, na = 0, nb = 0;
-        int check_hd = HD;  // Check full HD (all chunks)
+        int check_hd = HD;
         for (int t = 0; t < T; t++) {
             for (int d = 0; d < check_hd; d++) {
                 float a = bf16_to_f32_host(h_o[h * MAX_T * HD + t * HD + d]);