From 6ea7356fdde6708dbec041c51cd48c4f11e969ee Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 28 May 2026 15:07:55 +0000
Subject: [PATCH] Debug: print P values for HD=64

---
 tests/unit/test_fmha_hd64_smem_p.cu | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/unit/test_fmha_hd64_smem_p.cu b/tests/unit/test_fmha_hd64_smem_p.cu
index efed944d..06ccb152 100644
--- a/tests/unit/test_fmha_hd64_smem_p.cu
+++ b/tests/unit/test_fmha_hd64_smem_p.cu
@@ -126,6 +126,7 @@ test_fmha_hd64_smem_p(const bf16_t* q, const bf16_t* k, const bf16_t* v,
         }
         row_sum = wsum(row_sum);
         if (lane == 0) for (int j=0;j<SK;j++) s_vals[j] /= row_sum;
+        if (lane == 0) { printf("P[0..7]: "); for(int j=0;j<8;j++) printf("%.6f ", s_vals[j]); printf("\n"); }
         if (lane == 0) for (int j=0;j<SK;j++) s_p_vals[j] = s_vals[j];
     }
     __syncthreads();