From 75bdcbf728cecb9e3d0afdae8d49a67e2453b138 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 28 May 2026 14:46:21 +0000
Subject: [PATCH] Debug: override P with uniform 1/128

---
 tests/unit/test_fmha_v5.cu | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/tests/unit/test_fmha_v5.cu b/tests/unit/test_fmha_v5.cu
index 92fdfd7c..6bdfbc31 100644
--- a/tests/unit/test_fmha_v5.cu
+++ b/tests/unit/test_fmha_v5.cu
@@ -95,6 +95,8 @@ test_fmha_v5(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("%.4f ", s_vals[j]); printf("\n"); }
+        // DEBUG: Override P with known values
+        if (lane == 0) for (int j=0;j<SK;j++) s_vals[j] = 1.0f / SK;  // Uniform P
         // Write P values to shared memory
         if (lane == 0) for (int j=0;j<SK;j++) s_p_vals[j] = s_vals[j];
     }