From 8eb569e31c1bea796452a377c93f2a319f1eb9dd Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Thu, 21 May 2026 19:10:47 +0000
Subject: [PATCH] BREAKTHROUGH: Found the real bug!

Each epilogue thread owns only 32 of 128 columns per row (4 warps partition
the 128 columns). The scalar row_max is max of 32 elements (not global row max).
The scalar row_sum is sum of 32 P values (not all 128).

Evidence: kernel row_sum=5.5 for row 0, correct=29.22, ratio=5.3x
(128/32 = 4, but the P values arent uniformly distributed so not exact).

Fix: need warp reduce across the 4 threads that own the same rows columns.
CUTLASS FMHA uses utils.sm100 warp-reduce helpers for this exact reason.