From 6d35135d82710d8388bd0de38b916a9d38a24d74 Mon Sep 17 00:00:00 2001
From: Carlo Bertolli <Carlo.Bertolli@amd.com>
Date: Tue, 28 Oct 2025 15:46:37 -0500
Subject: [PATCH] [AMDGPU] Enable expensive unroll trip count. This patch
 enables unrolling innermost loop of Pytorch reduce function in
 Normalization.cuh.

---
 llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index 03d16fdd54c42..100c66f77c557 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -271,6 +271,8 @@ void AMDGPUTTIImpl::getUnrollingPreferences(
     if (L->isInnermost() && BB->size() < UnrollMaxBlockToAnalyze)
       UP.MaxIterationsCountToAnalyze = 32;
   }
+
+  UP.AllowExpensiveTripCount = true;
 }
 
 void AMDGPUTTIImpl::getPeelingPreferences(Loop *L, ScalarEvolution &SE,