From 1be9b9f81694617edd85e28f03cab07b3864f7e1 Mon Sep 17 00:00:00 2001
From: Changpeng Fang <changpeng.fang@gmail.com>
Date: Thu, 9 Mar 2017 00:07:00 +0000
Subject: [PATCH] AMDGPU/SI: Disable unrolling in the loop vectorizer if the
 loop is not vectorized.

Reviewers:
  arsenm

Differential Revision:
  http://reviews.llvm.org/D30719

llvm-svn: 297328
---
 .../AMDGPU/AMDGPUTargetTransformInfo.cpp      |  4 +++
 .../LoopVectorize/AMDGPU/lit.local.cfg        |  2 ++
 .../AMDGPU/unroll-in-loop-vectorizer.ll       | 28 +++++++++++++++++++
 3 files changed, 34 insertions(+)
 create mode 100644 llvm/test/Transforms/LoopVectorize/AMDGPU/lit.local.cfg
 create mode 100644 llvm/test/Transforms/LoopVectorize/AMDGPU/unroll-in-loop-vectorizer.ll

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index a780a76c86a60..4a6d12bd883e7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -155,6 +155,10 @@ bool AMDGPUTTIImpl::isLegalToVectorizeStoreChain(unsigned ChainSizeInBytes,
 }
 
 unsigned AMDGPUTTIImpl::getMaxInterleaveFactor(unsigned VF) {
+  // Disable unrolling if the loop is not vectorized.
+  if (VF == 1)
+    return 1;
+
   // Semi-arbitrary large amount.
   return 64;
 }
diff --git a/llvm/test/Transforms/LoopVectorize/AMDGPU/lit.local.cfg b/llvm/test/Transforms/LoopVectorize/AMDGPU/lit.local.cfg
new file mode 100644
index 0000000000000..2a665f06be72e
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/AMDGPU/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'AMDGPU' in config.root.targets:
+    config.unsupported = True
diff --git a/llvm/test/Transforms/LoopVectorize/AMDGPU/unroll-in-loop-vectorizer.ll b/llvm/test/Transforms/LoopVectorize/AMDGPU/unroll-in-loop-vectorizer.ll
new file mode 100644
index 0000000000000..85ba95cab1603
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/AMDGPU/unroll-in-loop-vectorizer.ll
@@ -0,0 +1,28 @@
+; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=fiji -loop-vectorize < %s | FileCheck %s
+
+
+; For AMDGPU, loop unroll in loop vectorizer is disabled when VF==1.
+;
+; CHECK-LABEL: @small_loop(
+; CHECK: store i32
+; CHECK-NOT: store i32
+; CHECK: ret
+define void @small_loop(i32* nocapture %inArray, i32 %size) nounwind {
+entry:
+  %0 = icmp sgt i32 %size, 0
+  br i1 %0, label %loop, label %exit
+
+loop:                                          ; preds = %entry, %loop
+  %iv = phi i32 [ %iv1, %loop ], [ 0, %entry ]
+  %1 = getelementptr inbounds i32, i32* %inArray, i32 %iv
+  %2 = load i32, i32* %1, align 4
+  %3 = add nsw i32 %2, 6
+  store i32 %3, i32* %1, align 4
+  %iv1 = add i32 %iv, 1
+;  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %cond = icmp eq i32 %iv1, %size
+  br i1 %cond, label %exit, label %loop
+
+exit:                                         ; preds = %loop, %entry
+  ret void
+}