From 4506fcb3c2763f2a5468587d2276082172b5b336 Mon Sep 17 00:00:00 2001
From: Dan Gohman <gohman@apple.com>
Date: Mon, 22 Feb 2010 03:59:54 +0000
Subject: [PATCH] When emitting an instruction which depends on both a
 post-incremented induction variable value and a loop-variant value, don't
 force the insert position to be at the post-increment position, because it
 may not be dominated by the loop-variant value. This fixes a use-before-def
 problem noticed on PPC.

llvm-svn: 96774
---
 .../Transforms/Scalar/LoopStrengthReduce.cpp  |  6 ++--
 llvm/test/CodeGen/PowerPC/lsr-postinc-pos.ll  | 32 +++++++++++++++++++
 2 files changed, 36 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/CodeGen/PowerPC/lsr-postinc-pos.ll

diff --git a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index cf3d702d633f..a4bc0170edb3 100644
--- a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -2866,8 +2866,10 @@ Value *LSRInstance::Expand(const LSRFixup &LF,
       if (AR->getLoop() == LF.PostIncLoop) {
         Reg = SE.getAddExpr(Reg, AR->getStepRecurrence(SE));
         // If the user is inside the loop, insert the code after the increment
-        // so that it is dominated by its operand.
-        if (L->contains(LF.UserInst))
+        // so that it is dominated by its operand. If the original insert point
+        // was already dominated by the increment, keep it, because there may
+        // be loop-variant operands that need to be respected also.
+        if (L->contains(LF.UserInst) && !DT.dominates(IVIncInsertPos, IP))
           IP = IVIncInsertPos;
         break;
       }
diff --git a/llvm/test/CodeGen/PowerPC/lsr-postinc-pos.ll b/llvm/test/CodeGen/PowerPC/lsr-postinc-pos.ll
new file mode 100644
index 000000000000..f441e42da2f1
--- /dev/null
+++ b/llvm/test/CodeGen/PowerPC/lsr-postinc-pos.ll
@@ -0,0 +1,32 @@
+; RUN: llc < %s -print-lsr-output |& FileCheck %s
+
+; The icmp is a post-inc use, and the increment is in %bb11, but the
+; scevgep needs to be inserted in %bb so that it is dominated by %t.
+
+; CHECK: %t = load i8** undef
+; CHECK: %scevgep = getelementptr i8* %t, i32 %lsr.iv.next
+; CHECK: %c1 = icmp ult i8* %scevgep, undef
+
+target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128-n32"
+target triple = "powerpc-apple-darwin9"
+
+define void @foo() nounwind {
+entry:
+  br label %bb11
+
+bb11:
+  %i = phi i32 [ 0, %entry ], [ %i.next, %bb ] ; <i32> [#uses=3]
+  %ii = shl i32 %i, 2                       ; <i32> [#uses=1]
+  %c0 = icmp eq i32 %i, undef                ; <i1> [#uses=1]
+  br i1 %c0, label %bb13, label %bb
+
+bb:
+  %t = load i8** undef, align 16                ; <i8*> [#uses=1]
+  %p = getelementptr i8* %t, i32 %ii ; <i8*> [#uses=1]
+  %c1 = icmp ult i8* %p, undef          ; <i1> [#uses=1]
+  %i.next = add i32 %i, 1                        ; <i32> [#uses=1]
+  br i1 %c1, label %bb11, label %bb13
+
+bb13:
+  unreachable
+}