RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-18 19:28:16 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

InstCombine: Stop two transforms dueling

Browse Source 
InstCombineMulDivRem has:
// Canonicalize (X+C1)*CI -> X*CI+C1*CI.

InstCombineAddSub has:
// W*X + Y*Z --> W * (X+Z)  iff W == Y

These two transforms could fight with each other if C1*CI would not fold
away to something simpler than a ConstantExpr mul.

The InstCombineMulDivRem transform only acted on ConstantInts until
r199602 when it was changed to operate on all Constants in order to
let it fire on ConstantVectors.

To fix this, make this transform more careful by checking to see if we
actually folded away C1*CI.

This fixes PR20079.

llvm-svn: 211258
This commit is contained in:
David Majnemer 2014-06-19 07:14:33 +00:00
parent c246cb10ff
commit a95def3a31
2 changed files with 18 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
lib/Transforms/InstCombine/InstCombineMulDivRem.cpp
View File

@ -203,8 +203,11 @@ Instruction *InstCombiner::visitMul(BinaryOperator &I) {
Value *X;
Constant *C1;
if (match(Op0, m_OneUse(m_Add(m_Value(X), m_Constant(C1))))) {
Value *Add = Builder->CreateMul(X, Op1);
return BinaryOperator::CreateAdd(Add, Builder->CreateMul(C1, Op1));
Value *Mul = Builder->CreateMul(C1, Op1);
// Only go forward with the transform if C1*CI simplifies to a tidier
// constant.
if (!match(Mul, m_Mul(m_Value(), m_Value())))
return BinaryOperator::CreateAdd(Builder->CreateMul(X, Op1), Mul);
}
}
}

13
test/Transforms/InstCombine/pr20079.ll Normal file
View File

@ -0,0 +1,13 @@
; RUN: opt -S -instcombine < %s | FileCheck %s
@b = internal global [1 x i32] zeroinitializer, align 4
@c = internal global i32 0, align 4
; CHECK-LABEL: @fn1
; CHECK: [[ADD:%.*]] = add i32 %a, -1
; CHECK-NEXT: [[AND:%.*]] = and i32 [[ADD]], sub (i32 0, i32 zext (i1 icmp eq (i32* getelementptr inbounds ([1 x i32]* @b, i64 0, i64 0), i32* @c) to i32))
; CHECK-NEXT: ret i32 [[AND]]
define i32 @fn1(i32 %a) {
%xor = add i32 %a, -1
%mul = mul nsw i32 %xor, zext (i1 icmp eq (i32* getelementptr inbounds ([1 x i32]* @b, i64 0, i64 0), i32* @c) to i32)
ret i32 %mul
}