RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2025-03-06 03:47:44 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[DAGCombiner] Preserve the exact bit when simplifying SRA to SRL.

Browse Source 
Allows more aggressive folding of ashr/shl pairs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@240788 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Benjamin Kramer 2015-06-26 14:51:49 +00:00
parent 3791d56da6
commit 875007d3c4
2 changed files with 17 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
lib/CodeGen/SelectionDAG/TargetLowering.cpp
View File

@ -771,10 +771,13 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
// If the input sign bit is known to be zero, or if none of the top bits
// are demanded, turn this into an unsigned shift right.
if (KnownZero.intersects(SignBit) || (HighBits & ~NewMask) == HighBits)
return TLO.CombineTo(Op, TLO.DAG.getNode(ISD::SRL, dl, VT,
Op.getOperand(0),
Op.getOperand(1)));
if (KnownZero.intersects(SignBit) || (HighBits & ~NewMask) == HighBits) {
SDNodeFlags Flags;
Flags.setExact(cast<BinaryWithFlagsSDNode>(Op)->Flags.hasExact());
return TLO.CombineTo(Op,
TLO.DAG.getNode(ISD::SRL, dl, VT, Op.getOperand(0),
Op.getOperand(1), &Flags));
}
int Log2 = NewMask.exactLogBase2();
if (Log2 >= 0) {

10
test/CodeGen/X86/shift-combine.ll
View File

@ -37,6 +37,16 @@ define i32* @test_exact2(i32 %a, i32 %b, i32* %x) {
ret i32* %gep
}
define i32* @test_exact3(i32 %a, i32 %b, i32* %x) {
; CHECK-LABEL: test_exact3:
; CHECK-NOT: sarl
%sub = sub i32 %b, %a
%shr = ashr exact i32 %sub, 2
%gep = getelementptr inbounds i32, i32* %x, i32 %shr
ret i32* %gep
}
define i32* @test_exact4(i32 %a, i32 %b, i32* %x) {
; CHECK-LABEL: test_exact4:
; CHECK: shrl %