forked from OSchip/llvm-project
Make use of the exact bit when optimizing '(X >>exact 3) << 1' to eliminate the
'and' that would zero out the trailing bits, and to produce an exact shift ourselves. llvm-svn: 147391
This commit is contained in:
parent
1c65a21ec4
commit
b59008c694
|
@ -536,12 +536,11 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
|
|||
if (ShiftAmt1 == 0) return 0; // Will be simplified in the future.
|
||||
Value *X = ShiftOp->getOperand(0);
|
||||
|
||||
uint32_t AmtSum = ShiftAmt1+ShiftAmt2; // Fold into one big shift.
|
||||
|
||||
IntegerType *Ty = cast<IntegerType>(I.getType());
|
||||
|
||||
// Check for (X << c1) << c2 and (X >> c1) >> c2
|
||||
if (I.getOpcode() == ShiftOp->getOpcode()) {
|
||||
uint32_t AmtSum = ShiftAmt1+ShiftAmt2; // Fold into one big shift.
|
||||
// If this is oversized composite shift, then unsigned shifts get 0, ashr
|
||||
// saturates.
|
||||
if (AmtSum >= TypeBits) {
|
||||
|
@ -603,9 +602,16 @@ Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
|
|||
// (X >>? C1) << C2 --> X >>? (C1-C2) & (-1 << C2)
|
||||
if (I.getOpcode() == Instruction::Shl &&
|
||||
ShiftOp->getOpcode() != Instruction::Shl) {
|
||||
Value *Shift = Builder->CreateBinOp(ShiftOp->getOpcode(), X,
|
||||
ConstantInt::get(Ty, ShiftDiff));
|
||||
|
||||
ConstantInt *ShiftDiffCst = ConstantInt::get(Ty, ShiftDiff);
|
||||
if (ShiftOp->isExact()) {
|
||||
// (X >>?exact C1) << C2 --> X >>?exact (C1-C2)
|
||||
BinaryOperator *NewShr = BinaryOperator::Create(ShiftOp->getOpcode(),
|
||||
X, ShiftDiffCst);
|
||||
NewShr->setIsExact(true);
|
||||
return NewShr;
|
||||
}
|
||||
Value *Shift = Builder->CreateBinOp(ShiftOp->getOpcode(),
|
||||
X, ShiftDiffCst);
|
||||
APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt2));
|
||||
return BinaryOperator::CreateAnd(Shift,
|
||||
ConstantInt::get(I.getContext(),Mask));
|
||||
|
|
|
@ -542,3 +542,21 @@ define i32 @test45(i32 %a) nounwind {
|
|||
; CHECK-NEXT: %y = lshr i32 %a, 5
|
||||
; CHECK-NEXT: ret i32 %y
|
||||
}
|
||||
|
||||
define i32 @test46(i32 %a) {
|
||||
%y = ashr exact i32 %a, 3
|
||||
%z = shl i32 %y, 1
|
||||
ret i32 %z
|
||||
; CHECK: @test46
|
||||
; CHECK-NEXT: %z = ashr exact i32 %a, 2
|
||||
; CHECK-NEXT: ret i32 %z
|
||||
}
|
||||
|
||||
define i32 @test47(i32 %a) {
|
||||
%y = lshr exact i32 %a, 3
|
||||
%z = shl i32 %y, 1
|
||||
ret i32 %z
|
||||
; CHECK: @test47
|
||||
; CHECK-NEXT: %z = lshr exact i32 %a, 2
|
||||
; CHECK-NEXT: ret i32 %z
|
||||
}
|
||||
|
|
Loading…
Reference in New Issue