6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-13 04:30:23 +00:00
Code Issues Projects Releases Wiki Activity

InstCombine: Detect when llvm.umul.with.overflow always overflows

Browse Source 
We know overflow always occurs if both ~LHSKnownZero * ~RHSKnownZero
and LHSKnownOne * RHSKnownOne overflow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225077 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
David Majnemer 2015-01-02 07:29:47 +00:00
parent 25e8e79fab
commit 5e9c6212a8
3 changed files with 31 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
lib/Analysis/ValueTracking.cpp
View File

@ -2686,10 +2686,11 @@ OverflowResult llvm::computeOverflowForUnsignedMul(Value *LHS, Value *RHS,
// Ref: "Hacker's Delight" by Henry Warren // Ref: "Hacker's Delight" by Henry Warren
unsigned BitWidth = LHS->getType()->getScalarSizeInBits(); unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
APInt LHSKnownZero(BitWidth, 0); APInt LHSKnownZero(BitWidth, 0);
APInt LHSKnownOne(BitWidth, 0);
APInt RHSKnownZero(BitWidth, 0); APInt RHSKnownZero(BitWidth, 0);
APInt TmpKnownOne(BitWidth, 0); APInt RHSKnownOne(BitWidth, 0);
computeKnownBits(LHS, LHSKnownZero, TmpKnownOne, DL, /*Depth=*/0, AT, CxtI, DT); computeKnownBits(LHS, LHSKnownZero, LHSKnownOne, DL, /*Depth=*/0, AT, CxtI, DT);
computeKnownBits(RHS, RHSKnownZero, TmpKnownOne, DL, /*Depth=*/0, AT, CxtI, DT); computeKnownBits(RHS, RHSKnownZero, RHSKnownOne, DL, /*Depth=*/0, AT, CxtI, DT);
// Note that underestimating the number of zero bits gives a more // Note that underestimating the number of zero bits gives a more
// conservative answer. // conservative answer.
unsigned ZeroBits = LHSKnownZero.countLeadingOnes() + unsigned ZeroBits = LHSKnownZero.countLeadingOnes() +
@ -2705,9 +2706,17 @@ OverflowResult llvm::computeOverflowForUnsignedMul(Value *LHS, Value *RHS,
// We know the multiply operation doesn't overflow if the maximum values for // We know the multiply operation doesn't overflow if the maximum values for
// each operand will not overflow after we multiply them together. // each operand will not overflow after we multiply them together.
bool Overflow; bool MaxOverflow;
LHSMax.umul_ov(RHSMax, Overflow); LHSMax.umul_ov(RHSMax, MaxOverflow);
if (!MaxOverflow)
return OverflowResult::NeverOverflows;
return Overflow ? OverflowResult::MayOverflow // We know it always overflows if multiplying the smallest possible values for
: OverflowResult::NeverOverflows; // the operands also results in overflow.
bool MinOverflow;
LHSKnownOne.umul_ov(RHSKnownOne, MinOverflow);
if (MinOverflow)
return OverflowResult::AlwaysOverflows;
return OverflowResult::MayOverflow;
} }

2
lib/Transforms/InstCombine/InstCombineCalls.cpp
View File

@ -443,6 +443,8 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
OverflowResult OR = computeOverflowForUnsignedMul(LHS, RHS, II); OverflowResult OR = computeOverflowForUnsignedMul(LHS, RHS, II);
if (OR == OverflowResult::NeverOverflows) { if (OR == OverflowResult::NeverOverflows) {
return CreateOverflowTuple(II, Builder->CreateNUWMul(LHS, RHS), false); return CreateOverflowTuple(II, Builder->CreateNUWMul(LHS, RHS), false);
} else if (OR == OverflowResult::AlwaysOverflows) {
return CreateOverflowTuple(II, Builder->CreateMul(LHS, RHS), true);
} }
} // FALL THROUGH } // FALL THROUGH
case Intrinsic::smul_with_overflow: case Intrinsic::smul_with_overflow:

13
test/Transforms/InstCombine/intrinsics.ll
View File

@ -231,6 +231,19 @@ define i32 @umultest4(i32 %n) nounwind {
; CHECK: umul.with.overflow ; CHECK: umul.with.overflow
} }
define %ov.result.32 @umultest5(i32 %x, i32 %y) nounwind {
%or_x = or i32 %x, 2147483648
%or_y = or i32 %y, 2147483648
%mul = call %ov.result.32 @llvm.umul.with.overflow.i32(i32 %or_x, i32 %or_y)
ret %ov.result.32 %mul
; CHECK-LABEL: @umultest5(
; CHECK-NEXT: %[[or_x:.*]] = or i32 %x, -2147483648
; CHECK-NEXT: %[[or_y:.*]] = or i32 %y, -2147483648
; CHECK-NEXT: %[[mul:.*]] = mul i32 %[[or_x]], %[[or_y]]
; CHECK-NEXT: %[[ret:.*]] = insertvalue %ov.result.32 { i32 undef, i1 true }, i32 %[[mul]], 0
; CHECK-NEXT: ret %ov.result.32 %[[ret]]
}
define void @powi(double %V, double *%P) { define void @powi(double %V, double *%P) {
entry: entry:
%A = tail call double @llvm.powi.f64(double %V, i32 -1) nounwind %A = tail call double @llvm.powi.f64(double %V, i32 -1) nounwind