maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

InstCombine: Don't claim to be able to evaluate any shl in a zexted type. 

The shift amount may be larger than the type leading to undefined behavior.
Limit the transform to constant shift amounts. While there update the bits to
clear in the result which may enable additional optimizations.

PR15959.

llvm-svn: 181604
This commit is contained in:
Benjamin Kramer 2013-05-10 16:26:37 +00:00
parent 4ea23b56c5
commit 14e915f7b4
2 changed files with 39 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
View File

@ -677,7 +677,6 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) {
case Instruction::Add:
case Instruction::Sub:
case Instruction::Mul:
case Instruction::Shl:
if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear) ||
!CanEvaluateZExtd(I->getOperand(1), Ty, Tmp))
return false;
@ -701,6 +700,17 @@ static bool CanEvaluateZExtd(Value *V, Type *Ty, unsigned &BitsToClear) {
// Otherwise, we don't know how to analyze this BitsToClear case yet.
return false;
case Instruction::Shl:
// We can promote shl(x, cst) if we can promote x. Since shl overwrites the
// upper bits we can reduce BitsToClear by the shift amount.
if (ConstantInt *Amt = dyn_cast<ConstantInt>(I->getOperand(1))) {
if (!CanEvaluateZExtd(I->getOperand(0), Ty, BitsToClear))
return false;
uint64_t ShiftAmt = Amt->getZExtValue();
BitsToClear = ShiftAmt < BitsToClear ? BitsToClear - ShiftAmt : 0;
return true;
}
return false;
case Instruction::LShr:
// We can promote lshr(x, cst) if we can promote x. This requires the
// ultimate 'and' to clear out the high zero bits we're clearing out though.

28
llvm/test/Transforms/InstCombine/cast.ll
View File

@ -898,3 +898,31 @@ define double @test81(double *%p, float %f) {
%l = load double* %r
ret double %l
}
define i64 @test82(i64 %A) nounwind {
%B = trunc i64 %A to i32
%C = lshr i32 %B, 8
%D = shl i32 %C, 9
%E = zext i32 %D to i64
ret i64 %E
; CHECK: @test82
; CHECK-NEXT: [[REG:%[0-9]*]] = shl i64 %A, 1
; CHECK-NEXT: %E = and i64 [[REG]], 4294966784
; CHECK-NEXT: ret i64 %E
}
; PR15959
define i64 @test83(i16 %a, i64 %k) {
%conv = sext i16 %a to i32
%sub = add nsw i64 %k, -1
%sh_prom = trunc i64 %sub to i32
%shl = shl i32 %conv, %sh_prom
%sh_prom1 = zext i32 %shl to i64
ret i64 %sh_prom1
; CHECK: @test83
; CHECK: %sub = add nsw i64 %k, 4294967295
; CHECK: %sh_prom = trunc i64 %sub to i32
; CHECK: %shl = shl i32 %conv, %sh_prom
}