Auto merge of #103299 - nikic:usub-overflow, r=wesleywiser
Don't use usub.with.overflow intrinsic The canonical form of a usub.with.overflow check in LLVM are separate sub + icmp instructions, rather than a usub.with.overflow intrinsic. Using usub.with.overflow will generally result in worse optimization potential. The backend will attempt to form usub.with.overflow when it comes to actual instruction selection. This is not fully reliable, but I believe this is a better tradeoff than using the intrinsic in IR. Fixes #103285.
This commit is contained in:
commit
f42b6fa7ca
2 changed files with 24 additions and 5 deletions
|
@ -365,11 +365,14 @@ impl<'a, 'll, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'll, 'tcx> {
|
|||
Int(I64) => "llvm.ssub.with.overflow.i64",
|
||||
Int(I128) => "llvm.ssub.with.overflow.i128",
|
||||
|
||||
Uint(U8) => "llvm.usub.with.overflow.i8",
|
||||
Uint(U16) => "llvm.usub.with.overflow.i16",
|
||||
Uint(U32) => "llvm.usub.with.overflow.i32",
|
||||
Uint(U64) => "llvm.usub.with.overflow.i64",
|
||||
Uint(U128) => "llvm.usub.with.overflow.i128",
|
||||
Uint(_) => {
|
||||
// Emit sub and icmp instead of llvm.usub.with.overflow. LLVM considers these
|
||||
// to be the canonical form. It will attempt to reform llvm.usub.with.overflow
|
||||
// in the backend if profitable.
|
||||
let sub = self.sub(lhs, rhs);
|
||||
let cmp = self.icmp(IntPredicate::IntULT, lhs, rhs);
|
||||
return (sub, cmp);
|
||||
}
|
||||
|
||||
_ => unreachable!(),
|
||||
},
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue