The fix is fairly simple, but is says something unpleasant about the usage and testing of invariant.start/end scopes that this went undetected. To put this in perspective, *any* invariant.end in a loop flowing through LICM crashed. I haven't bothered to figure out just how far back this goes, but it's not caused by any of the recent changes. We're probably talking months if not years.
llvm-svn: 339936
Arg, cast<ConstantInt>(II->getArgOperand(0))->getZExtValue(), AATags);
case Intrinsic::invariant_end:
+ // The first argument to an invariant.end is a "descriptor" type (e.g. a
+ // pointer to a empty struct) which is never actually dereferenced.
+ if (ArgIdx == 0)
+ return MemoryLocation(Arg, 0, AATags);
assert(ArgIdx == 2 && "Invalid argument index");
return MemoryLocation(
Arg, cast<ConstantInt>(II->getArgOperand(1))->getZExtValue(), AATags);
store i32 0, i32* %ptr
%scope = call {}* @llvm.invariant.start.p0i32(i64 4, i32* %ptr)
%val = load i32, i32* %ptr
-;; NOTE: despite being correct syntax, uncommenting this line causes
-;; a crash in the optimizer. FIXME
-;; call void @llvm.invariant.end.p0i32({}* %scope, i64 4, i32* %ptr)
+ call void @llvm.invariant.end.p0i32({}* %scope, i64 4, i32* %ptr)
%x.inc = add i32 %x, %val
br label %loop
}