[llvm-dev] Bug in pass 'ipsccp' on function attribute 'argmemonly'? (original) (raw)

Fangqing Du via llvm-dev llvm-dev at lists.llvm.org
Fri Jul 10 19:04:41 PDT 2020

Hi all,

Let's see an example (from Alexandre Isoard) first:

; RUN: opt -ipsccp -deadargelim -licm -O2 -S < %s

@g = internal global i32 0

; Function Attrs: argmemonly define internal void @foo(i32* nonnull dereferenceable(4) %arg, i32 %val) #0 { entry: store i32 %val, i32* %arg ret void }

define i32 @bar(i32 %n) { entry: store i32 1, i32* @g br label %loop

loop: ; preds = %bb1, %bb %i = phi i32 [ %i.inc, %loop ], [ 0, %entry ] %g.val = load i32, i32* @g %g.inc = add nuw i32 %g.val, 1 tail call void @foo(i32* @g, i32 %g.inc) %i.inc = add nuw i32 %i, 1 %cond = icmp eq i32 %i.inc, %n br i1 %cond, label %exit, label %loop

exit: ; preds = %bb1 ret i32 %g.val }

declare void @llvm.assume(i1)

attributes #0 = { argmemonly }

With opt cmd '-ipsccp -deadargelim -licm -O2', function @bar will return constant value 1, instead of correct value %n. This is crazy, right? Let's see what happens here. Due to pass 'ipsccp' replaced the argument of function '@foo' with global variable '@g', but keeps attr 'argmemonly' there, so pass 'licm' will hoist the load of '@g' before the loop (as the value of @g isn't changed, but it is changed), and will cause function return wrong constant value '1', instead of '%n' (the correct value) , like following:

; Function Attrs: nofree norecurse nounwind writeonly define i32 @bar(i32 %n) local_unnamed_addr #0 { entry: ret i32 1 }

attributes #0 = { nofree norecurse nounwind writeonly }

And if remove the 'argmemonly' attribute, function @bar will return the correct value:

; Function Attrs: nofree norecurse nounwind define i32 @bar(i32 %n) local_unnamed_addr #0 { entry: ret i32 %n }

So if function attribute 'argmemonly' on function @foo is correct, then there's a bug in pass 'ipsccp'. When it replaces the function local value with global variable, then it shoud remember to remove this function attribute.

Thanks, Fangqing -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200710/71e12c6e/attachment.html>

More information about the llvm-dev mailing list