miscompile from arm64 backend with (icmp ult (sub -6, -8) 3)

[regehr]

here's a function that needs to return true:

define i1 @f() {
  %1 = sub i4 -6, -8
  %2 = icmp ult i4 %1, 3
  ret i1 %2
}

however, since the 14 release the arm64 backend got a regression causing it to return false:

regehr@john-home:~/tmp$ ~/clang+llvm-14.0.0-x86_64-linux-gnu-ubuntu-18.04/bin/llc foo.ll -o - -march=arm64
	.text
	.file	"foo.ll"
	.globl	f                               // -- Begin function f
	.p2align	2
	.type	f,@function
f:                                      // @f
	.cfi_startproc
// %bb.0:
	mov	w0, #1
	ret
.Lfunc_end0:
	.size	f, .Lfunc_end0-f
	.cfi_endproc
                                        // -- End function
	.section	".note.GNU-stack","",@progbits
regehr@john-home:~/tmp$ ~/llvm-project/for-alive/bin/llc foo.ll -o - -march=arm64
	.text
	.file	"foo.ll"
	.globl	f                               // -- Begin function f
	.p2align	2
	.type	f,@function
f:                                      // @f
	.cfi_startproc
// %bb.0:
	mov	w0, wzr
	ret
.Lfunc_end0:
	.size	f, .Lfunc_end0-f
	.cfi_endproc
                                        // -- End function
	.section	".note.GNU-stack","",@progbits
regehr@john-home:~/tmp$ 

cc @ornata @nunoplopes @ryan-berger @nbushehri @zhengyang92 @aqjune @Hatsunespica

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

[topperc]

Is this the same as #55342?

[regehr]

Is this the same as #55342?

well it sure looks related but no, this one is still wrong using an LLVM that has your patch to that one

[regehr]

if y'all prefer, we can sit on these ones that are highly non-canonical, but I feel like it's hard to prove that there aren't canonical triggers for this issue

[topperc]

Looks like a bug in the same code as 55342. When the safewrap instruction is a subtract, instead of sign extending the constant, we need to negate it, sign extend it, then negate it again, I think. I think the sign extend by itself is only correct for add.

[topperc]

Candidate patch https://reviews.llvm.org/D125653

[regehr]

ok here's a trigger for this issue that, while not being canonical, at least doesn't contain a trivial fold opportunity

define i1 @f(i4 %0) {
  %2 = sub i4 %0, 6
  %3 = icmp ult i4 -5, %2
  ret i1 %3
}

[topperc]

Fixed by 8d3894f