On Wed, 2024-04-10 at 17:40 +0800, LIU Hao wrote:
> 在 2024-04-10 17:19, Xi Ruoyao via Gcc-help 写道:
> > $ cc t.c -O2 -fsanitize=undefined
> > $ ./a.out
> > t.c:7:15: runtime error: signed integer overflow: 65535 * 65535
> > cannot be represented in type 'int'
> > fffe0001
>
> Undefined behavior is not a valid point, as it never happens.
You only get a "different result" when an undefined behavior happens,
thus it **is** a valid point to say there is no wrong-code issue.
> It's a real bug. There are many PRs on bugzilla.
You may argue it's a missed-optimization, but we were discussing about
wrong-code or not.
> The sign extension could have been eliminated completely:
>
> movzx eax, si
> movzx edi, di
> imul eax, edi
> ret
>
>
> And here is a similar issue:
>
> typedef unsigned long long int u64;
> typedef unsigned int u32;
>
> u64 foo(u64 a) {
> return (u32) __builtin_ctzll(a);
> }
>
> which results in
>
> xor eax, eax
> rep bsf rax, rdi // effectively `tzcnt rax, rdi`
> cdqe // unnecessary sign-extension
> ret
--
Xi Ruoyao <xry111@xxxxxxxxxxx>
School of Aerospace Science and Technology, Xidian University
