https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95529
Gabriel Ravier <gabravier at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|Failure to reuse flags |Failure to reuse flags
|generated by trailing zeros |generated by TZCNT for
|instruction for cmovcc |cmovcc on BMI-capable
| |targets
Resolution|WONTFIX |---
Status|RESOLVED |UNCONFIRMED
--- Comment #3 from Gabriel Ravier <gabravier at gmail dot com> ---
Also, I misread your message, so I can see that LLVM doesn't seem to have
invalid behaviour (unless I misread it twice), but I don't think it should make
the bug report invalid altogether.
I've reopened the bug report and changed the name to make it clearer. GCC emits
this for -O3 -mbmi :
f(int):
xor eax, eax
mov edx, 24
tzcnt eax, edi
test edi, edi
cmove eax, edx
ret
LLVM emits this :
f(int): # @f(int)
tzcnt ecx, edi
mov eax, 24
cmovae eax, ecx
ret
If you believe that GCC shouldn't consider it safe to assume the processor is
BMI-capable when -mbmi is used (which seems rather odd but ok), feel free to
close this bug report again, otherwise I believe it should stay open.
