https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98957
Bug ID: 98957
Summary: [x86] Odd code generation for 8-bit left shift
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: gabravier at gmail dot com
Target Milestone: ---
bool f(uint8_t m)
{
return m >> 7;
}
With -O3, LLVM outputs this:
f(unsigned char):
mov eax, edi
shr al, 7
ret
With -march=[some-amd-machine-type], GCC outputs this:
f(unsigned char):
mov eax, edi
shr ax, 7
and eax, 1
ret
Otherwise, it outputs the same thing as LLVM. I took a long look at
x86-tune.def to see if there was anything related to this triggered by either
from m_AMD_MULTIPLE or m_ZNVER, but couldn't find anything. Also, even if this
is normal (8-bit shr is bad and 16-bit shr is better?? Since when ?? I've
searched for a while and found nothing about this), GCC 10 outputs this:
f(unsigned char):
movzx eax, dil
shr ax, 7
ret
making this look at the very least like a regression to me.
