https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97759
--- Comment #3 from Hongtao.liu <crazylht at gmail dot com> ---
for testcase:
---
#include<stdbool.h>
bool
is_power2_popcnt (int a)
{
return __builtin_popcount (a) == 1;
}
bool
is_power2_arithmetic (int a)
{
return !(a & (a - 1)) && a;
}
---
gcc -O2 -mavx2 -S got
---
.file "test.c"
.text
.p2align 4
.globl is_power2_popcnt
.type is_power2_popcnt, @function
is_power2_popcnt:
.LFB0:
.cfi_startproc
popcntl %edi, %edi
cmpl $1, %edi
sete %al
ret
.cfi_endproc
.LFE0:
.size is_power2_popcnt, .-is_power2_popcnt
.p2align 4
.globl is_power2_arithmetic
.type is_power2_arithmetic, @function
is_power2_arithmetic:
.LFB1:
.cfi_startproc
leal -1(%rdi), %eax
testl %edi, %eax
sete %al
testl %edi, %edi
setne %dl
andl %edx, %eax
ret
.cfi_endproc
---
Latency of popcnt is 3, others is 1.
Can't tell which version is better from static observation.
