https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126973
Bug ID: 126973
Summary: Worse X86 codegen for arbitrary
__builtin_clzg/__builtin_ctzg compared to clang.
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: kaelfandrew at gmail dot com
Target Milestone: ---
https://godbolt.org/z/zTrYKvvxK
```
#define N 16
int
clzg_uN (unsigned _BitInt(N) x)
{
return __builtin_clzg (x);
}
int
ctzg_uN (unsigned _BitInt(N) x)
{
return __builtin_ctzg (x);
}
```
With -O2 and mostly N <= 64, Clang produce much better code:
N = 16
``` GCC
"clzg_uN":
movzx edi, di
bsr eax, edi
xor eax, 31
sub eax, 16
ret
"ctzg_uN":
movzx eax, di
rep bsf eax, eax
ret
```
``` Clang
clzg_uN:
bsr ax, di
xor eax, 15
movzx eax, ax
ret
ctzg_uN:
rep bsf eax, edi
ret
```
N = 37
```GCC
"clzg_uN":
movabs rax, 137438953471
and rdi, rax
bsr rax, rdi
xor rax, 63
sub eax, 27
ret
"ctzg_uN":
movabs rax, 137438953471
and rdi, rax
xor eax, eax
rep bsf rax, rdi
ret
```
```Clang
clzg_uN:
shl rdi, 27
bsr rax, rdi
xor eax, 63
ret
ctzg_uN:
rep bsf rax, rdi
ret
```
