https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120683
H.J. Lu <hjl.tools at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2025-06-16
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Summary|vector_loop generates |vector_loop generates
|horrible prologue and |horrible prologue and
|epilogue on memset |epilogue on memset/memcpy
--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> ---
memcpy isn't much better:
---
#include <sys/types.h>
void
foo (void *dest, void *src, size_t len)
{
__builtin_memcpy (dest, src, len);
}
---
-O2 -mmemcpy-strategy=vector_loop:256:noalign,libcall:-1:noalign
-minline-all-stringops
generates:
foo:
.LFB6:
.cfi_startproc
movq %rdi, %r8
movq %rsi, %rax
cmpq $64, %rdx
jnb .L13
.L2:
andl $63, %edx
je .L1
xorl %ecx, %ecx
.L5:
movzbl (%rax,%rcx), %esi
movb %sil, (%r8,%rcx)
addq $1, %rcx
cmpq %rdx, %rcx
jb .L5
.L1:
ret
.p2align 4,,10
.p2align 3
.L13:
movq %rdx, %rcx
xorl %eax, %eax
andq $-64, %rcx
.L3:
movdqu (%rsi,%rax), %xmm3
movdqu 16(%rsi,%rax), %xmm2
movdqu 32(%rsi,%rax), %xmm1
movdqu 48(%rsi,%rax), %xmm0
movups %xmm3, (%rdi,%rax)
movups %xmm2, 16(%rdi,%rax)
movups %xmm1, 32(%rdi,%rax)
movups %xmm0, 48(%rdi,%rax)
addq $64, %rax
cmpq %rcx, %rax
jb .L3
leaq (%rdi,%rax), %r8
addq %rsi, %rax
jmp .L2
.cfi_endproc
.LFE6:
.size foo, .-foo
