https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95887
Bug ID: 95887
Summary: suboptimal memcmp with embedded zero bytes
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: msebor at gcc dot gnu.org
Target Milestone: ---
Similar to pr95886, the memcmp expansion into compare-by-pieces is less than
optimal for sequences containing embedded null bytes. For example, in the test
case below, the memcmp call in f() is expanded into what looks like a more
efficient sequence than the equivalent memcmp call in g(). The only difference
between the two is that the former copies a sequence of non-zero bytes while
among the bytes copied by the latter is a null byte. Clang emits the same code
for g() as GCC does for f().
I believe the root cause of the problem in both cases is working with
nul-terminated strings (using the result of c_getstr() without the size of what
it points to) instead of with arbitrary byte sequences.
$ cat z.c && gcc -O2 -S -Wall -fdump-tree-optimized=/dev/stdout -o/dev/stdout
z.c
const char a[8] = { 1, 2, 3, 4, 5, 6, 7, 8 };
const char b[8] = { 0, 1, 2, 3, 4, 5, 6, 7 };
int f (void *d)
{
return __builtin_memcmp (d, a, 8);
}
int g (void *d)
{
return __builtin_memcmp (d, b, 8);
}
.file "z.c"
.text
;; Function f (f, funcdef_no=0, decl_uid=1932, cgraph_uid=1, symbol_order=2)
f (void * d)
{
int _3;
<bb 2> [local count: 1073741824]:
_3 = __builtin_memcmp (d_2(D), &a, 8); [tail call]
return _3;
}
.p2align 4
.globl f
.type f, @function
f:
.LFB0:
.cfi_startproc
movl $8, %edx
movl $a, %esi
jmp memcmp
.cfi_endproc
.LFE0:
.size f, .-f
;; Function g (g, funcdef_no=1, decl_uid=1935, cgraph_uid=2, symbol_order=3)
g (void * d)
{
int _3;
<bb 2> [local count: 1073741824]:
_3 = __builtin_memcmp (d_2(D), &b, 8); [tail call]
return _3;
}
.p2align 4
.globl g
.type g, @function
g:
.LFB1:
.cfi_startproc
movzbl (%rdi), %eax
ret
.cfi_endproc
.LFE1:
.size g, .-g
.globl b
.section .rodata
.align 8
.type b, @object
.size b, 8
b:
.string ""
.ascii "\001\002\003\004\005\006\007"
.globl a
.align 8
.type a, @object
.size a, 8
a:
.ascii "\001\002\003\004\005\006\007\b"
.ident "GCC: (GNU) 10.1.1 20200527"
.section .note.GNU-stack,"",@progbits
