https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87941
Bug ID: 87941
Summary: by_pieces infra does not use movmisalign optab
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: amonakov at gcc dot gnu.org
Target Milestone: ---
by_pieces code uses only mov_optab, never checking movmisalign_optab, so on
STRICT_ALIGNMENT targets such as arm it does not use available misaligned
load/store patterns. It results in suboptimal code for e.g.
void f(char *c)
{
__builtin_memcpy(c, "foo", 4);
}
where with -O2 -march=armv6t2 gcc emits
f:
movw r3, #:lower16:.LANCHOR0
mov r2, r0
movt r3, #:upper16:.LANCHOR0
ldr r0, [r3]
str r0, [r2] @ unaligned
bx lr
.size f, .-f
.section .rodata
.align 2
.set .LANCHOR0,. + 0
.LC0:
.ascii "foo\000"
while optimal code is emitted for the equivalent
void f(char *c)
{
int t;
__builtin_memcpy(&t, "foo", 4);
__builtin_memcpy(c, &t, 4);
}
f:
movw r3, #28518
movt r3, 111
str r3, [r0] @ unaligned
bx lr
.size f, .-f
