https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128467
Bug ID: 128467
Summary: Missed register allocation optimization around stack
allocated addresses
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: ra
Severity: normal
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: avieira at gcc dot gnu.org
Target Milestone: ---
Target: aarch64
Using the following testcase:
test.c:
typedef unsigned int flags_t;
extern void use_buffers(char *, char *, long);
extern int predicate(long);
__attribute__((noinline, used)) flags_t *
stack_flags(long a0, long a1, long a2, long a3, long a4, long a5, long a6,
long a7, flags_t *flags) {
char min_key[4096];
char max_key[4096];
if (predicate(a6))
*flags |= 0x10;
use_buffers(min_key, max_key, a0);
}
Compiled with
$ gcc -O3 -march=armv8-a -fomit-frame-pointer -g0 test.c -S
you'll see:
ldr x0, [sp, 8208]
ldr x1, [sp, 8208]
ldr w0, [x0]
orr w0, w0, 16
str w0, [x1]
where the same address get's loaded off the stack twice, back-to-back, though
there is no need as the same code could've been written as:
ldr x0, [sp, 8288]
ldr w1, [x0]
orr w1, w1, 16
str w1, [x0]
