https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68143
Bug ID: 68143
Summary: [5/6 Regression][ARM] Wrong code initialising struct
member after zeroing out the whole struct
Product: gcc
Version: 6.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: ktkachov at gcc dot gnu.org
Target Milestone: ---
Target: arm
The testcase:
#define NULL 0
struct stuff
{
int a;
int b;
int c;
int d;
int e;
char *f;
int g;
};
void __attribute__ ((noinline))
bar (struct stuff *x)
{
if (x->g != 2)
__builtin_abort ();
}
void
main (int argc, char** argv)
{
struct stuff x = {0, 0, 0, 0, 0, NULL, 0};
x.a = 100;
x.d = 100;
x.g = 2;
/* Struct should now look like {100, 0, 0, 100, 0, 0, 0, 2}. */
bar (&x);
}
fails on arm targets with options:
-O3 -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard
The compiler does the initial zeroing-out of the struct using two vst
instructions but the initialisation of the 'g' field to 2 gets move before
them.
The setmem expansion in arm_block_set_aligned_vect is suspect.
