[Bug target/38488] New: [4.4 Regression] x86_64 generates much larger and slightly slower code for memset

[jakub at gcc dot gnu dot org]

typedef __SIZE_TYPE__ size_t;
extern void *memset (void *, int, size_t);
struct A { char a, b, c, d; int i, j, k, l; char p[64]; };
void __attribute__((noinline)) bar (struct A *p)
{
  asm volatile ("" : : "r" (p) : "memory");
}
void foo (void)
{
  struct A a;
  memset (&a.b, '\0', sizeof (a) - 1);
  bar (&a);
}
int
main (void)
{
  int i;
  for (i = 0; i < 100000000; i++)
    foo ();
  return 0;
}

at -O2 on x86_64-linux gives:

$ ~/timing ./test-4.3 # 4.3.2 20081105
Strip out best and worst realtime result
minimum: 1.319280371 sec real / 0.000023862 sec CPU
maximum: 1.856280790 sec real / 0.000048852 sec CPU
average: 1.476243858 sec real / 0.000025087 sec CPU
stdev  : 0.222421870 sec real / 0.000001019 sec CPU
$ ~/timing ./test-4.4 # 4.4.0 20081211
Strip out best and worst realtime result
minimum: 1.729227419 sec real / 0.000021394 sec CPU
maximum: 1.746533792 sec real / 0.000042904 sec CPU
average: 1.737414022 sec real / 0.000025143 sec CPU
stdev  : 0.003746136 sec real / 0.000000764 sec CPU
$ ~/timing ./test-4.4-patched # 4.4.0 20081211 + patch
Strip out best and worst realtime result
minimum: 1.243328664 sec real / 0.000023905 sec CPU
maximum: 1.256095660 sec real / 0.000046869 sec CPU
average: 1.250533713 sec real / 0.000025471 sec CPU
stdev  : 0.002965601 sec real / 0.000001744 sec CPU

Patched is current trunk with the
http://gcc.gnu.org/ml/gcc-patches/2008-12/msg00706.html
patch.


-- 
           Summary: [4.4 Regression] x86_64 generates much larger and
                    slightly slower code for memset
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jakub at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38488