gcc version 4.3.0 20070704 (experimental)
test case derived from linux kernel
struct x {
long a,b,c,d,e,f;
char array[32];
};
void f(struct x *p)
{
p->a = 0;
p->b = 0;
p->c = 0;
p->d = 0;
p->e = 0;
p->f = 0;
memset(&p->array, 0, 32);
}
compiled with -O2 or -Os gives
movq $0, (%rdi)
movq $0, 8(%rdi)
movl $8, %ecx
movq $0, 16(%rdi)
movq $0, 24(%rdi)
xorl %eax, %eax
movq $0, 32(%rdi)
movq $0, 40(%rdi)
addq $48, %rdi
rep
stosl
ret
This shows several problems:
- the zero in eax should have been used for all the initializations
replacing the immediate giving shorter code [especially with -Os,
but it would have been a win even with -O2 e.g on decode limited CPUs]
In a more test complex case the xorl was even before all the field
initializations.
- the rep ; stosl should be rep ; stosq because 8 byte alignment is guaranteed
by the type
--
Summary: missing CSE for constant in registers / inefficient
memset
Product: gcc
Version: 4.3.0
Status: UNCONFIRMED
Severity: enhancement
Priority: P3
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: ak at muc dot de
GCC host triplet: x86_64-linux
GCC target triplet: x86_64-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32629
