This function generates the following asm under -O3 in version 4.5
#include <stdarg.h>
void va_overload2(int p1, int p2);
void va_overload3(int p1, int p2, int p3);
void va_overload(int p1, int p2, ...)
{
if (p2 == 7)
{
va_list v;
va_start(v, p2);
int p3 = va_arg(v, int);
va_end(v);
va_overload3(p1, p2, p3);
return;
}
va_overload2(p1, p2);
}
Dump of assembler code for function va_overload:
0x0000000000400520 <+0>: sub $0x58,%rsp
0x0000000000400524 <+4>: cmp $0x7,%esi
0x0000000000400527 <+7>: mov %rdx,0x30(%rsp)
0x000000000040052c <+12>: je 0x400540 <va_overload+32>
0x000000000040052e <+14>: callq 0x4004f0 <va_overload2>
0x0000000000400533 <+19>: add $0x58,%rsp
0x0000000000400537 <+23>: retq
0x0000000000400538 <+24>: nopl 0x0(%rax,%rax,1)
0x0000000000400540 <+32>: lea 0x60(%rsp),%rax
0x0000000000400545 <+37>: mov 0x30(%rsp),%edx
0x0000000000400549 <+41>: movl $0x18,(%rsp)
0x0000000000400550 <+48>: mov %rax,0x8(%rsp)
0x0000000000400555 <+53>: lea 0x20(%rsp),%rax
0x000000000040055a <+58>: mov %rax,0x10(%rsp)
0x000000000040055f <+63>: callq 0x400500 <va_overload3>
0x0000000000400564 <+68>: add $0x58,%rsp
0x0000000000400568 <+72>: retq
This could be replaced with the much more compact:
cmp $0x7, %esi
je 1f
jmp va_overload2
1: jmp va_overload3
since the third parameter is passed in a register, and will still be there
after the comparison. (Actually, if it were passed on the stack it still
wouldn't matter, because we can tail-call here.)
--
Summary: va_list usage missed optimization.
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: enhancement
Priority: P3
Component: c
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: svfuerst at gmail dot com
GCC build triplet: x86_64-linux
GCC host triplet: x86_64-linux
GCC target triplet: x86_64-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44262
