[Bug target/69986] New: smaller code possible with -Os by using push/pop to spill/reload

["peter at cordes dot ca" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69986

            Bug ID: 69986
           Summary: smaller code possible with -Os by using push/pop to
                    spill/reload
           Product: gcc
           Version: 5.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: minor
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: peter at cordes dot ca
  Target Milestone: ---
            Target: x86-64-*-*

#include <unistd.h>
int f(int a) { close(a); return a; }

 push   rbx
 mov    ebx,edi
 call   400490 <close@plt>
 mov    eax,ebx
 pop    rbx
 ret    

with gcc 5.3 -Os.

It could be smaller:

 push   rbi
 call   400490 <close@plt>
 pop    rax
 ret    

saving 4 bytes (mov reg,reg is two bytes).

More generally, push/pop are 1 byte each, much smaller than mov [rsp-8], edi or
something.

This might not be a desirable optimization, though, because a round-trip
through memory increases latency.  It's one of those code-size optimizations
that will might often have a negative impact on performance in the case where
the function is already hot in L1 I-cache.

It would be nice if there was a way to optimize a bit for code-size without
making bad performance sacrifices, and also another option to optimize for code
size without much regard for performance.  -Oss vs. -Os?  Or -OS?  I assume
tuning these options is a lot of work.