optimization/420: 68k dbCC peephole fails in simple test case

[tz@execpc.com]

>Number:         420
>Category:       optimization
>Synopsis:       68k dbCC peephole fails in simple test case
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    unassigned
>State:          open
>Class:          pessimizes-code
>Submitter-Id:   net
>Arrival-Date:   Mon Jul 17 21:46:00 PDT 2000
>Closed-Date:
>Last-Modified:
>Originator:     Tom Zerucha
>Release:        gcc-2.9.5 (palmos/prc-tools-2.0, m68k-coff embedded, probably any m68k)
>Organization:
>Environment:
Linux 2.4.0-test1.
>Description:
testcase.c:
void test(register void *dst, register void *src, register int num) {
  while ((*((char *) dst)++ == *((char *) src)++) && num--); }
compile:
m68k-palmos-gcc -O5 -S testcase.c
testcase.s: (excerpt)
.L5:
        cmpm.b (%a0)+,(%a1)+
        jbne .L4
        dbra %d0,.L5
.L4:

In gcc/config/m68k/m68k.md, it says under ";; dbCC peepholes" it
should optimize this to:
 cmpm.b ...
 dbne %d0,.L5
 jbne .L4
.L4

Of course it should simply drop the jbne .L4 altogether.

A more elaborate version to force the jbne target not to be the next instruction also fails:

void test(register char *dst, register char *src, register int num) {
volatile *port = 0x1235;
  for(;;) {
    *port=1;
    if (*((char *) dst)++ == *((char *) src)++)
      continue;
    if( !num--)
      break;
    *port=2;
 }
}
>How-To-Repeat:
See description.  I can't get it to do the peephole described in m68k.mk
>Fix:
Unknown.
>Release-Note:
>Audit-Trail:
>Unformatted: