Too optimized and not so much optimized code

[mitr@email.cz]

I came across a bit strange egcs-1.1.2 behaviour
-----------------------------
$ cat > tst.c
-----------------------------
unsigned
expand (char s)
{
  unsigned x;
  x = 0;
  *(char *)&x = s;
  return x;
}

unsigned
mult_256 (char s)
{
  unsigned x;
  x = 0;
  ((char *)&x)[1] = s;
  return x;
}
-----------------------------
$ egcs -O2 -fomit-frame-pointer -S tst.c
$ cat tst.s
-----------------------------

        .file   "tst.c"
        .version        "01.01"
gcc2_compiled.:
.text
        .align 4
.globl expand
        .type    expand,@function
expand:
        movl 4(%esp),%eax
        ret
.Lfe1:
        .size    expand,.Lfe1-expand
        .align 4
.globl mult_256
        .type    mult_256,@function
mult_256:
        subl $4,%esp
        movl 8(%esp),%eax
        movl $0,(%esp)
        movb %al,1(%esp)
        movl (%esp),%eax
        popl %ecx
        ret
.Lfe2:
        .size    mult_256,.Lfe2-mult_256
        .ident  "GCC: (GNU) egcs-2.91.66 19990314/Linux (egcs-1.1.2
release)"
------------------------------
(Yes, I know that it is not portable)
1) don't you think that expand () is overoptimized ? IMHO it should look
like
   this:
expand:
        movzbl  4(%esp),%eax
        ret
2) the mult_256 () is here for two reasons:
 a) it is compiled reasonably, so one would conclude that expand () should
too
 b) instead of the last three insns the following should be generated
        popl    %eax
        ret
Thanks for the time you spend while developing egcs.
        Mirek


--- CREATED BY ATC O..R..G..A..N..I..Z..E..R
--- http://email.cz <--- Get Your Free Email