[cairo] Speeding up _cairo_fixed_from_double

[Behdad Esfahbod]

On Mon, 2006-10-30 at 01:42 -0500, Jamey Sharp wrote:
> On Sun, Oct 29, 2006 at 08:12:48PM -0500, Jim Gettys wrote:
> > It should be.  The man page refers to dead code elimination (dce) being
> > the default at -O or higher.
> 
> Sadly, GCC isn't recognizing that as dead code, so it can't eliminate
> it. With -O, the version of GCC I tested [1] does some remarkable tricks
> to eliminate branches, but still checks endianness at runtime. I tested
> this C:
> 
> 	int endian = 1;
> 	if (*(char *) &endian)
> 		return '\154'; /* 'l' */
> 	else
> 		return '\102'; /* 'B' */
> 
> GCC produced this x86 assembly:
> 
> 	movl    $1, %eax
> 	movl    %eax, 12(%esp)
> 	cmpb    $1, 12(%esp)
> 	sbbl    %eax, %eax
> 	andl    $-42, %eax
> 	addl    $108, %eax
> 
> Impressive, but not what was desired.
> 
> --Jamey
> 
> [1] GCC: (GNU) 4.1.2 20060901 (prerelease) (Debian 4.1.1-13)

Cannot reproduce.

This code:

void
main (void)
{
  int endian = 1;
  if (*(char *) &endian)
    return '\154'; /* 'l' */
  else
    return '\102'; /* 'B' */
}

generates:
$ gcc -O -S a.c

        leal    4(%esp), %ecx
        andl    $-16, %esp
        pushl   -4(%ecx)
        pushl   %ebp
        movl    %esp, %ebp
        pushl   %ecx
        popl    %ecx
        popl    %ebp
        leal    -4(%ecx), %esp
        ret

This code:

void
main (void)
{
  return '\154'; /* 'l' */
}

generates:

        leal    4(%esp), %ecx
        andl    $-16, %esp
        pushl   -4(%ecx)
        pushl   %ebp
        movl    %esp, %ebp
        pushl   %ecx
        popl    %ecx
        popl    %ebp
        leal    -4(%ecx), %esp
        ret

(exactly the same as the previous one.)

and this code:

void
main (void)
{
  volatile int endian = 1;
  if (*(char *) &endian)
    return '\154'; /* 'l' */
  else
    return '\102'; /* 'B' */
}

generates:

        leal    4(%esp), %ecx
        andl    $-16, %esp
        pushl   -4(%ecx)
        pushl   %ebp
        movl    %esp, %ebp
        pushl   %ecx
        subl    $16, %esp
        movl    $1, -8(%ebp)
        addl    $16, %esp
        popl    %ecx
        popl    %ebp
        leal    -4(%ecx), %esp
        ret

(this is exactly what I get if I compile the first code with -O0)

Without trying to understand what the generated code means, this
suggests that gcc is doing the right thing.

$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)


-- 
behdad
http://behdad.org/

"Commandment Three says Do Not Kill, Amendment Two says Blood Will Spill"
        -- Dan Bern, "New American Language"