On Wed, Oct 08, 2003 at 11:51:43AM -0700, cgd@broadcom.com wrote:
> At Wed, 8 Oct 2003 18:38:21 +0000 (UTC), "Andrey Petrov" wrote:
> > We use gcc, and even though I tried on sparc64 I'd be surprised if mips port
> > does better with this regard. I also would argue because of driver which is used
> > for all ports.
> 
> GCC 3.2.3 seems to do OK on MIPS.  (the local mods we've made to it
> wouldn't change this behaviour.)
> 
> I'd expect 3.3.1 and later to do similarly well, but haven't checked.
> 
> 
> 
> cgd
> --
> 16 [ldt-sj3-010] xx % cat /tmp/x.c
> 
> unsigned char foo(int *x)
> {
>         return ((x[1] >> 8) & 0xff);
> }


I used 3.3.1 and this example is all right on sparc64 too.
here is my test

int test2(int *p)
{

	if (((*p >> 16) & 0xffff) == 123)
		return 1;

	return 2;
}


	.file	"t2.c"
	.section	".text"
	.align 4
	.global test2
	.type	test2, #function
	.proc	04
test2:
	!#PROLOGUE# 0
	!#PROLOGUE# 1
	sethi	%hi(64512), %g1
	ldsh	[%o0], %g2
	or	%g1, 1023, %g1
	mov	1, %o0
	and	%g2, %g1, %g2
	cmp	%g2, 123
	retl
	movne	%icc, 2, %o0
	.size	test2, .-test2
	.ident	"GCC: (GNU) 3.3.1"

Interesting that gcc indeed does access it as a halfword but still wants
to apply mask. It certainly looks like gcc optimizes this, just not
always successfully. Seems that I got to take back my performance argument:-)

	Andrey