Re: [RFC] arm: use built-in byte swap function

[Kim Phillips]

On Wed, 20 Feb 2013 23:29:58 -0500
Nicolas Pitre <nico@xxxxxxxxxxx> wrote:

> On Wed, 20 Feb 2013, Kim Phillips wrote:
> 
> > On Wed, 20 Feb 2013 10:43:18 -0500
> > Nicolas Pitre <nico@xxxxxxxxxxx> wrote:
> > 
> > > On Wed, 20 Feb 2013, Woodhouse, David wrote:
> > > > On Wed, 2013-02-20 at 09:06 -0500, Nicolas Pitre wrote:
> > > > > ... in which case there is no harm shipping a .c file and trivially 
> > > > > enforcing -O2, the rest being equal.
> > > > 
> > > > For today's compilers, unless the wind changes.
> > > 
> > > We'll adapt if necessary.  Going with -O2 should remain pretty safe anyway.
> > 
> > Alas, not so for gcc 4.4 - I had forgotten I had tested
> > Ubuntu/Linaro 4.4.7-1ubuntu2 here:
> > 
> > https://patchwork.kernel.org/patch/2101491/
> > 
> > add -O2 to that test script and gcc 4.4 *always* emits calls to
> > __bswap[sd]i2, even with -march=armv6k+.

argh, sorry - that script was testing support for 
__builtin_bswap{16,32,64} directly, which isn't the same as testing
code generation of a byte swap pattern in C.

> Crap.  OK, assembly code is the way to go then.
> 
> > I'll try working on an assembly version given it probably
> > makes more sense, future-gcc-immunity-wise.
> 
> Agreed.

I'll still try the assembly approach - gcc 4.4's armv6 output looks
worse than both the pre-armv6 and post-armv6 __arch_swab32
implementations currently in use:

mov     ip, sp
push    {fp, ip, lr, pc}
sub     fp, ip, #4
and     r2, r0, #65280  ; 0xff00
lsl     ip, r0, #24
orr     r1, ip, r0, lsr #24
and     r0, r0, #16711680       ; 0xff0000
orr     r3, r1, r2, lsl #8
orr     r0, r3, r0, lsr #8
ldm     sp, {fp, sp, pc}

Kim

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/