On Tue, May 08, 2012 at 09:09:24AM +0200, Dmitry Tarnyagin wrote:
> Yes, correct, you mentioned all the 3 problems with the minstrel.
> "Buggy" is not the best word here, minstrel just behaves in this way,
> and the behavior is specified. It can be changed (at least last two),
> but carefully and with respect to other minstrel clients.
OK, thanks. I (and others) had also noticed these issues a while ago;
I guess we should take a look at fixing it in the core.
> As I remember compiler changes divide to a shift only for
> unsigned divisions, isn't it? Or do modern gcc-s changes it for signed
> divs as well?
It's easy enough to check. I can't speak for ARM, but on gcc 4.6.0
for x86, I get the following:
signed x / 2 compiles to (%eax = dividend):
mov %eax,%edx ; put x into edx
shr $0x1f,%edx ; edx >>= 31, so sign of x is LSB of edx
add %edx,%eax ; eax += 1 (round up) if eax is signed
sar %eax ; arithmetic (sign-preserving) shift right eax
This sequence corrects for, say, -3 / 2 == -1 vs -3 >> 1 == -2.
unsigned x / 2 compiles to:
shr %eax ; same as x >> 1
Gcc has some neat tricks -- even something like "x * 22/7" gets turned
into a pair of multiplies.
> Well, it's datapath. This code is executed for every SDU in data traffic,
> with frequency ~3kHz on target throughput. For me it's a good reason
> to explicitly eliminate divisions :)
Fair enough :)
--
Bob Copeland %% www.bobcopeland.com
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
