Messages in this thread |  | | Subject | Re: Athlon possible fixes | Date | Sat, 5 May 2001 18:26:30 +0200 (MEST) | From | (Rogier Wolff) |
| |
> + __asm__ __volatile__ ( > + " movq (%0), %%mm0\n" > + " movq 8(%0), %%mm1\n" > + " movq 16(%0), %%mm2\n" > + " movq 24(%0), %%mm3\n" > + " movq %%mm0, (%1)\n" > + " movq %%mm1, 8(%1)\n" > + " movq %%mm2, 16(%1)\n" > + " movq %%mm3, 24(%1)\n" > + " movq 32(%0), %%mm0\n" > + " movq 40(%0), %%mm1\n" > + " movq 48(%0), %%mm2\n" > + " movq 56(%0), %%mm3\n" > + " movq %%mm0, 32(%1)\n" > + " movq %%mm1, 40(%1)\n" > + " movq %%mm2, 48(%1)\n" > + " movq %%mm3, 56(%1)\n" > : : "r" (from), "r" (to) : "memory"); > from+=64; > to+=64;
As all this is trying to avoid bus turnarounds (i.e. switching from reading to writing), wouldn't it be fastest to just trust that the CPU has at least 4k worth of cache? (and hope for the best that we don't get interrupted in the meanwhile).
void copy_page (char *dest, char *source) { long *dst = (long *)dest, *src=(long *)source, *end= (long *)(source+PAGE_SIZE); #if 1 register int i; long t=0; static long tt;
for (i=0;i<PAGE_SIZE/sizeof (long);i += cache_line_size()/sizeof(long)) /* Actually the innards of this loop should be: (void) from[i]; however, the compiler will probably optimize that away. */ t += src[i];
tt = t; #endif while (src < end) *dst++ = *src++;
}
So, this is 15 lines of C, and it'd be interesting to benchmark this against the assembly.
I'm assuming that the "loop variable handling" is not going to influence the overall performance: that would run at 500 - 1000MHz, and around 1 clock cycle (1-2ns) per loop. Set this against the stalls against the memory unit whose output buffer is full, and memory writes that take on the order of 30 ns per 64bits.
At least, that's what I expect, however, I haven't been optimizing cycles for quite a while....
Roger.
-- ** R.E.Wolff@BitWizard.nl ** http://www.BitWizard.nl/ ** +31-15-2137555 ** *-- BitWizard writes Linux device drivers for any device you may have! --* * There are old pilots, and there are bold pilots. * There are also old, bald pilots. - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
|  |