| Re: [AD] 16-bit MMX Clear DONE ! |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
varobert@xxxxxxxxxx writes:
> It's a full replacement for iblit16.s
Great stuff, thanks.
> Anyway, I wonder how portable the code is (to Linux/XWin/BeOS that is).
> Any takers ?
I'm sure you'll be hearing plenty of complaints over the next few weeks if
it turns out not to be :-) (for example as soon as I get around to merging
it and find that it doesn't work on my box). I suspect it will be fine,
though: this sort of asm is very portable.
> Note: I don't think a 8bpp version would help as it would require too much
> overhead for alignments and such to make it through the 32-byte copy loop.
It would probably be worth trying, though, if you got decent speedups from
the 16 bit version. The alignment overhead for MMX code isn't especially
greater than what it already has to do in order to align the standard 8 bit
copy code: it's the same thing as for 16 bit alignment, only with an extra
level of shifting/masking tacked on the end, so if MMX turns out faster for
16 bits, I'd expect it would for 8 as well...
The only real difference is that fewer bytes are being copied per scanline,
so if the setup costs are hugely greater than the line copying cost, it
might not turn out to be so useful. I find it hard to believe that this
would be the case, though.
--
Shawn Hargreaves - shawn@xxxxxxxxxx - http://www.talula.demon.co.uk/
"A binary is barely software: it's more like hardware on a floppy disk."