| Re: [AD] Allegro x86 clear and blit optimizations - update |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
Esaelon@xxxxxxxxxx <Esaelon@xxxxxxxxxx> sent in a patch.
Here are my results. Sorry about the delay.
Pentium 133 (Voodoo Banshee; Linux 2.2.16; gcc 2.96 20000731)
Fbcon 320x200x8bit - ssb ssba msb msba msmb
before: 1953 2134 11173 23545 10949
after: 1974 2166 11735 25707 11498
change: +1.1% +1.5% +5.0% + 9.2% +5.0%
Fbcon 640x480x8bit - ssb ssba msb msba msmb
before: 1947 2125 11141 23236 10928
after: 1976 2158 11799 25985 11573
change: +1.5% +1.6% +5.9% +11.8% +5.9%
Fbcon 320x200x24bit - ssb ssba msb msba msmb
before: 656 726 4035 10131 3524
after: 659 725 4118 10847 4253
change: +0.5% -0.1% +2.0% + 7.1% +20.7%
(this caused a bug [1])
Fbcon 320x200x32bit - ssb ssba msb msba msmb
before: 543 544 7813 7828 8228
after: 544 545 8300 8299 8000
change: +0.2% +0.2% +6.2% + 6.0% -2.8%
Pentium 233 (Neomagic something; Linux 2.2.16; gcc 2.95.2)
VGA 320x200x8bit - ssb ssba msb msba msmb
before: 2289 2681 7828 25423 12231
after: 2292 2694 7852 25586 12467
change: +0.1% +0.5% +0.3% + 0.6% +1.9%
SVGAlib 640x480x8bit - ssb ssba msb msba msmb
before: 2141 2552 7658 20224 12201
after: 2153 2607 7687 20366 12289
change: +0.6% +2.2% +0.4% + 0.7% +0.7%
There seem to be improvements across the board, so I'd say the
patch did make a difference. If the differences were _purely_ due
to experimental error, there would probably be more minus signs in
there.
Somebody should compare all the results and conclude whether the
patch should be committed: (1) in its entirety, (2) in the areas
with substantial improvement, or (3) not at all.
[1] bug appears when you first go into the blit test, when it says
"Testing overlapping bits", on the left. Eric's patch was applied.