On Fri, 2003-02-21 at 03:58, James Simmons wrote:

quoted

I was for five weeks in U.S., so I did not do anything with
matroxfb during that time. I plan to use fillrect and copyrect
from generic code 

I have ported the accelerated functions to the new api. What is left is to 
deal with the loadfont and putcs issue which I'm working on the code right 
now.

quoted

(although it means unnecessary multiply on
generic side, and division in matroxfb, 

????

quoted

but well, if we gave
up on reasonable speed for fbdev long ago...). 

This is not true. Several benchmarks have shown a large performance 
improvement in 2.5.X.

2.5.x might be a bit slower with bpp8 but at higher color depths is
significantly faster. And this is done with a single generic color
exapnd function that replaces the entire fbcon-cfb*.c in 2.4.  And it
will theoretically still draw correctly whatever the condition is (any
bpp from  1-32,  unaligned origin, pitch, width, etc). 

Drivers with accelerated color expansion, if done correctly, _should_
perform better whatever the color depth.  

However, using fonts with widths not divisible by 8 will be several
folds slower.  This should be helped if we add some form of tile/texture
blitting support to fbdev.

Note: I cannot test with 12x22 fonts in 2.4 because some/most drivers do
not support it.

Tony

no accel
scrollmode: yredraw
font: 8x16
visual: packed pixels

time cat /usr/src/linux/MAINTAINERS

linux-2.4.20

bpp8
----
real    0m2.499s
user    0m0.000s
sys     0m2.500s

bpp16
-----
real    0m8.324s
user    0m0.000s
sys     0m8.320s

bpp24
-----
real    0m12.364s
user    0m0.000s
sys     0m12.370s

bpp32
-----
real    0m16.274s
user    0m0.000s
sys     0m16.280s

linux-2.5.62

bpp8
----
real    0m2.557s
user    0m0.003s
sys     0m2.553s

bpp16
-----
real    0m4.051s
user    0m0.002s
sys     0m4.050s

bpp24
-----
real    0m9.520s
user    0m0.000s
sys     0m9.520s

bpp32
-----
real    0m7.496s
user    0m0.002s
sys     0m7.494s