Dear Ian, thanks for your quick and comprehensive answer!
> I'm running on a Marvell PXA310 processor at 624 Mhz and 208 Mhz > system bus which is an Intel XScale derivative (and further derived up > an ARM architecture). Below is the software rendering results. That processor is also on my list of candidates, but from several reasons we would prefer the AVR32 or AT91SAM (ARM9) if they turn out to be fast enough. > I would look into the architecture on your processor, as on my CPU the > main bus carries all data from memory to GCU to LCD to drive access, > i/o. I found that going from 640x480 output on the LCD to 320x240 > produced dramatic results in my frame rates due to there being 4 times > less data being sent across the bus. Also the LCD output on the PXA > will support 25 bit RBGT format, but we also had to reduce the size to > 16 bit RGB. This provided a great speed boost benefit as well to the > CPU rendering rates. Good to know! I will try with a depth of 16 Bits! > What could be happening is that the bus is wasting a lot of the CPU > time transporting the data from the CPU to the GCU to LCD output, or > combination there of. Yes, the Bus is definitively at its limit... But we need the resolution... The AT91SAM actually has two busses of which one can be exclusively dedicated to the LCD controller. RAM access however could still be the bottleneck... > Despite the architecture being 32 bit and 208 mhz (theoretical > bandwidth of 793 MB/s) we found that 640x480 @ 30 fps @ 24bit depth > should utilize about 210.95 MB/s or 27% of the bus bandwidth where as > 320x240 @ 30 fps @ 16 bit depth should utilize about 35.16 MB/s or a > little under 5% of the bus bandwidth. A dramatic difference as the > CPU doesn't need to wait on memory access or other timing issues > related to the bus being occupied transporting frames. > > Anti-aliased Text 3.015 secs ( 42.985 > KChars/sec) > Anti-aliased Text (blend) 3.106 secs ( 13.908 > KChars/sec) > Fill Rectangle 3.036 secs ( 153.262 > MPixel/sec) > Fill Rectangle (blend) 4.454 secs ( 2.942 > MPixel/sec) > Fill Rectangles [10] 3.058 secs ( 171.448 > MPixel/sec) > Fill Rectangles [10] (blend) 22.455 secs ( 2.918 > MPixel/sec) > Fill Triangles 3.034 secs ( 76.681 > MPixel/sec) > Fill Triangles (blend) 3.419 secs ( 2.875 > MPixel/sec) > Draw Rectangle 3.007 secs ( 4.389 > KRects/sec) > Draw Rectangle (blend) 3.019 secs ( 1.059 > KRects/sec) > Draw Lines [10] 3.033 secs ( 16.815 > KLines/sec) > Draw Lines [10] (blend) 3.057 secs ( 4.579 > KLines/sec) > Fill Spans 3.019 secs ( 117.222 > MPixel/sec) > Fill Spans (blend) 4.532 secs ( 2.892 > MPixel/sec) > Blit 3.083 secs ( 55.268 > MPixel/sec) > Blit colorkeyed 3.053 secs ( 23.612 > MPixel/sec) > Blit destination colorkeyed 3.854 secs ( 5.101 > MPixel/sec) > Blit with format conversion 3.925 secs ( 6.678 > MPixel/sec) > Blit with colorizing 3.860 secs ( 5.093 > MPixel/sec) > Blit from 32bit (blend) 5.625 secs ( 2.330 > MPixel/sec) > Blit from 32bit (blend) with colorizing 3.680 secs ( 1.780 > MPixel/sec) > Stretch Blit 3.379 secs ( 42.781 > MPixel/sec) > Stretch Blit colorkeyed 3.738 secs ( 31.673 > MPixel/sec) > > I know this doesn't answer your question about a specific driver for > the ATMEL chip, but it may help increase your frame rate. I hope this > is helpful to you. Those benchmark results are much higher than ours... At what resolution, bit depth and refresh rate were they measured? Ours were taken at 24 Bit 800x480. Kind regards, Robert > > Ian -- Robert Feld, Dipl.-Ing. Research and Development Corscience GmbH & Co. KG Henkestr. 91 D-91052 Erlangen Germany Tel: +49 9131 977986-511 Fax: +49 9131 977986-59 e-mail: [email protected] <mailto:[email protected]> Internet: www.corscience.de <http://www.corscience.de/> ---------------------------------------- Corscience GmbH & Co. KG Sitz der Gesellschaft/Place of business: Erlangen Amtsgericht/Local court: Fürth Handelsregisternummer/Commercial Register No.: HRA 7510 Geschäftsführer/Managing director: Prof. Dr. Armin Bolz, Dr. Karl-Andreas Feldhahn, Dipl.-Volksw. Marc Griefahn CONFIDENTIALITY: This e-mail and any attachments are confidential and may also be privileged. If received in error, please do not disclose the contents to anyone, but notify us immediately by return e-mail and delete this e-mail and any attachments from your system. Thank you.
_______________________________________________ directfb-users mailing list [email protected] http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-users
