Toshiba e740 GAPI blit

by **Sergey Chaban** » Dec 25, 2002 @ 6:21pm

Ok, so I figured out enough of the ATI API to implement RGB blits.
I updated blitmark.zip to use ATI routines (see URL in the previous post). Surprisingly, it seems that accessing vmem of the offscreen surface is slightly faster than accessing frontbuffer's memory, so I always get better results when ATI blit is used.
I included two precompiled binaries in the archive:
BlitMark-e740-GAPI.exe (GAPI + memcpy)
BlitMark-e740-ATI.exe (ATI overlay + xs_blit_full_ldrd)
The latter is always slightly faster on my e740.

SC

by **sponge** » Dec 25, 2002 @ 7:26pm

by **Sergey Chaban** » Dec 25, 2002 @ 8:45pm

by **Another guest is asking** » Jan 8, 2003 @ 8:43pm

so is the performance difference on the e740 mainly in the blitting stage?

Would that explain why its pocketQuake score is decent (low FPS)? The higher the FPS, the more significant the time wasted in the blitting stage becomes.

Thanks!

by **James S** » Jan 8, 2003 @ 8:48pm

More like, the more of the screen that needs to be blitted each frame the slower it is. The e740 is inordinately slower in landscape PocketQuake than other PocketPCs, whereas it's slightly faster (after the ROM update) than other PocketPCs in portrait because only half of the screen is being blitted each frame in portrait.

by **Another Guest remembers** » Jan 8, 2003 @ 9:01pm

I remember having this discussion with the Pocket TV guy.

When the e740 came out, PTV performance in landscape was 50% slower than in portrait (if not even worse). I speculated that the extra pixels in landscape could not have caused that much of a slowdown for the following simple reason:

If blitting constitutes 20% of the CPU time, than a 50% increase in pixels would only increase the blitting CPU time to say 30%. But that would only translate to a small decrease (10%) in total performance.

Shortly after a new version of PTV was released WITHOUT any ATI specific API calls in which Landscape performance was 10-20% slower than portrait (as opposed to 50% earlier). From what I remember he mentionned something along the lines of "the e740 video buffer needs to be filled in a certain way to prevent a huge performance drop".

I'm paraphrasing, but I thought I'd throw it out there anyway in case it might help a developper. I think this might also explain why some games seem to run fine on the e740 and some are MUCH slower (depending on how they fill the video buffer).

by **Another Guest quotes** » Jan 8, 2003 @ 9:10pm

by **James S** » Jan 8, 2003 @ 9:26pm

by **Another guest responds** » Jan 8, 2003 @ 9:36pm

by **Sergey Chaban** » Jan 8, 2003 @ 9:56pm

by **Sergey Chaban** » Jan 8, 2003 @ 9:58pm

Toshiba e740 GAPI blit

e740 benchmark

Sort

Forum Description

Moderators:

Forum permissions