Radeon Page-Flipping Patches Come About

[agd5f]

new patch up.

[darkbasic]

Damn, I started compiling few minutes ago -.-

[monraaf]

The good news is that it now passes the width test (both 1680), the bad news is that I'm now back to having a corrupted image . So they must have a different stride and pitch. And just like before it works again after using xrandr to switch to a different resolution and then back.

[monraaf]

Okay I think I found the inconsistency with regard to stride after xrandr resize. I think the initial stride of 1792 is correct, and the 1728 after xrandr not.

In drmmode_xf86crtc_resize:

pitch = RADEON_ALIGN(width, drmmode_get_pitch_align(scrn, cpp, tiling_flags) / cpp);

aligns to 64 bytes (256/cpp) Change it to:

pitch = RADEON_ALIGN(width, drmmode_get_pitch_align(scrn, cpp, tiling_flags));

and xrandr now also gives a stride of 1792. Both same stride, bot same corruption with glxgears -fullscreen . Now there must be something changed in exa pixmap code to fix this I think.

Btw There's also some pitch calculation code in RADEONSetPitch, maybe can be unified as well.

[nanonyme]

Btw, are you testing with drm-next on top of 2.6.36? There's some regression there that leads to corrupted textures in at least Nexuiz, unrelated to color tiling or page flipping. git bisecting it here atm, sevenish iterations to go but kinda slow.

[monraaf]

No this is definitely pageflip problem, as

vblank_mode=0 glxgears -fullscreen

gives no corruption

[nanonyme]

If it's vblank-related, fine, might be completely separate. I run vblank off anyhow.

[nanonyme]

(is this the modern way to say I run with scissors?)

[agd5f]

Ok, I think I fixed all the allocations. new commit to xf86-video-ati master and udpated ddx patches:

http://people.freedesktop.org/~agd5f/pflip/

[darkbasic]

Ok, I think I fixed all the allocations. new commit to xf86-video-ati master and udpated ddx patches:
http://people.freedesktop.org/~agd5f/pflip/

I still see corruption in openarena with RV670.