[Bug 38800] glXSwapBuffersMscOML is slow on AMD Fusion but not on Intel 945 w/Atom

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=38800

Michel Dänzer <michel at daenzer.net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mario.kleiner at tuebingen.mpg
                   |                            |.de

--- Comment #18 from Michel Dänzer <michel at daenzer.net> 2011-07-06 03:15:51 PDT ---
AFAICT one problem is:

[drm:drm_calc_vbltimestamp_from_scanoutpos], crtc 0 : v 5 p(1154,2)@
1309862042.323318 -> 1309862042.323284 [e 3 us, 0 rep]

The value 5 does not include DRM_SCANOUTPOS_INVBL (2), so apparently scanout is
outside of vertical blank when the flip is processed, so the flip only happens
on the next vertical blank.

p(1154,2) looks like it missed the vertical blank period by 2 scanlines. On an
RS880 laptop here, this is -10 or lower, and the test program runs at full
speed.

Simon, are you using reduced blanking modes?

Alex / Mario, what was the reason for waiting for a vblank interrupt before
emitting the flip to the hardware? Wouldn't it be possible to emit it
immediately, and only if scanout is outside of vertical blank, wait for the
next vblank interrupt before signalling flip completion to userspace?

-- 
Configure bugmail: https://bugs.freedesktop.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.