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

From: bugzilla-daemon@freedesktop.org
To: dri-devel@lists.freedesktop.org
Subject: [Bug 38800] glXSwapBuffersMscOML is slow on AMD Fusion but not on Intel 945 w/Atom
Date: Wed,  6 Jul 2011 14:25:54 -0700 (PDT)	[thread overview]
Message-ID: <20110706212554.5DCB613004D@annarchy.freedesktop.org> (raw)
In-Reply-To: <bug-38800-502@http.bugs.freedesktop.org/>

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

--- Comment #23 from Jerome Glisse <glisse@freedesktop.org> 2011-07-06 14:25:54 PDT ---
So what exactly the timestamp should correspond to ? For me vblank is good
enough and looking at specification its says that it's when the hw register are
using the new value. So either you consider that's when vblank occur or that's
at the end of vblank when the new buffer start being scanout.

Still i think this is achievable with Simon patch which simplify this whole
process to:

So :
pflip_ioctl
  if (invblank)
    wait;
  update crtc reg
  enable irq and ask for vblank or pageflip completed irq

irq handler:
  report proper timestamp

For older hw we could either accept the error margin due to interrupt happening
on vblank or we could compute how much time is needed for vblank to end and add
it to timestamp.

This change would remove complexity or least confine it to the timestamp
computation. I don't think we can count too much on fast irq handling.

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