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

[bugzilla-daemon@freedesktop.org]

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

           Summary: glXSwapBuffersMscOML is slow on AMD Fusion but not on
                    Intel 945 w/Atom
           Product: Mesa
           Version: unspecified
          Platform: x86 (IA32)
        OS/Version: Linux (All)
            Status: NEW
          Severity: normal
          Priority: medium
         Component: Drivers/Gallium/r600
        AssignedTo: dri-devel@lists.freedesktop.org
        ReportedBy: simon.farnsworth@onelan.co.uk


Created an attachment (id=48588)
 --> (https://bugs.freedesktop.org/attachment.cgi?id=48588)
My test program to demonstrate the performance issues

I'm trying to get an in-house OpenGL compositor up to the same performance on
AMD Fusion as it currently achieves on Intel graphics, and I've found that
glXSwapBuffersMscOML is slower (with a trivial test program) on AMD than it is
on Intel.

I've attached my test program (swapbufferstest.c), which I've been compiling
with:

gcc -O2 -ggdb -Wall -Werror -o swapbufferstest swapbufferstest.c -lX11 -lGL
-lGLEW -lGLU

I'm running identical Fedora 15 based software stacks on an Intel Atom D510
(integrated GMA 950 class graphics) and on an AMD Fusion board (GPU PCI ID
[1002:9802]).

On Intel, I get output like:
# ./swapbufferstest 
Screen is 1920x1200
Longest frame time 17.308969 ms
Biggest delta 3 frames, averaging 1.034188 frames per swap
Average swap interval 17.118282 ms
121 frames, 117 swaps

On AMD, I get:
# ./swapbufferstest 
Screen is 1920x1200
Longest frame time 16.751312 ms
Biggest delta 3 frames, averaging 3.000000 frames per swap
Average swap interval 49.326561 ms
123 frames, 41 swaps

I've confirmed that I've got a direct rendering context in both cases; on
Intel, I've been able to confirm that I'm doing page-flipping, but I can't yet
work out how to do that on the AMD driver; however, only hitting one frame in 3
doesn't seem right to me, especially since it's not reaching 10% CPU load. I
can't however see whether I'm actually hitting the pageflip codepath, or
whether I'm in a slow path.

Versions of interesting components are:

kernel-PAE-2.6.38.8-91.fc15.i686
mesa-dri-drivers-7.11-0.11.20110525.0.fc15.i686
libdrm-2.4.26-1.fc15.i686
xorg-x11-server-Xorg-1.10.2-1.fc15.i686
xorg-x11-drv-ati-6.14.1-1.20110504gita6d2dba6.fc15.i686
xorg-x11-drv-intel-2.15.0-3.fc15.i686

I'm able to rebuild RPMs with extra patches if needed for testing, and I have
both systems to hand for testing as needed.

I'll attach dmesg and Xorg.log from the AMD system to this bug.

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