Bug 19177 - [UXA] glean/colored{Tex,Lit}Perf2 failed in UXA, while passed in EXA
Summary: [UXA] glean/colored{Tex,Lit}Perf2 failed in UXA, while passed in EXA
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/DRI/i965 (show other bugs)
Version: unspecified
Hardware: Other Linux (All)
: high normal
Assignee: Eric Anholt
QA Contact:
Depends on:
Blocks: 18841 intel-3d-release
  Show dependency treegraph
Reported: 2008-12-19 02:05 UTC by zhao jian
Modified: 2009-06-29 22:50 UTC (History)
2 users (show)

See Also:
i915 platform:
i915 features:

xorg.conf (3.82 KB, text/plain)
2008-12-19 02:05 UTC, zhao jian
glean tests results on gm45 with the 4th quarter release (16.29 KB, text/plain)
2009-01-18 23:42 UTC, zhao jian

Description zhao jian 2008-12-19 02:05:39 UTC
Created attachment 21311 [details]

System Environment:
Host:        x-gm45
Arch:          i386
OSD:            Fedora release 10 (Cambridge)
Kernel:    2.6.28-rc8
Libdrm:         (master)c86d431fe6174b1c2de531929213ea7dbd92326d
Mesa_stable:            (intel-2008-q4)0d5b1e591b7fb2cf3109b7e147bb3ea6aa8f8b15
GEM Kernel: (for-airlied branch)  88493410cbe68eb267c540fbb0f6809a1a4124d6

Bug Description:
more than 10 glean cases such as makeCurrent run failed on 945gm and gm45 in UXA mode, and in EXA mode these 10 cases will pass.

Reproduce steps:
1.enable UXA
3. run glean cases
Comment 1 Gordon Jin 2008-12-19 23:24:03 UTC
Jian, you need list the specific list.
Comment 2 zhao jian 2008-12-22 03:46:40 UTC
These cases include: basicPerf, makeCurrent, blendFunc, logicOp, readPixSanity, rgbTriStrip, scissor, texCombine, texEnv, texRect.  And coloredTexPerf2 and coloredLitPerf2 had the result of warn. 
Comment 3 Eric Anholt 2009-01-06 11:29:08 UTC
Please retest since

commit 9a5082d2920c1a212fe935b5a093013e8035c321
Author: Eric Anholt <eric@anholt.net>
Date:   Mon Jan 5 23:28:50 2009 -0800

    Disable DRI2 buffer tiling on non-965, as those need fence regs for 2D blits
    This fixes glReadPixels failure on single-channel 915GM, as the software cod
    for readpixels was actually the only code in the driver doing tiling against
    these buffers (everything else says "rely on fence registers", since the 2D
    blits don't have a "don't rely on fence registers" option).
Comment 4 Gordon Jin 2009-01-10 22:12:08 UTC
Jian, any update?
Comment 5 zhao jian 2009-01-18 23:34:44 UTC
I tested with the RC4's code, and it still failed. The log is in attachment. 
With the following configuration:
Mesa:            (intel-2008-q4)200fa9165d7078a6f36c5c9d3e0c997c2438bde3
Xorg:           7.2
Xserver:         (server-1.6-branch)251d0d8090322b2c9dc0c8b7bef001f338d19433
Xf86_video_intel:                (xf86-video-intel-2.6-branch)5cd65d965c8ed388275fe2084553302aad601d4a

Comment 6 zhao jian 2009-01-18 23:42:13 UTC
Created attachment 22091 [details]
glean tests results on gm45 with the 4th quarter release
Comment 7 Gordon Jin 2009-02-24 05:05:14 UTC
*** Bug 20291 has been marked as a duplicate of this bug. ***
Comment 8 martin 2009-02-24 05:13:45 UTC
The glean/makeCurrent test case not only fails but it causes a SEGV in the xserver (at least on my G45, 29-rc6 + ubuntu user land machine).
Comment 9 liuhaien 2009-04-22 19:58:09 UTC
with the latest driver,these cases as below have got pass now:
we test them agaist below commit:
Libdrm:		(master)a1e3ab9e55047c08a4006ec389c1a99b72bc672c
Mesa:		(mesa_7_4_branch)e8807a14a61a0b9389aa2f2a113da24ab22a364d
Xserver:	(server-1.6-branch)11db545a86c8933c638a0bc1fcd4f2c65279f617
Xf86_video_intel:		(2.7)296a986e5258e2fd13ec494071b7063bd639cd68
Kernel:       (qa-branch)ba1d2a9be507cda299c15740ff7e2bb3705a4792
Comment 10 Gordon Jin 2009-06-09 12:10:38 UTC
Jian, could you give an updated list for those passed with EXA and now still fail with UXA?
Comment 11 zhao jian 2009-06-10 02:39:17 UTC
(In reply to comment #10)
> Jian, could you give an updated list for those passed with EXA and now still
> fail with UXA?

UXA nearly has a same performance compared with EXA now. Only coloredTexPerf2, coloredLitPerf2 are passed with EXA and failed with UXA. And texCombine is timeout with both UXA and EXA. 
Comment 12 Eric Anholt 2009-06-20 11:45:07 UTC
This is half-fixed now, and the tests don't crash but they do read wrong results and complain:

commit df70d3049a396af3601d2a1747770635a74120bb
Author: Eric Anholt <eric@anholt.net>
Date:   Fri Jun 19 22:12:52 2009 -0700

    intel: Also get the DRI2 front buffer when doing front buffer reading.

idr, there's a piglit test called read-front now that catches the glean regression without having to suffer through running the interminable glean test sequence to do so.  Any chance you could take a look at it? (be sure to log read-front.c)
Comment 13 Eric Anholt 2009-06-23 19:37:14 UTC
commit 5ca800e1006555ea1c5dcbbc56c35838c9f04994
Author: Eric Anholt <eric@anholt.net>
Date:   Mon Jun 22 16:33:29 2009 -0700

    dri2: Refresh the fake front contents after glXSwapBuffers().
    Bug #19177.
    Reviewed by: Ian Romanick <ian.d.romanick@intel.com>
Comment 14 zhao jian 2009-06-29 22:50:40 UTC
Verified with the following commits:
Libdrm:         (master)790097c51330090b2b7b90429b9ab8ddf259fd8e
Mesa:           (mesa_7_5_branch)bb8f3090ba37aa3f24943fdb43c4120776289658
Xserver:        (server-1.6-branch)dbac41b624e4aa86a6a184b7ebb52bfdd367bbf0
Xf86_video_intel:    (master)f53b3239dbc0ed723774e386e07ac9d8ce96bb89

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.