FFe: new upstream release 9.1.1

[Timo Aaltonen]

The Ubuntu X team would like to push mesa 9.1.1 for raring. It has been on a staging ppa for roughly two months now, and we've tested it on a variety of hw. During this time we've identified a regression with the Intel driver having a slow blur effect, causing issues with kwin and Unity dash. This was why we didn't upload it to raring before it got fixed upstream.

For me personally, it also fixes a bug in the Intel driver where compiz hangs after returning from another user session.

Upstream release notes for 9.1 and 9.1.1:

9.1 New features

Note: some of the new features are only available with certain drivers.

    GL_ANGLE_texture_compression_dxt3
    GL_ANGLE_texture_compression_dxt5
    GL_ARB_ES3_compatibility
    GL_ARB_internalformat_query
    GL_ARB_map_buffer_alignment
    GL_ARB_shading_language_packing
    GL_ARB_texture_buffer_object_rgb32
    GL_ARB_texture_cube_map_array
    GL_EXT_color_buffer_float
    GL_OES_depth_texture_cube_map
    OpenGL 3.1 core profile support on Radeon HD2000 up to HD6000 series
    Multisample anti-aliasing support on Radeon X1000 series
    OpenGL ES 3.0 support on Intel HD Graphics 2000, 2500, 3000, and 4000

Bug fixes

TBD -- This list is likely incomplete.

Changes

    Removed VAAPI state tracker (unmaintained and broken)
    Removed i965's broken hardware implementation of GL_NV_vertex_program
    Removed swrast support for GL_NV_vertex_program
    Removed swrast support for GL_NV_fragment_program
    Removed OpenVMS support (unmaintained and broken)
    Removed makedepend build dependency

9.1.1 New features

None.

Bug fixes

This list is likely incomplete.

    Bug 30232 - [GM45] mesa demos spriteblast render incorrectly
    Bug 32429 - [gles2] Ironlake: gl_PointCoord takes no effect for point sprites
    Bug 38086 - Mesa 7.11-devel implementation error: Unexpected program target in destroy_program_variants_cb()
    Bug 57121 - [snb] corrupted GLSL built-in function results when using Uniform Buffer contents as arguments
    Bug 58042 - [bisected] Garbled UI in Team Fortress 2 and Counter-Strike: Source
    Bug 58960 - Texture flicker with fragment shader
    Bug 59495 - [i965 Bisected]Oglc fbblit(advanced.blitFb-3d-cube.mirror.both) fails
    Bug 59783 - [IVB bisected] 3DMMES2.0 Taiji performance reduced by ~13% with gnome-session enable compositing
    Bug 60121 - build - libvdpau_softpipe fails at runtime.
    Bug 60143 - gbm_dri_bo_create fails to initialize bo->base.base.format
    Bug 60802 - Corruption with DMA ring on cayman
    Bug 60848 - [bisected] r600g: add htile support cause gpu lockup in Dishonored wine.
    Bug 60938 - [softpipe] piglit interpolation-noperspective-gl_BackColor-flat-fixed regression
    Bug 61012 - alloc_layout_array tx * ty assertion failure when making pbuffer current
    Bug 61026 - Segfault in glBitmap when called with PBO source

=== TESTED ON

== Intel
haswell GT2 - tjaalton, sarvatt
haswell GT3 (CRW) - sarvatt
HD4000 (ivybridge) - tjaalton, sarvatt, marcin-jedrzejewski85, chrisle
HD3000 (sandybridge) - tjaalton, timo-jyrinki
HD2000 (sandybridge) - bdrung
HD "core" (arrandale/ironlake) - fabiomarconi, carla-sella, matttbe, pitti
GM45 (gen4) - anthony-balico, bryce
965GM (gen4) - tjaalton

== AMD/radeon

HD6750M (NI) - mlankhorst
HD6700 (NI) - nskaggs
HD6450 (NI) - tjaalton
HD4650 (RV730) - rapier

== NVIDIA/nouveau

GTX460m (NVC0/GF106) - sarvatt
GTX480 (NVC0) - mlankhorst
GF8600GT (NV50/G84) - tjaalton
GF8400m (NV50/G86) - mc3man
GF6600LE (NV43) - mlankhorst

== swrast (llvm)

virtualbox - saqman2060

[Timo Aaltonen (tjaalton)]

I've uploaded it to another staging ppa:

https://launchpad.net/~ubuntu-x-swat/+archive/x-staging

[Bryce Harrington (bryce)]

I've been testing the staging ppa with this, and have found it to be stable on my hardware. So +1 here.

[Maarten Lankhorst (mlankhorst)]

Piglit results for nouveau:
 9.0.3: 10118/11275
 9.1.0: 10297/11330
Some new tests went from skip to partially succeeding.
Some EXT_framebuffer_multisample tests started failing, but those seem to be flaky on all drivers, those failures can be ignored entirely.
A new nouveau specific crash occured in EXT_texture_compression_s3tc, it needs to be investigated, but looks like it could be fixed in a straightforward way.
Some vertex tests need looking at to determine whether they're flaky, but overall a lot of fixes have happened for nouveau.

[Maarten Lankhorst (mlankhorst)]

Piglit results for radeon:
 9.0.3: 10380/11420
 9.1.x: 10985/11449

Some regressions, mostly seem to be flaky tests or glsl failures, nothing big though, and expected because of the move towards the llvm shader compiler.

[Maarten Lankhorst (mlankhorst)]

piglit results for i915:
 9.0.3: 10750/11418
 9.1.x: 11146/11533

Similar results to radeon, some small new glsl failures, but no major new regressions. I have some patches to cherry pick to the 9.1 branch to fix the nouveau failures, which leaves only the s3tc crash as an outstanding issue.

[Maarten Lankhorst (mlankhorst)]

I tested the nouveau added the copy regressions to the git tree, and I verified they fixed the image copy regressions in piglit.

+1 from me now.

[Matthieu Baerts (matttbe)]

Hello,

@Timo Aaltonen: Is is maybe possible to include this tiny patch too?
    vbo: fix crash found with shared display lists: http://cgit.freedesktop.org/mesa/mesa/commit/?id=c2665aacdd14beec9580246736c602e3ff04d7ae

It fixes crashes in Cairo-Dock and Avogadro UI and it's already in the 9.1 branch but not in the 9.1.1 version: http://cgit.freedesktop.org/mesa/mesa/log/?h=9.1

[Timo Jyrinki (timo-jyrinki)]

Working fine with mobile Sandy Bridge (Asus Zenbook).

[Timo Aaltonen (tjaalton)]

Matthieu: yep, added to the git tree.

[Matthieu Baerts (matttbe)]

@Timo: thank you :)

[Benjamin Drung (bdrung)]

I have been using mesa from ppa:ubuntu-x-swat/x-staging since the weekend without any issues. I am using a Sandy Bridge system (Core i5-2400S).

[Timo Aaltonen (tjaalton)]

Thanks for the testing. Pitti reported today that blur on gen5 (ILK) has also regressed with the new version,. I've filed a new bug upstream to track it

https://bugs.freedesktop.org/show_bug.cgi?id=63283

so, the question is if it's acceptable to regress ILK until the bug is fixed. I also asked pitti how bad the performance was:

14:36 < pitti> tjaalton: it's certainly bearable, but it looks quite ugly

[Nicholas Skaggs (nskaggs)]

Community Testing for this can be seen here: http://packages.qa.ubuntu.com/qatracker/milestones/266/builds/41730/testcases/1515/results

[Timo Jyrinki (timo-jyrinki)]

Interestingly there's an Ironlake user there as well who didn't have problems.

[Timo Aaltonen (tjaalton)]

that's because the version being tested has the offending commit reverted, instead of pulling many commits from upstream that should've fixed it for all. I believe it's the best solution for now.

[Maarten Lankhorst (mlankhorst)]

Tested piglit on my fermi (nvc0) too. Similar results to my other card from nv50 series, except that with the copy commits already cherry picked I didn't hit those regressions first. :-)

[Adam Conrad (adconrad)]

Accepted.

[Launchpad Janitor (janitor)]

This bug was fixed in the package mesa - 9.1.1-0ubuntu2

---------------
mesa (9.1.1-0ubuntu2) raring; urgency=low

  * control, rules: Add llvm-3.2-dev to build-depends for powerpc, enable
    gallium llvm and shared libs (FTBFS).
 -- Timo Aaltonen <email address hidden> Wed, 17 Apr 2013 01:42:38 -0600