Interestingly, copying the i965_dri.so from 10.2.8 fixes the problem, so it seems like a regression with 10.3.

Hi, same issue here.

Video capture: https://www.youtube.com/watch?v=jACO53o1CwU

Game: unvanquished 0.33.0-1~deb from http://www.unvanquished.net/download

System:
- Debian GNU/Linux testing
- Linux sark 3.16.0-4-amd64 #1 SMP Debian 3.16.7-2 (2014-11-06) x86_64 GNU/Linux
- xserver-xorg 1.16.1.901
- xserver-xorg-video-intel 2.21.15

$ glxinfo | grep OpenGL | grep string
OpenGL vendor string: Intel Open Source Technology Center
OpenGL renderer string: Mesa DRI Intel(R) Haswell Mobile 
OpenGL core profile version string: 3.3 (Core Profile) Mesa 10.3.2
OpenGL core profile shading language version string: 3.30
OpenGL version string: 3.0 Mesa 10.3.2
OpenGL shading language version string: 1.30
OpenGL ES profile version string: OpenGL ES 3.0 Mesa 10.3.2
OpenGL ES profile shading language version string: OpenGL ES GLSL

I just bisected this between mesa-10.2 and mesa-10.3
afe5db3293d4d04d7dae342c537f18db00ad09a5 is the first bad commit
commit afe5db3293d4d04d7dae342c537f18db00ad09a5
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Thu Aug 7 20:07:25 2014 -0700

    i965: Calculate start/base_vertex_location after preparing vertices.
    
    Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
    Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
    (cherry picked from commit c89306983c07e5a88c0d636267e5ccf263cb4213)

:040000 040000 0113d7047ea4c7b74d743ccca1bfc92efe07c74c 13c5da0511572cd784054746c05afef567b52f2c M      src



git bisect log:
git bisect start
# good: [f836ef63fdbb4abada29299e226eed735094736c] Bump version to 10.2 (final)
git bisect good f836ef63fdbb4abada29299e226eed735094736c
# bad: [1b12af300dfa77c24088780e88200703653293d3] docs: Update 10.3 release notes
git bisect bad 1b12af300dfa77c24088780e88200703653293d3
# good: [a06c9791d1b7fcedfb56ecbdc601d42fab196916] docs: Add missing release notes for ARB_separate_shader_objects
git bisect good a06c9791d1b7fcedfb56ecbdc601d42fab196916
# good: [282b783a158a495a86537a8a4858c8c1c55f71fb] st/mesa: add some missing MESA/PIPE_FORMAT_R10G10B10A2_UNORM switch cases
git bisect good 282b783a158a495a86537a8a4858c8c1c55f71fb
# good: [58007aec4193e3bdb38cafcfb2188dcbd8f18def] i965/fs: don't read from uninitialized memory while assigning registers
git bisect good 58007aec4193e3bdb38cafcfb2188dcbd8f18def
# good: [fbfcd671a1c74eebc00c56f609a8792fedb0a45d] i965/fs: Add support for non-const sampler indices in generator
git bisect good fbfcd671a1c74eebc00c56f609a8792fedb0a45d
# good: [498dc676ea7efac9bce490006a4f5b7f81e9e458] r600g: copy IA_MULTI_VGT_PARAM programming from radeonsi for Cayman
git bisect good 498dc676ea7efac9bce490006a4f5b7f81e9e458
# good: [c546523b4df5d613f4a97fc2068315d50cdb365c] r600g: fix alpha-test with HyperZ enabled, fixing L4D2 tree corruption
git bisect good c546523b4df5d613f4a97fc2068315d50cdb365c
# good: [73192345c333cb7cfcc6210bf085b6d87bdee423] mesa: Add SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
git bisect good 73192345c333cb7cfcc6210bf085b6d87bdee423
# bad: [0f4dc098076cd6ccac6a117cabf3df0194558709] glsl: Speed up constant folding for swizzles.
git bisect bad 0f4dc098076cd6ccac6a117cabf3df0194558709
# bad: [10aee701ae10f2057f8d24792b56c8f832830a99] i965: Make gl_BaseVertex available in a buffer object.
git bisect bad 10aee701ae10f2057f8d24792b56c8f832830a99
# good: [09a763bea551a880449265e558384392214ecf96] mesa: Replace string comparisons with SYSTEM_VALUE enum checks.
git bisect good 09a763bea551a880449265e558384392214ecf96
# good: [d9df31cc6edf9a01d5093d40e0fcebb591693cd8] i965: Handle SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
git bisect good d9df31cc6edf9a01d5093d40e0fcebb591693cd8
# bad: [afe5db3293d4d04d7dae342c537f18db00ad09a5] i965: Calculate start/base_vertex_location after preparing vertices.
git bisect bad afe5db3293d4d04d7dae342c537f18db00ad09a5
# first bad commit: [afe5db3293d4d04d7dae342c537f18db00ad09a5] i965: Calculate start/base_vertex_location after preparing vertices.

Created attachment 110839 [details]
A patch that "fixes" the issue

The patch just partially reverts the original commit that introduced the regression. It is probably not a proper fix but my help provide clues about where the problem actually occurs.

Ken I cannot see this commit nominated for stable yet it seems that Ian has explicitly picked it up. Would you have any ideas why this is happening ?

Guys to avoid chasing ghosts can you confirm if the problem exists with mesa 10.3.5 and 10.4.0 ?

And by "why this is happening" I'm referring about the bug :)

I can confirm it happens on both and the current master.

(In reply to Emil Velikov from comment #5)
> Ken I cannot see this commit nominated for stable yet it seems that Ian has
> explicitly picked it up.

It was part of a series that was needed for OpenGL ES 3.0 conformance.

(In reply to vcelestialragev from comment #4)
> Created attachment 110839 [details]
> A patch that "fixes" the issue
> 
> The patch just partially reverts the original commit that introduced the
> regression. It is probably not a proper fix but my help provide clues about
> where the problem actually occurs.

I'm pretty sure your patch is correct, actually.

After ~2 hours of staring at the code, and adding some assertions, I realized that if we have multiple _mesa_prims and somehow don't flag BRW_NEW_VERTICES on the second one, brw_try_draw_prims resets start/base_vertex_location to prim[i].{start,basevertex}, but we don't call brw_prepare_shader_draw_parameters, so we fail to add in brw->vb.start_vertex_bias and brw->ib.start_vertex_offset when processing subsequent primitives.

Your patch makes us no longer add brw->vb.start_vertex_bias to the value of gl_BaseVertexARB.  However, I think we aren't *supposed* to add it in the first place, so the patch actually fixes two bugs.

I'd like to revert more, restoring more of the older/simpler code.  I'll send out a patch shortly.

Fixed by:

commit 929b48510b083919a434f01afe464e2bb7f26ac0
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Thu Dec 18 04:45:40 2014 -0800

    i965: Fix start/base_vertex_location for >1 prims but !BRW_NEW_VERTICES.