Bug 67046 - [radeonsi] pixmark_piano benchmark from geeks3D gputest fail with a llvm error output
Summary: [radeonsi] pixmark_piano benchmark from geeks3D gputest fail with a llvm erro...
Status: RESOLVED DUPLICATE of bug 63132
Alias: None
Product: Mesa
Classification: Unclassified
Component: Mesa core (show other bugs)
Version: git
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: mesa-dev
QA Contact:
URL:
Whiteboard:
Keywords:
: 68568 (view as bug list)
Depends on:
Blocks:
 
Reported: 2013-07-18 13:56 UTC by Laurent carlier
Modified: 2013-08-26 13:07 UTC (History)
1 user (show)

See Also:
i915 platform:
i915 features:

Attachments

Description Laurent carlier 2013-07-18 13:56:16 UTC 
gputest is available here:
http://www.geeks3d.com/gputest/download/

* Advanced Micro Devices, Inc. [AMD/ATI] Pitcairn XT [Radeon HD 7870 GHz Edition]
* Mesa 9.2.0-devel (git-e4fdf1b)
* llvm-3.4svn rev 186383
* kernel 3.10.0

[lordh@archMain GpuTest_Linux_x64_0.5.0]$ ./start_pixmark_piano_windowed_1024x640.sh
LLVM ERROR: Cannot select: 0x7ff8111889c0: ch = store 0x7ff811188ac0, 0x7ff81115d980, 0x7ff811178f20, 0x7ff81115d680<ST4[%580]> [ORD=661] [ID=580]
  0x7ff81115d980: i32 = Constant<0> [ID=11]
  0x7ff811178f20: i64 = add 0x7ff81115f5a0, 0x7ff811178d10 [ORD=405] [ID=84]
    0x7ff81115f5a0: i64 = FrameIndex<0> [ID=20]
    0x7ff811178d10: i64 = Constant<96> [ID=51]
  0x7ff81115d680: i64 = undef [ID=9]
In function: main
Comment 1 Michel Dänzer 2013-07-19 10:32:27 UTC 
AFAICT the real problem is this, visible with a debugging build of Mesa:

../../../../src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c:2190:lp_emit_declaration_soa: Assertion `idx < 256' failed.

This can be avoided by bumping LP_MAX_TGSI_TEMPS, but it looks like the app and/or glsl-to-tgsi aren't paying attention to the maximum number of temporaries advertised by the driver.
Comment 2 Aaron Watry 2013-07-19 13:30:45 UTC 
Yup, in this case it looks like it's trying to use 2403 temporaries... which is most definitely > 255

I bumped the max temporaries on my machine (Mint 15 amd64, Radeon 7850) to 4192 as a test, and the piano benchmark now renders (slowly, but it renders).

This is with git mesa/llvm, and a few not-yet-upstreamed patches to llvm which are targeted at OpenCL but may possibly affect this code.
Comment 3 Johannes Obermayr 2013-07-26 15:39:33 UTC 

*** This bug has been marked as a duplicate of bug 63132 ***
Comment 4 Michel Dänzer 2013-08-26 13:07:21 UTC 
*** Bug 68568 has been marked as a duplicate of this bug. ***

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.