Bug 108321 - rv610: corrupt shader output with SB if_conversion optimization pass
Summary: rv610: corrupt shader output with SB if_conversion optimization pass
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/Gallium/r600 (show other bugs)
Version: git
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: Default DRI bug account
QA Contact: Default DRI bug account
Depends on:
Reported: 2018-10-11 01:57 UTC by Nicholas Bishop
Modified: 2019-09-18 19:26 UTC (History)
0 users

See Also:
i915 platform:
i915 features:

rv610 sbdump (42.47 KB, text/x-matlab)
2018-10-11 01:57 UTC, Nicholas Bishop
screenshot of the confetti graphics bug (75.42 KB, image/png)
2018-10-11 20:36 UTC, Nicholas Bishop

Description Nicholas Bishop 2018-10-11 01:57:19 UTC
Created attachment 141984 [details]
rv610 sbdump

I'm debugging some graphics corruption in Chromium on an iMac7,1 with rv610 graphics. Basically an area of the application has randomish blocks of color, like it's reading from an uninitialized texture.

I've narrowed the problem down to a fragment shader:

    varying mediump vec2 _uv_texCoord;
    uniform lowp sampler2D _ulut_texture;
    uniform mediump float _ulut_size;

    mediump vec4 _uLUT(in lowp sampler2D _usampler, in mediump vec3 _upos, in mediump float _usize){
      (_upos *= (_usize - 1.0));
      mediump float _ulayer = min(floor(_upos.z), (_usize - 2.0));
      (_upos.xy = ((_upos.xy + vec2(0.51234001, 0.51234001)) / _usize));
      (_upos.y = ((_upos.y + _ulayer) / _usize));
      return mix(texture2D(_usampler, _upos.xy), texture2D(_usampler, (_upos.xy + vec2(0, (1.0 / _usize)))), (_upos.z - _ulayer));

    void main(){
      mediump vec2 _utexCoord = _uv_texCoord;
      mediump vec4 _utexColor = texture2D(_us_texture, _utexCoord);
      if ((_utexColor.w > 0.0))
        (_utexColor.xyz /= _utexColor.w);
      (_utexColor.xyz = _uLUT(_ulut_texture, _utexColor.xyz, _ulut_size).xyz);
      (_utexColor.xyz *= _utexColor.w);
      (gl_FragColor = vec4(_utexColor.xyz, 1.0));

(Note that in the original shader "0.51234001" is just "0.5", I just changed it to make grepping easier.) I played around with changes to the _uLUT function, and I found that any mixing of the two samples seems to trigger the bug, e.g. I can mix with a constant 0.5 and the bug still happens. I also tried replacing "mix" with an explicit a*(x-1)+b*x; no change.

I found that disabling SB fixed the issue, and in particular just commenting out the "if_conversion" SB pass fixes the issue.

I've attached the full sbdump output. Please let me know if there are more details I can provide that would help.
Comment 1 Nicholas Bishop 2018-10-11 20:36:58 UTC
Created attachment 141999 [details]
screenshot of the confetti graphics bug
Comment 2 GitLab Migration User 2019-09-18 19:26:22 UTC
-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/mesa/mesa/issues/646.

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.