Chrome Version: (copy from chrome://version)
OS: Android 5.0+ (persists in most recent update, not present in 4.4.4)
Device: Nexus 7, Adreno 320

    Android 4.4.4: OpenGL ES 3.0 V@53.0 AU@ (CL@)
    Android 5.0:   ?

This test fails:
https://www.khronos.org/registry/webgl/sdk/tests/conformance/glsl/bugs/varying-arrays-should-not-be-reversed.html
in Chrome but NOT in Firefox.

Prior discussion here:
https://github.com/KhronosGroup/WebGL/pull/2358

Initially it seems likely that there is a driver bug workaround difference between 4.4.4 and 5.0, but this doesn't appear to be the case. There is no difference in the generated shader in Chrome between 4.4.4 and 5.0, but there is a difference between Chrome and Firefox:

  varying highp float webgl_fb198836d046c6c3[3];
  uniform highp vec3 webgl_75c5f78948595750;
  attribute highp vec3 webgl_74509a83309904df;
  void main(){
+   (gl_Position = vec4(0.0, 0.0, 0.0, 0.0)); // these lines only appear in Chrome
+   (webgl_fb198836d046c6c3[2] = 0.0);
+   (webgl_fb198836d046c6c3[1] = 0.0);
+   (webgl_fb198836d046c6c3[0] = 0.0);
    (gl_Position = vec4(webgl_74509a83309904df, 1.0));
    (webgl_fb198836d046c6c3[0] = webgl_75c5f78948595750.x);
    (webgl_fb198836d046c6c3[1] = webgl_75c5f78948595750.y);
    (webgl_fb198836d046c6c3[2] = webgl_75c5f78948595750.z);
  }

The detail of the result is that instead of mapping x,y,z to 0,1,2 in the array, they instead map to 2,1,0. So there could be a regression in 5.0+ related to this fact. If we can modify the shader translator to reverse the order of SH_INIT_OUTPUT_VARIABLES, perhaps it would workaround the issue.


 

Chrome Version: 57.0.2987.132
OS: Android 5.0+ (persists in most recent update, not present in 4.4.4)
Device: Nexus 7, Adreno 320

    Android 4.4.4: OpenGL ES 3.0 V@53.0 AU@ (CL@)
    Android 5.0:   ?

This test fails:
https://www.khronos.org/registry/webgl/sdk/tests/conformance/glsl/bugs/varying-arrays-should-not-be-reversed.html
in Chrome but NOT in Firefox.

Prior discussion here:
https://github.com/KhronosGroup/WebGL/pull/2358

Initially it seems likely that there is a driver bug workaround difference between 4.4.4 and 5.0, but this doesn't appear to be the case. There is no difference in the generated shader in Chrome between 4.4.4 and 5.0, but there is a difference between Chrome and Firefox:

  varying highp float webgl_fb198836d046c6c3[3];
  uniform highp vec3 webgl_75c5f78948595750;
  attribute highp vec3 webgl_74509a83309904df;
  void main(){
+   (gl_Position = vec4(0.0, 0.0, 0.0, 0.0)); // these lines only appear in Chrome
+   (webgl_fb198836d046c6c3[2] = 0.0);
+   (webgl_fb198836d046c6c3[1] = 0.0);
+   (webgl_fb198836d046c6c3[0] = 0.0);
    (gl_Position = vec4(webgl_74509a83309904df, 1.0));
    (webgl_fb198836d046c6c3[0] = webgl_75c5f78948595750.x);
    (webgl_fb198836d046c6c3[1] = webgl_75c5f78948595750.y);
    (webgl_fb198836d046c6c3[2] = webgl_75c5f78948595750.z);
  }

The detail of the result is that instead of mapping x,y,z to 0,1,2 in the array, they instead map to 2,1,0. So there could be a regression in 5.0+ related to this fact. If we can modify the shader translator to reverse the order of SH_INIT_OUTPUT_VARIABLES, perhaps it would workaround the issue.