Bug 89647

Summary:	dEQP-GLES3: mix does not produce correct results with infinities
Product:	Mesa	Reporter:	Iago Toral <itoral>
Component:	Drivers/DRI/i965	Assignee:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Status:	RESOLVED WONTFIX	QA Contact:	Intel 3D Bugs Mailing List <intel-3d-bugs>
Severity:	normal
Priority:	medium	CC:	itoral
Version:	git
Hardware:	Other
OS:	All
Whiteboard:
i915 platform:		i915 features:

Description Iago Toral 2015-03-18 08:07:25 UTC

List of affected tests:

dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_vertex.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_vertex.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_vertex.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_vertex.vec4
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_fragment.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_fragment.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_fragment.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.lowp_fragment.vec4
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_vertex.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_vertex.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_vertex.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_vertex.vec4
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_fragment.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_fragment.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_fragment.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.mediump_fragment.vec4
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_vertex.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_vertex.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_vertex.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_vertex.vec4
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_fragment.scalar
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_fragment.vec2
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_fragment.vec3
dEQP-GLES3.functional.shaders.builtin_functions.precision.mix.highp_fragment.vec4

This is the same problem described for modf in bug 89645 but for mix.

In gen6+, mix() is implemented as an lrp hardware operation, so I guess there is no way around this other than adding conditionals to check for infinities and resolve them directly.

An alternative would be to implement it as gen5 (when lrp was not available in hardware) by doing x*(1-a)+y*a, which works better in some cases but still produces NaN for some (specifically when we need to multiply infinity with 0). This method should have a performance penalty.

Comment 1 Matt Turner 2017-02-13 19:46:16 UTC

These tests now all pass after

commit 6c1b3db5fb439806b9b67442f1042615ffa32816
Author: Pyry Haulos <phaulos@google.com>
Date:   Fri May 29 14:06:53 2015 -0700

    Allow alternative mix() implementation
    
    Some compilers translate mix(a, b, x) to a + (b - a)*x which produces
    slightly different results than the form a*(1 - x) + b*x suggested by
    the GLSL specification. This change relaxes tests to allow both forms.
    
    Bug: 21110253
    Change-Id: I3af056145ef8e64c0fc998302770fd38d105481a

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.