Bug 29784

Summary:	[regression] No shadows in Raytraced Shadows demo
Product:	Mesa	Reporter:	Sven Arvidsson <sa>
Component:	Mesa core	Assignee:	Ian Romanick <idr>
Status:	RESOLVED FIXED	QA Contact:
Severity:	normal
Priority:	medium
Version:	git
Hardware:	Other
OS:	All
URL:	http://www.humus.name/index.php?page=3D&ID=53
Whiteboard:
i915 platform:		i915 features:
Attachments:	Dump of shaders from Raytraced Shadows demo

Description Sven Arvidsson 2010-08-24 10:55:41 UTC

The glsl2 merge seems to have introduced a regression in the Raytraced Shadows demo from Humus, the shadows are no longer visible.

I have only tried this on r300g, as the demo fails to run on i965 with "Variable indexing of shader outputs unsupported".

http://www.humus.name/index.php?page=3D&ID=53

Comment 1 Marek Olšák 2010-08-24 11:03:35 UTC

Variable indexing of shader outputs is not supported by r300 either. I believe this has something to do with missing loop unrolling that was part of the previous GLSL compiler.

Comment 2 Sven Arvidsson 2010-08-24 12:50:21 UTC

Oh, in that case it's probably the same as bug 29654: It refuses to run with the same error on i965, and runs (but not correctly) on r300g.

Will retry this stuff once loop unrolling lands.

Comment 3 Ian Romanick 2010-08-30 14:50:02 UTC

I've been playing with this demo a bit today.  With my loop unrolling branch, setting SPHERE_COUNT to 1 (in Main.h) works.  Setting it to 2 or higher results in the whole box interior being shadowed.  Using MESA_GLSL=dump, the IR generated looks correct, so I'm not sure what's going wrong.

I'll keep digging.

Comment 4 Ian Romanick 2010-08-30 15:12:56 UTC

Created attachment 38323 [details]
Dump of shaders from Raytraced Shadows demo

The problem appears to be improper placement of varying arrays.  In the Mesa IR for linked fragment program 1 notice that INPUT[15] is used for both lightVec and sphereVec[1].

Comment 5 Ian Romanick 2010-08-30 16:20:58 UTC

With the commit listed below and the reset of my glsl2-loops branch, this demo renders correctly with all 7 spheres.  I will leave this bug open until glsl2-loops is merged to master.

commit df869d916308759fbacb227f60b1b6eda73131e2
Author: Ian Romanick <ian.d.romanick@intel.com>
Date:   Mon Aug 30 15:37:44 2010 -0700

    linker: Handle varying arrays, matrices, and arrays of matrices
    
    Fixes piglit test case glsl-array-varying-01.

Comment 6 Ian Romanick 2010-09-03 12:06:03 UTC

glsl2-loops branch is now merged, and this bug is fixed (at least on GM45 with i965_dri.so).

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.