| Bug ID | 109265 |
|---|---|
| Summary | [regression, bisected] arb_texture_multisample texelfetch pig= lit test failing on NIR backend |
| Product | Mesa |
| Version | git |
| Hardware | Other |
| OS | All |
| Status | NEW |
| Severity | normal |
| Priority | medium |
| Component | Drivers/Gallium/radeonsi |
| Assignee | dri-devel@lists.freedesktop.org |
| Reporter | t_arceri@yahoo.com.au |
| QA Contact | dri-devel@lists.freedesktop.org |
For example: R600_DEBUG=3Dnir ./bin/texelFetch fs sampler2DMS 4 1x71-501x71 -auto -fbo Bisected to (on VEGA 64): commit f7ffa504a065dc2631fd38cc5fe885b277f4e7e7 (refs/bisect/bad) Author: Marek Ol=C5=A1=C3=A1k <marek.olsak@amd.com> Date: Sat Jul 29 01:40:48 2017 +0200 ac/surface: compute tile swizzle for GFX9 Tested-by: Dieter N=C3=BCtzel <Dieter@nuetzel-hh.de>
The commit has no effect on shaders. The bisecting has most li= kely been unsuccessful. Setting tile_swizzle =3D 0 in ac_surface.c reverts the commit= .
Note that f7ffa504a065dc2631fd38cc5fe885b277f4e7e7 indeed brok= e MSAA, but it was fixed by c5c6e0187fd.
The test passes with MESA_GLSL_CACHE_DISABLE=3D1.
This is odd. As far as I can tell the shader variant produced =
after loading
from the case is the same as the one produced when the cache is disabled.
Its also a fairly simple program. From tests/texturing/shaders/texelFetch.c=
the
offending program is:
int vs =3D piglit_compile_shader_text(GL_VERTEX_SHADER,
"#version 130\n"
"#extension GL_ARB_explicit_attrib_location: require\n&quo=
t;
"layout(location=3D0) in vec4 pos;\n"
"void main() {\n"
" gl_Position =3D pos;\n"
"}\n");
(void)!asprintf(&fs_code,
"#version 130\n"
"#extension GL_ARB_explicit_attrib_location: require\n&quo=
t;
"#extension GL_ARB_fragment_coord_conventions: require\n&q=
uot;
"uniform int layer;\n"
"uniform int si;\n"
"layout(pixel_center_integer) in vec4 gl_FragCoord;\n"
"layout(location=3D0) out %s o;\n"
"void main() {\n"
" o =3D %s(%sgl_FragCoord.x, gl_FragCoord.y, layer, si);\=
n"
"}\n",
sampler.return_type,
sampler.return_type,
sampler.data_type =3D=3D GL_UNSIGNED_INT ? "" : "=
;-");
We must not be setting a flag somewhere, but I've looked over the code that
handles the cache multiple time now and can't spot any real difference bet=
ween
tgsi and nir.
You were right it was a bad bisect due to that patch breaking = MSAA also. I found the real issue, fix sent to the list. https://patchwo= rk.freedesktop.org/patch/288934/
| What | Removed | Added |
|---|---|---|
| Status | NEW | RESOLVED |
| Resolution | --- | FIXED |
Fixed by: commit 7536af670b7501228628a8c90f9e8456b5aec9e1 Author: Timothy Arceri <tar= ceri@itsqueeze.com> Date: Wed Feb 27 18:26:07 2019 +1100 glsl: fix shader cache for packed param list Some types of params such as some builtins are always padded. We need to keep track of this so we can restore the list correctly. Here we also remove a couple of cache entries that are not actually required as they get rebuilt by the _mesa_add_parameter() calls. This patch fixes a bunch of arb_texture_multisample and arb_sample_shading piglit tests for the radeonsi NIR backend. Fixes: edded1237607 ("mesa: rework ParameterList to allow packing&= quot;) Reviewed-by: Marek Ol=C5=A1=C3=A1k <marek.olsak@amd.com>