From: Roland Scheidegger
The spec says with msaa disabled gl_SampleMaskIn is always 1.
This is not particularly related to arb_sample_shading, but drivers may
do different workarounds depending on the state of sample shading and the
presence of bits in the shader which would
From: Roland Scheidegger
I believe querying that information for GL_TEXTURE_BUFFFER via internal format
query should return the correct values, but it's definitely impossible if just
ARB_texture_buffer_object is supported but not GL 3.1. Hence just pretend it
succeeded in
From: Roland Scheidegger
The test fed the equivalent_pname into the get_unsupported_response()
function, which led to nonsense logged errors as this only recognizes
the original pnames.
Refactor pname/equivalent_pname translation so always the original pnames
are fed into
From: Roland Scheidegger
-128 and -127 represent exactly the same value (-1.0) for snorm8 values,
as do -32768/-32767 for snorm16. Therefore it seems quite reasonable that an
implementation would return the other representation (in particular r600 will
do this with a blit,
From: Roland Scheidegger
This just verifies that sampler arrays indexed with a uniform work with TBOs.
(Broken on r600 evergreen pending a fix.)
v2: fix compiler warnings, directly assign index uniform in shader
---
tests/all.py | 1 +
From: Roland Scheidegger
This just verifies that sampler arrays indexed with a uniform work with TBOs.
(Broken on r600 evergreen pending a fix.)
---
tests/all.py | 1 +
.../arb_texture_buffer_object/CMakeLists.gl.txt| 1 +
From: Roland Scheidegger
The max size reported by GL_MAX_TEXTURE_BUFFER_SIZE only guarantees that
buffers which contain that many texels can be sampled from. It does not mean
it is guaranteed the corresponding huge buffers can be allocated in the first
place. (r600 for
From: Roland Scheidegger
This is a problem of all texturing tests, they cannot exercise the tesselation
stages. (But I'm too lazy to fix the others, and too lazy to hack something up
for tcs stage, I only need it to verify a bug/fix with r600 buffer textures.)
---
From: Roland Scheidegger
The existing repeat and clamp modes are easy to implement. The
mirror_repeat (GL_MIRRORED_REPEAT) and mirror_clamp
(GL_MIRROR_CLAMP_TO_EDGE) modes however have some very interesting
differences to "gather is just the same as bilinear filtering without
From: Roland Scheidegger
The existing fbo-blending-formats test is mostly useless for anything but
unorm formats, since it does not test any values outside [0,1] (neither for
src, dst, intermeidate calculations, blend result).
I tried to enhance it but it got too complex, in
From: Roland Scheidegger
We expect non-nan results for these (according to d3d10 rules, and at least
for min/max, also according to ieee rules). The tests will not actually fail
with other results (since GL's NaN behavior is all undefined), albeit some
apps may rely on this.
From: Roland Scheidegger
Tolerance was added for the tests a while ago, but it looks like it was
forgotten for the nearest_biased test (the nearest one has it).
Also, for the linear test, add tolerance too when comparing x and y lodq
results - the values should probably be
From: Roland Scheidegger
The problem with the chosen depth values was that the depth values
naturally got viewport-transformed to values below 0.5 (for those
quads drawn) or above 0.5 (for those not drawn).
This allowed buggy implementations (in particular llvmpipe) to pass
From: Roland Scheidegger
The spec doesn't really say this should work in older versions. It was first
added in glsl 4.30, mentioning it was forgotten (initially part of
EXT_gpu_shader4, hence should have been added with 1.30), but with the wrong
syntax. Finally fixed in glsl
From: Roland Scheidegger
The test cannot work without this (and thus always failed if not present).
---
tests/spec/arb_viewport_array/render_viewport_2.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/tests/spec/arb_viewport_array/render_viewport_2.c
From: Roland Scheidegger
At least mesa/st fails this right now (always uses the initial format for mip
gen).
v2: don't use piglit_display (not actually drawing anything), suggested
by Ilia Mirkin.
---
tests/all.py | 1 +
From: Roland Scheidegger
At least mesa/st fails this right now (always uses the initial format for mip
gen).
---
tests/all.py | 1 +
tests/spec/arb_texture_view/CMakeLists.gl.txt | 1 +
tests/spec/arb_texture_view/mipgen.c | 102
From: Roland Scheidegger
The rect halves comparison is in fact not valid in general. The reason is that
the same geometry does not have the same precision even if it it just shifted
by some fixed amount in one direction.
As an example, some calculated x value near 7 will be
From: Roland Scheidegger
The rect halves comparison is in fact not valid in general. The reason is that
the same geometry does not have the same precision even if it it just shifted
by some fixed amount in one direction.
As an example, some calculated x value near 7 will be
From: Roland Scheidegger
This test is quite mean to mesa's draw pipe. llvmpipe/softpipe fail even if
front and back fill mode is the same (for points and aa lines as the latter
get rendered as tris finally again). Most everything else fails too, before
the driver dies in a
From: Roland Scheidegger
Previously, the test was actually using all-zeros in the end for the image,
so unsurprisingly it passed, except when border color was used.
Add the special stencil case handling in a couple more cases.
This will still not test depth/stencil textures
From: Roland Scheidegger
This is more of a clip test involving the system-interpreted values (which
cannot be interpolated). This exposes some problems in gallium for drivers
using draw, where clipping such values was completely broken, thus layered
rendering being generally
From: Roland Scheidegger srol...@vmware.com
Just using tri strip instead of tri fan.
Exposing more bugs in llvmpipe/draw...
---
...imitive-id-no-gs-strip-first-vertex.shader_test | 62 ++
.../execution/primitive-id-no-gs-strip.shader_test | 60 +
2 files
From: Roland Scheidegger srol...@vmware.com
The test takes ages (somewhere in the order of an hour) on llvmpipe, because
the driver announces support for one million indirections (some drivers
actually say billions even). It never gets compiled even, that hour is fully
spent on string
From: Roland Scheidegger srol...@vmware.com
GL_REPEAT is not legal for the rectangle targets, and because the test uses
sampler objects the default GL_REPEAT value will be used even for rectangle
target because unlike texture objects they can't be initialized to the legal
CLAMP_TO_EDGE value.
From: Roland Scheidegger srol...@vmware.com
line loops are difficult to handle because when trying to accumulate vertex
data in a buffer the loop must be closed at the end - this is problematic if
using any kind of fixed size buffer. With immediate mode api (presumably
display lists have similar
From: Zack Rusin za...@vmware.com
Increase the subpixel precision to 8 bits. This requires
changing some of the code to 64 bits to avoid overflows.
v2 (sroland): Query GL for the number of subpixel bits and use that
instead of a fixed 8 bits. Note that mesa drivers don't set this yet
From: Roland Scheidegger srol...@vmware.com
By the looks of it default is not required to appear as last statement
in a switch expression, and c rules should be followed (which is probably
a mess to implement thanks to fallthrough).
Seems to fail with mesa glsl compiler (at least with tgsi
From: Roland Scheidegger srol...@vmware.com
Similar to arb_shader_texture_lod-texgrad, but using cube maps instead.
Given the somewhat undefined behavior of explicit gradients with cube
maps, the main purpose of the test is really to test that those work at
all, as there doesn't seem to be any
From: Roland Scheidegger srol...@vmware.com
This is pretty rough and doesn't really test all that much (just
textureOffsetLod, but there's other texturing functions with offsets),
doesn't test different wrap modes, NPOT sizes etc., but there's no
other test using ordinary texture opcodes and
From: Roland Scheidegger srol...@vmware.com
Similar to glsl-fs-main-return (and glsl-vs-main-return), this is testing
using return in main. Contrary to the these other tests, this hits both
the cases where the return path is and is NOT taken (the gallivm code
got it wrong and always did an early
From: Roland Scheidegger srol...@vmware.com
Similar to glsl-fs-main-return (and glsl-vs-main-return), this is testing
using return in main. Contrary to the these other tests, this hits both
the cases where the return path is and is NOT taken (the gallivm code
got it wrong and always did an early
From: Roland Scheidegger srol...@vmware.com
This is pretty rough and doesn't really test all that much (just
textureOffsetLod, but there's other texturing functions with offsets),
doesn't test different wrap modes, NPOT sizes etc., but there's no
other test using ordinary texture opcodes and
From: Roland Scheidegger srol...@vmware.com
This is pretty rough and doesn't really test all that much (just
textureOffsetLod, but there's other texturing functions with offsets),
doesn't test different wrap modes, NPOT sizes etc., but there's no
other test using ordinary texture opcodes and
From: Roland Scheidegger srol...@vmware.com
core GL specifies out-of-bound texel fetches return undefined results,
except in a robust context with ARB_robust_buffer_access_behavior supported
(which is core in 4.3), in which case texel fetch will return 0 (OpenGL 4.3
compatibility profile, page
From: Roland Scheidegger srol...@vmware.com
core GL specifies out-of-bound texel fetches return undefined results,
except in a robust context with ARB_robust_buffer_access_behavior supported
(which is core in 4.3), in which case texel fetch will return 0 (OpenGL 4.3
compatibility profile, page
From: Roland Scheidegger srol...@vmware.com
Make sure clipping is needed sometimes, and more often use small index counts,
to expose issues and excercise more paths in mesa's draw module.
---
tests/spec/arb_robustness/draw-vbo-bounds.c |4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
37 matches
Mail list logo