Age | Commit message (Collapse) | Author |
|
tracking. Generates the JSON and dispatch files under /out rather than in tree.
|
|
so they can be used for glslparstertest_gles2
|
|
tests. These tests are a processed copy of
tests/glslparsertest/shaders to support glsl es 1.00.
|
|
and associated testcases which are modified copies of
tests/glslparsertest/glsl2 to support glsl es 1.00.
|
|
to the tests built.
|
|
variable-index-write.sh. Default precision added.
|
|
Android into the data partition.
|
|
support android with precision default settings and glsl es 1.00
|
|
tests/spec/glsl-es-1.00/compiler/Android.mk
|
|
|
|
by variable-index-read.sh & variable-index-write.sh
|
|
|
|
files onto the data partition.
|
|
Adds Android.mk to build invalid-es3-queries & minmax. Test
binaries are installed into /system/xbin/piglit/
|
|
Adds Android.mk to copy the glsl es 1.00 shader test
sanity.shader_test to /data/shader-data/execution/
|
|
Adds Android.mk to build shader_runner.
|
|
Adds Android.mk to build libpiglitutil_gles2 as well as
autogenerate config.h, and generated_dispatch.ch
|
|
This are Adrian M Negreanu's changes for piglit to support Android.
These are included in this patch series for the purposes of completeness.
|
|
|
|
Add top of the tree Android.mk, sets piglit_top
|
|
|
|
strchrnul is used in shader_runner.c (and no where else in all
of piglit). Unfortunately bionic, Android's c library does not
include this function. I've writen an implementation of
strchrnul which is only used when compiling for Android.
|
|
On Android -Werror=return-type is set for C code, as such there
are two cases were I've added a return 0 after an assert.
|
|
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 exit which got unnoticed by the
existing tests, see https://bugs.freedesktop.org/show_bug.cgi?id=62357).
v2: use mod() instead of integer arithmetic suggested by Ian Romanick.
This gets rid of the glsl 1.30 requirement. And do minor simplifications.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
|
|
It will be added based on the GLSL requirement instead.
Also make it clearer that these shader_test files are automatically generated.
Same number of passes on Sandy Bridge running all.tests without glean.
v2: Add missing comment #
Signed-off-by: Stuart Abercrombie <sabercrombie@chromium.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Tom Gall <tom.gall@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
|
|
This test provides a complex test of using interface blocks.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Stuart Abercrombie <sabercrombie@chromium.org>
|
|
A few tests to make sure that interface blocks are defined
the same during intrastage and interstage linking.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Stuart Abercrombie <sabercrombie@chromium.org>
|
|
VISUAL_RGBA is apparantly lies
|
|
Catches an assertion failure/rendering failure in i965.
|
|
|
|
Adds gl_MaxCombinedTextureImageUnits.shader_test
gl_MaxDrawBuffers.shader_test
gl_MaxFragmentUniformVectors.shader_test
gl_MaxTextureImageUnits.shader_test
gl_MaxVaryingVectors.shader_test
gl_MaxVertexAttribs.shader_test
gl_MaxVertexTextureImageUnits.shader_test
gl_MaxVertexUniformVectors.shader_test
These check the built-in constants of the same name from
section 7.4 of the glsl es 1.0.17 spec.
v2: squash patch, address one by one spec reference.
Signed-off-by: Tom Gall <tom.gall@linaro.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
|
|
Signed-off-by: Brian Paul <brianp@vmware.com>
|
|
It's only called from main(), which was already disabled.
Signed-off-by: Brian Paul <brianp@vmware.com>
|
|
Signed-off-by: Brian Paul <brianp@vmware.com>
|
|
waffles-decompressed-etc2-r11-64x32-miptree.ktx contains per pixel RGBA data
generated by ericsson's etcpack tool. While generating decompressed image
for COMPRESSED_R11_EAC format, RED color data for each pixel is copied to
GREEN and BLUE channels as well. Using this image for texturing in OpenGL ES
3.0 (internalFormat=GL_R8) causes GL_INVALID_OPERATION. This is because
glTexImage2D() in GL ES 3.0 doesn't allow internalFormat = GL_R8 with
format= GL_RGBA. To workaround this issue use internalFormat = GL_RGBA and
mask all color channels except Red.
Other option is to edit the waffles-decompressed-etc2-r11-64x32-miptree.ktx
file and set all GREEN, BLUE components to zero. Considering the hassle involved
in editing image file, I won't prefer that.
This fixes the piglit failure on gles3 with COMPRESSED_R11_EAC format.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
|
|
Use piglit_check_gl_error(). Note: we're getting an INVALID_OPERATION
error from glBlitFramebuffer() with NVIDIA's driver. That'll be
investigated/fixed separately. At least the test now reports FAIL
instead of dying on an assertion.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
|
|
Don't allocate 1MB arrays on the stack. Use heap allocations instead.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
|
|
Use piglit_check_gl_error().
Reviewed-by: José Fonseca <jfonseca@vmware.com>
|
|
NVIDIA's later GPUs/drivers report 192 combined texture units.
This causes us to read outside the too-small arrays and segfault.
Increase the array sizes and add a check for the future.
Reviewed-by: José Fonseca <jfonseca@vmware.com>
|
|
|
|
When a mipmapped depth texture's size satisfies size > 1 and (size %
16) == 1, this causes miplevel 1 to have a size that is a multiple of
8, but not to be aligned on a multiple of 8 boundary. This turns out
to be an important corner case for the i965 driver.
Note: in an ideal world, I'd modify depthstencil-render-miplevels to
simply test every possible buffer size from 1 to some reasonably large
size (perhaps 256), but this would produce an impractically large
number of test cases. So instead I'm just testing the sizes that have
been known to cause problems.
|
|
No functional change.
Reviewed-by: Eric Anholt <eric@anholt.net>
|
|
v2:
- Generate tests in the generated_tests directory tree at compile time
|
|
Functions tested:
abs, abs_diff, add_sat, hadd, rhadd, clz, clamp, mad_hi,
mad_sat, max, min, mul_hi, rotate, sub_sat, upsample, mad24, mul24
Data Types tested:
char, uchar, short, ushort, int, uint, long, ulong
Sizes tested:
Scalar, Vector sizes 2/4/8/16
v2:
- Move file into generated_tests and rename to generate-cl-int-builtins.py
- Generate tests in cl/builtin/int (creating dir if necessary)
|
|
|
|
Clear the large FBO to a non-black color and probe for it in the
final test.
This uncovers a bug in Mesa's llvmpipe driver where we ran out of
bin commmand memory and skipped some tiles, leaving them black.
Setting the clear color to white (and actually clearing the FBO)
helps to detect that.
|
|
|
|
|
|
Move the re.compile into the core.Enivironment constructor, which
reduces code duplication. It also allows us to pass environment data on
initilization of the object, rather that having edit it's attributes
individually.
V2: - Does not remove deprecated options, only marks them as such
V3: - Fixes deperecated warning for tests from V2 always being triggered
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
|
|
This change gets us cleaner, more readable argument parsing code.
Another advantage of this approach is that it automatically generates
help menus, which means options will not be implemented without a help
entry (like --resume)
V2: - Does not remove deprecated options, only marks them as deprecated
V3: - Fixes deprecated warning for tests from V2 always being triggered
Signed-off-by: Dylan Baker <baker.dylan.c@gmail.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
|