Since VkRunner commit 6b1d2f51efad6, there are now alternative names
for the commands for setting push constants and UBO data. These names
are a bit less confusing so this patch converts all of the VkRunner
test scripts to use them.
---
.../face-forward-double.vk_shader_test| 80
This will make it easier to distinguish tests written for Vulkan. It
makes a bit of an inconsistency because the GL and CL tests don’t have
any prefix. Ideally maybe we would add a prefix for those too, but
changing the test names at this late stage would probably cause a lot
of hassle.
Adding
gl_BaseVertex has slightly different semantics under Vulkan and GL so
this is worth testing.
---
.../base-vertex.vk_shader_test| 66 +++
1 file changed, 66 insertions(+)
create mode 100644
tests/vulkan/shader_draw_parameters/base-vertex.vk_shader_test
diff --git
All but one of these are currently broken on anv.
See: https://patchwork.freedesktop.org/series/40797/
---
.../block-layout-location.vk_shader_test | 117 +
...lock-member-layout-location.vk_shader_test | 68 ++
...block-mixed-layout-location.vk_shader_test | 120
These are unique to SPIR-V so they are interesting to test.
---
.../unordered-comparison.vk_shader_test | 780 ++
1 file changed, 780 insertions(+)
create mode 100644 tests/vulkan/shaders/unordered-comparison.vk_shader_test
diff --git
For a while there was a bug in anv which meant that the interpolate
functions wouldn’t work if the thing being interpolated contained a
vertex component selector. Adding these tests will help make sure we
don’t regress that.
See: https://patchwork.freedesktop.org/patch/237095/
---
Looks good to me
Reviewed-by: Neil Roberts
- Neil
Alejandro Piñeiro writes:
> Equivalent to the GLSL tests added with the series "Test cases for
> large GLSL local arrays". Quoting from the cover-letter of that
> series:
>
>"Drivers typically need sp
Previously the filename was the full path to the test. This would
break if the filename was serialized and installed to a different
location.
The code is based on what is done for ASMParserTest. However I have a
suspicion that this won’t work for running generated tests from an
out-of-tree build.
---
CMakeLists.txt | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index f7d23fee1..b416cacb2 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -537,7 +537,7 @@ install (
install (
DIRECTORY tests
DESTINATION
---
framework/profile.py| 4 +++-
tests/CMakeLists.no_api.txt | 11 +++
tests/find_static_tests.py | 5 -
tests/serializer.py | 7 ++-
4 files changed, 24 insertions(+), 3 deletions(-)
diff --git a/framework/profile.py b/framework/profile.py
index
This will be retrieved when generating the XML profile for the
VkRunner tests.
---
framework/test/piglit_test.py | 2 ++
1 file changed, 2 insertions(+)
diff --git a/framework/test/piglit_test.py b/framework/test/piglit_test.py
index c80e355d1..4d3008139 100644
---
As requested by Dylan in the review for the patch series to add
VkRunner as an external dependency¹, this series adds the missing
serialization support.
1. https://lists.freedesktop.org/archives/piglit/2018-November/025336.html
Neil Roberts (4):
Cmake: Install vk_shader_test files
framework
This searches for files named *.vk_shader_test in the tests/vulkan
directory and runs them with VkRunner. VkRunner is executed as an
external dependency. It is found either with the vkrunner:bin config
option, by setting the PIGLIT_VKRUNNER_BINARY environment variable, or
just in the search path.
for the build system.
Neil Roberts (2):
framework: Add a vulkan tests profile
vulkan: Add some tests for glsl450 builtin functions using doubles
framework/test/piglit_test.py | 17
tests/vulkan.py | 33 +++
.../face-forward-double.vk_shader_test
This searches for files named *.vk_shader_test in the tests/vulkan
directory and runs them with VkRunner. VkRunner is executed as an
external dependency. It is found either in the path or by setting the
PIGLIT_VKRUNNER_BINARY environment variable.
v2: Move VkShaderTest to piglit_test.py and
---
.../face-forward-double.vk_shader_test| 88 +++
.../glsl450/frexp-double.vk_shader_test | 61 +
.../glsl450/isinf-double.vk_shader_test | 81 +
.../glsl450/reflect-double.vk_shader_test | 55
ry hits each pixel and count
it that way rather than relying on downsampling a high-res test image.
Regards,
- Neil
> -Mark
>
> Neil Roberts writes:
>
>> The unlit and totally_lit stats are supposed to count the pixels where
>> either a primitive completely covers the pixel
Neil Roberts <nrobe...@igalia.com> writes:
> Eventually we could probably get a lot of testing fairly cheaply by
> modifying some of the generated_tests scripts to also generate Vulkan
> tests. It might even be worth semi-automatically converting existing
> GL shader tests using
This searches for files named *.vk_shader_test in the tests/vulkan
directory and runs them with VkRunner.
Reviewed-by: Dylan Baker
v2: Move VkShaderTest to piglit_test.py and rename to VkRunnerTest.
Add future imports. Remove unused import.
---
Note that the script is not run automatically as part of the build
process and instead vr-format-table.h is checked into git.
---
I’ve split the script out to make it easier to review.
Thanks for the feedback Dylan.
tests/vulkan/vkrunner/make-formats.py | 137 ++
The sixth patch seems to have gotten stuck in the moderator queue for
being too big. In the meantime I’ve also pushed a branch here:
https://github.com/Igalia/piglit/commits/nroberts/vkrunner
- Neil
Neil Roberts <nrobe...@igalia.com> writes:
> This series sets up a framework for testi
This searches for files named *.vk_shader_test in the tests/vulkan
directory and runs them with VkRunner.
---
framework/test/vk_shader_test.py | 39 +++
tests/vulkan.py | 30 ++
2 files changed, 69 insertions(+)
---
.../glsl450/face-forward-double.vk_shader_test | 88 ++
tests/vulkan/glsl450/frexp-double.vk_shader_test | 61 +++
tests/vulkan/glsl450/isinf-double.vk_shader_test | 81
tests/vulkan/glsl450/reflect-double.vk_shader_test | 55
This defaults to building only on Linux as with the GLES platforms
because later patches add code using fork/exec which will need porting
to compile on Windows.
---
CMakeLists.txt | 7 +++
cmake/piglit_util.cmake| 2 +-
cmake/target_api/CMakeLists.txt
They will be equally useful on Vulkan.
---
tests/util/piglit-util-gl.c | 90 -
tests/util/piglit-util-gl.h | 29 ---
tests/util/piglit-util.c| 89
tests/util/piglit-util.h| 28
/
Neil Roberts (8):
util: Add a PIGLIT_STRINGIFY macro
util: Add a piglit_fatal utility
util: Add a libutil_vulkan library
util: Move the half float functions from util-gl to util
framework: Add a Vulkan platform
Import VkRunner
framework: Add a vulkan tests profile
vulkan: Add some tests
This prints a formatted error and then aborts.
---
tests/util/piglit-util.c | 16
tests/util/piglit-util.h | 2 ++
2 files changed, 18 insertions(+)
diff --git a/tests/util/piglit-util.c b/tests/util/piglit-util.c
index ed7d21c95..a79db3fb7 100644
--- a/tests/util/piglit-util.c
---
tests/util/piglit-util.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/tests/util/piglit-util.h b/tests/util/piglit-util.h
index 0a5eab810..3dfb5d6c5 100644
--- a/tests/util/piglit-util.h
+++ b/tests/util/piglit-util.h
@@ -162,6 +162,8 @@ piglit_register_subtests(const char *names[]);
This is currently no different from libutil but it is added anyway to
match the other APIs.
---
tests/util/CMakeLists.vulkan.txt | 15 +++
tests/util/piglit-util-vulkan.c | 1 +
2 files changed, 16 insertions(+)
create mode 100644 tests/util/CMakeLists.vulkan.txt
create mode
These tests set up either UBOs or SSBOs containing different values in
order to render squares with different colors, sizes and rotations.
However the test was only probing the center pixel of the square so it
wouldn’t actually matter if the rotation or size was wrong. The
rotation is the only
The test can now take an extra argument ‘indirect’ which will cause
it to emit the draw calls via an indirect buffer. This is useful to
test at least on i965 because emitting the gl_BaseVertex takes a
different path in that case.
---
tests/all.py | 10 +-
The ‘first’ parameter should not affect the value of gl_BaseVertex put
it should be added to gl_VertexID. This patch adds an extra rectangle
to the drawn area to test a non-zero ‘first’ parameter. This is worth
testing because Mesa is currently getting this wrong.
---
Oops, I used the wrong message ID in --in-reply-to. That’s what I get
for trying to be too clever with git-send-email. Sorry about that.
Regards,
- Neil
Neil Roberts <nrobe...@igalia.com> writes:
> Timothy Arceri <tarc...@itsqueeze.com> writes:
>
>> Thanks for the
Timothy Arceri writes:
> Thanks for the test! The idea is great, the only problem I see is
> that the minimum GL version seems too high. This test would not run
> on the i915 driver which is a problem IMO. Can you lower the
> requirements?
Good point. Here is a v2 that
If a bound program is relinked unsuccessfully it is supposed to keep
the executable and the program state such as the uniforms. This adds a
test for that which updates a uniform, does a failed relink and
verifies that a render succeeds with the new uniform value. This is
currently causing a
The previous comparison values were calculated by shifting ~0 by the
invocation index. This would end up including bits higher than
gl_SubGroupSizeARB.
---
.../arb_shader_ballot/execution/fs-builtin-variables.shader_test | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git
Oh I didn't notice you'd written this test when I replied to your
message on mesa-dev. Looks good to me.
Reviewed-by: Neil Roberts <n...@linux.intel.com>
- Neil
Ilia Mirkin <imir...@alum.mit.edu> writes:
> From the ARB_sample_shading spec:
>
>"gl_NumSamples
Ilia Mirkin writes:
>> + "imageStore(img,\n"
>> + " ivec2(gl_FragCoord.xy),\n"
>> + " uvec4(gl_NumSamples));\n"
>
> Could I convince you to use ssbo instead? Not only is it supported by
> nouveau [unlike
Adds a test that creates a framebuffer with no attachments and sets
the default sample count to 3. It then verifies that GL_SAMPLES for
the fbo is at least 3 and GL_SAMPLE_BUFFERS is 1. It also checks the
value of gl_NumSamples in the shader by writing to an SSBO with the
value to verify that it
Adds a test that creates a framebuffer with no attachments and sets
the default sample count to 3. It then verifies that GL_SAMPLES for
the fbo is at least 3. It also checks the value of gl_NumSamples in
the shader by doing an image store with the value to verify that it
has the same value.
This creates a cubemap where one of the faces either has a different
size or a different format from the other faces and then checks that
it is incomplete by rendering with it and veryifying that the sampler
returns solid black.
The format test currently fails on Mesa.
---
tests/all.py
The i965 driver currently has a bug where if no window system surface
is bound then it will crash when glViewport is called. This patch adds
a test case which just sets up a simple context and binds it with no
surface before calling glViewport.
As far as I can tell this is the first explicit test
Ian Romanick writes:
>> As far as I can tell this is the first test for surfaceless contexts
>> so it also adds the directory and the test group.
>
> There are a couple other tests that use surfaceless contexts
> (tests/egl/spec/egl-1.4/egl-terminate-then-unbind-context.c
Previously there were a couple of places directly searching for
extensions in EGL_EXTENSIONS via strstr. This doesn't check the
terminator for the extension name properly and there is already a
helper function for this so we might as well use it.
---
tests/egl/egl-util.c
The i965 driver currently has a bug where if no window system surface
is bound then it will crash when glViewport is called. This patch adds
a test case which just sets up a simple context and binds it with no
surface before calling glViewport.
As far as I can tell this is the first test for
"Pohjolainen, Topi" <topi.pohjolai...@intel.com> writes:
> On Tue, Dec 01, 2015 at 02:18:59PM +, Neil Roberts wrote:
>> "Pohjolainen, Topi" <topi.pohjolai...@intel.com> writes:
>>
>> >> + if (
"Pohjolainen, Topi" <topi.pohjolai...@intel.com> writes:
> On Wed, Nov 25, 2015 at 06:11:53PM +0100, Neil Roberts wrote:
>> On SKL in the i965 driver there is some special handling of the fast
>> clear optimisation when GL_FRAMEBUFFER_SRGB is enabled bec
"Pohjolainen, Topi" writes:
>> +if (single_sample) {
>> +glTexParameteri(tex_target,
>> +GL_TEXTURE_MAG_FILTER,
>> +GL_NEAREST);
>> +glTexParameteri(tex_target,
>> +
test_once was ignoring the return value of draw_and_test_layer so the
test would report a pass even if it failed.
Reported-by: Marek Olšák
---
Oops, sorry about that. Thanks for reporting it.
"Pohjolainen, Topi" writes:
>> glBindFramebuffer(GL_FRAMEBUFFER, piglit_winsys_fbo);
>> piglit_draw_rect(offset * 16 * 2.0f / piglit_width - 1.0f,
>
> This is a question regarding the existing logic. Earlier the test
> calls
The test colors now include negative values and values greater than
one. Instead of rendering the results into the window system buffer a
floating point texture is now created so that it can store values that
haven't been clamped to [0,1].
---
.../spec/ext_framebuffer_multisample/fast-clear.c |
On SKL in the i965 driver there is some special handling of the fast
clear optimisation when GL_FRAMEBUFFER_SRGB is enabled because the
hardware can't cope with the fast clear buffer in that case and the
color buffer needs to be resolved. This test just tries varies
different clear colors and
ext_framebuffer_multisample-fast-clear can now take a parameter on the
command line to make it test a single-sample buffer instead. This is
worth testing at least on i965 because fast clears are handled
differently when multisampling is not used.
---
tests/all.py
This is similar to ext_framebuffer_multisample-fast-clear except that
it tests a regular single-sampled texture. This is worth testing at
least on i965 because fast clears are handled differently when
multisampling is not used.
---
tests/all.py| 5 +
tests/fbo/CMakeLists.gl.txt
On SKL in the i965 driver there is some special handling of the fast
clear optimisation when GL_FRAMEBUFFER_SRGB is enabled because the
hardware can't cope with the fast clear buffer in that case and the
color buffer needs to be resolved. This test just tries varies
different clear colors and
Neil Roberts <n...@linux.intel.com> writes:
> +g(['fast-clear-blend'])
Oops, this should be 'arb_framebuffer_srgb-fast-clear-blend'. I've fixed
this locally.
- Neil
___
Piglit mailing list
Piglit@lists.freedesktop.org
http://lists.freede
Ian Romanick writes:
> +# Group EXT_shader_samples_identical
> +with profile.group_manager(
> +PiglitGLTest,
> +grouptools.join('spec', 'EXT_shader_samples_identical')) as g:
> +g(['ext_shader_samples_identical', '2'])
> +
Looks good to me, thanks.
Reviewed-by: Neil Roberts <n...@linux.intel.com>
Regards,
- Neil
Vinson Lee <v...@freedesktop.org> writes:
> Fixes "uninitialized scalar variable" defect reported by Coverity.
>
> Signed-off-by: Vinson Lee <v...@fr
On Gen7+ in the i965 driver of Mesa there is an optimisation so that
when the color buffer is cleared with certain allowable colors a
separate smaller buffer is used to keep track of which pixels are
using this clear color and the clear color is stored outside the
buffer in the surface state. That
---
tests/fbo/fbo-formats.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tests/fbo/fbo-formats.h b/tests/fbo/fbo-formats.h
index 655d149..d73e282 100644
--- a/tests/fbo/fbo-formats.h
+++ b/tests/fbo/fbo-formats.h
@@ -274,7 +274,7 @@ static const struct format_desc
Ben Widawsky writes:
> - glReadPixels(0, 0, 3, 1, GL_STENCIL_INDEX, GL_FLOAT, p);
> + glReadPixels(0, 0, 256, 256, GL_STENCIL_INDEX, GL_UNSIGNED_BYTE, p);
> piglit_present_results();
>
> /* we hardcode 129 in the shader */
> - if (p[0] !=
Ben Widawsky writes:
> diff --git a/tests/spec/arb_shader_stencil_export/CMakeLists.gl.txt
> b/tests/spec/arb_shader_stencil_export/CMakeLists.gl.txt
> new file mode 100644
> index 000..86d55a9
> --- /dev/null
> +++
uj.pho...@gmail.com>
+ * Neil Roberts <n...@linux.intel.com>
*/
#include "piglit-test-pattern.h"
@@ -50,24 +58,70 @@ PIGLIT_GL_TEST_CONFIG_BEGIN
PIGLIT_GL_TEST_CONFIG_END
const int pattern_width = 256; const int pattern_height = 256;
-Fbo src_fbo, dst_fbo;
+Fbo src_fbo, d
There is currently a bug with the i965 driver in Mesa with the fast
clear optimisation when there are missing components in the
framebuffer. If the texture that was rendered to is later sampled it
will use the fast clear color without overriding the missing
components to 0 or 1. This patch adds a
When the -fbo option is given to the fbo-depth-array test it tries to
render to a range of different depth/stencil buffer sizes with
dimensions of up to 8192. It verifies the rendering was correct by
binding the depth or stencil buffer as a texture source, rendering the
texture into the Piglit
With 16x MSAA on the i965 driver the accuracy test was failing. It is
possible to fix it by shifting the position of the supersampling grid
in the reference image down and left by 1/16th of a pixel so that the
multisample positions used in the hardware coincide with those of the
supersampling
The unlit and totally_lit stats are supposed to count the pixels where
either a primitive completely covers the pixel or no primitive touches
it at all. However this is effectively only determined by checking
whether any primitive has intersected one of the supersample positions
of the reference
Neil Roberts <n...@linux.intel.com> writes:
> This fixes false negatives for hardware using the sampling positions
> described by D3D on 16x MSAA. Please see the comment explaining the
> issue.
> ---
> tests/spec/ext_framebuffer_multisample/common.cpp | 37
> +
This fixes false negatives for hardware using the sampling positions
described by D3D on 16x MSAA. Please see the comment explaining the
issue.
---
tests/spec/ext_framebuffer_multisample/common.cpp | 37 ---
tests/spec/ext_framebuffer_multisample/common.h | 3 +-
2 files
Preivously the frag coord tests were effectively trying to assert that
the results for gl_FragCoord were aligned to whole numbers, ie that it
reports the bottom left edge of the pixel. However by default
gl_FragCoord actually reports the center of the pixel, so the bottom
left pixel would be
Previously the test was trying to convert a sample number to the range
[0,1] by dividing by 10.0. This doesn't work when testing 16x MSAA
because it ends up with values greater than 1.0 which get clamped.
This patch makes it divide by miplevels so that it will work whatever
the number of samples
Looks good to me.
Reviewed-by: Neil Roberts n...@linux.intel.com
- Neil
Vinson Lee v...@freedesktop.org writes:
Fix resource leak defect reported by Coverity.
Signed-off-by: Vinson Lee v...@freedesktop.org
---
tests/general/point-vertex-id.c | 2 ++
1 file changed, 2 insertions
Adds a test which is similar to interpolateAtSample-different except
that it draws a triangle which covers more than one fragment and makes
sure to use a different sample ID for each fragment so that it won't
be dynamically uniform. The GLSL spec doesn't mention that the sample
ID has to be
Adds a new test for interpolateAtSample which captures the results
from the function with each different sample value and ensures that
they are all different. This can be done either with a constant or
non-constant value depending on whether a command line argument is
given.
The other tests for
Adds an option to use an instanced attribute so that it can be tested
in combination with glPolygonMode. This demonstrates a bug on the Mesa
i965 driver on BDW which was reported as bug #91292. The ‘divisor’
argument can be added in addition to any of the other arguments so
that all of the
The point-vertex-id test was originally created to test gl_VertexID in
combination with glPolygonMode(GL_POINT) because of a bug with the
i965 driver in Mesa. The same bug exists with gl_InstanceID and the
code for BDW has some extra complications in this case so I think it's
worth extending the
Looks good to me. Thanks for doing that.
Reviewed-by: Neil Roberts n...@linux.intel.com
Regards,
- Neil
Timothy Arceri t_arc...@yahoo.com.au writes:
Previously the shaders were using a checkerboard texture with
identical colours for each part of the grid.
That meant that it didn't actually
Hi,
I've just pushed this patch which changes the other
sampler_array_indexing tests so that the textures are not just a solid
colour. The idea is that that way it will ensure that the texture
coordinates are being used correctly.
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *Neil Roberts n...@linux.intel.com
+ *
+ */
+
+/** @file zero-tex-coord.c
+ *
+ * Tests various texture
The copyteximage test tries to render each layer of a 3D texture by
rendering a quad with the same z value in the texture coordinates for
each vertex. It was picking this value by just dividing the layer
number by the depth of the texture. I think this is wrong because it
will effectively point to
If texsubimage is passed cube_map_array on the command line it will
try updating a subregion of a cube map array. All of the faces of all
of the layers of the texutre are rendered using a special vertex
shader to modify the texture coordinates so that they pick a
particular face based on the z
Adds a test which tries setting a uint attrib value to 0x7f817f81.
If this value were interpreted as a float then it would look like a
signalling NaN. If this is passed through an x87 register then it
would get converted to a quiet NaN on store.
Currently Mesa does this on 32-bit builds (possibly
to it use the regular glTexImage*(...,NULL)
functions first. It would also make sense to squash this patch into the
first patch instead of pushing both versions.
Assuming those changes are made,
Reviewed-by: Neil Roberts n...@linux.intel.com
Regards,
- Neil
Laura Ekstrand la...@jlekstrand.net writes
If texsubimage is passed cube_map_array on the command line it will
try updating a subregion of a cube map array. All of the faces of all
of the layers of the texutre are rendered using a special vertex
shader to modify the texture coordinates so that they pick a
particular face based on the z
If the 'array' command line option is passed to the texsubimage test
it will now try updating subregions of 1D and 2D array textures. This
requires a shader to render.
---
tests/all.py | 1 +
tests/texturing/texsubimage.c | 122 +-
2
Today I pushed some patches to the original version of this test which
change it quite a lot. Previously it updated a sub-region of the texture
with the exact same contents as were already in the texture so an
implementation that did nothing on glTexSubImage[123]D would actually
pass the test. It
Previously the test was updating sub regions of the test texture with
the same values that were already in the texture. Therefore an
implementation that does nothing on glTexSubImage2D would pass which
doesn't seem very useful. This patch makes it create two reference
images with different values.
Previously when testing a 3D texture the test would just draw a single
image with the width and height of the texture and the z coordinates
set to span across the depth. This wouldn't end up drawing all of the
texels in the texture so instead it will now render all of the images
in a vertical
There is nothing in the GLSL spec which says you can have a float
literal without an exponent or a decimal point even if you add the ‘f’
suffix. This patch replaces that literal with some other variations
that should be accepted according to the spec.
---
OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *Neil Roberts n...@linux.intel.com
This currently exposes a bug in the i965 driver.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=84677
---
tests/all.py| 1 +
tests/general/CMakeLists.gl.txt | 1 +
tests/general/point-vertex-id.c | 154
3 files changed, 156
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *Neil Roberts n...@linux.intel.com
+ *
+ */
+
+/** @file glx-context-flush-control.c
+ *
+ * Tests
OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *Neil Roberts n...@linux.intel.com
+ *
+ */
+
+/** @file glx-context-flush-control-performance.c
+ *
+ * Creates a pair of contexts each with its own window and renders to
+ * both of them
Hi,
I had a look at this patch because I'm about to post patches to Mesa to
implement the extension. It would be good to get this patch in before
the patches land in Mesa but I think it currently has some issues.
The test probes for a pixel value immediately after calling
glXSwapBuffers but as
Ilia Mirkin imir...@alum.mit.edu writes:
Just bikeshedding here, but isn't the common way of doing this to just have
(1ULL bits) - 1
?
Yes, that's probably more clear. One advantage of doing the shift in the
other direction is if we ever end up supporting textures with 64-bit
ints then we
The un_to_float function was trying to get the maximum value given a
number of bits by shifting ~0ul by the number of bits. For the
GL_UNSIGNED_INT type this function was also being used to get a
maximum value for a 32-bit quantity. However on a 32-bit build this
would mean that it is shifting a
Actually I think this might be a slightly cleaner way to do the shift
because it doesn't depend on any particular size of unsigned int and
there are fewer ~s.
- Neil
--- 8 --- (use git am --scissors to automatically chop here)
The un_to_float function was trying to get the
Hi Ken,
Ken Phillis Jr kphilli...@gmail.com writes:
- unsigned int max = ~(~0ul bits);
+ unsigned int max = ~0u (sizeof max * 8 - bits);
why not use, sizeof(unsigned int)? the call sizeof max may not always
work depending on compiler
Could you explain which compilers “sizeof
This changes one of the values of the integer clear color to
0x7f817f81. If this were interpreted as a float value then it would be
a signalling NaN. If this passes through an x87 register then the
store instruction will convert it to a quiet NaN by setting bit 22.
The integer value would then be
IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *Neil Roberts n...@linux.intel.com
+ *
+ */
+
+/** @file bptc-modes.c
+ *
+ * Tests a BPTC-compressed half-float texture
1 - 100 of 115 matches
Mail list logo