Use the shared helper in order to avoid duplication. This will also
ensure that any format added to the test will also be usable by other
tests and clients.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
As required by the spec when accessing dmabufs from the CPU, see
https://docs.kernel.org/driver-api/dma-buf.html
While effectively a no-op on amd64, this fixes test failures on various
aarch64 devices, including:
- rk3588 (Panfrost, Rock5)
- rk3399 (Panfrost, PineBook Pro)
- Broadcom BCM2712 (V3D, RPi5)
- Qualcomm SDM845 (freedreno, OnePlus6)
Fixes: 303d88448 (tests: client-buffer: Add dmabuf support)
Signed-off-by: Robert Mader <robert.mader@collabora.com>
This function is required for producing color transformations from
parametric image descriptions.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Now that we are testing the more robust average error tolerance pretty
tightly, we can relax the tolerances on the maximum error which is more
flaky between GL drivers and hardware.
This makes sRGB->adobeRGB MAT test pass on radeonsi Polaris 11.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Maximum error is fragile: radeonsi on Polaris 11 has one error spike
that would force the the tolerances for everything to be much higher
than desired, and it masks changes to usual error levels.
Add a new condition to opaque_pixel_conversion: average two-norm error
statistic. The average is not as susceptible to outliers as maximum is.
We still keep the maximum error tolerance as well, to detect gross
outliers that might not sway the average enough to cause a test failure.
The new average error tolerances have been collected from the worst case
among the following + 0.01:
- GL renderer: Mesa Intel(R) HD Graphics 4600 (HSW GT2)
- GL renderer: AMD Radeon RX 550 Series (radeonsi, polaris11, ACO,
DRM 3.61, 6.12.35+deb13-amd64)
- GL renderer: llvmpipe (LLVM 19.1.7, 256 bits)
(There is no need to separately print matrix vs. clut, it is already in
the fixture name.)
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
The concise table style was nice, but is getting more unwieldy as more
fields will be added. The new style won't fall apart.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
When the same buffer is reused for both the first and second screenshot,
this can result in a deadlock. Avoid the issue by using a new buffer for
the second screenshot.
Signed-off-by: Deborah Brouwer <deborah.brouwer@collabora.com>
lua-shell is a new meta-shell for Weston. It does nothing in and of
itself, but is defined to be user-scriptable and configurable. A
supplied Lua script will be interpreted and executed by Weston, allowing
full control over window management in response to events.
This includes a shell.lua example script which behaves as a tiling WM.
Co-authored-by: Michael Tretter <m.tretter@pengutronix.de>
Co-authored-by: Marius Vlad <marius.vlad@collabora.com>
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
I don't know when this was improved, I just noticed that we can reduce
the tolerance here.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
The sRGB expected display behavior uses the pure power-law with exponent
2.2, not the two-piece sRGB transfer function.
cmsCreate_sRGBProfileTHR() used the two-piece TF, now we use the proper
display TF.
This is particularly meaningful when implicit sRGB content is converted
to HDR formats, in order to maintain the stimuli reproduction near zero.
cmlcms_send_image_desc_info() is already sending this, it doesn't need
fixing.
Changing the curve also changes the error tolerances. The change is
theoretically a no-op, but the curve and its inverse and temporary
rounding add error. The new curve is more prone to error, so it is not
surprising we need to raise the tolerance. The color transformation does
end up as power-2.2 analytical form and I do not think it is ever
lowered to a LUT in alpha-blending test, so there is no obvious fix
improving the accuracy. The worst case point in alpha-blending still
occurs at the very same point as before.
The test reference images are updated for the same reason, they would
fail otherwise.
Both alpha-blending and color-icc-output contain the same sRGB-optical
sub-test, hence the same error tolerance.
It is surprising to have to increase the ICC roundtrip error tolerance
in color-icc-output test, given that the curves are passed as parametric
to LittleCMS, and adobeRGB case works with the old tolerance even. I did
not investigate further.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Refactoring data about a thing into one place to simplify code.
Also this function is never expected to fail, so no need for the boolean
return.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Collect all information of an item into a table entry in one place. Will
be easier to add new items this way.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
The sRGB display uses a power-2.2 transfer characteristic. These enum
values refer to the two-piece sRGB transfer function instead, so they
should not be named "EOTF".
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Adjust the no-GL jobs to also do no-Vulkan and only create a couple more
to cover disabling the individual renderers.
Copy the hack for obscure LLVM leak reports so that tests can pass in CI
with lavapipe for a Vulkan driver.
Disable vulkan-renderer build on debian lts for now as unsupported.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Containining the pre-compiler conditionals inside tiny functions makes
the code easier to read.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
These cases were never meant to exist, but because the harness assumes
all tests must run on all fixtures, we have to return something.
This is a temporary measure to improve readability. Eventually this test
program should be split:
- shm RGB + YUV
- dmabuf RGB without force-yuv-fallback
- dmabuf YUV with and without force-yuv-fallback
or
- shm RGB + YUV
- dmabuf RGB + YUV without force-yuv-fallback
- dmabuf YUV with force-yuv-fallback
or
- shm + dmabuf; RGB + YUV without force-yuv-fallback
- dmabuf YUV with force-yuv-fallback
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This test has only supported the GL renderer, with a must pass list of
formats which matches what the GL renderer advertises and supports.
This commit prepares the test to support other renderers by moving the
must pass list to the test setup declaration, so that it can be defined
per renderer. This is in preparation to add more renderers here as a
followup.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
And make use of it in kiosk-shell-test. There's no functional change,
this is for other shells to re-use the code here.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Add a status return value to every test definition. There are no
behavioral changes, previously the RESULT_OK was simply assumed.
The benefit of this is that in the future individual tests can also
return RESULT_SKIP, so that we can keep statistics of skipped tests.
ivi-layout-internal-test.c has the only case where a test function may
return early. That one is set to return RESULT_HARD_ERROR to match the
compositor exit code already there.
Also documentation is updated.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Using the udmabuf allocator which is usually available these days,
both on CI runners and dev mashines. The buffers have the the same
content are tested against the same reference images as in the shm
case.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Allow querying supported formats and modifiers.
We use version 3 as it is easier to implement and also works
if no main device is present - such as when using llvmpipe.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
So we can test them on CI. For completeness and because they
are commonly used by SW decoders.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Add stride alignment support make various cleanups in
order to support both shm and dmabufs. No changes in
test results intended.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
We have the following error message in parse_color_characteristics():
name=%s: reserved name. Do not use ':' character in the name.
It gets confusing when "name" ends with ':'. So rephrase it to clarify
the error message in such case.
Signed-off-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Replace an ad hoc 3x3 matrix, that was supposed to match LittleCMS matrix
but did not, with our new consistent weston_mat3f API.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This converts weston_matrix and weston_vector to use linalg-types
internally. All direct accesses to members had to be converted
everywhere, mostly in the simplest form possible which leaves some
trivially reducable code around.
The intention is that we can now gradually migrate away from
weston_matrix in the future.
Look like one trailing space got accidentally annihilated in
compositor.c.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
These are the same as linalg-4 cases, except the matrices have been
generated anew, and checked with Octave that the condition numbers and
determinant values hold.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
The 24 bpp RGB and BGR SHM formats are meant to be stored little
endian, not big endian.
Fixes a regression in commit 815ccaff.
Signed-off-by: Loïc Molinari <loic.molinari@collabora.com>
Found out in https://github.com/mm2/Little-CMS/issues/483
This should be a complete no-op change, but there is one difference.
Previously matrix_inf_norm() summed over rows. Now it sums over columns
as it should be. This shouldn't hurt because it is only used to identify
identity matrices. The bug was actually in is_identity_matrix_stage()
because it deliberately passed what was assumed to be a transpose of the
matrix.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
