Refer [1], always-inline is not suggested to be used if you have indirect
calls. so replace force_inline with inline to fix error like:
In function ‘combine_inner’,
inlined from ‘combine_soft_light_ca_float’ at ../pixman/pixman-combine-float.c:655:511:
../pixman/pixman-combine-float.c:655:211: error: inlining failed in call to ‘always_inline’ ‘combine_soft_light_c’: function not considered for inlining
Test with gcc-9 and gcc-14, both works well
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115679
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Temporary version pin of gcovr due to errors in coverage report
generation when running with newly released version 8.x.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
https://github.com/mesonbuild/meson/pull/13604 got merged and released
with Meson 1.6.0, which we already use in the Docker images, so the
workaround can be dropped.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
Some targets require different version of LLVM, so now it's possible to
set it in the target's environment. Mind that the highest available
version depends on the base Debian image.
The change bumps LLVM version for all Linux targets:
- by default from 14 to 16,
- from 16 to 18 for riscv64 (based on Sid; for now, LLVM 19 doesn't have
libomp packaged),
- mipsel stays at 14 as there seem to be some missing packages for
higher versions.
Windows targets stay the same, as they use a different source of LLVM
(MinGW-compatible, which is currently version 18).
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
To ensure that the runtime discovery works correctly, and RVV code is
disabled for target without RVV extension.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
RVV compilation will be enabled for RVV implementation alone, similar to
other platforms. This prevents introducing autovectorized code in the
main library, thus making pixman compatible with RISC-V targets without
RVV.
If rule condition for selectively running Docker image builds was ill
formed. It resulted in build of all images even when not all targets
were selected with ACTIVE_TARGET_PATTERN variable.
If the MR doesn't modify the Docker context, the pipeline should use the
image from upstream.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
This commit unifies codecov and pltcov build and test stages as single
parametrizable GitLab job templates. This cleans up the pipeline flow in
preparation for LLVM support in the pipeline.
Each target has now a Meson cross file, even when using a native
compiler, so that the job template can be better generalized. This also
allows to move architecture-specific build configuration to the cross
file instead of using the additional Meson flags in the job declaration.
This commit merges codecov and pltcov Dockerfiles into a single,
multi-stage Dockerfile. This results in more streamlined Docker image
builds with some common layers which can be reused by multiple images.
Also, by making a common Dockerfile, all common dependencies have the
same exact description, which decreases disparity between different
images for all the supported architectures. Mind that package version
disparity cannot be prevented 100%, as different base images may be used
for different architectures (e.g., Debian Sid for riscv64 instead of
Bookworm).
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
It replaces CODE- and PLT- specific target enable variables. It is a
ground work for unification of codecov and pltcov flows.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
It makes per-targe environment declaration more extensible, as it's
possible now to set custom env variables only for the selected target
for the entire pipeline workflow in a centralized way.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
There was a missing wildcard for Docker directory
change detection, so basically this rule was not
checked correctly.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
Platform coverage checks if the code builds and executes properly for
architectures that are not officially supported by Debian. They don't
contribute to general code coverage report but provide a valuable
insight if all supported platforms are working correctly.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
Add images providing an environment for architecture coverage tests.
There is a separate build for Linux and Windows, as the Windows image is
really large compared to Linux one. It decreases the execution time of
both targets, as the images needed to be pulled by runners are smaller.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
This commit introduces a build and test CI workflow, which tests the
correctness of execution for nearly all configurations supported by
pixman. The notable exception is ARM iWMMXt, which is omitted as it's
soon to be deprecated as mentioned in #98.
The build and test stage is separated, as a single build can be used to
test multiple configurations for a given platform (e.g., MMX, SSE2,
SSSE3 for x86).
Execution is performed using multi-arch Docker images built in the
`docker` stage. The important thing to note is that the runner needs to
have a relatively recent version of Docker and QEMU, and needs to have
the qemu-user-static+binfmt execution enabled.
Once all tests are complete, coverage reports are merged together in the
`summary` stage. Then the result can be used in a GitLab-native coverage
report summary.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
Used to force feature discovery in CI where /proc/cpuinfo is unreliable.
It can happen, e.g., if executed in qemu-user-static mode.
For such a build, MIPS-specific features need to be manually disabled by
using `PIXMAN_DISABLE` env variable.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
DSPr2 can be available for targets other than mips32. Some distros
(e.g., Debian) don't support mips32 but still support mipsel. Extending
the check enables use of such images for testing.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
The image is used in CI pipeline to build and test on different
architectures.
This commit introduces more extensible GitLab CI scheme borrowed from
qemu project.
Signed-off-by: Marek Pikuła <m.pikula@partner.samsung.com>
This avoids callers to have to optimize this codepath, in case this scenario happens.
And definitely it may happen when the function is not explicitly called.
