When masking out the x bits, blitter-test would make the incorrect
assumption that the they were always in the topmost position. This is
not correct for formats of type PIXMAN_TYPE_BGRA.
The code and overall method is mostly based on scaling-test. This one
focuses on trying to stress as many different color formats and types
of composition operations as possible.
This is an initial implementation which may need more tuning. Also
not all color format and operator combinations are actually used.
When cpu specific optimizations are disabled, this test provides
identical deterministic results on x86, PPC and ARM.
Script blitters-test-bisect.rb now works in non-stop mode, until
it finds any problem. This allows to run it for example overnight
in order to test a lot more variants of pixman calls and increase
chances of detecting problems in pixman. Just like with scaling-test,
running blitters-test binary alone with no command line arguments
runs a small predefined number of tests and compares checksum
with a reference value for quick verification.
Bug 22844 demonstrates that it is not sufficient to play tricks with
the clip regions to work around the bogus images from the X
server. The problem there is that if the operation hits the general
path and the destination has a different format than a8r8g8b8, the
destination pixels will be fetched into a temporary array. But because
those pixels would be outside the clip region, they would be fetched
as black. The previous workaround was relying on fast paths fetching
those pixels without checking the clip region.
In the new scheme we work around the problem at the
pixman_image_composite() level. If an image is determined to need a
work around, we translate both the bits pointer, the coordinates, and
the clip region, thus effectively undoing the X server's broken
computation.
Old X servers rely on out-of-bounds accesses when they are asked
to composite with a window as the source. They create a pixman image
pointing to some bogus position in memory, but then they set a clip
region to the position where the actual bits are.
Due to a bug in old versions of pixman, where it would not clip
against the image bounds when a clip region was set, this would
actually work.
The workaround added by this commit is to try and detect whether a
source drawable is actually a window without a client clip set. Such a
window will generally have a clip region that corresponds exactly to
the hierarchy clip in the server, whereas pixmaps will have a clip
region that is an exact match to the drawable.
When we detect such a window, we allow a fast path to run that would
normally be rejected due to the sources not completely subsuming the
composite region.
Fixed X servers should call the new function
pixman_disable_out_of_bounds_workaround() to disable the workaround.
This was reported in bug 22484.
The changes in pixman behavior justifying this are:
- New clipping rules
- Bug fixes in region code. In particular, when
pixman_region_init_rects() is called on these two boxes:
{ 2, 6, 7, 6 }
{ 4, 1, 6, 7 }
it now ignores the first one, which is empty and produces
{ 4, 1, 6, 7 }
Previously, it would produce:
{ 2, 1, 7, 7 }
for some reason.
Now scaling test should reliably detect problems in new scaling code.
Maximum image size reduced to improve performance (more tests can be
run per second) and also simplify detected errors analysis.
According to the RENDER spec, the origin of the alpha map is
interpreted relative to the origin of the drawable of the image, not
the origin of the drawable of the alpha map.
This commit fixes that and adds an alpha-test.c test program.
The only use of alpha maps I have been able to find is in Qt and they
don't use a non-zero alpha origin.
Now test provides better coverage for various image scaling
cases. Unused byte for x8r8g8b8 format is ignored. Running
the test program without any command line options provides
a PASS/FAIL verdict based on the precalculated crc32 value
for using pixman with all the fastpath functions disabled,
which should simplify testing for correcteness.
This test script can help in finding regressions in image scaling
fastpath implementations. It uses test program compiled with
and without fastpath code and can compare results of execution
for different pseudorandom compositing operations involving scaling.
Signed-off-by: Søren Sandmann Pedersen <sandmann@redhat.com>
