These functions can simply be passed as arguments to the various pixel
fetchers. We don't need to store them. Since they are known at compile
time and the pixel fetchers are force_inline, this is not a
performance issue.
Also temporarily make all pixel access go through the alpha path.
This commit fixes two separate problems: 1. Incorrect computation of
the FAST_PATH_SAMPLES_COVER_CLIP flag, and 2. FAST_PATH_16BIT_SAFE is
a nonsensical thing to compute.
== 1. Incorrect computation of SAMPLES_COVER_CLIP:
Previously we were using pixman_transform_bounds() to compute which
source samples would be used for a composite operation. This is
incorrect for several reasons:
(a) pixman_transform_bounds() is transforming the integer bounding box
of the destination samples, where it should be transforming the
bounding box of the samples themselves. In other words, it is too
pessimistic in some cases.
(b) pixman_transform_bounds() is not rounding the same way as we do
during sampling. For example, for a NEAREST filter we subtract
pixman_fixed_e before rounding off to the nearest sample so that a
transformed value of 1 will round to the sample at 0.5 and not to the
one at 1.5. However, pixman_transform_bounds() would simply truncate
to 1 which would imply that the first sample to be used was the one at
1.5. In other words, it is too optimistic in some cases.
(c) The result of pixman_transform_bounds() does not account for the
interpolation filter applied to the source.
== 2. FAST_PATH_16BIT_SAFE is nonsensical
The FAST_PATH_16BIT_SAFE is a flag that indicates that various
computations can be safely done within a 16.16 fixed-point
variable. It was used by certain fast paths who relied on those
computations succeeding. The problem is that many other compositing
functions were making similar assumptions but not actually requiring
the flag to be set. Notably, all the general compositing functions
simply walk the source region using 16.16 variables. If the
transformation happens to overflow, strange things will happen.
So instead of computing this flag in certain cases, it is better to
simply detect that overflows will happen and not try to composite at
all in that case. This has the advantage that most compositing
functions can be written naturally way.
It does have the disadvantage that we are giving up on some cases that
previously worked, but those are all corner cases where the areas
involved were very close to the limits of the coordinate
system. Relying on these working reliably was always a somewhat
dubious proposition. The most important case that might have worked
previously was untransformed compositing involving images larger than
32 bits. But even in those cases, if you had REPEAT_PAD or
REPEAT_REFLECT turned on, you would hit bits_image_fetch_transformed()
which has the 16 bit limitations.
== Fixes
This patch fixes both problems by introducing a new function called
analyze_extents() that has the responsibility to reject corner cases,
and to compute flags based on the extents.
It does this through a new compute_sample_extents() function that will
compute a conservative (but tight) approximation to the bounding box
of the samples that will actually be needed. By basing the computation
on the positions of the _sample_ locations in the destination, and by
taking the interpolation filter into account, it fixes problem one.
The same function is also used with a one-pixel expanded version of
the destination extents. By checking if the transformed bounding box
will overflow 16.16 fixed point, it fixes problem two.
This extends scaling-crash-test to test some more things:
- All combinations of NEAREST/BILINEAR/CONVOLUTION filters and
NORMAL/PAD/REFLECT repeat modes.
- Tests various scale factors very close to 1/7th such that the source
area is very close to edge of the source image.
- The same things, only with scale factors very close to 1/32767th.
- Enables the commented-out tests for accessing memory outside the
source buffer.
Also there is now a border around the source buffer which has a
different color than the source buffer itself so that if we sample
outside, it will show up.
Finally, the test now allows the destination buffer to not be changed
at all. This allows pixman to simply bail out in cases where the
transformation too strange.
This format is used on PXA framebuffer with some boards. It uses only 18 bits
from the 32 bit framebuffer to interpret color.
Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Negative scale factors are now also tested. A small additional
translate transform helps to stress the use of fractional
coordinates better.
Also the number of iterations to run by default increased in order
to compensate increased variety of operations to be tested.
This test tries to exploit some corner cases and previously known
bugs in nearest neighbor scaling fast path code, attempting to
crash pixman or cause some other nasty effect.
Instead of relying on preprocessor version checks to see if a
some compiler flags are supported, actually try to compile and
link a test program with the flags.
This patch adds extra guards around our use of
OpenMP pragmas and checks that the pragmas won't
cause link errors. This fixes the build on
Tru64 and Solaris with the native compilers and clang.
The x888 suggests that they have something to do with the x8r8g8b8
formats, but that's not the case; they are assuming a8r8g8b8
formats. (Although in some cases they also work for x8r8g8b8 type
formats).
The palettes for indexed formats must satisfy the condition that if
some index maps to a color C, then the 15 bit version of that color
must map back to the index. This ensures that the destination operator
is always a no-op, which seems like a reasonable assumption to make.
Protect the arguments to the combiner macros with parentheses, and
postfix their temporary variables with underscores to avoid name space
collisions with the surrounding code.
Alexander Shulgin pointed out that underscore-prefixed identifiers are
reserved for the C implementation, so we use postfix underscores
instead.
Under the assumption that pixman gradients are supposed to match
QConicalgradient, described here:
http://doc.trolltech.com/4.4/qconicalgradient.html
this patch fixes two separate bugs in pixman-conical-gradient.c.
The first bug is that the output of atan2() is in the range of [-pi,
pi], which means the parameter into the gradient can be negative. This
is wrong since a QConicalGradient always interpolates around the
center from 0 to 1. The fix for that is to (a) make sure the given
angle is between 0 and 360, and (b) add or subtract 2 * M_PI if the
computed angle ends up outside [0, 2 * pi].
The other bug is that we were interpolating clockwise, whereas
QConicalGradient calls for a counter-clockwise interpolation. This is
easily fixed by subtracting the parameter from 1.
Finally, this patch encapsulates the computation in a new force-inline
function so that it can be reused in both the affine and non-affine
case.
Back in the day, the mask_bits argument was used to distinguish
between masks used for component alpha (where it was 0xffffffff) and
masks for unified alpha (where it was 0xff000000). In this way, the
fetchers could check if just the alpha channel was 0 and in that case
avoid fetching the source.
However, we haven't actually used it like that for a long time; it is
currently always either 0xffffffff or 0 (if the mask is NULL). It also
doesn't seem worthwhile resurrecting it because for premultiplied
buffers, if alpha is 0, then so are the color channels
normally.
This patch eliminates the mask_bits and changes the fetchers to just
assume it is 0xffffffff if mask is non-NULL.
This patch comes from the mozilla central tree. See
http://hg.mozilla.org/mozilla-central/rev/89338a224278 for the
original changeset.
Signed-off-by: Jeff Muizelaar <jmuizelaar@mozilla.com>
Signed-off-by: Egor Starkov <egor.starkov@nokia.com>
Signed-off-by: Rami Ylimaki <ext-rami.ylimaki@nokia.com>
Signed-off-by: Siarhei Siamashka <siarhei.siamashka@nokia.com>
Some of the tests are quite heavy CPU users and may benefit from
using multiple CPU cores, so the programs from 'test' directory
are now built with OpenMP support. OpenMP is easy to use, portable
and also takes care of making a decision about how many threads
to spawn.
