Siarhei Siamashka [Mon, 29 Nov 2010 22:31:06 +0000 (00:31 +0200)]
Fix for potential unaligned memory accesses
The temporary scanline buffer allocated on stack was declared
as uint8_t array. As a result, the compiler was free to select
any arbitrary alignment for it (even though there is typically
no reason to use really weird alignments here and the stack is
normally at least 4 bytes aligned on most platforms). Having
improper alignment is non-portable and can impact performance
or even make the code misbehave depending on the target platform.
Using uint64_t type for this array should ensure that any possible
memory accesses done by pixman code are going to be handled correctly
(pixman-combine64.c can access this buffer via uint64_t * pointer).
Some alignment related problem was reported in:
http://lists.freedesktop.org/archives/pixman/2010-November/000747.html
Siarhei Siamashka [Thu, 25 Nov 2010 00:28:29 +0000 (02:28 +0200)]
ARM: added 'neon_src_rpixbuf_8888' fast path
With this optimization added, pixman assisted conversion from
non-premultiplied to premultiplied alpha format is now fully
NEON optimized (both with and without R/B color components
swapping in the process).
Siarhei Siamashka [Mon, 29 Nov 2010 07:11:29 +0000 (09:11 +0200)]
ARM: added 'neon_composite_in_n_8' fast path
Siarhei Siamashka [Mon, 29 Nov 2010 07:00:46 +0000 (09:00 +0200)]
ARM: added flags parameter to some asm fast path wrapper macros
Not all types of operations can be skipped when having transparent
solid source or transparent solid mask. Add an extra flags parameter
for providing this information to the wrappers.
Siarhei Siamashka [Mon, 29 Nov 2010 01:31:32 +0000 (03:31 +0200)]
ARM: added 'neon_composite_add_8888_n_8888' fast path
Siarhei Siamashka [Mon, 29 Nov 2010 00:38:52 +0000 (02:38 +0200)]
ARM: added 'neon_composite_add_n_8_8888' fast path
Siarhei Siamashka [Mon, 29 Nov 2010 00:10:22 +0000 (02:10 +0200)]
ARM: better NEON instructions scheduling for add_8888_8888_8888
Provides a minor performance improvement by using pipelining and hiding
instructions latencies. Also do not clobber d0-d3 registers (source
image pixels) while doing calculations in order to allow the use of
the same macro for add_n_8_8888 fast path later.
Benchmark from ARM Cortex-A8 @500MHz:
== before ==
add_8888_8888_8888 = L1: 95.94 L2: 42.27 M: 25.60 (121.09%)
HT: 14.54 VT: 13.13 R: 12.77 RT: 4.49 (48Kops/s)
add_8888_8_8888 = L1: 104.51 L2: 57.81 M: 36.06 (106.62%)
HT: 19.24 VT: 16.45 R: 14.71 RT: 4.80 (51Kops/s)
== after ==
add_8888_8888_8888 = L1: 106.66 L2: 47.82 M: 27.32 (129.30%)
HT: 15.44 VT: 13.96 R: 12.86 RT: 4.48 (48Kops/s)
add_8888_8_8888 = L1: 107.72 L2: 61.02 M: 38.26 (113.16%)
HT: 19.48 VT: 16.72 R: 14.82 RT: 4.80 (51Kops/s)
Siarhei Siamashka [Sun, 28 Nov 2010 20:05:53 +0000 (22:05 +0200)]
ARM: added 'neon_composite_add_8888_8_8888' fast path
Siarhei Siamashka [Sat, 27 Nov 2010 13:53:54 +0000 (15:53 +0200)]
ARM: added 'neon_composite_over_0565_n_0565' fast path
Siarhei Siamashka [Sat, 27 Nov 2010 02:47:39 +0000 (04:47 +0200)]
ARM: reuse common NEON code for over_{n_8|8888_n|8888_8}_0565
Renamed suppementary macros from 'over_n_8_0565' to 'over_8888_8_0565',
because they can actually support all variants of this operation:
over_8888_8_0565/over_n_8_0565/over_8888_n_0565.
Also 'over_8888_8_0565' now uses more optimized common code instead of its
own variant, improving performance a bit. Even though this operation is
still memory bandwidth limited, scaled variants of these fast paths may
put more stress on CPU later.
Benchmarked on ARM Cortex-A8 @500MHz:
== before ==
over_8888_8_0565 = L1: 67.10 L2: 53.82 M: 44.70 (105.17%)
HT: 18.73 VT: 16.91 R: 14.25 RT: 4.80 (52Kops/s)
== after ==
over_8888_8_0565 = L1: 77.83 L2: 58.14 M: 44.82 (105.52%)
HT: 20.58 VT: 17.44 R: 15.05 RT: 4.88 (52Kops/s)
Siarhei Siamashka [Sat, 27 Nov 2010 01:53:12 +0000 (03:53 +0200)]
ARM: added 'neon_composite_over_8888_n_0565' fast path
Siarhei Siamashka [Fri, 26 Nov 2010 15:06:58 +0000 (17:06 +0200)]
ARM: better NEON instructions scheduling for over_n_8_0565
Code rearranged to get better instructions scheduling for ARM Cortex-A8/A9.
Now it is ~30% faster for the pixel data in L1 cache and makes better use
of memory bandwidth when running at lower clock frequencies (ex. 500MHz).
Also register d24 (pixels from the mask image) is now not clobbered by
supplementary macros, which allows to reuse them for the other variants
of compositing operations later.
Benchmark from ARM Cortex-A8 @500MHz:
== before ==
over_n_8_0565 = L1: 63.90 L2: 63.15 M: 60.97 ( 73.53%)
HT: 28.89 VT: 24.14 R: 21.33 RT: 6.78 ( 67Kops/s)
== after ==
over_n_8_0565 = L1: 82.64 L2: 75.19 M: 71.52 ( 84.14%)
HT: 30.49 VT: 25.56 R: 22.36 RT: 6.89 ( 68Kops/s)
Siarhei Siamashka [Sun, 28 Nov 2010 19:45:06 +0000 (21:45 +0200)]
ARM: introduced 'fetch_mask_pixblock' macro to simplify code
This macro hides the implementation details of pixels fetching
for the mask image just like 'fetch_src_pixblock' does for the
source image. This provides more possibilities for reusing the
same code blocks in different compositing functions.
This patch does not introduce any functional changes and the
resulting code in the compiled object file is exactly the same.
Siarhei Siamashka [Fri, 26 Nov 2010 06:55:49 +0000 (08:55 +0200)]
ARM: added 'neon_composite_over_n_8_8' fast path
Siarhei Siamashka [Mon, 15 Nov 2010 16:26:43 +0000 (18:26 +0200)]
C fast path for a1 fill operation
Can be used as one of the solutions to fix bug
https://bugs.freedesktop.org/show_bug.cgi?id=31604
Alan Coopersmith [Sun, 21 Nov 2010 19:42:22 +0000 (11:42 -0800)]
Sun's copyrights belong to Oracle now
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Cyril Brulebois [Wed, 17 Nov 2010 15:16:56 +0000 (16:16 +0100)]
Fix argument quoting for AC_INIT.
One gets rid of this accordingly:
| autoreconf -vfi
| autoreconf: Entering directory `.'
| autoreconf: configure.ac: not using Gettext
| autoreconf: running: aclocal --force
| configure.ac:61: warning: AC_INIT: not a literal: "pixman@lists.freedesktop.org"
| autoreconf: configure.ac: tracing
| configure.ac:61: warning: AC_INIT: not a literal: "pixman@lists.freedesktop.org"
Signed-off-by: Cyril Brulebois <kibi@debian.org>
Søren Sandmann Pedersen [Tue, 16 Nov 2010 22:14:47 +0000 (17:14 -0500)]
Post-release version bump to 0.21.3
Søren Sandmann Pedersen [Tue, 16 Nov 2010 21:43:26 +0000 (16:43 -0500)]
Pre-release version bump
Søren Sandmann Pedersen [Wed, 3 Nov 2010 03:38:10 +0000 (23:38 -0400)]
Generate {a,x}8r8g8b8, a8, 565 fetchers for nearest/affine images
There are versions for all combinations of x8r8g8b8/a8r8g8b8 and
pad/repeat/none/normal repeat modes. The bulk of each function is an
inline function that takes a format and a repeat mode as parameters.
Andrea Canciani [Tue, 2 Nov 2010 16:04:35 +0000 (17:04 +0100)]
Improve conical gradients opacity check
Conical gradients are completely opaque if all of their stops are
opaque and the repeat mode is not 'none'.
Andrea Canciani [Tue, 2 Nov 2010 16:02:01 +0000 (17:02 +0100)]
Fix opacity check
Radial gradients are "conical", thus they can have some non-opaque
parts even if all of their stops are completely opaque.
To guarantee that a radial gradient is actually opaque, it needs to
also have one of the two circles containing the other one. In this
case when extrapolating, the whole plane is completely covered (as
explained in the comment in pixman-radial-gradient.c).
Andrea Canciani [Sun, 31 Oct 2010 15:59:45 +0000 (16:59 +0100)]
Remove unused stop_range field
Siarhei Siamashka [Sun, 3 Oct 2010 22:56:59 +0000 (01:56 +0300)]
ARM: optimization for scaled src_0565_0565 with nearest filter
The performance improvement is only in the ballpark of 5% when
compared against C code built with a reasonably good compiler
(gcc 4.5.1). But gcc 4.4 produces approximately 30% slower code
here, so assembly optimization makes sense to avoid dependency
on the compiler quality and/or optimization options.
Benchmark from ARM11:
== before ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=34.86 MPix/s
== after ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=36.62 MPix/s
Benchmark from ARM Cortex-A8:
== before ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=89.55 MPix/s
== after ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=94.91 MPix/s
Siarhei Siamashka [Tue, 2 Nov 2010 14:12:42 +0000 (16:12 +0200)]
ARM: NEON optimization for scaled src_0565_8888 with nearest filter
Benchmark from ARM Cortex-A8 @720MHz:
== before ==
op=1, src_fmt=
10020565, dst_fmt=
20028888, speed=8.99 MPix/s
== after ==
op=1, src_fmt=
10020565, dst_fmt=
20028888, speed=76.98 MPix/s
== unscaled ==
op=1, src_fmt=
10020565, dst_fmt=
20028888, speed=137.78 MPix/s
Siarhei Siamashka [Tue, 2 Nov 2010 13:25:51 +0000 (15:25 +0200)]
ARM: NEON optimization for scaled src_8888_0565 with nearest filter
Benchmark from ARM Cortex-A8 @720MHz:
== before ==
op=1, src_fmt=
20028888, dst_fmt=
10020565, speed=42.51 MPix/s
== after ==
op=1, src_fmt=
20028888, dst_fmt=
10020565, speed=55.61 MPix/s
== unscaled ==
op=1, src_fmt=
20028888, dst_fmt=
10020565, speed=117.99 MPix/s
Siarhei Siamashka [Tue, 2 Nov 2010 12:39:02 +0000 (14:39 +0200)]
ARM: NEON optimization for scaled over_8888_0565 with nearest filter
Benchmark from ARM Cortex-A8 @720MHz:
== before ==
op=3, src_fmt=
20028888, dst_fmt=
10020565, speed=10.29 MPix/s
== after ==
op=3, src_fmt=
20028888, dst_fmt=
10020565, speed=36.36 MPix/s
== unscaled ==
op=3, src_fmt=
20028888, dst_fmt=
10020565, speed=79.40 MPix/s
Siarhei Siamashka [Tue, 2 Nov 2010 12:29:57 +0000 (14:29 +0200)]
ARM: NEON optimization for scaled over_8888_8888 with nearest filter
Benchmark from ARM Cortex-A8 @720MHz:
== before ==
op=3, src_fmt=
20028888, dst_fmt=
20028888, speed=12.73 MPix/s
== after ==
op=3, src_fmt=
20028888, dst_fmt=
20028888, speed=28.75 MPix/s
== unscaled ==
op=3, src_fmt=
20028888, dst_fmt=
20028888, speed=53.03 MPix/s
Siarhei Siamashka [Tue, 2 Nov 2010 17:16:46 +0000 (19:16 +0200)]
ARM: performance tuning of NEON nearest scaled pixel fetcher
Interleaving the use of NEON registers helps to avoid some stalls
in NEON pipeline and provides a small performance improvement.
Siarhei Siamashka [Tue, 2 Nov 2010 12:26:13 +0000 (14:26 +0200)]
ARM: macro template in C code to simplify using scaled fast paths
This template can be used to instantiate scaled fast path functions
by providing main loop code and calling NEON assembly optimized
scanline processing functions from it. Another macro can be used
to simplify adding entries to fast path tables.
Siarhei Siamashka [Mon, 1 Nov 2010 08:03:59 +0000 (10:03 +0200)]
ARM: nearest scaling support for NEON scanline compositing functions
Now it is possible to generate scanline processing functions
for the case when the source image is scaled with NEAREST filter.
Only 16bpp and 32bpp pixel formats are supported for now. But the
others can be also added later when needed. All the existing NEON
fast path functions should be quite easy to reuse for implementing
fast paths which can work with scaled source images.
Siarhei Siamashka [Mon, 1 Nov 2010 03:10:34 +0000 (05:10 +0200)]
ARM: NEON: source image pixel fetcher can be overrided now
Added a special macro 'pixld_src' which is now responsible for fetching
pixels from the source image. Right now it just passes all its arguments
directly to 'pixld' macro, but it can be used in the future to provide
a special pixel fetcher for implementing nearest scaling.
The 'pixld_src' has a lot of arguments which define its behavior. But
for each particular fast path implementation, we already know NEON
registers allocation and how many pixels are processed in a single block.
That's why a higher level macro 'fetch_src_pixblock' is also introduced
(it's easier to use because it has no arguments) and used everywhere
in 'pixman-arm-neon-asm.S' instead of VLD instructions.
This patch does not introduce any functional changes and the resulting code
in the compiled object file is exactly the same.
Siarhei Siamashka [Tue, 2 Nov 2010 20:53:55 +0000 (22:53 +0200)]
ARM: fix 'vld1.8'->'vld1.32' typo in add_8888_8888 NEON fast path
This was mostly harmless and had no effect on little endian systems.
But wrong vector element size is at least inconsistent and also
can theoretically cause problems on big endian ARM systems.
Siarhei Siamashka [Fri, 24 Sep 2010 13:36:16 +0000 (16:36 +0300)]
Do CPU features detection from 'constructor' function when compiled with gcc
There is attribute 'constructor' supported since gcc 2.7 which allows
to have a constructor function for library initialization. This eliminates
an extra branch for each composite operation and also helps to avoid
complains from race condition detection tools like helgrind.
The other compilers may or may not support this attribute properly.
Ideally, the compilers should fail to compile the code with unknown
attribute, so the configure check should do the right job. But in
reality the problems are surely possible. Fortunately such problems
should be quite easy to find because NULL pointer dereference should
happen almost immediately if the constructor fails to run.
clang 2.7:
supports __attribute__((constructor)) properly and pretends to be gcc
tcc 0.9.25:
ignores __attribute__((constructor)), but does not pretend to be gcc
Søren Sandmann Pedersen [Sun, 31 Oct 2010 05:40:57 +0000 (01:40 -0400)]
Delete the source_image_t struct.
It serves no purpose anymore now that the source_class_t field is gone.
Søren Sandmann Pedersen [Sat, 30 Oct 2010 21:20:22 +0000 (17:20 -0400)]
[mmx] Mark some of the output variables as early-clobber.
GCC assumes that input variables in inline assembly are fully consumed
before any output variable is written. This means it may allocate the
variables in the same register unless the output variables are marked
as early-clobber.
From Jeremy Huddleston:
I noticed a problem building pixman with clang and reported it to
the clang developers. They responded back with a comment about
the inline asm in pixman-mmx.c and suggested a fix:
"""
Incidentally, Jeremy, in the asm that reads
__asm__ (
"movq %7, %0\n"
"movq %7, %1\n"
"movq %7, %2\n"
"movq %7, %3\n"
"movq %7, %4\n"
"movq %7, %5\n"
"movq %7, %6\n"
: "=y" (v1), "=y" (v2), "=y" (v3),
"=y" (v4), "=y" (v5), "=y" (v6), "=y" (v7)
: "y" (vfill));
all the output operands except the last one should be marked as
earlyclobber ("=&y"). This is working by accident with gcc.
"""
Cc: jeremyhu@apple.com
Reviewed-by: Matt Turner <mattst88@gmail.com>
Søren Sandmann Pedersen [Fri, 29 Oct 2010 00:14:03 +0000 (20:14 -0400)]
Remove workaround for a bug in the 1.6 X server.
There used to be a bug in the X server where it would rely on
out-of-bounds accesses when it was asked to composite with a
window as the source. It would create a pixman image pointing
to some bogus position in memory, but then set a clip region
to the position where the actual bits were.
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. So when the pixman bug was fixed, a workaround was
added to allow certain out-of-bound accesses.
However, the 1.6 X server is so old now that we can remove this
workaround. This does mean that if you update pixman to 0.22 or later,
you will need to use a 1.7 X server or later.
Siarhei Siamashka [Sat, 30 Oct 2010 12:51:30 +0000 (15:51 +0300)]
Fixed broken configure check for __thread support
Somehow the patch from [1] was not applied correctly, fixing that.
1. http://lists.cairographics.org/archives/cairo/2010-September/020826.html
Søren Sandmann Pedersen [Mon, 1 Nov 2010 21:52:29 +0000 (17:52 -0400)]
COPYING: Stop saying that a modification is currently under discussion.
Also put the copyright text into a C comment for easier cut and paste.
Søren Sandmann Pedersen [Wed, 27 Oct 2010 21:21:06 +0000 (17:21 -0400)]
Version bump 0.21.1.
The previous bump to 0.20.1 was a mistake; it belongs on the 0.20 branch.
Søren Sandmann Pedersen [Wed, 27 Oct 2010 20:58:29 +0000 (16:58 -0400)]
Post-release version bump to 0.20.1
Søren Sandmann Pedersen [Wed, 27 Oct 2010 20:51:40 +0000 (16:51 -0400)]
Pre-release version bump to 0.20.0
Scott McCreary [Wed, 27 Oct 2010 19:31:27 +0000 (12:31 -0700)]
Added check to find pthread on Haiku.
Jon TURNEY [Sun, 24 Oct 2010 14:58:39 +0000 (15:58 +0100)]
Plug another leak in alphamap test
Even after commit
e46be417cebac984a858da05e61d924889695c9e alphamap
test is still leaking the alphamap pixmap, leading to mmap() failures
on cygwin
Signed-off-by: Jon TURNEY <jon.turney@dronecode.org.uk>
Søren Sandmann Pedersen [Wed, 20 Oct 2010 20:31:57 +0000 (16:31 -0400)]
Post-release version bump to 0.19.7
Søren Sandmann Pedersen [Wed, 20 Oct 2010 20:25:55 +0000 (16:25 -0400)]
Pre-release version bump to 0.19.6
Andrea Canciani [Tue, 12 Oct 2010 13:38:20 +0000 (15:38 +0200)]
Fix an overflow in the new radial gradient code
huge-radial in the cairo test suite pointed out an undocumented
overflow in the radial gradient code.
By casting to pixman_fixed_48_16_t before doing the operations,
the overflow can be avoided.
Søren Sandmann Pedersen [Wed, 20 Oct 2010 20:09:44 +0000 (16:09 -0400)]
Remove the class field from source_image_t
The linear gradient was the only image type that relied on the class
being stored in the image struct itself. With the previous changes, it
doesn't need that anymore, so we can delete the field.
Andrea Canciani [Wed, 20 Oct 2010 19:24:32 +0000 (21:24 +0200)]
Remove unused enum value
The new linear gradient code doesn't use SOURCE_IMAGE_CLASS_VERTICAL
anymore and it was not used anywhere else.
Andrea Canciani [Mon, 18 Oct 2010 20:21:52 +0000 (22:21 +0200)]
Make classification consistent with rasterization
Use the same computations to classify the gradient and to
rasterize it.
This improves the correctness of the classification by
avoiding integer division.
Andrea Canciani [Wed, 11 Aug 2010 07:58:05 +0000 (09:58 +0200)]
Improve precision of linear gradients
Integer division (without keeping the remainder) can discard a lot
of information. Doing the division maths in floating point (and
paying attention to error propagation) allows to greatly improve
the precision of linear gradients.
Andrea Canciani [Tue, 12 Oct 2010 07:52:53 +0000 (09:52 +0200)]
Add comments about errors
Explain how errors are introduced in the computation performed for
radial gradients.
Andrea Canciani [Sun, 15 Aug 2010 07:07:33 +0000 (09:07 +0200)]
Draw radial gradients with PDF semantics
Change radial gradient computations and definition to reflect the
radial gradients in PDF specifications (see section 8.7.4.5.4,
Type 3 (Radial) Shadings of the PDF Reference Manual).
Instead of having a valid interpolation parameter value for every
point of the plane, define it only for points withing the area
covered by the family of circles generated by interpolating or
extrapolating the start and end circles.
Points outside this area are now transparent black (rgba 0 0 0 0).
Points within this area have the color assiciated with the maximum
value of the interpolation parameter in that point (if multiple
solutions exist within the range specified by the extend mode).
Søren Sandmann Pedersen [Fri, 8 Oct 2010 11:44:20 +0000 (07:44 -0400)]
Plug leak in the alphamap test.
The images are being created with non-NULL data, so we have to free it
outselves. This is important because the Cygwin tinderbox is running
out of memory and produces this:
mmap failed on 20000 1507328
mmap failed on 40000 1507328
mmap failed on 20000 1507328
mmap failed on 40000 1507328
mmap failed on 40000 1507328
mmap failed on 40000 1507328
http://tinderbox.x.org/builds/2010-10-05-0014/logs/pixman/#check
Søren Sandmann Pedersen [Wed, 6 Oct 2010 06:40:39 +0000 (02:40 -0400)]
Add no-op combiners for DST and the CA versions of the HSL operators.
We already exit early for DST, but for the HSL operators with
component alpha, we crash at the moment. Fix that by adding a dummy
combine_dst() function.
Søren Sandmann Pedersen [Tue, 5 Oct 2010 15:05:25 +0000 (11:05 -0400)]
test: Add some more colors to the color table in composite.c
Specifically, add transparent black and superluminescent white with
alpha = 0.
Søren Sandmann Pedersen [Tue, 5 Oct 2010 13:49:45 +0000 (09:49 -0400)]
test: Parallize composite.c with OpenMP
Each test uses the test number as the random number seed; if it
didn't, all the threads would run the same tests since they would all
start from the same seed.
Søren Sandmann Pedersen [Sun, 7 Mar 2010 16:26:16 +0000 (11:26 -0500)]
test: Change composite so that it tests randomly generated images
Previously this test would try to exhaustively test all combinations
of formats and operators, which meant that it would take hours to run.
Instead, generate images randomly and test compositing those.
Cc: chris@chris-wilson.co.uk
Søren Sandmann Pedersen [Sun, 7 Mar 2010 16:24:30 +0000 (11:24 -0500)]
test: Fix eval_diff() so that it provides useful error values.
Previously, this function would evaluate the error under the
assumption that the format was 565 or wider. This patch changes it to
take the actual format into account.
With that fixed, we can turn on testing for the rest of the formats.
Cc: chris@chris-wilson.co.uk
Søren Sandmann Pedersen [Sun, 7 Mar 2010 15:31:04 +0000 (10:31 -0500)]
test: Fix bug in color_correct() in composite.c
This function was using the number of bits in a channel as if it were
a mask, which lead to many spurious errors. With that fixed, we can
turn on testing for all formats where all channels have 5 or more
bits.
Cc: chris@chris-wilson.co.uk
Søren Sandmann Pedersen [Tue, 5 Oct 2010 15:08:42 +0000 (11:08 -0400)]
Remove broken optimizations in combine_disjoint_over_u()
The first broken optimization is that it checks "a != 0x00" where it
should check "s != 0x00". The other is that it skips the computation
when alpha is 0xff. That is wrong because in the formula:
min (1, (1 - Aa)/Ab)
the render specification states that if Ab is 0, the quotient is
defined to positive infinity. That is the case even if (1 - Aa) is 0.
Siarhei Siamashka [Mon, 4 Oct 2010 01:49:08 +0000 (04:49 +0300)]
ARM: restore fallback to ARMv6 implementation from NEON in the delegate chain
After fast path cache introduction, the overhead of having this fallback is
insignificant. On the other hand, some of the ARM assembly optimizations (for
example nearest neighbor scaling) do not need NEON.
Siarhei Siamashka [Wed, 8 Sep 2010 06:30:23 +0000 (09:30 +0300)]
Use more unrolling for scaled src_0565_0565 with nearest filter
Benchmark from Intel Core i7 860:
== before ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=1335.29 MPix/s
== after ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=1550.96 MPix/s
== performance of nonscaled src_0565_0565 operation as a reference ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=2401.31 MPix/s
Benchmark from ARM Cortex-A8:
== before ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=81.79 MPix/s
== after ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=89.55 MPix/s
== performance of nonscaled src_0565_0565 operation as a reference ==
op=1, src_fmt=
10020565, dst_fmt=
10020565, speed=197.44 MPix/s
Siarhei Siamashka [Thu, 23 Sep 2010 20:41:50 +0000 (23:41 +0300)]
ARM: added 'neon_composite_out_reverse_8_0565' fast path
== before ==
outrev_8_0565 = L1: 22.91 L2: 22.40 M: 18.75 ( 10.47%)
HT: 12.62 VT: 12.22 R: 11.32 RT: 5.30 ( 58Kops/s)
== after ==
outrev_8_0565 = L1: 176.27 L2: 151.70 M:108.79 ( 60.81%)
HT: 50.43 VT: 37.16 R: 32.26 RT: 9.62 ( 97Kops/s)
Siarhei Siamashka [Thu, 23 Sep 2010 19:28:55 +0000 (22:28 +0300)]
ARM: added 'neon_composite_add_0565_8_0565' fast path
== before ==
add_0565_8_0565 = L1: 14.05 L2: 14.03 M: 11.57 ( 12.94%)
HT: 8.31 VT: 8.10 R: 7.47 RT: 3.64 ( 42Kops/s)
== after ==
add_0565_8_0565 = L1: 123.36 L2: 94.70 M: 74.36 ( 83.15%)
HT: 31.17 VT: 23.97 R: 21.06 RT: 6.42 ( 70Kops/s)
Siarhei Siamashka [Fri, 21 May 2010 13:31:03 +0000 (16:31 +0300)]
ARM: NEON: added forgotten cache preload for over_n_8888/over_n_0565
Prefetch provides up to 40-50% better performance when working
with large images and/or when having lots of L2 cache misses
on ARM Cortex-A8 @ 720MHz:
== before ==
over_n_8888 = L1: 225.83 L2: 181.02 M: 55.57 ( 41.41%)
HT: 38.96 VT: 36.92 R: 32.84 RT: 14.15 ( 123Kops/s)
over_n_0565 = L1: 153.91 L2: 149.69 M: 83.17 ( 30.95%)
HT: 50.41 VT: 49.15 R: 40.56 RT: 15.45 ( 131Kops/s)
== after ==
over_n_8888 = L1: 222.39 L2: 170.95 M: 76.86 ( 57.27%)
HT: 58.80 VT: 53.03 R: 45.51 RT: 14.13 ( 124Kops/s)
over_n_0565 = L1: 151.87 L2: 149.54 M:125.63 ( 46.80%)
HT: 67.85 VT: 57.54 R: 50.21 RT: 15.32 ( 130Kops/s)
Mika Yrjola [Fri, 1 Oct 2010 13:17:50 +0000 (16:17 +0300)]
Fix "syntax error: empty declaration" warnings.
These minor changes should fix a large number of
macro declaration - related "syntax error: empty declaration" warnings
which are seen while compiling the code with the Solaris Studio
compiler.
Søren Sandmann Pedersen [Tue, 28 Sep 2010 04:51:07 +0000 (00:51 -0400)]
Delete simple repeat code
This was supposedly an optimization, but it has pathological cases
where it definitely isn't. For example a 1 x n image will cause it to
have terrible memory access patterns and to generate a ton of modulus
operations.
Since no one has ever measured whether it actually is an improvement,
and since it is doing the repeating at the wrong the stage in the
pipeline, and since with the previous commit it can't be triggered
anymore because we now require SAMPLES_COVER_CLIP for regular fast
paths, just delete it.
Søren Sandmann Pedersen [Tue, 28 Sep 2010 04:42:25 +0000 (00:42 -0400)]
Fix bug in FAST_PATH_STD_FAST_PATH
The standard fast paths deal with two kinds of images: solids and
bits. These two image types require different flags, but
PIXMAN_STD_FAST_PATH uses the same ones for both.
This patch makes it so that solid images just get the standard flags,
while bits images must be untransformed contain the destination clip
within the sample grid.
This means that the old FAST_PATH_COVERS_CLIP flag is now not used
anymore, so it can be deleted.
Dmitri Vorobiev [Tue, 28 Sep 2010 11:42:02 +0000 (14:42 +0300)]
Some clean-ups in fence_malloc() and fence_free()
This patch removes an unnecessary typecast of MAP_FAILED,
replaces an erroneous free() by the correct munmap() in the
error path for a failing mprotect(), and, finally, removes
redundant calls to mprotect() that aren't necessary, because
munmap() doesn't call for any specific memory protection.
Søren Sandmann Pedersen [Tue, 28 Sep 2010 06:52:02 +0000 (02:52 -0400)]
Fix search-and-replace issue in lowlevel-blt-bench.c
Søren Sandmann Pedersen [Fri, 17 Sep 2010 13:21:09 +0000 (09:21 -0400)]
Rename all the fast paths with _8000 in their names to _8
This inconsistent naming somehow survived the refactoring from a while
back.
Liu Xinyun [Sat, 25 Sep 2010 06:56:38 +0000 (14:56 +0800)]
Remove cache prefetch code.
The performance is decreased with cache prefetch, especially for
ATOM. So remove these code. Following is the experiment.
old: 0.19.5-with-cache-prefetch
new: 0.19.5-without-cache-prefetch
CPU: Intel Atom N270@1.6GHz
OS: MeeGo (32 bits)
Speedups
========
image-rgba poppler-0 17125.68 (17279.58 0.92%) -> 14765.36 (15926.49 3.54%): 1.16x speedup
image-rgba ocitysmap-0 9008.25 (9040.41 7.50%) -> 8277.94 (8343.09 5.44%): 1.09x speedup
image-rgba xfce4-terminal-a1-0 18020.76 (18230.68 0.97%) -> 16703.77 (16712.42 1.22%): 1.08x speedup
image-rgba gnome-terminal-vim-0 25081.38 (25133.38 0.24%) -> 23407.47 (23652.98 0.54%): 1.07x speedup
image-rgba firefox-talos-gfx-0 57916.97 (57973.20 0.11%) -> 54556.64 (54624.55 0.39%): 1.06x speedup
image-rgba firefox-planet-gnome-0 102377.47 (103496.63 0.70%) -> 96816.65 (97075.54 0.15%): 1.06x speedup
image-rgba swfdec-giant-steps-0 12376.24 (12616.84 1.02%) -> 11705.30 (11825.20 1.06%): 1.06x speedup
CPU: Intel Core(TM)2 Duo CPU T9600@2.80GHz
OS: Ubuntu 10.04 (64bits)
Speedups
========
image-rgba ocitysmap-0 2671.46 (2691.82 8.55%) -> 2296.20 (2307.26 5.77%): 1.16x speedup
image-rgba swfdec-giant-steps-0 1614.55 (1615.18 1.68%) -> 1532.84 (1538.52 0.72%): 1.05x speedup
Signed-off-by: Liu Xinyun <xinyun.liu@intel.com>
Signed-off-by: Chen Miaobo <miaobo.chen@intel.com>
Dmitri Vorobiev [Wed, 22 Sep 2010 09:34:57 +0000 (12:34 +0300)]
Use <sys/mman.h> macros only when they are available
Not all systems are regular Unices, so let's be careful with the
mmap()-related stuff, which might be unavailable. This patch makes
sure that mmap() and friends is used only when the <sys/mman.h>
header is found.
Søren Sandmann Pedersen [Tue, 21 Sep 2010 18:20:43 +0000 (14:20 -0400)]
Revert "add enable-cache-prefetch option"
Revert this accidentally committed patch.
This reverts commit
19ea0e16b958e5abe491365c203293ab372f3586.
Søren Sandmann Pedersen [Tue, 21 Sep 2010 18:12:00 +0000 (14:12 -0400)]
If MAP_ANONYMOUS is not defined, define it to MAP_ANON.
This hopefully fixes the build failure on OS X.
Liu Xinyun [Tue, 21 Sep 2010 16:15:10 +0000 (00:15 +0800)]
add enable-cache-prefetch option
OK. here is the work to clear all cache prefetch. Please review it. 3x
On Tue, Sep 21, 2010 at 11:36:30PM +0800, Soeren Sandmann wrote:
> Liu Xinyun <xinyun.liu@intel.com> writes:
>
> > This patch is to add a new configuration option: enable-cache-prefetch,
> > which is default yes.
> >
> > Here is a link which talks on cache issue.
> > http://lists.freedesktop.org/archives/pixman/2010-June/000218.html
> >
> > When disable it on Atom CPU(configured with --enable-cache-prefetch=no),
> > it will have a little performance gain. Here is the patch.
>
> I think the cache prefetch code should just be deleted outright. No
> benchmarks that I'm aware of show it to be an improvement.
>
>
> Thanks,
> Soren
>From
bca2192ef524bcae4eea84d0ffed9e8c4855675f Mon Sep 17 00:00:00 2001
From: Liu Xinyun <xinyun.liu@intel.com>
Date: Wed, 22 Sep 2010 00:11:56 +0800
Subject: [PATCH] remove cache prefetch
Søren Sandmann Pedersen [Tue, 21 Sep 2010 14:18:44 +0000 (10:18 -0400)]
Post-release version bump to 0.19.5
Søren Sandmann Pedersen [Tue, 21 Sep 2010 14:11:34 +0000 (10:11 -0400)]
Pre-release version bump to 0.19.4
Søren Sandmann Pedersen [Tue, 21 Sep 2010 14:05:52 +0000 (10:05 -0400)]
compute_composite_region32: Zero extents before returning FALSE.
If the extents of the composite region are broken such that x2 <= x1
or y2 <= y1, then we need to zero the extents before returning so that
the region won't be completely broken when calling
pixman_region32_fini().
Jonathan Morton [Fri, 17 Sep 2010 14:52:23 +0000 (17:52 +0300)]
Add a lowlevel blitter benchmark
This test is a modified version of Siarhei's compositor throughput
benchmark. It's expanded with explicit reporting of memory bandwidth
consumption for the M-test, and with an additional 8x8-random test
intended to determine peak ops/sec capability. There are also quite a
lot more operations tested for.
Dmitri Vorobiev [Fri, 17 Sep 2010 14:52:22 +0000 (17:52 +0300)]
Add noinline macro
This patch adds a noinline macro, which expands to compiler-dependent
keywords that tell the compiler to never inline a function.
Dmitri Vorobiev [Fri, 17 Sep 2010 14:52:21 +0000 (17:52 +0300)]
Add gettime() routine to test utils
Impending benchmark code will need a function to get current time
in seconds, and this patch introduces such routine. We try to use
the POSIX gettimeofday() function when available, and fall back to
clock() when not.
Dmitri Vorobiev [Fri, 17 Sep 2010 14:52:20 +0000 (17:52 +0300)]
Move aligned_malloc() to utils
The aligned_malloc() routine will be used in more than one test utility.
At least, a low-level blitter benchmark needs it. Therefore, let's make
this function a part of common test utilities code.
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:33:23 +0000 (10:33 -0400)]
Enable bits_image_fetch_bilinear_affine_normal_r5g6b5
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:33:10 +0000 (10:33 -0400)]
Enable bits_image_fetch_bilinear_affine_reflect_r5g6b5
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:33:00 +0000 (10:33 -0400)]
Enable bits_image_fetch_bilinear_affine_none_r5g6b5
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:32:44 +0000 (10:32 -0400)]
Enable bits_image_fetch_bilinear_affine_pad_r5g6b5
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:32:27 +0000 (10:32 -0400)]
Enable bits_image_fetch_bilinear_affine_normal_a8
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:32:12 +0000 (10:32 -0400)]
Enable bits_image_fetch_bilinear_affine_reflect_a8
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:31:57 +0000 (10:31 -0400)]
Enable bits_image_fetch_bilinear_affine_none_a8
Søren Sandmann Pedersen [Thu, 16 Sep 2010 14:31:45 +0000 (10:31 -0400)]
Enable bits_image_fetch_bilinear_affine_pad_a8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:41:20 +0000 (02:41 -0400)]
Enable bits_image_fetch_bilinear_affine_normal_x8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:41:08 +0000 (02:41 -0400)]
Enable bits_image_fetch_bilinear_affine_reflect_x8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:40:56 +0000 (02:40 -0400)]
Enable bits_image_fetch_bilinear_affine_none_x8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:40:46 +0000 (02:40 -0400)]
Enable bits_image_fetch_bilinear_affine_pad_x8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:40:16 +0000 (02:40 -0400)]
Enable bits_image_fetch_bilinear_affine_normal_a8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:40:03 +0000 (02:40 -0400)]
Enable bits_image_fetch_bilinear_affine_reflect_a8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:39:51 +0000 (02:39 -0400)]
Enable bits_image_fetch_bilinear_affine_none_a8r8g8b8
Søren Sandmann Pedersen [Sat, 28 Aug 2010 06:39:37 +0000 (02:39 -0400)]
Enable bits_image_fetch_bilinear_affine_pad_a8r8g8b8