Initial NEON optimization for bilinear scaling. Can be probably
improved more.
Benchmark on ARM Cortex-A8:
Microbenchmark (scaling 2000x2000 image with scale factor close to 1x):
before: op=1, src=20028888, dst=20028888, speed=6.70 MPix/s
after: op=1, src=20028888, dst=20028888, speed=44.27 MPix/s
A primitive naive implementation of bilinear scaling using SSE2 intrinsics,
which only handles one pixel at a time. It is approximately 2x faster than
pixman general compositing path. Single pass processing without intermediate
temporary buffer contributes to ~15% and loop unrolling contributes to ~20%
of this speedup.
Benchmark on Intel Core i7 (x86-64):
Using cairo-perf-trace:
before: image firefox-planet-gnome 12.566 12.610 0.23% 6/6
after: image firefox-planet-gnome 10.961 11.013 0.19% 5/6
Microbenchmark (scaling 2000x2000 image with scale factor close to 1x):
before: op=1, src=20028888, dst=20028888, speed=70.48 MPix/s
after: op=1, src=20028888, dst=20028888, speed=165.38 MPix/s
Individual correctness check for the new bilinear scaling related
supplementary function. This test program uses a bit wider range
of input arguments, not covered by other tests.
Can be used for implementing SIMD optimized fast path
functions which work with bilinear scaled source images.
Similar to the template for nearest scaling main loop, the
following types of mask are supported:
1. no mask
2. non-scaled a8 mask with SAMPLES_COVER_CLIP flag
3. solid mask
PAD repeat is fully supported. NONE repeat is partially
supported (right now only works if source image has alpha
channel or when alpha channel of the source image does not
have any effect on the compositing operation).
MSVC does not notice non-returning functions (abort() / assert(0))
and warns about paths which end with them in non-void functions:
c:\cygwin\home\ranma42\code\fdo\pixman\test\fetch-test.c(114) :
warning C4715: 'reader' : not all control paths return a value
c:\cygwin\home\ranma42\code\fdo\pixman\test\stress-test.c(133) :
warning C4715: 'real_reader' : not all control paths return a value
c:\cygwin\home\ranma42\code\fdo\pixman\test\composite.c(431) :
warning C4715: 'calc_op' : not all control paths return a value
These warnings can be silenced by adding a return after the
termination call.
The Microsoft C compiler cannot handle subobject initialization and
Win32 does not provide snprintf.
Work around these limitations by using normal struct initialization
and using sprintf (a manual check shows that the buffer size is
sufficient).
Previously 'make check' would compile and run tests first, and only
then proceed to compiling demos. Which is not very convenient
because of the need to scroll back console output to see the
tests verdict. Swapping order of SUBDIRS variable entries in
Makefile.am resolves this.
This allows some more code to be deleted from the X server. The
implementation consists of converting to trapezoids, and is shared
with pixman_composite_triangles().
The Render X extension can draw triangles as well as trapezoids, but
the implementation has always converted them to trapezoids. This patch
moves the X server's triangle conversion code into pixman, where we
can reuse the pixman_composite_trapezoid() code.
This function is an implementation of the X server request
Trapezoids. That request is what the X backend of cairo is using all
the time; by moving it into pixman we can hopefully make it faster.
This separates the test suite from the random gtk+ using test
programs. "demos" is somewhat misleading because the programs there
are not particularly exciting (with the possible exception of
composite-test which shows off all the compositing operators).
Scaling function now gets an extra boolean argument, which is set
to TRUE when we are fetching padding pixels for NONE repeat. This
allows to make a decision whether to interpret alpha as 0xFF or 0x00
for such pixels when working with formats which don't have alpha
channel (for example x8r8g8b8 and r5g6b5).
In addition to the most common case of not having any mask at all, two
variants of scaling with mask show up in cairo traces:
1. non-scaled a8 mask with SAMPLES_COVER_CLIP flag
2. solid mask
This patch extends the nearest scaling main loop template to also
support these cases.
Depending on CPU architecture, performance is in the range of 1.5 to 4 times
slower than simple nonrotated copy (which would be an ideal case, perfectly
utilizing memory bandwidth), but still is more than 7 times faster if
compared to general path.
This implementation sets a performance baseline for rotation. The use
of SIMD instructions may further improve memory bandwidth utilization.
Split this function into two, one that has a mask, and one that
doesn't. This is a fairly substantial speed-up in many cases.
New output of lowlevel-blt-bench over_x888_8_0565:
over_x888_8_0565 = L1: 63.76 L2: 62.75 M: 59.37 ( 21.55%) HT: 45.89 VT: 43.55 R: 34.51 RT: 16.80 ( 201Kops/s)
New output of lowlevel-blt-bench over_x888_8_0565:
over_x888_8_0565 = L1: 57.85 L2: 56.80 M: 54.14 ( 19.50%) HT: 42.64 VT: 40.56 R: 32.67 RT: 16.22 ( 195Kops/s)
Based in part on code by Steve Snyder from
https://bugs.freedesktop.org/show_bug.cgi?id=21173
