Hi Sergey,
It looks fine to me.
If you are having trouble getting consistent results due to
multi-tasking on the machines, it may make sense to look at the "best"
time for each configuration even though I think in general using the
default averaging with more a quieter testing environment is really the
only sure way...
...jim
On 5/12/14 6:00 PM, Sergey Bylokhov wrote:
Hello.
Please review the fix for jdk 9. This a second iteration of the fix for
8041129, because it was changed to cover 8017626 as well.
Description of the problem:
8041129: [OGL] surface->sw blit is extremely slow
- The problem is in the glReadPixels(), which is really slow and we
call it very often for each image to make a flip.
Solution:
- We call glReadPixels() only once and then flip is done using memcpy
after the whole image was moved from gpu to the memory.
- glPixelStorei were added/removed when necessary.
- Note: actually on my mac the slow path and the fast path(memcpy)
works in the same time. I left it as is, because on other system it can
be faster.
8017626: [OGL] Translucent VolatileImages don't paint correctly
- The problem is in OGLSurfaceToSwBlit. It was implemented with
assumption that the source(ogl) surface contain argb color format, this
is wrong because it use premultiplied alpha.
Solution:
- The new ogl blit was added to cover the old behavior
OGLSurfaceToSwBlit(SurfaceType.IntArgbPre,OGLSurfaceData.PF_INT_ARGB_PRE),
- The blit was changed to properly restore the color components if the
source is transparent and destination does not use premultiplied alpha.
Bugs:
https://bugs.openjdk.java.net/browse/JDK-8041129
https://bugs.openjdk.java.net/browse/JDK-8017626
Webrev can be found at: http://cr.openjdk.java.net/~serb/8041129/webrev.10
Benchmarks:
Note that each benchmark was executed 3 times(25 seconds each) to be
protected from such spread:
D3D-base-nvidia-windows: 61796.32109 (var=308.77%) (5857.64%)
|*************
|*********************************************************
|**************************************************
So check the full report for additional information, since the spread
for most of the test is small.
==========================================================
Windows 7 x64, lenovo T410, nvidia NVS 3100M
http://cr.openjdk.java.net/~serb/8041129/webrev.10/benchmarks/results_win.txt
Summary:
OGL-base-nvidia-windows:
Number of tests: 40
Overall average: 1538.6158105111376
Best spread: 0.26% variance
Worst spread: 51.66% variance
(Basis for results comparison)
D3D-base-nvidia-windows:
Number of tests: 40
Overall average: 25477.2292224911
Best spread: 0.03% variance
Worst spread: 308.77% variance
Comparison to basis:
Best result: 11192.65% of basis
Worst result: 9.13% of basis
Number of wins: 28
Number of ties: 0
Number of losses: 12
OGL-fix-nvidia-windows:
Number of tests: 40
Overall average: 125053.84134106003
Best spread: 0.08% variance
Worst spread: 11.96% variance
Comparison to basis:
Best result: 12631.78% of basis
Worst result: 84.45% of basis
Number of wins: 33
Number of ties: 6
Number of losses: 1
==========================================================
OSX 10.8.5, macbook pro retina, nvidia GeForce GT 650M 1024 MB
http://cr.openjdk.java.net/~serb/8041129/webrev.10/benchmarks/results_osx_nvidia.txt
Summary:
OGL-base-nvidia-osx:
Number of tests: 20
Overall average: 714.8688065382015
Best spread: 0.16% variance
Worst spread: 2.12% variance
(Basis for results comparison)
OGL-fix-nvidia-osx:
Number of tests: 20
Overall average: 10497.046970040716
Best spread: 0.11% variance
Worst spread: 5.55% variance
Comparison to basis:
Best result: 4479.08% of basis
Worst result: 172.49% of basis
Number of wins: 20
Number of ties: 0
Number of losses: 0
==========================================================
OSX 10.8.5, macbook pro retina, Intel HD Graphics 4000
http://cr.openjdk.java.net/~serb/8041129/webrev.10/benchmarks/results_osx_intel.txt
Summary:
OGL-base-intel-osx:
Number of tests: 20
Overall average: 3912.1181498636274
Best spread: 0.25% variance
Worst spread: 22.52% variance
(Basis for results comparison)
OGL-fix-intel-osx:
Number of tests: 20
Overall average: 9647.033915964947
Best spread: 0.2% variance
Worst spread: 23.32% variance
Comparison to basis:
Best result: 457.06% of basis
Worst result: 104.3% of basis
Number of wins: 20
Number of ties: 0
Number of losses: 0
