| fe871ecd00d11c56b6764c077944a8cb7801a18c |
|
06-Jun-2014 |
Daniel Kurtz <djkurtz@chromium.org> |
glbench: add additional texture types to the texture upload tests
glbench already has a set of texture upload tests that vary the texture
size, API used for upload, and test different texture use cases
(upload/update/reuse).
The texture format used by all of these tests, however, is GL_LUMINANCE,
which is an 8-bit-per-texel format.
In reality, however, Chrome uses several other texture types much more
often than 8-bit, in particular the 32-bit (RGBA/BGRA) formats. At least
on some drivers, the 32-bit and 8-bit texture upload take significantly
different code paths with different performance characteristics, as do the
different component orderings.
Therefore, there is value in benchmarking all of the texture formats.
Note: this patch is much scarier than it appears because it changes the
names of the image files by inserting a 'format' is between test case and
gl command:
<test_case>_<format>_<texture_upload_command>_size.pixmd5-<md5>.png
Signed-off-by: Daniel Kurtz <djkurtz@chromium.org>
BUG=none
TEST=test_that graphics_GLBench
Change-Id: Ia34de4182fb29a70823d1eea718ba5906236afb7
Reviewed-on: https://chromium-review.googlesource.com/202478
Reviewed-by: Ilja Friedel <ihf@chromium.org>
Commit-Queue: Daniel Kurtz <djkurtz@chromium.org>
Tested-by: Daniel Kurtz <djkurtz@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| aace01b7d49e41d1bd26be3767305f57e036e0d7 |
|
17-Jun-2014 |
Daniel Kurtz <djkurtz@chromium.org> |
glbench: refactor how tests specify the unit of their measurement
Instead of hardcoding the unit into subtest name strings, let's set the
unit on a per test basis, and have testbase append the unit as required.
Signed-off-by: Daniel Kurtz <djkurtz@chromium.org>
BUG=none
TEST=test_that graphics_GLBench
Change-Id: Ieb06ad1dd6b2bee8fb0af039e65dc357aed4df1c
Reviewed-on: https://chromium-review.googlesource.com/202886
Reviewed-by: Ilja Friedel <ihf@chromium.org>
Tested-by: Daniel Kurtz <djkurtz@chromium.org>
Commit-Queue: Daniel Kurtz <djkurtz@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| 3a35b3b43d43d34e249172178ef61e6a8c6382ad |
|
03-Jun-2014 |
Daniel Kurtz <djkurtz@chromium.org> |
glbench: Add missing '\n' to error message.
Signed-off-by: Daniel Kurtz <djkurtz@chromium.org>
BUG=none
TEST=none
Change-Id: If8313ed90162382a5d040b3a0188d8a8fd53a96d
Reviewed-on: https://chromium-review.googlesource.com/202475
Reviewed-by: Ilja Friedel <ihf@chromium.org>
Tested-by: Daniel Kurtz <djkurtz@chromium.org>
Commit-Queue: Daniel Kurtz <djkurtz@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| 3907fd1607fdc5ae5065adad80b35c7733fbbf7a |
|
15-Apr-2014 |
Ilja Friedel <ihf@chromium.org> |
GLBench: add new fillrate test of variable FBO size.
* Add fbofill test: We want to measure memory throughput (in this case
bilinear copy) depending on texture size. This is interesting to study
cache behavior.
* Also changes slightly thermal throttling if we can't read temperatures
(by sleeping for a full 30 seconds between tests to ensure cold state).
* Finally add a command line option called "-hasty". This option will
run each test at at most 512x512 resolution, run it only for a fraction
of the loops we normally do, and not cool down the machine between
tests. The hasty option will be used to verify MD5s in BVT and might
also be useful for debugging test failures. Currently
glbench -save -hasty
runs in less than 30 seconds on daisy.
* This change fixes an old standing bug with the yuv_shader images. The
references for this and the new fbofill test are updated.
* We will run "glbench -hasty" on the BVT suite. It will run a subset of
tests in less than a minute total and verify MD5s of the generated
images. We will not upload any performance results from BVT.
Notice that we might need to collect more references from the lab as the
yuv images differ depending on the GPU architecture. Hence BVT is only
experimental.
BUG=chromium:355202, chromium:361925
TEST=Ran locally on daisy and link. Ran cbuildbot on alex.
Change-Id: I54039cce967664d553c79d16ca2b1361953ab9fe
Reviewed-on: https://chromium-review.googlesource.com/194991
Commit-Queue: Ilja Friedel <ihf@chromium.org>
Tested-by: Ilja Friedel <ihf@chromium.org>
Reviewed-by: Ilja Friedel <ihf@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| 3bbf4f178422d0a234a8ac0d090f691ff6344765 |
|
09-Apr-2014 |
Ben Chan <benchan@chromium.org> |
glbench: Update to build against libchrome-242728
BUG=chromium:356431
CQ-DEPEND=CL:193874
TEST=`emerge-$BOARD autotest-tests autotest-deps-glbench`
TEST=Run graphics_GLBench test
Change-Id: I1b40642b8a0ccb5c7a43c7331f8855d875a8098f
Reviewed-on: https://chromium-review.googlesource.com/193882
Reviewed-by: Daniel Erat <derat@chromium.org>
Reviewed-by: Ilja Friedel <ihf@chromium.org>
Tested-by: Ben Chan <benchan@chromium.org>
Commit-Queue: Ben Chan <benchan@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| 6f72b2bf8d333b52f6808bcc0f1e43239de26fed |
|
11-Apr-2014 |
Ilja Friedel <ihf@chromium.org> |
Revert "GLBench: add new fillrate test of variable FBO size."
Some pictures have changed that should not have. Starting over.
This reverts commit b4efb4a15609c2af3248aa9a256a7d5e957afa78.
Change-Id: I15af5b4c4878f2f95b661b7c575a3fd71b81b075
Reviewed-on: https://chromium-review.googlesource.com/194135
Reviewed-by: Ilja Friedel <ihf@chromium.org>
Commit-Queue: Ilja Friedel <ihf@chromium.org>
Tested-by: Ilja Friedel <ihf@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| b4efb4a15609c2af3248aa9a256a7d5e957afa78 |
|
08-Apr-2014 |
Ilja Friedel <ihf@chromium.org> |
GLBench: add new fillrate test of variable FBO size.
We would like to measure memory throughput (in this case bilinear copy)
depending on texture size. This is interesting to study cache behavior.
Also changes slightly thermal throttling if we can't read temperatures
(by sleeping for a full 30 seconds between tests to ensure cold state).
Finally adding a command line option called "-hasty". This option will
run each test at 512x512 resolution only, run it only for a fraction of
the loops we normally do, and not cool down the machine between test.
The hasty option will be used to verify MD5s in BVT and might also be
useful for debugging test failures.
Currently glbench -save -hasty runs in 15 seconds on daisy.
BUG=chromium:355202
TEST=Ran locally on daisy and link.
Change-Id: I6a21565eaf112f3373b2bab26ac122ed7b1ae5cb
Reviewed-on: https://chromium-review.googlesource.com/193576
Commit-Queue: Ilja Friedel <ihf@chromium.org>
Tested-by: Ilja Friedel <ihf@chromium.org>
Reviewed-by: Stéphane Marchesin <marcheu@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|
| c6ae4d5c6c4c9f5f996278e7d4c29d457b6cfce6 |
|
27-Dec-2012 |
Simon Que <sque@chromium.org> |
glbench: create TextureTest class
Currently there is a TextureUpdateTest class. We want to extend this to
support other types of texture-related tests. This patch splits that
class into a base class and a derived class, TextureTest and
TextureUpdateTest, respectively.
BUG=chromium-os:36745
TEST=emerge autotest-deps-glbench, run graphics_GLBench test
Change-Id: If6a3f22715b393088e0f975d63ab51f1bc90c92a
Signed-off-by: Simon Que <sque@chromium.org>
Reviewed-on: https://gerrit.chromium.org/gerrit/40246
Reviewed-by: Stéphane Marchesin <marcheu@chromium.org>
/external/autotest/client/deps/glbench/src/texturetest.cc
|