|
b0e15ee4
|
2021-12-28T20:37:33
|
|
Decide GL_KHR_parallel_shader_compile in backends
GL_KHR_parallel_shader_compile was previously being enabled
unconditionally in the front end. However, some backends (Vulkan)
perform worse with parallel shader compilation. This CL moves the
decision of enabling GL_KHR_parallel_shader_compile to the backends.
To support single-threaded shader compilation without affecting the
generic worker thread pool, Context::mSingleThreadPool is added to own
the single-threaded WorkerThreadPool and can be returned by the new
function Context::getShaderCompileThreadPool(). Otherwise, if the
extension is enabled, the (renamed) Context::mMultiThreadPool is
returned.
Bug: angleproject:6748
Change-Id: Ic8d3a183f397608f3002a05480deb976dfe44792
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3360337
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Tim Van Patten <timvp@google.com>
|
|
e9f9fa17
|
2021-11-09T18:13:15
|
|
Add cpu time measurement to perf tests
Bug: angleproject:6667
Change-Id: I16570a7b51d363dd1c0f35789b985520cb76d488
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3270601
Commit-Queue: Lingfeng Yang <lfy@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
|
|
666fcf38
|
2021-10-08T12:19:15
|
|
Rename functions that overlap with Windows APIs.
Bug: angleproject:6283
Change-Id: Ifcd9ea9e3bf729fd2066178eb9429050b2f10518
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3212894
Reviewed-by: Bruce Dawson <brucedawson@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
62414cf8
|
2021-10-06T16:07:03
|
|
Fix GPU-time-tracking in WebGL perf tests
GL_EXT_disjoint_timer_query is not enabled by default in WebGL mode.
This change ensures that extension is enabled.
Bug: angleproject:4794
Change-Id: I1835f7ed2afd7a155ff9b63fad74a066df59ea4f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3210626
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
e024635d
|
2021-09-10T13:56:28
|
|
ANGLEPerfTest: Skip tests when missing color space support.
Make tests where a non-linear colorspace was requested on platforms
that do not support EGL_KHR_gl_colorspace non-fatal.
Add a initializeGLWithResult function to GLWindowBase, returning a
new GLWindowResult enum and wrap it in EGLWindow and WGLWindow.
Bug: angleproject:6366
Change-Id: Ib57327c4d988d82064272f229f8ad59287541623
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3151833
Commit-Queue: Lubosz Sarnecki <lubosz.sarnecki@collabora.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
247ee1c3
|
2021-09-10T09:38:25
|
|
Perf Tests: Output results in "msBestFitFormat" again.
The histogram bins for "ms" would cause excessive rounding for tests
that run very quickly. For example, 0.0012354 ms would be rounded to
0.001 ms. This would produce very flat graphs that don't accurately
represent the real results.
Bug: angleproject:6090
Change-Id: I9293b2083d7e54dc2663648c361f4883278806ec
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3152746
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
9c05f55a
|
2021-07-15T10:58:10
|
|
Capture/Replay: Add replay state validation.
We can use this to validate traces that have slight differences
when retracing. The valdation works by embedding the "expected"
JSON into a string at the end of each captured frame. The replay
also embeds a callback which fires right before the swap in the
replay harness. The harness then gets the "actual" JSON and runs
a comparison. On a mismatch it calls "diff" externally.
Currently the diff call is hard-coded to work on Linux only.
Note that when running validation it's important to replay on
SwiftShader since that's what we use to retrace.
Bug: angleproject:5133
Change-Id: Icbf0031d07be8bd916607c537dec235f9a512c43
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3066008
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
1b2febb9
|
2021-08-23T12:39:16
|
|
Trace Tests: Refactor calibration logic.
Instead of calling glFinish every step we now run a few warmup loops
to try and figure out the right step interval. On each run we refine
the step interval until we reach a number that completes in the
specific time after a single glFinish call. This method will use the
specified test time much more consistenty.
Also a couple other improvements. Will use more test time on the bots
because previously they were using a calibration/warmup time of 1
second per run loop, and this bumps it up to 3 per run loop.
Long explanation below.
What we did before:
step()
glFinish()
step()
glFinish()
.. repeat for "gCalibrationTime" seconds
then take the number of steps, the actual time ran, and compute an
estimate for "mStepsToRun".
When we run for real, we go:
step()
step()
step()
.. repeat for "mStepsToRun" steps.
glFinish()
this has an obvious performance difference, because after each step()
(one frame in a trace) you had a complete stall.
The new code first does the above to compute a tentative mStepsToRun,
then uses the results of the second run to compute a much more accurate
mStepsToRun by using the "actual" time it took to run the estimate.
Bug: angleproject:6090
Change-Id: I0ff19873580e13f2af5a145a21f5b1c59fb15795
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3114606
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
d240f104
|
2021-08-23T16:25:20
|
|
Perf Tests: Update histogram units.
Use "ms_smallerIsBetter" instead of "msBestFitFormat". This matches
the requirements of the online docs as well as ensuring the dashboard
knows which way to track the regressions.
https://chromium.googlesource.com/catapult/+/HEAD/docs/histogram-set-json-format.md
Bug: angleproject:6090
Change-Id: I0b234725b55d4abb1dd1e8153e132fdbcbad1c60
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3115225
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
13e52124
|
2021-08-23T08:44:13
|
|
ANGLEPerfTest: Fix a couple comment typos.
Bug: angleproject:6090
Change-Id: I2c1a2f4c6b25bb1d4e462bcb20bd788529e255e1
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3114605
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
d2d1f41f
|
2021-08-16T08:05:55
|
|
Add EGL extension to create window with a swap interval set.
On Vulkan this allows creating a window once without needing to
recreate the swapChain after we specify the swap interval.
Also adds a simple regression test and EGL enum assertion
printing formatting.
Bug: angleproject:5133
Change-Id: I72af124cb0e8f7cddfa810988a9862c0f36a0e46
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3097806
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
4a841e6a
|
2021-07-27T13:30:26
|
|
PerfTests: Replay EGL color spaces.
Add EGL_GL_COLORSPACE_KHR to the EGLWindow's surface attributes if
EGL_KHR_gl_colorspace is available.
Require the extension to be available if the color space differs from
the default EGL_COLORSPACE_LINEAR value.
Bug: angleproject:5857
Change-Id: Ib33cbed7d4b115979bd847418cb5cd2b96f4f173
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3056374
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Lubosz Sarnecki <lubosz.sarnecki@collabora.com>
|
|
61517a16
|
2021-08-02T16:49:32
|
|
Remove extra angle:: uses from ANGLEPestTest.cpp.
Bug: None
Change-Id: I7101398504c547b97b9184c56b252bef5198cecb
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3067510
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
48da1c35
|
2021-07-16T13:24:34
|
|
Vulkan: Prefer the local vulkan loader over the system one.
Load the Vulkan loader ourselves and give vkGetInstanceProcAddr to
volk. This allows us to always prefer loading from the current module
directory instead of using the platform-specific ordering.
Refactor angle::Library loading to use ModuleDir instead of
ApplicationDir.
CL originally authored by Geoff Lang.
Bug: chromium:1219969
Change-Id: I21d1926e90fd66e1c23cea7323991ae55f3d22d4
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3035444
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
51937ab3
|
2021-06-25T09:10:42
|
|
Re-land "Add SearchType to OpenSharedLibraryWithExtension."
This fixes a bug in SystemInfo_vulkan where we were using the
System search path for the Vulkan loader when we prefer using the
custom ANGLE loader.
Re-land fixes a bug where we would try to load the custom
libVulkan on Android and other platforms where we should be using
the system version.
Bug: chromium:1219969
Change-Id: I34b592fb87cbddfd02c837a17942cac54c85d9d1
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3007265
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
f8d5d5ed
|
2021-07-05T21:11:23
|
|
Revert "Add SearchType to OpenSharedLibraryWithExtension."
This reverts commit 18e99f4a2b37468b103da4a56c5b0fff25458062.
Reason for revert: breaks Mac, e.g.
https://ci.chromium.org/ui/p/chromium/builders/ci/mac-arm64-rel-tests/2197/overview
Original change's description:
> Add SearchType to OpenSharedLibraryWithExtension.
>
> This fixes a bug in SystemInfo_vulkan where we were using the
> System search path for the Vulkan loader when we prefer using the
> custom ANGLE loader.
>
> Bug: chromium:1219969
> Change-Id: Iedf0fd11fe9ed8cc020b445ea9e12a7936937361
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2988791
> Reviewed-by: Jamie Madill <jmadill@chromium.org>
> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Bug: chromium:1219969, chromium:1225040, chromium:1226675
Change-Id: I7a7e329181b69b0fb546e5245d8842723077126f
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3006320
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Yuly Novikov <ynovikov@chromium.org>
|
|
5f869b74
|
2021-06-28T18:02:42
|
|
Revert "Vulkan: Load custom Vk Loader."
This reverts commit e6e99c5472d2fe9beaa12840fa4a71191667b9a4.
Reason for revert: Seems to be flaking on linux-clang-rel:
https://ci.chromium.org/ui/p/angle/builders/try/linux-clang-rel/22602/overview
Original change's description:
> Vulkan: Load custom Vk Loader.
>
> Uses a GN copy rule to duplicate the loader. Also updates volk to
> load using the new custom loader. Once both are in place we can
> go back and remove the copy.
>
> Bug: chromium:1219969
> Change-Id: I8c48d168a842539f7cdba1ebfdaf3b08c3e1990d
> Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2982499
> Commit-Queue: Jamie Madill <jmadill@chromium.org>
> Reviewed-by: Geoff Lang <geofflang@chromium.org>
> Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
Bug: chromium:1219969
Change-Id: Ib29014e728c1ce63c5e5bf7136ce19de7692964d
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2987874
Bot-Commit: Rubber Stamper <rubber-stamper@appspot.gserviceaccount.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
e6e99c54
|
2021-06-23T08:25:58
|
|
Vulkan: Load custom Vk Loader.
Uses a GN copy rule to duplicate the loader. Also updates volk to
load using the new custom loader. Once both are in place we can
go back and remove the copy.
Bug: chromium:1219969
Change-Id: I8c48d168a842539f7cdba1ebfdaf3b08c3e1990d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2982499
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
|
|
18e99f4a
|
2021-06-25T09:10:42
|
|
Add SearchType to OpenSharedLibraryWithExtension.
This fixes a bug in SystemInfo_vulkan where we were using the
System search path for the Vulkan loader when we prefer using the
custom ANGLE loader.
Bug: chromium:1219969
Change-Id: Iedf0fd11fe9ed8cc020b445ea9e12a7936937361
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2988791
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
913bc473
|
2021-05-07T11:37:01
|
|
Tests: Add --fixed-test-time option
Similar to --max-steps-performed, don't do any calibration and
just run flat out for a fixed amount of time.
Also rename --test-time to --max-trial-time to reflect what the
parameter actually does, limiting each trial to a ceiling.
Also remove a redundant check in doRunLoop.
Test: angle_perftests --fixed-test-time 20
Test: angle_perftests --max-trial-time 2
Bug: angleproject:5935
Change-Id: I854799a734cd6c019399f5a1444b4ef5f659ddf4
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2878937
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Charlie Lao <cclao@google.com>
Commit-Queue: Cody Northrop <cnorthrop@google.com>
|
|
68f06888
|
2021-03-23T16:38:50
|
|
Perf Tests: Trigger test failure on API errors.
A prior refactor had broken the method we used to cause test steps to
fail on API errors. This CL restores the path. We detected this when
analyzing a particular GLES trace that contained invalid GLES calls.
Bug: angleproject:5788
Change-Id: I26940e49cf73ce6050ea6ee274d5c5748835a167
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2782008
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
78dde332
|
2021-03-29T17:31:52
|
|
Move restricted traces to CIPD.
All traces are now stored as DEPS entries in CIPD. The auto-generation
script generates the DEPS entries. Note that we don't include DEPS in
the list of generated outputs to simplify other rollers. Also we update
auto-generation to include full sources list to allow 'gn analyze' to
work successfully.
Usees a trace fixture for common code. This will enable a more compact
trace without as much repeated code. We must land a set of re-trace
traces to avoid breakage.
Also includes a python script for uploading new traces to CIPD. The
script first checks if traces are already present in the cloud, and if
so it skips the upload. It will take a while to complete as the number
of traces grows larger as it takes a few seconds per trace.
The traces in this patch are also re-traced to use the common fixture
code instead of including duplicated code in each trace. They now form
a simple common interface and the autogenerated cpp is now simply a list
of properties.
I've also updated the capture/replay tests to use the simpler common
entry point integration. There is less auto-generated glue code now.
We now use a new serialized string query extension instead of calling
directly into ANGLE's internals.
Also includes a docs update. The capture/replay sample is broken and
we'll need to update it in a follow-up CL. Also includes a few necessary
fixes to the retracing script.
Bug: angleproject:5811
Change-Id: I977bc6dc56843c8966377fc445ae97e91e17319a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2797833
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
2c685a41
|
2021-02-03T11:20:20
|
|
Fix the perf test runner calibration.
This path was broken for the white box unit tests. Also adds argparse
handling to the runner so we can more flexibly override command line
arguments.
Previously the broken calibration was causing some of the tests to run
only a single test iteration when measuring. This could lead to low
quality measurements.
Bug: angleproject:5573
Change-Id: Ic1cb2b2553774a361325f290440c40b2ff90db5e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2672702
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Mohan Maiya <m.maiya@samsung.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
ae5a2047
|
2020-11-20T16:31:32
|
|
Trace Tests: Screenshot within the render loop.
Previously we were using a separate mechanism to capture screen
shots. This would cause very long traces to replay the entire
trace before normal execution. The Maps trace used hundreds of
frames and this was timing out the test step.
Instead we can capture a screenshot after the first time we run
a trace. This will normally not interfere with performance work
because we don't usually capture screenshots and do performance
work at the same time.
Bug: b/172977194
Change-Id: I12cfe2d051d41f1e000a98ac8aef16eb153879c0
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2552925
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
202304e7
|
2020-11-17T17:33:47
|
|
Trace Tests: Add renderer string logging.
This outputs the GL_RENDERER string in each test when using the
--verbose-logging command-line flag.
Bug: b/173518650
Change-Id: Id966294d50aaba0c8a1f3e010f76d5560e6f1288
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2545403
Reviewed-by: Charlie Lao <cclao@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
8b76d37f
|
2020-11-17T17:33:47
|
|
Trace Tests: Run one step per frame.
Instead of trying to make one step take one sequence of frames,
one run step per frame always. This will scale better for traces
that are very long and cannot complete in the short times we use
on the bots.
Adds an explicit step round-up feature so that we can always run
chunks of full trace sequences when we are not running with a
fixed maximum number of steps.
Bug: b/172977194
Change-Id: I0f069a66a86d8c4a698ebffb66782d13843539a5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2545884
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
fb34c93d
|
2020-11-09T19:30:17
|
|
Perf Tests: Add mutex around trace events.
This prevents a race where two threads try to write events
simultaneously.
Bug: b/172704839
Change-Id: Idd2f3b1c2a393488ea2270b91d579dc8630e59d5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2533496
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
489f2c83
|
2020-11-06T17:24:28
|
|
Perf Tests: Trace multiple threads.
This uses a little helper list to map thread IDs into simple serials
that we output as 'tid' in our trace JSON file. We can use this to
analyze the behaviour of the command processor thread.
Bug: b/172704839
Change-Id: Ic1b5f3ec8427cb310833d1f589503daec4812681
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2524542
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
06eaaac5
|
2020-11-06T16:33:13
|
|
Fix trace writing in angle_perftests.
We can use this to produce AGI-like reports in standalone desktop.
Bug: b/172704839
Change-Id: Ifc510232e3da81210e22429b6cea9800e2723e06
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2524540
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
4323ed0c
|
2020-11-06T12:45:12
|
|
Perf Tests: Fix test time with max steps performed.
This lets the default test time be 10 minutes when using a maximum
steps value. This will prevent the current 10 second limit from
kicking in and aborting. Also changes the --test-time override to
be more clearly a calibration time, and adds a new --test-time that
lets the user override the actual trial test time.
Bug: b/172480147
Change-Id: I630aa8d3c9e118bb5f0100b38f37e717e12f3538
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2523371
Reviewed-by: Paul Thomson <paulthomson@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
af16c638
|
2020-11-06T09:50:51
|
|
Perf Tests: Add --max-steps-performed argument.
This will limit the total number of steps run in the test.
When using this argument we skip warmp & calibration.
Renames the prior --steps argument to --steps-per-trial.
Bug: b/172480147
Change-Id: Iae7cce26d66ca0bbdb21c1f511a6bfa246bd204e
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2519404
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
3e5b6f81
|
2020-10-31T12:33:28
|
|
Enable writing histogram-set-json-format.
This uses the protobuf histogram functionality to build up a histogram
set. The test suites then output the histograms to JSON. This is only
implemented for angle_perftests.
Bug: angleproject:5161
Change-Id: Ia5a7868e8d8dcf4f13d83115ae622828c63ef0d9
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2482295
Reviewed-by: Brian Sheedy <bsheedy@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
77defeb0
|
2020-10-27T18:31:53
|
|
Add the ability to re-trace existing traces.
This new script runs the ANGLE trace tests with capture enabled to
generate updated replay cpp files. This allows us to update our
traces files to a new file format in one step.
Trace metadata (currently only the default FBO format) is preserved
between re-traces.
Currently only desktop-based retracing is supported. This means a
couple traces that require specific extensions don't run on all
platforms.
Bug: angleproject:5134
Change-Id: I7c923d89e33c18285ab36a7cee91f2fb735758eb
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2488130
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
626a4185
|
2020-10-27T18:31:52
|
|
Trace Tests: Allow runs against SwiftShader.
SwiftShader can allow for a platform-indendent replay. This is useful
for re-tracing. By default disable the SwiftShader test configs unless
we pass in --enable-all-trace-tests.
Also disable the mock ICD trace tests by default.
Bug: angleproject:5134
Change-Id: Ibf1f3ebd6c17f15123705b7d4d9ddd6308f873f3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2502311
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
5eebf3a7
|
2020-10-18T11:44:27
|
|
Perf Tests: Call finish every step in calibration.
Calibration is only supposed to take one second of warmup time.
Instead we were queuing up a large amount of GPU work and then janking
on the finish call at the end of calibration.
Fix this by calling glFinish repeatedly during calibration.
Bug: chromium:1136900
Change-Id: Idb2fefe17fcb9acbe688cff5a36f051732fc5b59
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2483462
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
51ff0d05
|
2020-10-16T11:54:04
|
|
Trace Tests: Add vsync-limited mode with a GPU timer.
Local testing on Linux shows that ANGLE and native have
the same wall-time, and ANGLE has a 10 ms vs native 6 ms
GPU time.
Bug: angleproject:5183
Change-Id: I26da678b984af916bfb40e1db6416f5df4f85c08
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2480782
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
a22a19a4
|
2020-10-04T13:16:46
|
|
Add a --no-finish option to the perf tests.
We can use this to re-capture replays.
Bug: angleproject:5134
Change-Id: I765e0bbb66cfdd822ffa981f2b5533157d024cac
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2449161
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
974c2371
|
2020-10-04T12:22:38
|
|
Perf Tests: Run at least 1 step.
We could end up in some very slow scenes (e.g. when debugging). This
could produce a step count of zero which doesn't make sense. Noticed
while capturing the T-Rex replay.
Bug: angleproject:5134
Change-Id: Ibf2439961dfb3c01649dcab37acead7b6e2ab345
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2449160
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
|
|
f8e5b97e
|
2020-09-28T23:31:41
|
|
Additional perf test cleanup & functionality.
Adds several new command line arguments and documents all args in the
README. Changes the trace tests to output milliseconds.
Bug: b/169600938
Change-Id: Ie8c7840a6a8958951992c806979dde6dc7d8a709
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2438194
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
5ec560fb
|
2020-09-17T17:18:25
|
|
Add restricted trace golden image script test.
The script reads the list of tests from the restricted traces JSON and
runs angle_perftests once for each test. Running the tests in a batch
seems unstable on some platfroms so running each test separately works
around that limitation. It does not yet interact with Skia Gold. The
Gold integration will come later once this basic skeleton is running
on the bots.
Also updates the perf tests to exit more quickly with --one-frame-only.
This will speed up the screenshots.
Example run with trigger.py:
https://chromium-swarm.appspot.com/task?id=4ecb985330a33910
Bug: angleproject:4090
Bug: b/168049670
Test: tested with trigger.py
Change-Id: I52820dfba2acfbc5d638673c37297877e8476adb
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2417641
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
a2d670c5
|
2020-08-15T21:13:33
|
|
Run trace tests against Vulkan Mock ICD.
Required an upstream fix:
https://github.com/KhronosGroup/Vulkan-Tools/pull/437
Also a workaround filed as an issue:
https://github.com/KhronosGroup/Vulkan-Tools/issues/445
Bug: angleproject:4950
Change-Id: I21333464d2a8e52d5d85d35654f0e6372f9650db
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2358517
Reviewed-by: Tobin Ehlis <tobine@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
ba7f8913
|
2020-09-12T23:23:01
|
|
Samples: Enable Debug callback.
This moves the debug callback code into a common location. For ease of
access it's in shader_utils since that file has access to the GL API.
Bug: angleproject:5040
Change-Id: Iab9de47c2d520a5618ea6825852f8afa63565c8a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2408715
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
e89eb912
|
2020-08-12T11:21:07
|
|
Reduce spam in native perf tests.
The QC driver was repeating HIGH priority messages about map flags.
Bug: angleproject:4943
Change-Id: I927c38249f840544a7c1f5e595fa34e706e401bd
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2352472
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
48d951e2
|
2020-07-30T15:32:03
|
|
Perf Tests: Handle logError callback.
This will force errors when ANGLE runs into Vulkan Validation Layer
failures and other internal bugs. Currently we could sometimes fail
silently. Especially on Android where the error would only show up in
logcat.
Bug: angleproject:4909
Change-Id: Ie2888805c577d9fbad417d45a33e3269966b7b25
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2330091
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
2d0265fa
|
2020-08-02T12:01:46
|
|
Trace Tests: Scale results based on frames in capture.
This now reports averaged time per frame instead of per "step".
Scores in general will be divided by 11. Results still reported
in nanoseconds.
Bug: angleproject:4920
Change-Id: I6f9dca75885cce358c3c7d1e871441969d8aa3d3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2334095
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
bba8daea
|
2020-07-27T20:07:53
|
|
Perf Tests: Use timestamp queries for GPU time.
Testing shows that using timestamp queries and deferring the
query get calls until after the frame produces little overhead.
Also this fixes a missing GPU time reset between iterations.
Bug: angleproject:4879
Change-Id: I2a566548add1536aab689cd969594a15f0628da3
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2321573
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
4e577a2a
|
2020-07-15T15:50:06
|
|
Perf Tests: Make "offscreen" a common test param.
This will let us use it in the trace perf tests.
Bug: angleproject:4845
Change-Id: I6188a0765c6f482bb38842faa4c58209c70b0153
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2300206
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
4e7a6a69
|
2020-07-15T19:35:43
|
|
Perf Tests: Fix trace test calibration.
The trace tests were including the startup time in their calibration
calculations. Then was forcing them to always render one step. This
fixes the calibration to attempt to get a more consistent measurement.
The trace tests now render many more steps than they did before. It
should increase the stability and decrease variance in our measurements
as they will be runnnig many more frames than before.
Also adds a verbose logging flag to help debugging.
Noticed this bug when working on the trace test offscreen mode.
Bug: angleproject:4845
Change-Id: Iff0c987008a935e7051fca34ef12f4433eb46092
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2300205
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
1265dbe6
|
2020-06-24T15:12:17
|
|
Update JsonCpp usage
This patch updates usages of the JsonCpp library, including the
following changes:
1. Removed unused version variable from the DEPS file.
2. Removed deprecated writer and reader usages, in favor of the new
builder pattern inside JsonCpp.
3. Modernized usage of the Json::Value types, including some rewrites to
avoid unnecessary heap allocations, and using the new foreach
iterators.
Bug: chromium:983223
Change-Id: If26abc8be677d905183a23498fbb81256854525c
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2265093
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Commit-Queue: Geoff Lang <geofflang@chromium.org>
|
|
aad69df9
|
2020-06-11T13:27:52
|
|
Run TracePerfTests in WGL configs.
Uses the "native" naming pattern.
Bug: angleproject:4728
Change-Id: I28cafcdb8c5c1fb48bba51adeca641e88bfe3446
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2241619
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
6de7ee52
|
2020-05-25T17:00:01
|
|
Clean up overlay RenderPass count reporting.
This fixes the trace perf test to accurately report how many
RPs in each frame. Instead of counting the RPs on a flush we now
count only on a swap call. This won't work for offscreen surfaces
which is fine - the overlay doesn't really have the same use for
offscreen rendering. Also ignores the first frame in graph data so
we can ignore the first setup frame in the trace tests.
Also skips the redundant extra "flush" call that would generate an
empty space in the RP graph.
Gives a cleaner measurement for optimizing the XFB RP count.
Bug: angleproject:4622
Change-Id: I5762c500cdb216700247095984ae62b4f8741602
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2215309
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
6c4af0c3
|
2020-05-20T14:54:39
|
|
Tests: Add a common framebuffer config for perf tests
As we add traces tests, some were recorded expecting depth as an
attachment to framebuffer 0. This causes errors when they
unconditionally query info about depth attachments. Since perf tests
have never requested a config before (they are all DONT_CARE for each
componenent), go ahead and set one shared by all.
Test: angle_perftests
Bug: b/157158456
Change-Id: I83ef1a58f246bfbc98c3bc7e024aca01560107f7
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2210966
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
5b35c7f6
|
2020-05-07T13:40:55
|
|
Fix up screenshot saving for trace tests.
Now works when run in a sequence. Also saves RGB images to avoid
issues with the alpha being inconsistent and also flips images
vertically to fix the rendering.
Bug: angleproject:4615
Change-Id: I8d3b38c5d914e0ca2227320ac42a0e28acd12c4d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2187971
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
d08f1d8d
|
2020-05-05T12:54:07
|
|
Perf tests: Add screenshot capture mode.
This adds a "--screenshot-dir" argument to capture screenshots.
If we're running with screenshot capture then the test will
early exit after the first capture. The screenshots use the same
naming pattern as the test name:
TracePerfTest.Run/vulkan_trex_200 -> angle_vulkan_trex_200.png
Note the screenshot dir is relative to the test binary directory,
not the CWD. Also adds a PNG saving utility function.
Bug: angleproject:4615
Change-Id: I1de8ae6a6e6892586bb0b743e7b9a842f90f98e5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2184834
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
930b2641
|
2020-03-23T14:47:03
|
|
Allow tests to run on native EGL.
Adds support for Linux and Android native EGL testing.
This can be useful for doing performance comparisons of ANGLE vs
a native GL driver. Only enabled for the trace perf tests due to
limitations in the test harness.
Bug: angleproject:4596
Change-Id: Iba6d3ccd7c1275cf095893fab824a0ea33dc3a79
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2116254
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
78c77361
|
2020-03-20T13:50:24
|
|
Perf: Add a single Manhattan trace
Now that TRex is up and working and Manhattan runs cleanly:
* Add a single compressed Manhattan trace.
* Allow errors from tests to fail the test
* Restore the working directory between each trace test
Test: angle_perftest --gtest_filter=TracePerfTest*
Bug: b:151349786
Bug: angleproject:3630
Bug: angleproject:4496
Change-Id: I9e9d9342efb23431380f484befb9a19915c4bd0a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2113230
Commit-Queue: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
42c453fe
|
2020-03-15T18:23:15
|
|
Move perf test common files to a source set.
Similarly to a prior patch this organizes the common perf test sources
into a single target. This will allow us to more easily support setting
compiler flags on particular perf test sources.
This also fixes some build errors that popped up when new warnings were
enabled.
Bug: angleproject:3630
Change-Id: I866885199f93a5055e8d160e39f72d1f8209bc75
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2103086
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
1393bf9e
|
2020-02-28T22:03:27
|
|
Vulkan: Timestamp internal RenderPasses in perf tests.
This change book-ends RenderPasses in the Vulkan back-end with
timestamp queries. This allows us to write out a trace of GPU/CPU that
we can inspect in Chrome. To enable the GPU trace you can define
ANGLE_ENABLE_VULKAN_GPU_TRACE_EVENTS in ContextVk.cpp and run the
angle_perftests suite with --enable-trace.
Note that we ran into some issues with the implementation not returning
monotonic timestamps. This may be a driver bug but bears more
investigation.
Bug: angleproject:4433
Change-Id: I0e6a364367f15183068e55686549cb418aa94c2a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2081380
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
391f38c9
|
2020-03-05T12:18:34
|
|
Capture/Replay: Add Manhattan to TracePerfTest
This CL also adds the ability for a perf test to
elect to ignore GL errors during replay. Otherwise,
errors will cause the test to fail. We have known
failures with Manhattan that don't affect replay.
Test: angle_perftests.exe --gtest_filter="*Trace*"
Bug: angleproject:4091
Change-Id: I36268a9f80754f338e52208273e889d2404b7eed
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2090663
Commit-Queue: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
9acaf662
|
2020-02-28T14:58:17
|
|
Add timer query support to TracePerfTests.
This can allow us to get some profiling measurements on ANGLE vs a
native driver. We correlate host times with GL times using the blocking
timestamp query. We also time RenderPasses/FBO switches using the
trace instrumentation we added in an earlier patch combined with timer
queries.
Not currently instrumented for the current tests. We'll need to re-run
capture to get the callbacks working correctly.
Bug: angleproject:4433
Change-Id: I8f61774a3a090ac9460a378d34715a72954d1331
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2080597
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
c7f63a01
|
2020-02-23T16:46:41
|
|
Fix debug callback in perf tests.
Was not outputting correctly. Also print more information like the type
and severity of the message. Moves some stringification code out of
RendererGL into common/utilities.
Bug: angleproject:3630
Change-Id: I029e279764fa4ae5cab8d903bd0ef7d376b7b846
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2067630
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Courtney Goeltzenleuchter <courtneygo@google.com>
Reviewed-by: Tim Van Patten <timvp@google.com>
|
|
39087f4a
|
2020-02-16T11:53:23
|
|
Refactor DrawCallPerf test parameters.
Use the new combiners functions added in an earlier CL. Makes it easy
to maintain bigger lists of test combinations.
Also a few other changes:
- removes some D3D9 perf testing since we don't maintain this config
- removes the "validation only" tests. these were mostly redundant
- makes the tests permutation combinations more consistent
Bug: angleproject:3630
Change-Id: I175d887a01b21123f83f9fa4f64dacaa2644147a
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2059468
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
519163ee
|
2020-02-15T12:20:17
|
|
Run debug callback in perf render tests.
This can catch unexpected GL errors from ANGLE or the system driver.
Only enable when ASSERTs are enabled to ensure that we get full
performance when profiling.
Helpful to diagnose incorrect perf test rendering. In this case it
helped diagnose that our frame tracing and replay do not correctly
remap uniform locations.
Bug: angleproject:3630
Change-Id: Ifd9f47196381e48f74a810ea0c9f9c0fd7a13a22
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2057352
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tobin Ehlis <tobine@google.com>
|
|
080d711b
|
2020-01-08T13:04:44
|
|
Tests: Add a perf test using traces
This test will load traces and cycle through a set of frames.
The new tests are:
TracePerfTest.Run/gl_trex_200_210
TracePerfTest.Run/gl_trex_800_810
TracePerfTest.Run/gl_trex_900_910
TracePerfTest.Run/gl_trex_1300_1310
To download the traces, you must be granted access and authenticate
with the cloud. See the steps in RestrictedTraces.md for this.
Then add the following to the end of your .gclient file:
"custom_vars": {
"checkout_angle_internal":"True"
},
And run `glcient runhooks`.
To enable building the test, add the following GN arg:
build_angle_trace_perf_tests = true
See the new markdown document for ways to execute the tests.
Bug: angleproject:3630
Change-Id: Ic25cd94fb36b35fcf187e88d3bf9548c65f59755
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1949605
Commit-Queue: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
4ffc254e
|
2020-01-08T09:58:24
|
|
Only initialize the timestamp query pools if the extension is available.
Also added a few assertions to ensure that timestamp queries aren't attempted
when support is missing.
Bug: angleproject:4114
Change-Id: Ie6d7d5face59f9bc137aebd86c9d0e965773e6e6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1992184
Commit-Queue: Eric Binet <ericbinet@google.com>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
57b37b6b
|
2019-09-25T18:29:28
|
|
Rename util/system_utils to util/test_utils.
This removes a GN naming conflict between util/system_utils and
common/system_utils. This conflict was preventing us from adding
unit tests to utils' version of system_utils. Since these functions are
only useful to tests and samples rename them test_utils for simplicity.
Will enable further development of ANGLE's standalone testing harness.
Bug: angleproject:3162
Change-Id: I9e34fb69f96c5de6dc2453fce4148a0f285e15ed
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1825268
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
e54d0f90
|
2019-06-30T03:26:18
|
|
Vulkan: Debug overlay
A debug overlay system for the Vulkan backend designed with efficiency
and runtime configurability in mind. Overlay widgets are of two
fundamental types:
- Text widgets: A single line of text with small, medium or large font.
- Graph widgets: A bar graph of data.
Built on these, various overlay widget types are defined that gather
statistics. Five such types are defined with one widget per type as
example:
- Count: A widget that counts something. VulkanValidationMessageCount
is an overlay widget of this type that shows the number of validation
messages received from the validation layers.
- Text: A generic text. VulkanLastValidationMessage is an overlay
widget of this type that shows the last validation message.
- PerSecond: A value that gets reset every second automatically. FPS is
an overlay widget of this type that simply gets incremented on every
swap().
- RunningGraph: A graph of last N values. VulkanCommandGraphSize is an
overlay of this type. On every vkQueueSubmit, the number of nodes in
the command graph is accumulated. On every present(), the value is
taken as the number of nodes for the whole duration of the frame.
- RunningHistogram: A histogram of last N values. Input values are in
the [0, 1] range and they are ranked to N buckets for histogram
calculation. VulkanSecondaryCommandBufferPoolWaste is an overlay
widget of this type. On vkQueueSubmit, the memory waste from command
buffer pool allocations is recorded in the histogram.
Overlay font is placed in libANGLE/overlay/ which gen_overlay_fonts.py
processes to create an array of bits, which is processed at runtime to
create the actual font image (an image with 3 layers).
The overlay widget layout is defined in overlay_widgets.json which
gen_overlay_widgets.py processes to generate an array of widgetss, each
of its respective type, and sets their properties, such as color and
bounding box. The json file allows widgets to align against other
widgets as well as against the framebuffer edges.
Two compute shaders are implemented to efficiently render the UI:
- OverlayCull: This shader creates a bitset of Text and Graph widgets
whose bounding boxes intersect a corresponding subgroup processed by
OverlayDraw. This is done only when the enabled overlay widgets are
changed (a feature that is not yet implemented) or the surface is
resized.
- OverlayDraw: Using the bitsets generated by OverlayCull, values that
are uniform for each workgroup (set to be equal to hardware subgroup
size), this shader loops over enabled widgets that can possibly
intersect the pixel being processed and renders and blends in texts
and graphs. This is done once per frame on present().
Currently, to enable overlay widgets an environment variable is used.
For example:
$ export ANGLE_OVERLAY=FPS:VulkanSecondaryCommandBufferPoolWaste
$ ./hello_triangle --use-angle=vulkan
Possible future work:
- On Android, add settings in developer options and enable widgets based
on those.
- Spawn a small server in ANGLE and write an application that sends
enable/disable commands remotely.
- Implement overlay for other backends.
Bug: angleproject:3757
Change-Id: If9c6974d1935c18f460ec569e79b41188bd7afcc
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1729440
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
f0dd087e
|
2019-08-23T15:45:34
|
|
Move timer functionality from util/ to common/
The main timer functionality (get absolute time) is moved to common/ for
use in ANGLE itself (in upcoming overlay change). util/Timer.h is no
longer an abstract class and uses this functionality to implement the
timer.
Bug: angleproject:3757
Change-Id: I3fe418778d80d1089c9bfe43a9e8098e43236f18
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1769061
Reviewed-by: Cody Northrop <cnorthrop@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Tim Van Patten <timvp@google.com>
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
2f4a7518
|
2019-08-16T14:09:13
|
|
Refactor perf tests to fix metric/story swapping
Refactors the perf tests to fix the issue of metric and story being
swapped, which causes issues when trying to convert to histograms.
Specifically, does the following:
1. Rolls the version of src/tests/perf_tests/third_party/perf/ to
Chromium 476dae823269c8d05b544271af97ad1adb0db8ee
2. Switch to using PerfResultReporter instead of PrintResult directly.
3. Split RenderTestParams::suffix into backend and story; backend is
used as part of the metric, while story is used as the story.
4. Remove the "average" metric that was being automatically reported
by ANGLEPerfTest, as reported results are automatically averaged.
5. Update the reported metric to more clearly distinguish between
test, backend, and metric. It is now name_backend.metric. e.g.
DrawCallPerf_vulkan.wall_time.
Bug: chromium:923564,chromium:924618
Change-Id: I00cc191407052f23df57dbfa53b6fb088fc26960
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1762360
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
|
|
9d737966
|
2019-08-14T12:25:12
|
|
Standardize copyright notices to project style
For all "ANGLE Project" copyrights, standardize to the format specified
by the style guide. Changes:
- "Copyright (c)" and "Copyright(c)" changed to just "Copyright".
- Removed the second half of date ranges ("Y1Y1-Y2Y2"->"Y1Y1").
- Fixed a small number of files that had no copyright date using the
initial commit year from the version control history.
- Fixed one instance of copyright being "The ANGLE Project" rather than
"The ANGLE Project Authors"
These changes are applied both to the copyright of source file, and
where applicable to copyright statements that are generated by
templates.
BUG=angleproject:3811
Change-Id: I973dd65e4ef9deeba232d5be74c768256a0eb2e5
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1754397
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
69194e5f
|
2019-07-17T15:35:10
|
|
Reduce variance in TextureUploadPerf.
This reduces the test time to run in several ms instead of almost
seconds per iteration. It allows us to use the perf test runner harness
to check test performance more accurately by increasing the sample
count.
It first reduces the test iteration count to reduce total test time.
It also refactors the test contents to not allocate new objects and
memory with every test iteration. This gives a better sampling of what
the test is obsering: texture upload performance.
Also allows tests that like to track GPU time to be used with the NULL
device option.
Bug: angleproject:3712
Change-Id: Idacc3e3b424f8882d7680769b27d1f04146ea65d
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1707112
Reviewed-by: Tobin Ehlis <tobine@google.com>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
663b0481
|
2019-07-11T12:37:35
|
|
Load correct opengl32.dll
Tests are still loading ANGLE's opengl32.dll instead of system's
This is because external drivers are trying to load opengl32.dll
Workaround: renamed to libGL.dll, need to change back when copying
dll over for loading
Bug: angleproject:3641
Change-Id: I051adc1fbb488a704c99718ad4a2f74bb7756cbd
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1697290
Commit-Queue: Clemen Deng <clemendeng@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
a71a8c66
|
2019-07-04T09:42:31
|
|
ANGLE tests loading wrong opengl32.dll
ANGLE tests try to load opengl32.dll from ANGLE directory instead of system
Bug: angleproject:3645
Change-Id: I3a8cea37252d13e915ff54ae6bbac920db16e4c4
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1688544
Commit-Queue: Clemen Deng <clemendeng@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Geoff Lang <geofflang@chromium.org>
|
|
b5a13bec
|
2019-06-28T14:11:17
|
|
Prepare for JsonCpp roll in Chromium
This patch is the first of a multistep process to upgrade JsonCpp
from its current legacy version to the latest version. A minor fix to
how Json::Value is used is included, as well as properly pulling in the
jsoncpp_config where jsoncpp is used, allowing for us to roll Chromium
while Angle uses the deprecated Reader/Writer classes.
After this patch is landed, the Angle dependency in Chromium can be
rolled as part of rolling JsonCpp, then Angle can update to the new
StreamWriter/CharReader classes.
Bug: 961513
Change-Id: Idd1d61270aa0a7e4a91983b4c22a24e2fdaafe49
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1682815
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
d5cef305
|
2019-06-19T14:21:33
|
|
Vulkan: Add more trace events.
This captures a bit more information about where the call time goes
during a flush. It can show at least on desktop NVIDIA that we spend a
fair amount of time in vkQueuePresentKHR.
Bug: angleproject:3117
Change-Id: I2d0195b9338bcac80e8dd8cfb550402271f286f9
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1650787
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
eee67c6e
|
2019-06-18T13:00:43
|
|
Fix Platform.h integration with ANGLE
A function was renamed which broke the platformMethods table strings.
This CL reverts the rename (originally from crrev/c/1660952)
Bug: angleproject:1621
Change-Id: I8121a4956ba1d3e3c5036f72f6f1fdf5509dc491
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1664792
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jonah Ryan-Davis <jonahr@google.com>
|
|
beb0eb2d
|
2019-06-14T15:10:33
|
|
Clean up workarounds/features to single location.
Rename all workarounds structs to features, and move the lists to a
shared location in include/platform (to help with documentation,
see:
https://cs.chromium.org/chromium/src/ui/gl/gl_switches.cc?sq=package:chromium&g=0&l=69)
Bug: angleproject:1621
Change-Id: I4069f08131db5e886047a007efb5d7764dfee5f2
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1660952
Commit-Queue: Jonah Ryan-Davis <jonahr@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
3089f92e
|
2019-05-01T15:11:46
|
|
Move EGL Display config into EGLPlatformParameters.
This CL moves all of the EGLDisplay configuration into an EGL platform
struct. Consolidating display configuration in the struct allows us to
move configuration out of the test constructor. Then when we filter
test configs we don't need to wait for the individual test setup.
Bug: angleproject:3393
Change-Id: I5bd06dcdc9f2867ebc43c1d4984077ada35cafc8
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1574674
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
0659c991
|
2019-04-30T16:14:43
|
|
Set swap interval explicitly.
The swap interval can be changed independent of the Surface config.
Thus it makes more sense to set it explicitly in test setup. This
simplifies the test config.
Also updates some of the API for GLWindowBase. Return an explicit
error from makeCurrent.
Bug: angleproject:3393
Change-Id: Ic62b33018e872bc0e38f2848e2427ed898b60749
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1574672
Reviewed-by: Geoff Lang <geofflang@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
91524d8c
|
2019-04-02T15:31:30
|
|
Disable Clear and TextureUpload perf tests on D3D11
The tests crash on Nvidia in the driver.
Bug: 945415
Change-Id: I0f635eb5d99ecaeeaff055793b1b5fd3a314b055
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1548440
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
50fb5cc9
|
2019-03-25T15:10:34
|
|
Disable GPU perf tracking on D3D
Causes crashes.
Bug: angleproject:3310
Bug: 945415
Change-Id: I59c5fee0afaf74c99ac19168fbfd96de04796add
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1534946
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
0f566fc7
|
2019-03-20T11:36:29
|
|
Introduce ConfigParameters test helper struct.
This allows us to more easily compare sets of parameters used in our
tests. The config parameters are stuff like the red / gree / blue bits
used in an EGL config. Or particular sets of extensions or other
EGL options.
This will more easily allow us to determine when we need to use a new
EGL display instead of reusing a prior.
Bug: angleproject:3261
Change-Id: Ia1f0ede988e0b4084fbb4d55097e94fd89ee4899
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1531535
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
f576a708
|
2019-02-04T16:52:12
|
|
Add glmark2 to angle_perftests
Bug: angleproject:3125
Change-Id: I9242743c6b5c6e18d0a23ff853ef6b9b370865a6
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1452956
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Cody Northrop <cnorthrop@google.com>
|
|
77874e2c
|
2019-02-13T09:15:06
|
|
Vulkan: Reduce Default Descriptor Pool Max Sets
The current value of kDefaultDescriptorPoolMaxSets is 2048, which
allocates too much memory on a Pixel device, causing Android's low
memory killer to terminate dEQP (leading to a test failure). This change
reduces the value of kDefaultDescriptorPoolMaxSets to 128 to reduce the
memory usage during the test. We suspect a later change may be required
to dynamically grow the descriptor pools and update the recycling
behavior.
This change also exposed a bug in the ANGLERenderTest where the swap()
call was being skipped for null devices. This was causing code that
ANGLE requires to clean up and re-use resources to be skipped. Enabling
the swap() call fixes this issue as well as improves performance for
the test overall. This does make comparing old performance values to new
difficult (if not entirely incorrect), so take care when doing so.
dEQP-EGL.functional.multicontext.non_shared_clear
Bug: angleproject:3056
Bug: angleproject:3135
Change-Id: Ibdfd3a5eb436a7ba4e45002ebef5809afe777af1
Reviewed-on: https://chromium-review.googlesource.com/c/1470830
Commit-Queue: Tim Van Patten <timvp@google.com>
Reviewed-by: Tobin Ehlis <tobine@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
3571ad49
|
2019-02-06T12:24:19
|
|
Vulkan:Adding Cmd Buffer Reset tests
Resetting cmd buffers in various ways to understand VK driver perf on
various devices.
Also includes some changes to get the chromium APK wrapper for
angle_white_box_perftests compiling.
Bug: angleproject:3137
Change-Id: I56bd163cdd6605ce09effe509a9c7ac5e0008e7a
Reviewed-on: https://chromium-review.googlesource.com/c/1456482
Commit-Queue: Tobin Ehlis <tobine@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
4704eb02
|
2019-01-25T13:25:52
|
|
Add possibility to track GPU time in perf tests
This paves the way for perf tests that measure GPU performance.
Bug: angleproject:2999
Change-Id: I9d49d3e1256d8d18514885ae63264a5fbc5c29d1
Reviewed-on: https://chromium-review.googlesource.com/c/1436839
Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
|
|
5fe7c5b9
|
2019-01-17T12:16:34
|
|
Include common/platform.h where used
Bug: 922443
Change-Id: I35b9e34266d4a15f8d0769c2770801b1b0511398
Reviewed-on: https://chromium-review.googlesource.com/c/1418091
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Yuly Novikov <ynovikov@chromium.org>
|
|
1f56ed2a
|
2019-01-03T15:24:22
|
|
Add WGLWindow and WGL test configs.
WGLWindow lets us use a Windows driver's bindings instead of ANGLE.
This only works if the underlying driver supports OpenGL ES
compatibility.
Also adds the WGL headers, WGL XML, and a specialized WGL loader.
Because of a small driver issue with NVIDIA I added a retry for the WGL
Window initialization.
Bug: angleproject:2995
Change-Id: Ie5148ece470dd03df33015f4919ad1fa79a859ec
Reviewed-on: https://chromium-review.googlesource.com/c/1366021
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
ad398ee8
|
2019-01-03T13:01:08
|
|
Free OSWindow and EGLWindow through helpers.
This cleans up any potential problems with allocating and freeing
resources in different shared objects or DLLs. Previously we were
using a dynamically linked allocation function and then calling
the standard delete function.
Also adds a base class helper for EGLWindow. Will base the WGL
Window class on this.
Needed for running ANGLE tests against native drivers.
Bug: angleproject:2995
Change-Id: Ic92b447649ebb32c547605c20086c07a601842f0
Reviewed-on: https://chromium-review.googlesource.com/c/1393443
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
ba319ba3
|
2018-12-29T10:29:33
|
|
Re-land "Load entry points dynamically in tests and samples."
Fixes the Android/ChromeOS/Fuchsia builds by using consistent EGL
headers.
This CL adds a dynamic loader generator based on XML files. It also
refactors the entry point generation script to move the XML parsing
into a helper class.
Additionally this includes a new GLES 1.0 base header. The new
header allows for function pointer types and hiding prototypes.
All tests and samples now load ANGLE dynamically. In the future this
will be extended to load entry points from the driver directly when
possible. This will allow us to perform more accurate A/B testing.
The new build configuration leads to some tests having more warnings
applied. The CL includes fixes for the new warnings.
Bug: angleproject:2995
Change-Id: I5a8772f41a0f89570b3736b785f44b7de1539b57
Reviewed-on: https://chromium-review.googlesource.com/c/1392382
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
9f088621
|
2018-12-29T20:46:15
|
|
Revert "Load entry points dynamically in tests and samples."
This reverts commit 03923558a7103827ffec6a4d2a1453ed91f01c6f.
Reason for revert: fails compilation on Android, ChromeOS and Fuchsia during roll https://chromium-review.googlesource.com/c/chromium/src/+/1392624
Original change's description:
> Load entry points dynamically in tests and samples.
>
> This CL adds a dynamic loader generator based on XML files. It also
> refactors the entry point generation script to move the XML parsing
> into a helper class.
>
> Additionally this includes a new GLES 1.0 base header. The new
> header allows for function pointer types and hiding prototypes.
>
> All tests and samples now load ANGLE dynamically. In the future this
> will be extended to load entry points from the driver directly when
> possible. This will allow us to perform more accurate A/B testing.
>
> The new build configuration leads to some tests having more warnings
> applied. The CL includes fixes for the new warnings.
>
> Bug: angleproject:2995
> Change-Id: I6726d4163f7a6e54d2482f094c0a952f59702a05
> Reviewed-on: https://chromium-review.googlesource.com/c/1359516
> Commit-Queue: Jamie Madill <jmadill@chromium.org>
> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
TBR=ynovikov@chromium.org,jmadill@chromium.org,syoussefi@chromium.org
Change-Id: I902bec2d733c2b879be29c02ab52a0b7d4eaa077
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: angleproject:2995
Reviewed-on: https://chromium-review.googlesource.com/c/1392381
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Yuly Novikov <ynovikov@chromium.org>
|
|
03923558
|
2018-12-29T10:29:33
|
|
Load entry points dynamically in tests and samples.
This CL adds a dynamic loader generator based on XML files. It also
refactors the entry point generation script to move the XML parsing
into a helper class.
Additionally this includes a new GLES 1.0 base header. The new
header allows for function pointer types and hiding prototypes.
All tests and samples now load ANGLE dynamically. In the future this
will be extended to load entry points from the driver directly when
possible. This will allow us to perform more accurate A/B testing.
The new build configuration leads to some tests having more warnings
applied. The CL includes fixes for the new warnings.
Bug: angleproject:2995
Change-Id: I6726d4163f7a6e54d2482f094c0a952f59702a05
Reviewed-on: https://chromium-review.googlesource.com/c/1359516
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
720ca449
|
2018-12-17T14:04:10
|
|
Pass GN header visibility check.
This fixes a few things:
* removes includes that weren't supposed to be present
* scopes some compiler code into white_box_perftests
* makes version.h/commit and angle_common id more visible
* roll zlib to a version that passes check
This should help prevent build problems from popping up in the
downstream Chromium build. We could also potentially look at
including gn check in our CQ recipe.
Bug: chromium:915429
Change-Id: I350f543e16de13c84eb2c43260f4966d47185114
Reviewed-on: https://chromium-review.googlesource.com/c/1380771
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
35cd7332
|
2018-12-02T12:03:33
|
|
Refactor test shader style.
This change enforces a lot more consistency. We pass const char * to
the Compile functions instead of std::string. Also fixes the
indentation of C++11 block comments to be more consistent.
Bug: angleproject:2995
Change-Id: Id6e5ea94055d8cbd420df4ea2e81b2d96cb5ce78
Reviewed-on: https://chromium-review.googlesource.com/c/1357103
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
b980c563
|
2018-11-27T11:34:27
|
|
Reformat all cpp and h files.
This applies git cl format --full to all ANGLE sources.
Bug: angleproject:2986
Change-Id: Ib504e618c1589332a37e97696cdc3515d739308f
Reviewed-on: https://chromium-review.googlesource.com/c/1351367
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
|
|
0e604f9f
|
2018-11-19T21:33:25
|
|
ANGLE Perf Test fixes.
Includes the following fixes:
* Removes obsolete global variable.
* Disables trace event collection when trace events disabled.
* Fixes calibrated steps getting stuck after first test.
* Scale calibrated steps with actual time after glFinish.
Should fix a few crashes and timeouts with the perf tests.
Bug: angleproject:2923
Change-Id: I9ba1c042dee31cf2400ccbeedca0d497ed52fb12
Reviewed-on: https://chromium-review.googlesource.com/c/1343677
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
f3acb8c1
|
2018-11-14T16:24:25
|
|
Retool perf test running.
This change does a few things involving the way we run the perf tests:
- Perf test runner can capture different metrics
Useful for capturing the new "nanoSecPerIteration" metric.
- Removes the "score" metric
We'll move to the new time-based metrics. These new metrics are scaled
correctly with iteration counts.
- Runs three trials per perf test
This gives more measurements per test. Each trial is approximately one
second. First the perf tests set a fixed number of iterations after
calibrating the number of steps that we can run in one second. After
that the three trials are run. This should give more stable results.
- Apply more CPU stabilization on Windows
Use SetPriorityClass to apply more CPU priority. Also upgrade
SetThreadPriority to the highest level.
- Always build the Vulkan command buffer test
This catches build regressions more easily. We still skip the test on
non-Android platforms.
Bug: angleproject:2923
Change-Id: I7da234c5af07775ba4a232bb8d65e0138ee7073f
Reviewed-on: https://chromium-review.googlesource.com/c/1330262
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
557a1ee4
|
2018-10-30T17:26:24
|
|
Make perf tests faster in correctness-only mode.
When running with "--one-frame-only" we can also skip the test warmup.
Also we can reduce the internal iteration count to 1 to make the tests
as fast as possible.
Bug: angleproject:2923
Change-Id: I2f82ae0dd237767ea7b15074e459ed1094ba9943
Reviewed-on: https://chromium-review.googlesource.com/c/1308737
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|
|
3402d523
|
2018-10-30T15:14:52
|
|
Try to reduce variance in angle_perftests.
This change does a few things:
- make perf test runner script print % variation instead of stddev
This makes it a bit more clear how much variance there is.
- stabilize CPU in the render perf tests
Setting a thread affinity and priority should stop from switching cores
during the run. Hopefully can prevent background noise from changing
the test results.
- warm up the benchmark with a few iterations
This should hopefully make the test results a bit more stable.
- output a new normalized perf result value
The new result is normalized against the number of iterations. So it
should hopefully be stable even if the number of iterations is changed.
- increases the iteration count in the draw call perf tests.
These tests were completely dominated by SwapBuffers time. Increasing
the iterations per step means we actually are bottlenecked on CPU time
instead.
Bug: angleproject:2923
Change-Id: I5ee347cf93df239ac33b83dc5effe4c21e066736
Reviewed-on: https://chromium-review.googlesource.com/c/1303679
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
|
|
0ea96210
|
2018-10-30T15:14:51
|
|
Vulkan: Enable more perf tests.
Also fixes the check for the correct UINT index extension that wasn't
available on Vulkan. Also includes a workaround for the mock ICD not
implementing buffer state for index ranges.
Bug: angleproject:2923
Change-Id: Iab35809d15f890525a9e658d4148272c46cf1320
Reviewed-on: https://chromium-review.googlesource.com/c/1308733
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>
|