jsoncpp causes some flaky failures because of inconsistencies in nested
DEPS. We barely use it for JSON serialization so the simplest solution
is to just manually write the serialization for perf test tracing data.
Bug: dawn:208
Change-Id: Ie5e4b5436e2c9e32e6817d64c6e95c774cb8751f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14720
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
Results should be printed METRIC: STORY, not the other way around.
Also, story names shouldn't have slashes as it will parse as separate
chart segments and won't be allowed when we switch to using histograms.
Bug: dawn:208, dawn:311
Change-Id: Ifc893e5aa94eddcb3a08c0d4aff66b7a0f41620b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14740
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
As part of moving to webgpu.h's version of the Instance, the setters
will be replaced with a descriptor, and getters don't usually exists for
things set with descriptors (except getBindGroupLayout).
BUG=dawn:22
Change-Id: I05be8ebf241b33d019d521e40bfef8e49cdab07d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14540
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
On Android printing to stdout doesn't show in logcat. Use InfoLog
instead of fprintf(stdout) so that the message get routed to
android_log.
BUG=dawn:286
Change-Id: I067a12d8a790e42d75617c760a2693a3d94ebda6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14481
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
The LogMessage::LogMessage constructor was redefining a symbol that
exists in re2 inside of Chromium. So we namespace Log inside dawn:: to
avoid the conflict.
BUG=dawn:302
Change-Id: Ida349208e2c6fe9ac032e1bd8cd442dff0b3f6bc
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14320
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
The Vulkan spec mandates support for only one or the other, which is
why we have the concept of a depth24plus format. This also adds a Toggle
to test both formats in DepthStencilStateTests.
Finally this renames ForceWorkarounds to ForceToggles because toggles
can be more than just workarounds.
BUG=dawn:286
Change-Id: I5b5dc582ffd4ee61c51e3e75563aec815c580511
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14103
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: David Turner <digit@google.com>
On Android iostream doesn't appear in logcat, the system log that's
often used for printf debugging. Introduce Chromium/ANGLE like logging
that looks like the following:
InfoLog() << stuff << stuff;
This makes sure the message is put in logcat on Android and removes
static initializers from <iostream>
BUG=dawn:286
Change-Id: Ie0d018f49bcac1a7b740739a6e59d45ae6728638
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14102
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: David Turner <digit@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
This patch factors resource usage tracking so it is separate from
command validation, allowing the bulk of command validation to be
completely skipped.
In DrawCallPerfRun/Vulkan_DynamicPipeline_DynamicBindGroup, disabling
validation cuts roughly 74 nanoseconds (20%) of CPU time, per draw.
In DrawCallPerfRun/Vulkan_DynamicBindGroup, disabling validation
cuts roughly 35 nanoseconds (17%) of CPU time, per draw.
In DrawCallPerfRun/Vulkan_MultipleBindGroups, disabling validation
cuts roughly 45 nanoseconds (14%) of CPU time, per draw.
Bug: dawn:271
Change-Id: I517b85840ba18c6a554b83f34a1d0aef1a8c56a6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/13520
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
The Metal driver calls command buffer completion callbacks on a
separate thread so enqueueing these trace events needs to be made
thread-safe. In the future, Dawn will probably have other threads
that also require thread-safe tracing.
In this CL, each thread records trace events into its own buffer,
and all buffers concatenated when trace events are acquired.
Bug: dawn:254, dawn:208
Change-Id: I0f1abd404568d838091066a8f27a3bf98fa764b9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/13080
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
These tests draw a simple triangle with many ways of encoding commands,
binding, and uploading data to the GPU. This tests the performance of Dawn's
validation, command recording, and submission by switching buffer
bindings, bind groups, and pipelines. Some test instantiations upload
per-draw data to test efficiency of resource transitions or pre-record
commands in a render bundle.
Change-Id: I9994cd96ef5988cca410462418792f28161c0526
Bug: dawn:208
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12460
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Explicit enums are simpler to use in Dawn and allow only specific
categories which the perf tests understand.
Also adds trace events around command recording and validation on
all backends.
Bug: dawn:208
Change-Id: I7859ffd6668b20893780c6081bf2c9019a7115e0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12781
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Previously, the trace event buffer was cleared after *all* tests
completed. In order to avoid allocating a ton of space to hold
traces, this patch factors the trace event recording so that the
buffer is processed and reset at the end of each test trial.
This patch also prepares for the future where trace events will be
processed at the end of each trial to compute additional metrics.
Bug: dawn:208
Change-Id: If2ed193ee47794c666df9f0b369ec1ce660b177f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12780
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Trace data can be used to build additional metrics which measure
validation costs, GPU time, etc. It will also be helpful to store in
the test output for later analysis.
This CL also adds jsoncpp as a DEP so we can dump trace file json
output.
Bug: dawn:208
Change-Id: Ia6c05ca90aecae308ee6a4fd11e5f43bb03b1dc9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12080
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
The current BufferUpload perf test calls WaitForGPU at the end of
every step to time how long the batch of uploads took. Future tests
will want to have multiple Steps or Frames in flight to mimic real
applications that use double or triple buffering. This patch removes
WaitForGPU and adds a |maxStepsInFlight| argument which a test can
set to configure how many steps can be running on the GPU.
For the BufferUpload tests, this is still 1.
Bug: dawn:208
Change-Id: I95ae12b2358e0283088205bd2ffd20532ea62871
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12302
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This patch adds a perf test harness for Dawn and a simple test of
buffer upload performance. The test harness is based off of ANGLE's
perf tests.
Because perf tests are parameterized to support multiple test
variants, this patch also adds DawnTestWithParams and ParamGenerator
to support instantiating tests with additional parameters.
Bug: dawn:208
Change-Id: I60df730e9f9f21a4c29fc21ea1a8315e4fff1aa6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/10340
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
