add MProfiler/trace performance tracing when using Pixar's batch renderer #260
Conversation
This category will be used to track viewport performance when using Pixar's "batch renderer" Hydra integration.
| MProfilingScope profilingScope( | ||
| ProfilerCategory, | ||
| MProfiler::kColorG_L3, | ||
| "Batch Renderer Adding Shape Adapter"); |
There was a problem hiding this comment.
Since you are in the batch renderer category, what's the advantage of adding Batch Renderer to every event name? It makes names very long.
There was a problem hiding this comment.
I was finding the full event name useful when looking at the CPU or thread views in the Profiler vs. the category view. That way I could see batch renderer performance in the context of a full scene render and not have to turn off any other categories.
There was a problem hiding this comment.
Some additional tricks related to profiler and category view. Explore the option to collapse events. This reduces the noise by collapsing all events to starting one on a thread. What will remain is a view of the current concurrency level.
You can find this option in the right-click menu, acting on selected events.
…nderer The MProfilerScopes added are helpful for tracking overall time spent in the batch renderer when using Maya's Profiler. The trace performance tracking is good for looking at the time spent in batch renderer setup and overhead versus the time spent in underlying Hydra and USD libraries, since those are also instrumented using trace.
…ProxyShapeBase It turns out that a significant amount of time is spent generating bounding boxes when using Pixar's batch renderer. Adding the trace in the proxy shape's boundingBox() function ensures that that time is captured when looking at performance data.
1a4f030 to
40ec8d6
Compare
This takes a cue from recent work in the VP2.0 render delegate and adds some performance tracking to Pixar's batch renderer Hydra integration. This is proving to be helpful as we work to deprecate our usage of the legacy viewport, and also as we look towards eventually adopting the VP2.0 render delegate.
The MProfilerScopes added are good for tracking overall time spent in the batch renderer when using Maya's Profiler. The trace performance tracking is good for looking at the time spent in batch renderer setup and overhead versus the time spent in underlying Hydra and USD libraries, since those are also instrumented using trace.
The one change here that affects users not using the batch renderer is the addition of a
TRACE_FUNCTION()call inMayaUsdProxyShapeBase::boundingBox(). A lot of time is currently spent generating bounding boxes (serially) using the batch renderer, so that may also be a current bottleneck for the VP2.0 render delegate.