Parallel shader compilation is broken

[shanumante-sc]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow.js): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): M1 Mac 13.3.1 (22E261)
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: Reproducible on any device
• TensorFlow.js installed from (npm or script link):
• TensorFlow.js version (use command below): 4.2.0
• Browser version: Chrome Version 112.0.5615.49
• Tensorflow.js Converter Version: N/A

Describe the current behavior

• We use ENGINE_COMPILE_ONLY to speed up initial model loading time by not blocking on shader compilation status checks.
• After Use VAOs for save+restore of vertexAttribPointer state between different webgl programs. #6913, even if we set ENGINE_COMPILE_ONLY env variable, we end up calling getAttributeLocation when we call bindVertexProgramAttributeStreams.
• This introduces a synchronization point for the graphics pipeline and we block on shader compilation. A side issue is that TFJS calls getAttributeLocation without first checking for gl.LINK_STATUS
• See screenshot below which contains Chrome trace of a model which is being run with ENGINE_COMPILE_ONLY (running on M1 mac Chrome)

Describe the expected behavior

• TFJS should check ENGINE_COMPILE_ONLY before calling bindVertexProgramAttributeStreams so that we do not stall GPU pipeline. We can set up the VAO in checkCompileCompletion.

Standalone code to reproduce the issue

• Run any network with ENGINE_COMPILE_ONLY set to true, capture chrome perf trace, and see that getAttributeLocation is called for each call to createProgram

Other info / logs Include any logs or source code that would be helpful to
diagnose the problem. If including tracebacks, please include the full
traceback. Large logs and files should be attached.

[Linchenn]

@shanumantesc Thanks for reporting this! I just verified the fix and it should work now.

[gaikwadrahul8]

Hi, @shanumantesc

I see this PR #7587 got merged to take care of your issue so if your issue got resolved could you please close this issue now ? Thank you!

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No