Improve call site detection in capture mode and add it to hw decoder bench

[rostan-t]

Category:

New feature (non-breaking change which adds functionality)

Description:

Currently, the following fails to capture properly:

for _ in range(WARMUP_EPOCHS):
    run_epoch(reader)

for _ in range(NUM_EPOCHS):
    run_epoch(reader)

Operators were first captured in the warmup loop and the full call chain is used to identify a node so operators in the second loop fail to capture and fall back to normal execution.

This PR fixes this by trimming the call chain to only the required size. It also adds capture mode to L1_decoder_perf.

Additional information:

Affected modules and functionalities:

Dynamic capture mode.

Key points relevant for the review:

Does call chain trimming introduce silent regressions that I didn't think of?

Tests:

• Existing tests apply
• New tests added
  • Python tests
  • GTests
  • Benchmark
  • Other
• N/A

Checklist

Documentation

• Existing documentation applies
• Documentation updated
  • Docstring
  • Doxygen
  • RST
  • Jupyter
  • Other
• N/A

DALI team only

Requirements

• Implements new requirements
• Affects existing requirements
• N/A

REQ IDs: N/A

JIRA TASK: N/A

[dali-automaton]

CI MESSAGE: [61565034]: BUILD STARTED

[greptile-apps]

Greptile Summary

The PR shortens dynamic-capture call chains while preserving the frames needed to identify operator sites, and enables capture mode in the hardware-decoder benchmark.

• Tracks source-analysis proof depth and compacts captured call-site tries before replay.
• Adds warmup and nested proof-frame capture coverage.
• Adds an NDD capture benchmark run to decoder-performance QA.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
dali/python/nvidia/dali/experimental/dynamic/_capture.py	Adds proof-preserving call-trie compaction and early terminal lookup; no eligible residual issue was established.
dali/python/nvidia/dali/experimental/dynamic/_source_analysis.py	Propagates the stack depth needed to prove invariant arguments across nested classifier frames.
dali/test/python/experimental_mode/test_capture.py	Adds capture coverage for repeated warmup epochs and nested proof frames.
internal_tools/hw_decoder_bench.py	Adds optional capture execution to the dynamic RN50 decoder benchmark.
qa/TL1_decoder_perf/test.sh	Runs capture-mode decoder performance as an additional required QA benchmark.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A["Trace dynamic operator call"] --> B["Classify arguments and required proof depth"]
    B --> C["Record full call chain in trie"]
    C --> D["Compact to shortest operator-unique prefix"]
    D --> E["Replay lookup walks runtime frames"]
    E --> F["Return captured node"]

Loading

_{Reviews (4): Last reviewed commit: "Update copyright year in hw_decoder_benc..." | Re-trigger Greptile}

[JanuszL]

Do you expect perf to match pipeline mode MIN_PERF?

[rostan-t]

When I measure on my workstation, even when tweaking parameters capture mode is consistently slower but the difference is under 3%.
This is not the same machine as used in CI, it's an RTX 6000 PRO but the CPU only has 16 cores.

I enabled L1 tests when triggering CI so we'll see.

[rostan-t]

Results from CI:

• Pipeline mode throughput: 21567.07 frames/sec
• Dynamic mode throughput: 17244.40 frames/sec
• Dynamic capture mode throughput: 21525.88 frames/sec

Looks like capture mode does match the performance of pipeline mode 🎉

[rostan-t]

Results on SBSA:

• Pipeline mode throughput: 30879.93 frames/sec
• Dynamic mode throughput: 23685.68 frames/sec
• Dynamic capture mode throughput: 30909.08 frames/sec

[dali-automaton]

CI MESSAGE: [61575050]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [61565034]: BUILD PASSED

[dali-automaton]

CI MESSAGE: [61575050]: BUILD PASSED

[dali-automaton]

CI MESSAGE: [61924029]: BUILD STARTED

[dali-automaton]

CI MESSAGE: [61924029]: BUILD PASSED