fix: Swap ReportStatistics timestamp args in PT2/AOTI backend

[hg1112]

Summary

TRITONBACKEND_ModelInstanceReportStatistics expects timestamps in the order
(exec_start_ns, compute_start_ns, compute_end_ns, exec_end_ns), but
ModelInstanceState::ProcessRequests in the PT2/AOTI instance state passed
exec_end_ns before compute_start_ns/compute_end_ns. Triton core then
interprets exec_end_ns as compute_input_end_ns, and the resulting
subtraction underflows as an unsigned 64-bit value, so
nv_inference_compute_infer_summary_us reports an enormous bogus number
(~18446744074s) instead of the real compute latency.

This reorders the arguments to match the documented header signature and the
legacy (non-PT2) PyTorch backend's call convention, which already does this
correctly. The batch-level TRITONBACKEND_ModelInstanceReportBatchStatistics
call a few lines below was already in the correct order, so no other changes
are needed.

Ref: triton-inference-server/server#8874

Test plan

• pre-commit run (clang-format, codespell, etc.) passes on the changed file
• Verified new argument order against TRITONBACKEND_ModelInstanceReportStatistics's
  declaration in triton-inference-server/core's tritonbackend.h

[greptile-apps]

Greptile Summary

Corrects the PT2/AOTI per-request statistics call so timestamps follow the expected execution and computation sequence.

• Moves exec_end_ns after the compute timestamps.
• Aligns PT2/AOTI reporting with the legacy backend and batch-statistics call convention.

Confidence Score: 5/5

The PR appears safe to merge because the corrected timestamp order is consistent with the established sibling statistics calls.

The change only reorders same-typed timestamp arguments and now aligns the PT2/AOTI per-request call with both the legacy backend and adjacent batch-statistics reporting.

Important Files Changed

Filename	Overview
src/pt2/model_instance_state.cc	Reorders the statistics timestamps to exec_start_ns, compute_start_ns, compute_end_ns, and exec_end_ns; no actionable issue found.

_{Reviews (1): Last reviewed commit: "Fix swapped ReportStatistics timestamp a..." | Re-trigger Greptile}

[mattwittwer]

Hi @hg1112 - Thank you for the report and the fix!

Confirmed on our side: the API expects (exec_start, compute_start, compute_end, exec_end), so core computed compute_infer as compute_start_ns - exec_end_ns, which underflows uint64_t. compute_input/compute_output and trace timestamps were also skewed. Your reordering matches the API contract.

Validated through our internal CI (Pipeline ID: 63772955, branch mwittwer/pr203_ci_validation mirroring your commit 3d7c7ee), including a new L0_torch_aoti stats-sanity regression test that fails without this fix and passes with it.

One prerequisite before we can merge: Triton requires all contributors (or their corporate entity) to have a signed Contributor License Agreement on file with triton-cla@nvidia.com. Do you know if you or your employer already have one? Happy to help get that sorted if not.

[hg1112]

Hi @mattwittwer , I have sent an email to triton-cla@nvidia.com.

I am doing this work from my personal interest in this project :)

[hg1112]

Hi @mattwittwer , @yinggeh

I have received approval from triton cla

[whoisj]

Waiting for CI 63772955 to complete before merging.