Skip to content

Latest commit

 

History

History
70 lines (53 loc) · 3.34 KB

File metadata and controls

70 lines (53 loc) · 3.34 KB

TraceLens documentation

TraceLens is an open-source Python library developed by AMD that automates performance analysis from GPU trace files. Instead of manually inspecting raw profiling data in tools such as Perfetto or Chrome Trace Viewer, TraceLens parses traces from PyTorch, JAX, and rocprofv3 and produces structured performance reports — including hierarchical GPU-timeline breakdowns, per-operator roofline analysis (TFLOP/s, TB/s), and multi-GPU communication diagnostics.

The TraceLens source code is hosted at github.com/AMD-AGI/TraceLens.

.. grid:: 2
   :gutter: 3

   .. grid-item-card:: Install

      * :doc:`Install TraceLens <install/install>`

   .. grid-item-card:: How to

      * :doc:`Generate reports <how-to/generate-reports>`

        * :doc:`PyTorch performance report <how-to/generate-perf-report-pytorch>`
        * :doc:`PyTorch inference performance report <how-to/generate-perf-report-pytorch-inference>`
        * :doc:`JAX performance report <how-to/generate-perf-report-jax>`
        * :doc:`rocprof performance report <how-to/generate-perf-report-rocprof>`
        * :doc:`Collective-communication report <how-to/collective-report>`
        * :doc:`Compare performance reports <how-to/compare-perf-reports>`

      * :doc:`Split traces into iterations, steady state, and phases <how-to/split-traces>`
      * :doc:`Compare two traces <how-to/compare-traces>`
      * :doc:`Replay a single operation <how-to/event-replay>`
      * :doc:`Fuse multi-rank traces <how-to/trace-fusion>`
      * :doc:`Analyze traces with the SDK <how-to/sdk-analysis>`
      * :doc:`Index a corpus of traces <how-to/trace-index>`
      * :doc:`Model op performance without a trace <how-to/perf-model-without-trace>`
      * :doc:`Analyze collective communication <how-to/nccl-analysis>`
      * :doc:`Estimate kernel times with Origami <how-to/origami-integration>`
      * :doc:`Agentic performance analysis with the TraceLens Agent <how-to/agent>`

   .. grid-item-card:: Concepts

      * :doc:`Trace2Tree data model <conceptual/trace2tree>`
      * :doc:`PyTorch traces <conceptual/torch-profiling-analysis>`
      * :doc:`Tensor shape metadata <conceptual/shape-metadata>`
      * :doc:`GEMM analysis <conceptual/gemm-analysis>`
      * :doc:`Inference performance analysis <conceptual/inference-analysis>`
      * :doc:`Triton kernel performance model <conceptual/triton-perf-model-walkthrough>`

   .. grid-item-card:: Reference

      * :doc:`API reference <reference/api-reference>`
      * :doc:`Performance report columns <reference/perf-report-columns>`
      * :doc:`Compatibility matrix <reference/compatibility>`

For information on contributing to TraceLens, see the Contributing guide.

TraceLens is released under the MIT License. For details, see the :doc:`License <about/license>` page.