XProf v2.21.1

What's Changed

• Update xprof version to 2.21.1. by @Matt-Hurd in #1899
• Automatically select the single available host when no host is specified. by @Matt-Hurd in #1892
• Add a new data service interface for getting source info for pallas call, not implemented yet, thus no-op. by @zzzaries in #1878

Full Changelog: xprof-v2.21.0...xprof-v2.21.1

XProf v2.21.0

What's Changed

• Change session variable to session_path by @muditgokhale2 in #1883
• Documentation review by @melissawm and @pavithraes in #1451

Full Changelog: xprof-v2.20.9...xprof-v2.21.0

XProf v2.20.9

What's Changed

• Update XLA dependency in WORKSPACE.bazel and release new version by @Matt-Hurd in #1873

Full Changelog: xprof-v2.20.8...xprof-v2.20.9

XProf v2.20.8

What's Changed

• Update version to 2.20.8. by @Matt-Hurd in #1865
• Support lazy fetching for trace_events and functionality to support trace_events searching in streaming_trace_viewer_processor by @muditgokhale2 in #1855
• Add store management support for session tags. by @zzzaries in #1866
• Refactored HloProtoMap::contains into ContainsOptimizedModule and ContainsOriginalModule to clearly distinguish between optimized and original HLO module lookups. in #1784
• Rename "Runs" to "Sessions" in the sidenav in #1858
• Replaces the use of RemapInstructionIds with HloModule::CreateFromProto without preserving unique id in xprof. in #1856
• Add MatButtonModule to SideNavModule. by @muditgokhale2 in #1860
• Extends stack-trace-snippet components to source mapper components. by @zzzaries in #1848
• This change includes the BSD-style license for plugin/third_party/webcomponentsjs/webcomponents.js within the xprof LICENSE file. by @zzzaries in #1854
• Add a submit button to the hosts sidenav for trace_viewer by @muditgokhale2 in #1851
• Refactor original/unoptimized xla::HloModuleProto --> xla::HloProto instances in #1838
• Add Memory Bound rule in #1852
• Add multi-threading support to the processing of tensor core planes. by @muditgokhale2 in #1809
• Add double counter event support trace viewer. in #1842
• Store trace event args for Cost Model Flop Counter event as part of the main trace events file. in #1845
• Load highlightjs stylesheets only once in index.html by @zzzaries in #1841
• Create Utilization Component in xprof frontend. in #1811
• Created MemoryViewerOption for memory viewer data processing consolidation by @zzzaries in #1770
• Enable module_name query parameter for Memory Viewer. by @zzzaries in #1823
• Add Perf Counters tool to Xprof frontend. in #1818
• Fix input pipeline analysis's device side input wait time chart. by @bmass02 in #1825
• Filter out pstate events with invalid arguments in trace viewer JSON. in #1832
• Project import generated by Copybara in #1813

Full Changelog: xprof-v2.20.7...xprof-v2.20.8

XProf v2.20.7

What's Changed

• Update version to 2.20.7 by @subhamsoni-google in #1810
• Optimize arena usage and remove unnecessary HLO proto generation. by Profiler Team in #1812
• Replace tsl::kint64max with std::numeric_limits<int64_t>::max(). by @majnemer in #1808
• No-op Changes. by @zzzaries in #1765
• Update dependency config to include recent xla changes by @zzzaries in #1802
• The conversion from XPlane to OpMetricsDb for host threads now skips events that have an empty name, as these events (if any) are not relevant for the host event analysis. by @zzzaries in #1789
• Add ConvertHloModuleProtoToStringView to convert HloModuleProto to string. by Profiler Team in #1782
• Add license for all the tool processors by @muditgokhale2 in #1786
• Extract graph rendering helpers to graphviz_helper.h by @zzzaries in #1767
• Add ConvertHloModuleProtoToModule to fix non-consecutive instruction IDs. by Profiler Team in #1780
• This change simplifies the XProf server by making the profile processor the default path and removing the --use_distributed_processing flag. by Profiler Team in #1779
• Make the max_heap_chart colors match the allocation timeline graphviz colors by Profiler Team in #1774
• Rephrase input pipeline recommendations title to deliver a clearer message for now. by @zzzaries in #1777
• Fix typo in comment by Profiler Team in #1771
• Project import generated by Copybara by Profiler Team in #1778
• Avoid override selectedModule with default value in moduleList if moduleName is already set from query params. by @zzzaries in #1773
• Support trace events searching in 3P trace viewer by Profiler Team in #1758
• Support lazy fetching of trace events in 3P Trace Viewer by Profiler Team in #1757
• Support original/unoptimized xla::HloModuleProto instances by Profiler Team in #1766
• Enable -g flag for builds with compilation_mode=dbg. by Profiler Team in #1764
• [Xprof] Add a dropdown menu to group HLO Op Profile, including by Provenance(new grouping). by Profiler Team in #1745
• Improve derived line ordering and fix tests. by @bmass02 in #1756
• Show the metrics. by Profiler Team in #1645
• Add a rule to remove LINT.IfChange and LINT.ThenChange lines. by Profiler Team in #1755
• Add fix for TPU idleness attribution due to input pipeline slowness for host loop based scheduling. by @bmass02 in #1750
• Fix optional deref crash due to custom XStats. by @bmass02 in #1748
• Integrate gRPC worker for distributed profile processing by @subhamsoni-google in #1741
• Make log directory optional to start Xprof server by Profiler Team in #1733
• Update .bazelrc config to properly use new hermetic C++ toolchain by @Matt-Hurd in #1738
• No-op changes. by @zzzaries in #1739
• Parallelize LevelDB trace event search. by Profiler Team in #1732
• Enable providing session and run path (that contains multiple session directories) dynamically in the query URL. by Profiler Team in #1724
• Create inference stats processor with API similar to the ProfileProcessor Interface. by @muditgokhale2 in #1740
• Add error parsing logic to show error message correctly for different error types. by @muditgokhale2 in #1736
• Create op profile processor with API similar to the ProfileProcessor Interface. by @muditgokhale2 in #1725
• Create processors for trace viewer with APIs similar to Profile Processor by @muditgokhale2 in #1730
• Create processor for memory profile tool with APIs similar to Profile Processor by @muditgokhale2 in #1731
• Create processor for graph viewer with APIs similar to the Profile Processor. by @muditgokhale2 in #1729
• Create processor for memory viewer with APIs similar to the Profile Processor. by @muditgokhale2 in #1728
• Show pipeline analysis with new processing logic, while maintaining backward compatibility by @zzzaries in #1647
• Add generic input pipeline op processing support and add traceme ops from xplane to op_stats.host_op_metrics_db by @zzzaries in #1709
• Improve memory profile error message. by @cliveverghese in #1726
• Register rules_ml_toolchain toolchains by @Matt-Hurd in #1717
• Fix url parsing in hlo stats and roofline model. by @zzzaries in #1722
• Add "available only in the new analysis" to the tooltips. by Profiler Team in #1727
• Create processor for megascale stats with APIs similar to the Profile Processor by @muditgokhale2 in #1696
• Make stackFrameName mutable. by @zzzaries in #1723
• Include source info in the symbol we copy into a deduplicated node. by Profiler Team in #1721

Full Changelog: xprof-v2.20.6...xprof-v2.20.7

XProf v2.20.6

What's Changed

• Update version to 2.20.6. by @Matt-Hurd in #1720
• Hermetic CUDA was moved from XLA to rules_ml_toolchain open sourced repository. The current Hermetic CUDA directory will be removed from XLA soon. by @Matt-Hurd in #1719
• Hermetic CUDA was moved from XLA to rules_ml_toolchain open sourced repository. The current Hermetic CUDA directory will be removed from XLA soon. by Profiler Team in #1694
• Replace hardcoded "all_hosts" with a named constant. by @subhamsoni-google in #1716
• Frontend changes to support all trace events search for leveldb backend and changes to support backward compatibility for older profiles. by Profiler Team in #1695
• Fix wrong event stack frame issue while fetching complete event from leveldb backend. by Profiler Team in #1703
• Support trace events prefix name search in leveldb backend. by Profiler Team in #1684
• Fix Python 3.9 version compatibility by @Matt-Hurd in #1708
• Add Python 3.13 support to WORKSPACE.bazel. by @Matt-Hurd in #1707
• Change PROFILER_PYTHON_VERSION to HERMETIC_PYTHON_VERSION. by @Matt-Hurd in #1705
• Fix checks for conflicting release and nightly installs by @Matt-Hurd in #1706
• [Memory Viewer] Respect selecting HLO Module from sidenav by @Matt-Hurd in #1570
• Update input pipeline analysis to handle new PyGrain traces; specifically identifying the output node of the pipeline to measure overall wait time causing device idleness. by @bmass02 in #1681
• Create and Persist trace events prefix trie as leveldb sstable file. by Profiler Team in #1673
• Fix Copybara submit that is failing due to the incompatibility of std::string::starts_with that was introduced in C++20. by Profiler Team in #1701
• Integrate Smart Suggestions into the Overview Page. by Profiler Team in #1700
• Implement prefix trie and persist it as leveldb table. It would be used to enable trace event search for leveldb backend. by Profiler Team in #1662
• Fix Multi-host FW Power Metrics on Overview Page by @Becker-ZH in #1693
• Upgrade XLA to include major deps upgrades by @Matt-Hurd in #1655
• Display smart suggestion on Overview page by @sannidhyachauhan in #1697
• Display smart suggestion on Overview page by Profiler Team in #1651
• Enhance Smart Recommendation UI and add GA by Profiler Team in #1669
• Relax Python Version on passthrough package to match release package by @Matt-Hurd in #1692
• Reverse the order of frames in stack-trace. by Profiler Team in #1688
• Project import generated by Copybara by Profiler Team in #1661
• Create processor for the following tools : input_pipeline, framework_op_stats, hlo_stats, pod_viewer, kernel_stats, roofline_model by @muditgokhale2 in #1652
• Ignore the stack-frame when the line number is -1. by Profiler Team in #1687
• Improve error handling. by Profiler Team in #1686
• Add source-location input to stack-trace-snippet element in Trace Viewer tool. by Profiler Team in #1683
• Add python 3.13 requirements.txt by @Matt-Hurd in #1682
• Add source-location input to stack-trace-snippet element in Graph Viewer tool. by Profiler Team in #1680
• Add source-location input to stack-trace-snippet element in Roofline Model tool. by Profiler Team in #1679
• Add source-location input to stack-trace-snippet element in HLO OP Stats tool. by Profiler Team in #1678
• Add source-location input to stack-trace-snippet element in Memory Viewer tool. by Profiler Team in #1677
• Use source-info when stack-trace is unavailable by Profiler Team in #1675
• Remove old recommendationView from overview page by Profiler Team in #1670
• Only expand the most important stack frame initially by Profiler Team in #1674
• Add smart suggestion module in front end part. by Profiler Team in #1649
• Remove confused comment by Profiler Team in #1631
• Set graph_type to default when it's unspecified by @zzzaries in #1667
• Flatten the source code panel by Profiler Team in #1660

Full Changelog: xprof-v2.20.5...xprof-v2.20.6

XProf v2.20.5

What's Changed

• Unify Data Conversion for MegaScale_stats @sannidhyachauhan in #1580
• Add changes to support writing trace events metadata to a separate leveldb file. in #1592
• This is a first step to migrate xprof tools to a new interface that facilitates distributed processing. in #1632
• Enable remote capture without tensorflow @Matt-Hurd in #1605
• Add Flop Rate Chart and Model Properties in Framework Op Stats Tool @muditgokhale2 in #1638
• Project import generated by Copybara in #1604
• Add changes to support ReadFullEvent in trace viewer for leveldb backend. in #1613
• Import _pywrap_profiler_plugin from the proper directory by @Matt-Hurd in #1636
• Project import generated by Copybara in #1640
• Proto changes to optionally show the uncapped flops utilization by @muditgokhale2 in #1635
• Add toggle to optionally cap flops utilization to 100% by @muditgokhale2 in #1641
• Cleaning up the unused tf_data_stats_proto_to_gviz library @sannidhyachauhan in #1644
• Use fsspec rather than epath implementation for glob by @Matt-Hurd in #1634
• Make firstLine a property of Address (instead of a field in Content). in #1639
• Filter counter events correctly after parsing from its new format @rnayar3 in #1637
• Add default graph options for quick start in graph viewer @zzzaries in #1648
• Visualize custom-call kernel metadata in OpProfile and TraceViewer by @bmass02 in #1598
• Add tool name for loading complete GA event logging in throbber by @zzzaries in #1654
• Add graph type argument to moduleList data service call by @zzzaries in #1653
• Add a formatted text cell in Table row for DataTable by @sannidhyachauhan in #1642
• Fix Event Not Found issue while fetching complete trace event from leveldb backend. in #1656
• Setting Counter Events Permanently to grouped format by @rnayar3 in #1623
• Increment version to 2.20.5 by @Matt-Hurd in #1659
• Fix XProf server not coming up on windows machine in #1643

Full Changelog: xprof-v2.20.4...xprof-v2.20.5

XProf v2.20.4

What's Changed

• Add unit tests for two input bottleneck rules by @copybara-service[bot] in #1612
• [xprof] Relax and explicitly specify gcsfs version in pip dependency. by @Matt-Hurd in #1616
• Use rglob where available for run discovery by @Matt-Hurd in #1617

Reverts

• #1621 reverts #1528

Full Changelog: xprof-v2.20.3...xprof-v2.20.4

XProf v2.20.3

What's Changed

• Allow Xprof frontend to use the query parameter use_saved_result=False which will skip the intermediate analysis and regenerate the tool data from XSpace by @adityakakarotin #1456
• Removing stale function signature references from tensorflow that rely on old options of type variant<int, string> by @adityakakarot in #1527
• Add todo to rename stack_frame to stack_trace. in #1602
• Add throbber to hlo_stats, input_pipeline, megascale_stats and roofline_model tools. by @muditgokhale2 in #1599
• Rename OP Name search box to XLA Op name by @muditgokhale2 in #1601
• Remove legacy data service by @zzzaries in #1594
• Add error handling for network error (Server crashes) by @muditgokhale2 in #1597
• Fix sidenav visibility in graph_viewer for Firefox users. by @muditgokhale2 in #1608
• Remove GCS rate limiting and also remove the logic to generate runs based on tfevent files. in #1603
• Group TranceCounter Events to send more counter events over the network for large cases. by @rnayar3 in #1579
• Add Input bottleneck rules in #1600
• Add Trace options and update trace filters for 3P based on trace options in #1584
• Propagate full_dma option in 3P trace viewer in #1590
• Add async lines back to 3P trace viewer filtered by full_dma option in #1591
• Restructuring the data_table_utils file by @sannidhyachauhan in #1607
• In this CL: by @adityakakarot in #1528

Full Changelog: xprof-v2.20.2...xprof-v2.20.3