rocm-systems/projects/rocprofiler-sdk/CHANGELOG.md

Changelog for ROCprofiler-SDK

Full documentation for ROCprofiler-SDK is available at rocm.docs.amd.com/projects/rocprofiler-sdk

ROCprofiler-SDK for AFAR I

Added

• HSA API tracing
• Kernel dispatch tracing
• Kernel dispatch counter collection
  • Instances reported as single dimension
  • No serialization

ROCprofiler-SDK for AFAR II

Added

• HIP API tracing
• ROCTx tracing
• Tracing ROCProf Tool V3
• Documentation packaging
• ROCTx control (start and stop)
• Memory copy tracing

ROCprofiler-SDK for AFAR III

Added

• Kernel dispatch counter collection. This includes serialization and multidimensional instances.
• Kernel serialization.
• Serialization control (on and off).
• ROCprof tool plugin interface V3 for counters and dimensions.
• Support to list metrics.
• Correlation-Id retirement
• HIP and HSA trace distinction:
  • --hip-runtime-trace For collecting HIP Runtime API traces
  • --hip-compiler-trace For collecting HIP compiler-generated code traces
  • --hsa-core-trace For collecting HSA API traces (core API)
  • --hsa-amd-trace For collecting HSA API traces (AMD-extension API)
  • --hsa-image-trace For collecting HSA API traces (image-extension API)
  • --hsa-finalizer-trace For collecting HSA API traces (finalizer-extension API)

ROCprofiler-SDK for AFAR IV

Added

API:

• Page migration reporting
• Scratch memory reporting
• Kernel dispatch callback tracing
• External correlation Id request service
• Buffered counter collection record headers
• Option to remove HSA dependency from counter collection

Tool:

• rocprofv3 multi-GPU support in a single-process

ROCprofiler-SDK for AFAR V

Added

API:

• Agent or device counter collection
• PC sampling (beta)

Tool:

• Single JSON output format support
• Perfetto output format support (.pftrace)
• Input YAML support for counter collection
• Input JSON support for counter collection
• Application replay in counter collection
• rocprofv3 multi-GPU support:
  • Multiprocess (multiple files)

Changed

• rocprofv3 tool now requires mentioning -- before the application. For detailed use, see Using rocprofv3

Resolved issues

• Fixed SQ_ACCUM_PREV and SQ_ACCUM_PREV_HIRE overwriting issue

ROCprofiler-SDK 0.4.0 for ROCm release 6.2 (AFAR VI)

Added

• OTF2 tool support
• Kernel and range filtering
• Counter collection definitions in YAML
• Documentation updates (SQ block, counter collection, tracing, tool usage)
• rocprofv3 option --kernel-rename
• rocprofv3 options for Perfetto settings (buffer size and so on)
• CSV columns for kernel trace
  • Thread_Id
  • Dispatch_Id
• CSV column for counter collection

ROCprofiler-SDK 0.5.0 for ROCm release 6.3 (AFAR VII)

Added

• Start and end timestamp columns to the counter collection csv output
• Check to force tools to initialize context id with zero
• Support to specify hardware counters for collection using rocprofv3 as rocprofv3 --pmc [COUNTER [COUNTER ...]]
• Memory Allocation Tracing
• PC sampling tool support with CSV and JSON output formats
• List supported PC Sampling Configurations

Changed

• --marker-trace option for rocprofv3 now supports the legacy ROCTx library libroctx64.so when the application is linked against the new library librocprofiler-sdk-roctx.so.
• Replaced deprecated hipHostMalloc and hipHostFree functions with hipExtHostAlloc and hipFreeHost for ROCm versions starting 6.3.
• Updated rocprofv3 --help options.
• Changed naming of "agent profiling" to a more descriptive "device counting service". To convert existing tool or user code to the new name, use the following sed: find . -type f -exec sed -i 's/rocprofiler_agent_profile_callback_t/rocprofiler_device_counting_service_callback_t/g; s/rocprofiler_configure_agent_profile_counting_service/rocprofiler_configure_device_counting_service/g; s/agent_profile.h/device_counting_service.h/g; s/rocprofiler_sample_agent_profile_counting_service/rocprofiler_sample_device_counting_service/g' {} +
• Changed naming of "dispatch profiling service" to a more descriptive "dispatch counting service". To convert existing tool or user code to the new names, the following sed can be used: -type f -exec sed -i -e 's/dispatch_profile_counting_service/dispatch_counting_service/g' -e 's/dispatch_profile.h/dispatch_counting_service.h/g' -e 's/rocprofiler_profile_counting_dispatch_callback_t/rocprofiler_dispatch_counting_service_callback_t/g' -e 's/rocprofiler_profile_counting_dispatch_data_t/rocprofiler_dispatch_counting_service_data_t/g' -e 's/rocprofiler_profile_counting_dispatch_record_t/rocprofiler_dispatch_counting_service_record_t/g' {} +
• FETCH_SIZE metric on gfx94x now uses TCC_BUBBLE for 128B reads.
• PMC dispatch-based counter collection serialization is now per-device instead of being global across all devices.
• Added output return functionality to rocprofiler_sample_device_counting_service
• Added rocprofiler_load_counter_definition.

Resolved issues

• Create subdirectory when rocprofv3 --output-file includes a folder path
• Fixed misaligned stores (undefined behavior) for buffer records
• Fixed crash when only scratch reporting is enabled
• Fixed MeanOccupancy metrics
• Fixed aborted-application validation test to properly check for hipExtHostAlloc command
• Fixed implicit reduction of SQ and GRBM metrics
• Fixed support for derived counters in reduce operation
• Bug fixed in max-in-reduce operation
• Introduced fix to handle a range of values for select() dimension in expressions parser
• Conditional aql::set_profiler_active_on_queue only when counter collection is registered (resolves Navi3 kernel tracing issues)

Removed

• Removed gfx8 metric definitions
• Removed rocprofv3 installation to sbin directory

ROCprofiler-SDK 0.6.0 for ROCm release 6.4

Added

• Support for select() operation in counter expression.
• reduce() operation for counter expression with respect to dimension.
• --collection-period feature in rocprofv3 to enable filtering using time.
• --collection-period-unit feature in rocprofv3 to control time units used in collection period option.
• Deprecation notice for ROCProfiler and ROCProfilerV2.
• Support for rocDecode API Tracing
• Usage documentation for ROCTx
• Usage documentation for MPI applications
• SDK: rocprofiler_agent_v0_t support for agent UUIDs
• SDK: rocprofiler_agent_v0_t support for agent visibility based on gpu isolation environment variables such as ROCR_VISIBLE_DEVICES and so on.
• Accumulation VGPR support for rocprofv3.
• Host-trap based PC sampling support for rocprofv3.
• Support for OpenMP tool.

ROCprofiler-SDK 1.0.0 for ROCm release 7.0

Added

• Support for rocJPEG API Tracing.
• Support for AMD Instinct MI350X and MI355X accelerators.
• rocprofiler_create_counter to facilitate adding custom derived counters at runtime.
• Support in rocprofv3 for iteration based counter multiplexing.
• Perfetto support for counter collection.
• Support for negating rocprofv3 tracing options when using aggregate options such as --sys-trace --hsa-trace=no.
• --agent-index option in rocprofv3 to specify the agent naming convention in the output:
  • absolute == node_id
  • relative == logical_node_id
  • type-relative == logical_node_type_id
• MI300 and MI350 stochastic (hardware-based) PC sampling support in ROCProfiler-SDK and rocprofv3.
• Python bindings for rocprofiler-sdk-roctx
• SQLite3 output support for rocprofv3 using --output-format rocpd.
• rocprofiler-sdk-rocpd package:
  • Public API in include/rocprofiler-sdk-rocpd/rocpd.h.
  • Library implementation in librocprofiler-sdk-rocpd.so.
  • Support for find_package(rocprofiler-sdk-rocpd).
  • rocprofiler-sdk-rocpd DEB and RPM packages.
• --version option in rocprofv3.
• rocpd Python package.
• Thread trace as experimental API.
• ROCprof Trace Decoder as experimental API:
  • Requires ROCprof Trace Decoder plugin.
• Thread trace option in the rocprofv3 tool under the --att parameters:
  • See using thread trace with rocprofv3
  • Requires ROCprof Trace Decoder plugin.
• rocpd output format documentation:
  • Requires ROCprof Trace Decoder plugin.
• Perfetto support for scratch memory.
• Support in the rocprofv3 avail tool for command-line arguments.
• Documentation for rocprofv3 advanced options.
• Support for multi dispatch ATT file added

Changed

• SDK to NOT to create a background thread when every tool returns a nullptr from rocprofiler_configure.
• vaddr-to-file-offset mapping in disassembly.hpp to use the dedicated comgr API.
• rocprofiler_uuid_t ABI to hold 128 bit value.
• rocprofv3 shorthand argument for --collection-period to -P (upper-case) while -p (lower-case) is reserved for later use.
• Default output format for rocprofv3 to rocpd (SQLite3 database).
• rocprofv3 avail tool to be renamed from rocprofv3_avail to rocprofv3-avail tool.
• rocprofv3 tool to facilitate thread trace and PC sampling on the same agent.

Resolved issues

• Fixed missing callbacks around internal thread creation within counter collection service.
• Fixed potential data race in the ROCprofiler-SDK double buffering scheme.
• Fixed usage of std::regex in the core ROCprofiler-SDK library that caused segfaults or exceptions when used under dual ABI.
• Fixed Perfetto counter collection by introducing accumulation per dispatch.
• Fixed code object disassembly for missing function inlining information.
• Fixed queue preemption error and HSA_STATUS_ERROR_INVALID_PACKET_FORMAT error for stochastic PC-sampling in MI300X, leading to stabler runs.
• Fixed the system hang issue for host-trap PC-sampling on MI300X.
• Fixed rocpd counter collection issue when counter collection alone is enabled. rocpd_kernel_dispatch table is updated to be populated by counters data instead of kernel_dispatch data.
• Fixed rocprofiler_*_id_t structs for inconsistency related to a "null" handle:
  • The correct definition for a null handle is .handle = 0 while some definitions previously used UINT64_MAX.
• Fixed kernel trace csv output generated by rocpd.

Removed

• Support for compilation of gfx940 and gfx941 targets.

ROCprofiler-SDK 1.1.0 for ROCm release 7.1

Added

• Dynamic process attachment- ROCprofiler-sdk and rocprofv3 now facilitate dynamic profiling of a running GPU applications by attaching to its process ID (PID), rather than launching the application through the profiler itself.
• Scratch-memory trace information to the Perfetto output in rocprofv3.
• New capabilities to the thread trace support in rocprofv3, including real-time clock support for thread trace alignment on gfx9 architecture. This enables high-resolution clock computation and better synchronization across shader engines. Additionally, MultiKernelDispatch thread trace support is now available across all ASICs.
• Documentation for dynamic process attachment.
• Documentation for rocpd summaries.

Optimized

• Improved the stability and robustness of the rocpd output.

ROCprofiler-SDK 1.1.0 for ROCm release 7.2

Added

• Counter collection support for gfx1150 and gfx1151.
• HSA Extension API v8 support.
• hipStreamCopyAttributes API implementation.

Optimized

• Improved process attachment and updated the corresponding documentation.
• Improved [Quick reference guide for rocprofv3] (https://rocm.docs.amd.com/projects/rocprofiler-sdk/en/latest/quick_guide.html).
• Updated installation documentation with links to the latest repository (https://rocm.docs.amd.com/projects/rocprofiler-sdk/en/latest/install/installation.html).

Resolved issues

• Fixed multi-GPU dimension mismatch.
• Fixed device lock issue for dispatch counters.
• Addressed OpenMP Tools task scheduling null pointer exception.
• Fixed stream ID errors arising during process attachment.
• Fixed issues arising during dynamic code object loading.