AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
088fa6a4ac6f80feb1d4e569db0c3c7467dd5de0
rocm-systems/projects/rocprofiler-compute/CHANGELOG.md
T
cfallows-amd 5079a1803f Datatype selection option for roofline (#624) 
Added command line option to specify which datatype(s) to capture into the roofline PDF(s).
All datatypes are still collected by roofline call if applicable, but only specific datatypes are plotted into PDF outputs. Will dump out all datatypes into one graph, but separate FP from Int into two graphs if needed. Will skip datatype and give error message if the datatype is not valid on a particular gpu arch.
Default is FP32

Reworked roofline calls and plotting to be general enough such that any new datatypes added into rocm-amdgpu-bench can easily be reflected in rocprof-compute with simple modifications in roofline_calc.py.

Adjusted ctest to reflect expected default pdf outputs from roofline.

---------

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

[ROCm/rocprofiler-compute commit: a492e92034]
2025-03-25 15:02:09 -04:00

8.2 KiB
Raw Blame History

Changelog for ROCm Compute Profiler

Full documentation for ROCm Compute Profiler is available at https://rocm.docs.amd.com/projects/rocprofiler-compute/en/latest/.

Unreleased

Added

  • Add Docker files to package the application and dependencies into a single portable and executable standalone binary file

  • Analysis report based filtering

    • -b option in profile mode now additionally accepts metric id(s) for analysis report based filtering
    • -b option in profile mode also accept hardware IP block for filtering, however, this support will be deprecated soon
    • --list-metrics option added in profile mode to list possible metric id(s), similar to analyze mode

  • Datatype selection option for roofline profiling

    • --roofline-data-type / -R option added to specify which datatypes the user wants to capture in the roofline PDF plot outputs
    • Default is FP32, but user can specify as many types as desired to overlay on the same plot output

Changed

  • Change normal_unit default to per_kernel
  • Change dependency from rocm-smi to amd-smi
  • Decrease profiling time by not collecting counters not used in post analysis

Resolved issues

  • Fixed option specs-correction
  • Fixed kernel name and kernel dispatch filtering when using rocprof v3
  • Fixed not collecting TCC channel counters in rocprof v3

Known issues

  • GPU id filtering is not supported when using rocprof v3

ROCm Compute Profiler 3.1.0 for ROCm 6.4.0

Added

  • Roofline support for Ubuntu 24.04
  • Experimental support rocprofv3 (not enabled as default)

Resolved issues

  • Fixed PoP of VALU Active Threads
  • Workaround broken mclk for old version of rocm-smi

ROCm Compute Profiler 3.0.0 for ROCm 6.3.0

Changed

  • Renamed Omniperf to ROCm Compute Profiler (#475)

Omniperf 2.0.1 for ROCm 6.2.1

Changed

  • enable rocprofv1 for MI300 hardware (#391)
  • refactoring and updating documemtation (#362, #394, #398, #414, #420)
  • branch renaming and workflow updates (#389, #404, #409)
  • bug fix for analysis output
  • add dependency checks on application launch (#393)
  • patch for profiling multi-process/multi-GPU applications (#376, #396)
  • packaging updates (#386)
  • rename CHANGES to CHANGELOG.md (#410)
  • rollback Grafana version in Dockerfile for Angular plugin compatibility (#416)
  • enable CI triggers for Azure CI (#426)
  • add GPU model distinction for MI300 systems (#423)
  • new MAINTAINERS.md guide for omniperf publishing procedures (#402)

Optimized

  • reduced running time of Omniperf when profiling (#384)
  • console logging improvements

Omniperf 2.0.1 for ROCm 6.2.0

Added

  • new option to force hardware target via OMNIPERF_ARCH_OVERRIDE global (#370)
  • CI/CD support for MI300 hardware (#373)
  • support for MI308X hardware (#375)

Optimized

  • cmake build improvements (#374)

Omniperf 2.0.0 (17 May 2024)

  • improved logging than spans all modes (#177) (#317) (#335) (#341)
  • overhauled CI/CD that spans all modes (#179)
  • extensible SoC classes to better support adding new hardware configs (#180)
  • --kernel-verbose no longer overwrites kernel names (#193)
  • general cleanup and improved organization of source code (#200) (#210)
  • separate requirement files for docs and testing dependencies (#205) (#262) (#358)
  • add support for MI300 hardware (#231)
  • upgrade Grafana assets and build script to latest release (#235)
  • update minimum ROCm and Python requirements (#277)
  • sort rocprofiler input files prior to profiling (#304)
  • new --quiet option will suppress verbose output and show a progress bar (#308)
  • roofline support for Ubuntu 22.04 (#319)

Omniperf 1.1.0-PR1 (13 Oct 2023)

  • standardize headers to use 'avg' instead of 'mean'
  • add color code thresholds to standalone gui to match grafana
  • modify kernel name shortener to use cpp_filt (#168)
  • enable stochastic kernel dispatch selection (#183)
  • patch grafana plugin module to address a known issue in the latest version (#186)
  • enhanced communication between analyze mode kernel flags (#187)

Omniperf 1.0.10 (22 Aug 2023)

  • critical patch for detection of llvm in rocm installs on SLURM systems

Omniperf 1.0.9 (17 Aug 2023)

  • add units to L2 per-channel panel (#133)
  • new quickstart guide for Grafana setup in docs (#135)
  • more detail on kernel and dispatch filtering in docs (#136, #137)
  • patch manual join utility for ROCm >5.2.x (#139)
  • add % of peak values to low level speed-of-light panels (#140)
  • patch critical bug in Grafana by removing a deprecated plugin (#141)
  • enhancements to KernelName demangeler (#142)
  • general metric updates and enhancements (#144, #155, #159)
  • add min/max/avg breakdown to instruction mix panel (#154)

Omniperf 1.0.8 (30 May 2023)

  • add --kernel-names option to toggle kernelName overlay in standalone roofline plot (#93)

  • remove unused python modules (#96)

  • fix empirical roofline calculation for single dispatch workloads (#97)

  • match color of arithmetic intensity points to corresponding bw lines

  • ux improvements in standalone GUI (#101)

  • enhanced readability for filtering dropdowns in standalone GUI (#102)

  • new logfile to capture rocprofiler output (#106)

  • roofline support for sles15 sp4 and future service packs (#109)

  • adding dockerfiles for all supported Linux distros

  • new examples for --roof-only and --kernel options added to documentation

  • enable cli analysis in Windows (#110)

  • optional random port number in standalone GUI (#111)

  • limit length of visible kernelName in --kernel-names option (#115)

  • adjust metric definitions (#117, #130)

  • manually merge rocprof runs, overriding default rocprofiler implementation (#125)

  • fixed compatibility issues with Python 3.11 (#131)

Omniperf 1.0.8-PR2 (17 Apr 2023)

  • ux improvements in standalone GUI (#101)
  • enhanced readability for filtering dropdowns in standalone GUI (#102)
  • new logfile to capture rocprofiler output (#106)
  • roofline support for sles15 sp4 and future service packs (#109)
  • adding dockerfiles for all supported Linux distros
  • new examples for --roof-only and --kernel options added to documentation

Omniperf 1.0.8-PR1 (13 Mar 2023)

  • add --kernel-names option to toggle kernelName overlay in standalone roofline plot (#93)
  • remove unused python modules (#96)
  • fix empirical roofline calculation for single dispatch workloads (#97)
  • match color of arithmetic intensity points to corresponding bw lines

Omniperf 1.0.7 (21 Feb 2023)

  • update documentation (#52, #64)
  • improved detection of invalid command line arguments (#58, #76)
  • enhancements to standalone roofline (#61)
  • enable Omniperf on systems with X-server (#62)
  • raise minimum version requirement for rocm (#64)
  • enable baseline comparison in CLI analysis (#65)
  • add multi-normalization to new metrics (#68, #81)
  • support alternative profilers (#70)
  • add MI100 configs to override rocprofiler's incomplete default (#75)
  • improve error message when no GPU(s) detected (#85)
  • separate CI tests by Linux distro and add status badges

Omniperf 1.0.6 (21 Dec 2022)

  • CI update: documentation now published via github action (#22)
  • better error detection for incomplete ROCm installs (#56)

Omniperf 1.0.5 (13 Dec 2022)

  • store application command-line parameters in profiling output (#27)
  • enable additional normalizations in CLI mode (#30)
  • add missing ubuntu 20.04 roofline binary to packaging (#34)
  • update L1 bandwidth metric calculations (#36)
  • add L1 <-> L2 bandwidth calculation (#37)
  • documentation updates (#38, #41)
  • enhanced subprocess logging to identify critical errors in rocprofiler (#50)
  • maintain git sha in production installs from tarball (#53)

Omniperf 1.0.4 (11 Nov 2022)

  • update python requirements.txt with minimum versions for numpy and pandas
  • addition of progress bar indicator in web-based GUI (#8)
  • reduced default content for web-based GUI to reduce load times (#9)
  • minor packaging and CI updates
  • variety of documentation updates
  • added an optional argument to vcopy.cpp workload example to specify device id

Omniperf 1.0.3 (07 Nov 2022)

  • initial Omniperf release
Reference in New Issue View Git Blame Copy Permalink