1. Release Highlights

• AMDuProfPCM

  • Support profiling multiple Process (comma separated values with -p option) or Threads (comma separated values with --tid option)

  • New CLI options --start-delay and --wait-for-signal

  • UMC metric support (for Zen-4 and Zen-5 servers)

  • Override events to collect for kernel/user-only mode

• New Platform Support

  • Krackan1

  • Krackan2

  • Strix-Halo

• Hotspots Analysis

  • Source level attribution for Python Profiling

  • Python profiling v3.10 support for “Attach to Process” and “Launch App”

• Threading Analysis

  • Support Analysis of Thread Waits and Thread dependencies to improve load balance (Wait Objects Analysis)

  • Improved granularity - CPU Utilization for every 10 milliseconds using mperf and tsc

• Profile Control APIs and Instrumentation APIs

  • Addition of Hottest Events section in CLI report

  • Support consumption of data generated by Intel’s open source ITT API Implementation

  • Add AMD version of ITT Event API

• CPU Profiler

  • Support for new OS - Windows Server® 2025, Windows® 11 - 25H2, FreeBSD® 15

  • Report samples being attributed to Python functions and source lines

  • C++ Inline function support on Windows OS

  • FreeBSD OS enhancements:

    • Support profiling already running processes

    • Support profiling on Platform with more than 256 cores

  • Support Source code and line information for applications compiled with compressed dwarf

  • Added support for separate DWARF debug info

• HPC

  • Added Efficiency metrics for both MPI and OpenMP application profiling

  • Add Support for OpenMP Spec 5.1, 5.2 and 6.0

  • Removed default traceable PR instance limit

• AMDuProfSys

  • Adding UMC INI input support

  • UMC support for obfuscation on Milan

• AMDuProf GUI

  • Wait Object Analysis for analyzing Thread wait states and transitions

  • Highlight task section in All thread timeline for configs profile with ITT

  • Task running flame graph in Per thread timeline for configs profile with ITT

  • Added Aggregrate Flamegraph in Per thread timeline

  • GUI’s Thread Timeline views to show CPU Utilization for every 10msecs

  • GUI to support Re-sizing of Columns in various views, sort columns by Ascending/Descending order, Hide/Un-Hide columns and re-arrange the columns

  • Enhanced tooltips in custom config, column names in views, All Thread and Per-Thread Timeline

  • Support Navigation between table rows in Function/CSS/Source views using Up/Down key-strokes

  • Enable Transition from MPI Communication Matrix to MPI Timeline

  • Added “Search Source code” and “Search Assembly and address” fields in Source view tab

  • Added Context menu options “Copy current cell data” and “Copy all row(s) in range selected” in applicable tables

• Other Updates

  • Upgrade Microsoft Visual C++ 2015-2022 Redistributable Packag to latest one

  • Increased uProf linker flag max-page-size to 2MB for FreeBSD OS

  • Upgraded multiple 3rd party libraries (Utf8cpp, Tinyxml2, sqlite, disassembler and libdwarf) to latest version

  • Numerous improvements in the GUI views