1.3.4

• remove readme from perfspect.tgz
• move cycles to its own group when in a VM due to it being hidden in most CSP's
• filter out MSR required events when inside VM

1.3.3

updated html report to:

• x-axis now contains actual timestamps
• diffs are now calculated more intuitively with coloring for up and now
• diff in separate column
• TMA and PSI both under bottlenecks tab
• rounded values
• table now fits content instead of 100%
• disclaimer explaining all the periods in TMA metric names
• help info now behind help icon

1.3.2

• added per transaction flag to postprocess
• added diff utility to html report

1.3.1

• fix old TMA version names
• update tma support detection to not require msr kernel module
• update html report
  • add metadata
  • add all metric averages
  • add metric descriptions

1.3.0

Breaking change

• changed --thread flag to --cpu to align with Linux norms

Non-breaking changes

• added more metrics to all platforms:
  • metric_locks retired per instr
  • metric_% Uops delivered from decoded Icache (DSB)
  • metric_% Uops delivered from legacy decode pipeline (MITE)
• dynamically detect vPMU driver support for TMA fixed events
• refactored perf-postprocess for readability
• fixed some typos

1.2.14

1. added PSI metrics
2. fix vm check

1.2.13

1. bug fix cstate events when hyperthreading off
2. add metadata to logs of perf-collect
3. better handle short collection periods

1.2.12

1. fixing bug with html report not getting read as utf-8
2. adding TMA version to readme

1.2.11

New in v1.2.11:

• updated CLX, SKX, BDX metrics and events to latest perfmon
• improved cid mode (auto-selects 5 containers with highest cpu util)
• optimized postprocessing speed
• changed output interval to 5 seconds and multiplexing to 125ms (eliminates perf overhead)

1.2.10

1. Update html report
2. Update uncore detection to support older kernels
3. Add option to fail postprocessing if anomalies detected in perfstat.csv
4. Expand TMA metrics for SPR, ICX, CLX and SKX
5. Updated similarity analysis to support more components and output new visualizations
6. add system wide to thread and socket