Interpret performance reports

This section explains how to interpret the reports produced by Linaro Performance Reports.

Reports are generated in both HTML and textual formats for each run of your application, by default. The information presented in both of these formats is the same.

If you want to combine Linaro Forge with other tools, consider using the CSV output format. See CSV performance reports for more details.

• HTML performance reports
• Report summary
  • Compute
  • MPI
  • I/O (Input/Output)
  • Python Interpreter
• CPU breakdown
  • Single-core code
  • OpenMP regions
  • Scalar numeric ops
  • Vector numeric ops
  • Memory accesses
  • Waiting for accelerators
• CPU metrics breakdown
  • Cycles per instruction
  • Stalled cycles
  • L2 cache misses
• MPI breakdown
  • Time in collective calls
  • Time in point-to-point calls
  • Effective process collective rate
  • Effective process point-to-point rate
• I/O breakdown
  • Time in reads
  • Time in writes
  • Effective process read rate
  • Effective process write rate
  • Lustre metrics
• OpenMP breakdown
  • Computation
  • Synchronization
  • Physical core utilization
  • System load
• Threads breakdown
  • Computation (Threads)
  • Synchronization (Threads)
  • Physical core utilization (Threads)
  • System load (Threads)
• Memory breakdown
  • Mean process memory usage
  • Peak process memory usage
  • Peak node memory usage
• Energy breakdown
  • CPU
  • Accelerator
  • System
  • Mean node power
  • Peak node power
  • Requirements
• Accelerator breakdown
  • GPU utilization
  • Mean GPU memory usage
  • Peak GPU memory usage
• Thread affinity breakdown
  • Mean utilization
  • Max load
  • Migration opportunity
• Textual performance reports
• CSV performance reports