0.6.6

Memory Usage

• Changed struct work name to be dynamically allocated, but still contiguous within the struct
  • Address now points to space after struct work
    • This is similar to, but not the same as, a flexible array member to avoid needing to compile with C++ extensions enabled
• Swap work items to storage if queue limit is hit
  • gufi_dir2index, gufi_dir2trace, gufi_trace2index, and gufi_query
  • use -M <bytes> to set queue_limit
    • <bytes> is divided across the number of threads (-n) and work size to produce queue_limit
  • use -s <prefix> to write swap files to a location that is not pwd

QueuePerThreadPool

• Large amounts of code reorganization and separating out code into functions
• API updates to support swapping, BottomUp updates, and to generally have better design

Miscellaneous

• CMake minimum version is now 3.5.0
  • Updated GitHub Actions Older CMake build
• Now using _at functions to reduce cost of path name resolution
• struct input now has dynamically allocated values - call input_fini to clean up
• descend
  • If struct dirent d_type is DT_UNKNOWN, call lstat(2)
  • Removed unnecessary arguments
• Individual trace files are now scouted in parallel to reduce likelihood of work generation being a bottleneck
• Now some executables dump VmHWM from /proc/self/status at the end
• Updated version string to print consistently across C and Python
• Split BottomUp code into multiple functions in case non-BottomUp functions need to be run with BottomUp
• Removed PRINT_CUMULATIVE_TIMES, PRINT_PER_THREAD_STATS, and PRINT_QPTPOOL_QUEUE_SIZE
  • Performance History Framework
  • gnuplot scripts
  • GitHub Actions debug builds
• Updated gpfs-scan-tool to compile with latest code

GitHub Actions

• Removed macOS 12 and remnants of 13
• Added macOS 15
• Python 2 build now runs on CentOS 8