Download

Changelog

• Resolves an issue that occurred when writing large .cubex files (over 8 GB per metric).
• Includes an example demonstrating CubeW’s integration within the Score-P workflow.
• Minor fixes.

Changelog

• Various bug fixes in library.
• Adds an advanced internal cache, which capacity is controled via env variable CUBE_CACHE_CAPACITY=<n elements> and variable CUBE_CACHE_MAX_MEMORY=<size>. CUBE_CACHE_CAPACITY controls number of elements kept in cache. CUBE_CACHE_MAX_MEMORY controls maximal memory footprint. Does content of cache reaches CUBE_CACHE_MAX_MEMORY, a warning is issued and cache is emptied, which leads to future recalculation of values. Fine control of memory footprint of CubeLib is combination of both settings.
• Adds an additional data loading strategy, “paranoid”, where only the data is kept in memory, which is currentely required for the calculation. Otherwise it gets removed from the memory. Might lead to the I/O related overhead as CubeLib will read the data from the filesystem more often.
• Adds a command line option for cube_part to partition threads.
• Adds a command line option for cube_part to create stripped cubex file instead of VOIDing the locations.
• Fixes Tau2Cube tool to produce more Cube-like output.
• Improved cube_server:
  • allow a disconnected connection to be reestablished;
  • added support for unix domain sockets.
• Added cmake options for webassembly build.
• Fixes the calculation of POP GPU Efficiency metrics.
• Calculates the POP GPU efficiencies for measurements involving the ‘omp target’ regions.
• Improved performance of the cube_pop_metrics tool.
• FileServer plugin of cube_server is disabled.

Changelog

• Adds support for WebAssembly and WebSockets

Changelog

• Fixes parsing cube files with GPU location and corresponding call paths parameters.