Skip to content

Latest commit

 

History

History
145 lines (124 loc) · 6 KB

CHANGELOG.md

File metadata and controls

145 lines (124 loc) · 6 KB
Raw

Changelog

v3.3.0

  • Bug fixes and improvements for architectures with weak memory models (i.e., non TSO).
  • Using the synchGetNCores functionality provided by primitives.h instead of numa_num_configured_cpus provided by libnuma.
  • codecov workflow removed.

v3.2.0

  • Adding support for various thread placement policies.
  • Improvements to the threadtools.h API; some basic NUMA functionality is added.
  • A patch on hsynch.c for supporting the various thread placement policies.
  • Improvements of the performance of synchResched for many multicore machines (e.g., AMD, Intel, etc.).
  • Improvements of the performance of the synchGetMachineModel function in the case of multiple calls.
  • The performance of algorithms based on hsynch.c is substantially improved (more than X2 in many cases).

v3.1.4

  • Documentation updates.

v3.1.3

  • Fix for the chlhash.c and dsmhash.c concurrent hash-tables.
  • Fix for the functionality of synchFastRandomRange32 and synchFastRandomRange provided by fastrand.c.
  • Improvements for the benchmarking infrastructure of hash-tables.
  • Adding code for validating the data-structures provided by clhhash.c and dsmhash.c.
  • Increasing code coverage by codecov.

v3.1.2

  • Expanding validate.sh to cover a few more concurrent objects.
  • Increasing code coverage by codecov.
  • Enhancing the API of threadtools.h.
  • Code cleanup for the provided benchmark files.

v3.1.1

  • Bug fix for the pool objects on the recycling functionality.

v3.1.0

  • A new high-performant, optimized implementation for the flat-combining synchronization technique. This new implementation provided by the Synch framework is written from the scratch.
  • A few API fixes.

v3.0.1

  • README.md updated.
  • Adding the JOSS paper to the repository.

v3.0.0

  • Header-files cleanup in order to provide a better and consistent API.
  • Adding the Synch prefix to all structs provided by the runtime/primitives to the end-user.
  • Adding the synch prefix to all functions provided by the runtime/primitives to the end-user.
  • Extensive documentation for the API.
  • Create a doxygen configuration file for auto-generating man-pages/documentation.
  • API documentation is now provided at GitHub pages: https://nkallima.github.io/sim-universal-construction/index.html.
  • A code coverage report is provided through codecov.io for the validation script.
  • Support for installing the framework.
  • Improvements on the build environment.
  • Expansion of the CONTRIBUTING.md.
  • Providing a short discussion (see PERFORMANCE.md) of the expected performance for the various objects provided by the framework.
  • The bench.sh and validate.sh scripts use similar arguments as the provided benchmarks.

v2.4.0

  • Adding the LCRQ Queue implementation (Adam Morrison and Yehuda Afek, PPoPP 2013, http://mcg.cs.tau.ac.il/projects/lcrq).
  • Memory reclamation for SimStack is supported.
  • Adding support for 128 bit Compare&Swap.
  • Documentation improvements; a memory reclamation section was added in README.md.

v2.3.1

  • Improvements on validate.sh for better error reporting.
  • Enhancements on README.md.
  • Improved logo.

v2.3.0

  • Introducing validate.sh, a smoke/validation script that verifies many of the provided data structures.
  • A few improvements on some benchmarks in order to be conformed with validate.sh.
  • Introducing the logo of the repository.
  • Enhancements on README.md.

v2.2.1

  • Executing Pause instructions on spinning-loops for the X86 architecture. Spin-locks and some lock-free algorithms may have performance benefits in SMT architectures.
  • Documentation improvements.

v2.2.0

  • A new implementation of the pool.c functionality. Most stack and queue implementations support memory reclamation, the only exception are the simstack, simqueue, lfstack and msqueue implementations.
  • Code clean-up for the Sim family of algorithms.
  • Improvements on the HSynch family of algorithms.
  • Enhanced NUMA support for the NUMA-aware data-structures.
  • Homogenized code for stack and queue implementations.
  • Better output for bench.sh script.
  • Numerous performance optimizations and bug-fixes, especially in machines with weak memory models.
  • Documentation improvements.

v2.1.1

  • Improving the output messages of build-system.
  • Adding support for clang in Makefile (i.e.,use make clang for building the sources with the clang compile).
  • A few bug-fixes and minor improvements.
  • Dropped support for Intel icc compiler and added support for Intel icx compiler. Tested with Intel(R) oneAPI DPC++ Compiler 2021.1.2.

v2.1.0

  • Significant enhancements on bench.sh.
  • New folder/file library structure.
  • ftime is replaced by clock_gettime.

v2.0.1

  • Removing dead-code from the unsupported sparc/solaris architecture.
  • PAPI and NUMA libraries automatically used depending user's settings in config.h.
  • Pid fix for the Sync family of algorithms.

v2.0

  • Initial support for ARM-V8 and RISC-V machine architectures.
  • Significant performance optimizations for the Synch family of algorithms on modern NUMA machines.
  • The produced binaries can run with any number of threads/fibers without re-compilation.
  • All the produced benchmark binaries can accept arguments, such as the number of threads, the number of benchmark's iterations, etc.
  • Enhancements on how the runtime handles the NUMA characteristics of modern multiprocessors.
  • Better thread affinity policy in NUMA machines.
  • A new script called run_all.sh is provided for automatically testing all the produced binaries.
  • A .clang-format file is provided in order to maintain the styling-consistency of the source code.
  • Numerous performance optimizations and bug-fixes.

v1.9.1

  • Testing has been performed in RISC-V and aarch64 machine architectures.
  • Bug fixes and performance improvements.

v1.9

  • Support for ARM-V8 and RISC-V machine architectures.

v1.8

  • The last version that supports solaris/sparc architecture. From now on sparc/solaris architecture is unsupported.