Version 4.2.1

Very minor revision to not throw an error for new versions of CUDA and to take out some regressions that do not need to be run.

Version 4.2.0

GPGPU-Sim Release Notes

Major Features

SST Integration

• Added support for integrating GPGPU-Sim with the Structural Simulation Toolkit (SST)
• Implemented SST-specific CUDA API functions and memory handling
• Created an SST-specific GPU simulation class (sst_gpgpu_sim) for seamless integration
• Added callbacks and synchronization mechanisms for SST communication

Stream-based Statistics

• Implemented per-stream statistics tracking for more granular performance analysis
• Modified cache statistics to support stream-specific data collection
• Updated statistics printing functions to allow filtering by stream ID

Improved GCC Version Detection

• Enhanced GCC version detection to support newer compiler versions (e.g., GCC 11+)
• Updated version detection logic in setup scripts and makefiles

Performance Improvements

• Modified L2 bandwidth calculation to account for different core and interconnect frequencies
• Optimized cache statistics collection and reporting

Bug Fixes

• Fixed SEGFAULT issue related to kernel tracking in shader cores
• Corrected stream ID handling in various simulator components
• Addressed issues with cache statistics aggregation and reporting

Other Enhancements

• Added support for indirect texture access in PTX simulation
• Improved handling of CUDA events and synchronization
• Enhanced power model integration and statistics collection
• Updated CUDA version detection for compatibility with newer CUDA toolkits

Developer Tools

• Added GitHub Actions workflow for SST integration testing

Notes for Developers

• The codebase now supports C++11 by default
• Several internal APIs have been modified to accommodate stream-based statistics and SST integration
• Developers should review changes to statistics collection and reporting when working with cache and memory subsystems

This release significantly enhances GPGPU-Sim's capabilities, particularly in the areas of integration with other simulation frameworks, performance analysis, and support for modern development environments. Users and developers are encouraged to review the detailed commit history for a complete understanding of all changes.

GPGPU-Sim 4.0.1

Minor updates.

GPGPU-Sim 4.0.0

This version of GPGPU-Sim includes changes to model the memory system and tensor cores in NVIDIA's Volta GPU, support for CUDA dynamic parallelism, enable execution of cuDNN, and work with the NVIDIA SASS trace-based simulation enabled by AccelSim.