GitLab

- Rename transfer fields for clarity and introduce separate methods for reporting non-P2P and P2P transfers.
- Add new P2P fields extraction and sorting logic to improve data presentation.
- Update method names and comments for better understanding of functionality.

- Introduce `rccl_log_parser.py` for parsing RCCL logs, extracting system information, user-defined environment variables, graph info, and transfer arguments.
- Add usage examples in `README.md` for running the parser as a wrapper and processing existing log files.

- Update run.sh to include new options for warmups and prompt stretching.
- Refactor test_evo2_generation_batched.py to improve trace output formatting and add support for warmup sequences.
- Adjust batch processing to include detailed profiling for each step.

- Remove prompt_stretch option from run.sh
- Adjust condition in test_evo2_generation_batched.py to allow prompt stretching based on batch size

- Update run.sh to include new command-line options for prompt stretching and token limits.
- Modify test_evo2_generation_batched.py to adjust profiling settings and improve output formatting.
- Add support for stretching prompts to the longest length for batch processing.

- Update run.sh to include trace logging options with gzip support.
- Modify test_evo2_generation_batched.py to add command-line arguments for trace log directory and gzip option.
- Refactor custom trace handler to utilize gzip compression for trace outputs.

- Remove gemv_export.cpp
- Update Makefile and README for compiler variable changes
- Adjust run-all.sh for consistent build commands

- Introduce kernel_launch_overhead.cu to measure kernel launch latency, system throughput, CPU dispatch overhead, and GPU dispatch time.
- Create Makefile for building the benchmark with support for nvcc and hipcc.
- Add run-all.sh script to execute the benchmark with specified device settings.

- Add fix-pt-trace.sh for repairing non-UTF-8 traces.
- Remove deprecated run-rocblas.sh.
- Update trace handler (worker names) and tune GPU bindings in run-all.sh.