The Agent Skills Directory

Dynamic Execution Patterns: The skill uses Python's importlib module to dynamically load and execute kernel packages and baseline reference files. This is found in scripts/benchmark_cpu.py and scripts/cpu_profiler.py and is a standard requirement for benchmarking and verifying various code implementations.
Command Line Execution: The scripts/cpu_profiler.py script utilizes subprocess.run to invoke the perf system utility. This is used to collect hardware performance counters (such as IPC and cache misses) to provide optimization guidance to the user.
Runtime Script Generation: To facilitate profiling, scripts/cpu_profiler.py generates a temporary Python script that executes the target kernel and runs it as a separate process. This pattern allows the profiler to isolate the kernel execution for accurate measurement.
File System Operations: The scripts/trial_manager.py utility performs file copying and directory management within a local trials/ folder to track different optimization attempts. This is an expected behavior for a workflow management tool.
Indirect Data Ingestion: The skill analyzes and executes user-provided PyTorch files (e.g., baseline.py) to establish performance targets. This represents a standard interface for developer-oriented tools where the tool operates on user-supplied source code.

cpu-kernels