gale-v2-refactor - Refactor gale v2 while preserving GPU performance.
Consolidate CPU and GPU DG performance optimizations into Sim/Device abstractions while defending measured benchmark baselines.
Tags
Updated: 2026-09-20Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Consolidate CPU and GPU optimizations
- Refactor Sim and Device abstractions
- Freeze performance benchmark baselines
- Profile GPU kernels and graphs
- Implement dual-impl building blocks
Inputs
- docs/gale-v2-refactor.md design ledger
- Performance benchmark baselines
- ncu and nsys profiling logs
- gale-gpu-perf skill guidelines
Outputs
- Refactored src/sim abstraction files
- Updated benches baseline suite
- Feature-flagged cargo build configurations
- Dual-implementation building block code
Requirements
- cuda-oxide toolchain
- NVIDIA GPU with sm_70 architecture support
- nsys and ncu profiling tools
- Cargo build system