# Changelog

## Baseline

- First survey — no prior run to compare against. CAI 65.7.

## Lenses

- Code Health 65.3
- Architecture 92.3
- Maturity 60.8
- Readiness 63.6
- Security 84.1
- Performance 85.0

## Changes since last survey

- 300 commits — 295 feature/other, 5 fixes

## By area

- hpt/src — 83 commits
- (repo) — 44 commits
- (root) — 37 commits
- hpt-dyn/src — 33 commits
- hpt-types/src — 16 commits
- hpt-cudakernels/src — 14 commits
- hpt-tests/src — 14 commits
- hpt-matmul/src — 13 commits
- docs/benchmarks — 10 commits
- docs/user_guide — 9 commits
- docs/.vuepress — 3 commits
- hpt-allocator/src — 3 commits
- hpt-dataloader/src — 3 commits
- hpt-examples/examples — 3 commits
- .cargo/config.toml — 2 commits
- hpt-bench/benches — 2 commits
- hpt-common/src — 2 commits
- src/main.rs — 2 commits
- .vscode/settings.json — 1 commit
- gemm/gemm-common — 1 commit

## Notable commits

- fix: Fix CUDA reduction kernel synchronization and add debug configuration printing
- fix: Fix race condition in CUDA reduce kernel synchronization
- fix: fix doc
- fix: fixed readme, initial gemm implementation
- fix: fixed wrong index calculation for matmul when input is transposed
- change: - accelerate f32 to f16 for Neon. - Use fma for Neon
- change: A
- change: Add CUDA feature to hpt dependency and update documentation imports
- change: Add CUDA support for logsumexp reduction and improve kernel launch configuration
- change: Add CUDA support for reducel3 and sum_square operations
- change: Add GEMM implementation and performance benchmark
- change: Add GEMM operation to Tensor and enhance ONNX execution
- change: Add Linear struct with CUDA tensor initialization
- change: Add crate.io link to documentation navigation
- change: Add documentation for 2D convolution operations
- change: Add new conv2d module and update conv function calls
- change: Add new job calculation function and enhance matmul template
- change: Add parse_args module to ONNX utilities
- change: Add prefetch module and enhance matmul kernel with bias support
- change: Add type casting header for CUDA kernels and update dataloader traits
- …and 280 more
