feat(v0.2.2): Ampere-optimized SGEMM with cp.async pipeline (18 TFLOPS)#37

Merged

m96-chan merged 8 commits intomainfrom

feature/v0.2.2-3090ti-tuning

Dec 13, 2025

Commits on Dec 12, 2025

feat(matmul): Add RTX 3090 Ti optimized kernel with TDD performance tests

m96-chan
and
claude
committed

Commits on Dec 13, 2025

feat(gemm): Add Ampere-optimized SGEMM kernel with cp.async pipeline

m96-chan
and
claude
committed
fix(lint): Fix unused variable warnings in benchmark

m96-chan
and
claude
committed
fix(ci): require SM >= 80 for cmake-check to support cp.async

m96-chan
and
claude
committed
fix(lint): remove unused imports and variables in test file

m96-chan
and
claude
committed
fix(tests): handle missing CUDA directory on Windows CI

m96-chan
and
claude
committed
perf(ci): add CUDA and ccache caching for cmake-check

m96-chan
and
claude
committed
chore(release): bump version to 0.2.2 and update README

m96-chan
and
claude
committed