Vectorized Linear Algebra Implementation for ACTS
- Mentors
- Paul Gessinger, Joana Niermann, Hadrien
- Organization
- CERN-HSF
- Technologies
- c++, simd, sse, avx, intrinsics
- Topics
- high performance computing, optimization, hpc, vectorization
This project is for implementing a vectorized linear algebra backend. The current linear algebra implementation uses scalar instructions, but we can achieve a ~4x speed up if we use vector instructions when the relevant instructions sets (SSE, AVX) are available.
At the end, we expect to have a working SIMD implementation of the linear algebra code (addition and multiplication) for the relevant sizes (4×4, 6×6, 6×8, 8×8), a benchmarking and testing suite, and documentation which explains the code and the API.