SYCL backend in ALIEN

Introduction

Alien provide a SYCL backend to handle NVidia, AMD and Intel GP-GPUs

Alien SYCL backend has been tested with:

• AdaptiveCPP (former hipSYCL) implementation of the SYCL 2020 API version 0.9.4 with CUDA and HIP-ROCM

• oneAPI 2024.0, dppc++

It depends on :

• LLVM and Clang

• CUDA 12 to handle NVidia GP-GPUs

• ROCM to handle AMD GP-GPUs

• OneAPI and DPC++ for Intel GP-GPUs

It provides a Block EllPack Matrix implementation and a Linear Algebra with all the Blas 1 and 2 operations required to implement the CG and BiCGStab krylov algorithms.

Some Matrix Vector Builders abd Accessors are provided to enable Matrix and Vector assembly directly on the device memory.