Skip to content

v0.3

Compare
Choose a tag to compare
@toxicscum toxicscum released this 05 Dec 13:56
· 111 commits to develop since this release
3304e09

Features in This Release

BLAS

  • Added NETLIB X86 CPU backend support in BLAS domain for DPC++, LLVM and hipSYCL compilers

  • Added Unified Shared Memory (USM) Interfaces support in BLAS domain

  • Extended support in BLAS domain with low-precision, mixed precision GEMMs and various batch implementations.

  • Extended support in BLAS domain with transposition routines.

  • Added cuBLAS Nvidia GPU backend support in BLAS domain for hipSYCL compiler

  • Added portBLAS X86 CPU, Intel GPU, Nvidia GPU and AMD GPU backend support in BLAS domain for DPC++ and LLVM compilers.

  • Added rocBLAS AMD GPU backend in BLAS domain for LLVM and hipSYCL compilers.

  • Extended MKLCPU and MKLGPU backend support in BLAS domain with pointers for scalar parameters.

  • Various updates in BLAS domain to align with SYCL 2020 spec and bug fixes

LAPACK

  • Introduced LAPACK SYCL APIs support for x86 CPU and Intel GPU with the Intel® oneAPI Math Kernel Library, Nvidia GPU with the cuSOLVER library and AMD GPU with the rocSOLVER library.

DFT 

  • Introduced DFT SYCL APIs support for x86 CPU and Intel GPU with the Intel® oneAPI Math Kernel Library, Nvidia GPU with the cuFFT library and AMD GPU with the rocFFT library.

RNG

  • Introduced RNG Device API support according to the oneAPI 1.3 specification

Service

  • Replaced FindMKL.cmake with MKLConfig.cmake from Intel oneMKL

  • Extended support for cmake fetchcontent module

Known Issues and Limitations

  • RNG Host API Philox4x32x10 engine may produce incorrect numbers in Skip_Ahead_Ex test in case of mklgpu backend and Intel oneMKL 2024.0 release

  • The LAPACK domain for x86 CPU with Intel(R) oneAPI Math Kernel Library backend has build failures with builds of the LLVM* compiler after 2023-09-27. Use builds from 2023-09-26 or earlier.