onemkl_interfaces_gemm_example

Repo contains examples for oneAPI Math Kernel Library Interfaces project

Reproducing SC21 paper results

1. Build oneMKL project for MKLCPU, MKLGPU, and CUBLAS backends (see build instructions here).
2. Make sure you source required compilers (Intel DPC++ for MKLCPU/MKLGPU backends and DPC++ with CUDA backend for CUBLAS backend) and third-party libraries (Intel oneMKL and NVIDIA cuBLAS)
3. Set ONEMKL path to oneMKL interfaces header files and libraries (GNU Makefile from the repo expects $ONEMKL/include path with oneMKL header files and $ONEMKL/lib with libraries)
4. Use GNU Makefile from the repo to build and run GEMM benchmark as shown below:

# build and run GEMM with MKLCPU backend
> make mklcpu

# build and run GEMM with CUBLAS backend
> make cublas

# build and run GEMM with MKLGPU backend
> make mklgpu

# clean the repo
> make clean