Skip to content

Latest commit

 

History

History
44 lines (33 loc) · 1.43 KB

README.md

File metadata and controls

44 lines (33 loc) · 1.43 KB
Raw

sycl-microbench

This is a micro-benchmark for testing the overhead of SYCL features, the following features are tested in this benchmark

  • memory allocation
  • parallelization
  • atomics
  • barriers
  • vectorization

Building

Make is used to build this benchmark.

make VENDOR=acpp|intel-llvm|<empty=DPCPP> BACKEND=omp|<empty=ocl> ARCH=x86|a64fx|graviton3

Depending on the Implementation choosen, paths to all the the binaries and libraries should be added to the standard environment variables, such as $PATH and $LD_LIBRARY_PATH

For optimal performance set OMP_PROC_BIND environment variable is set to true.

Example

./binary [-s size |-b blocksize <optional> |-I No. iterations | --print-system\n"
			--mat-mul: to run matrix multiplication 
			--mat-vec: to run matrix-vector multiplication 
			--triad: to run a triad benchmark 
			--outer-product: to run an outer product benchmark
			--cross-product: to run an cross product benchmark
			-i: for different routines in vectorization benchmark
			      1 - range with USM
			      2 - range with Buffer and Accessors
			      3 - nd_range with USM
			      4 - nd_range with Buffer and Accessor
			--mem-alloc: to alloc memory using SYCL and standard malloc 
			--reduction: to test reduction using atomics and sycl reduction construct
			--range: to test sycl range construct
			--ndrange : to test sycl nd_range construct
			--barrier: to test sycl barrier construct