[Lab proposal] offload computations to a GPU

[dendibakh]

The lab shouldn't necessarily be very complicated. The goal is to show students how to offload computations to a GPU.
SYCL + Intel's DPCPP compiler should be good candidates.

[0legmak]

I've got an idea for a lab. It is similar to the "loop tiling" lab but GPU-based. GPUs have a concept of local (shared) memory, similar to CPU caches, except it must be managed manually by a programmer. The idea is to optimize matrix multiplication by processing input matrices using small tiles. Here is a proposed implementation of the lab using OpenCL to interface a GPU: https://github.com/0legmak/perf-ninja/blob/1779f209b3d40a2986bf4ee39e37d4375ea108b1/labs/core_bound/matmul/solution.cpp
@dendibakh Please let me know what you think.