Conjugate Gradient implementation on Meluxina FPGAs

The following project collects the code employed in the benchmarking of the Conjugate Method on Meluxina FPGAs, in the context of the "Advanced Computer Architectures" course held at PoliMi during the a.y. 2023/2024.

Compilation

The code was intended to be specifically tailored to the author needs, but still it can be compiled to be used from other Meluxina users. The framework employed is Intel HLS tools along with OpenCL.

Device code/Kernels compilation

To compile the FPGA kernels into binaries, follow the instructions in compilation_utils/compile_fpga.sh. This step could take from 5 to 10 hours to complete.

Host code compilation

In src/main.cpp, specify the directory containing the matrices and rhs following the requested file name format. In src/host_code.cpp, define in both prepare methods the path to the FPGA binaries (.aocx files).

Run

cd opencl
mkdir build
make -j
./cg_opencl {size_of_matrix} {repetitions of runs}