Advanced Micro Devices has just placed their translation tool, GPUFORT, into open source, which allows for more transparency from the company for developers and enthusiasts to utilize the code and build off of it.
This latest development from AMD comes in response to NVIDIA’s hold on the parallel computing industry with their tool, CUDA. GPUFORT “aims to offer a translation tool so that large codebases in CUDA can work outside of the closed ecosystem of the green giant.” GPUFORT is utilized under the Radeon Open Ecosystem (ROCm).
NVIDIA has kept its CUDA proprietary which has placed limitations on many developers as they brought stuck with NVIDIA’s tool without having any alternatives. AMD recognized this opportunity and provided users with another option while bringing mechanisms to utilize specific CUDA coding to something more compatible with their Radeon technology.
AMD’s previous projects utilized both C and C++ programming, but the GPUFORT is different in that as it translates from “source-to-source … of CUDA Fortran and Fortran code based on OpenACC to OpenMP 4.5+ for execution on GPU or Fortran + HIP C++ code.”
Currently, GPUFORT, along with the Python-based coding, is only capable of performing minor syntax checks instead of full translation. As it currently is, the GPUFORT is only meant to translate HPC applications into compatible code formats which are supported by the AMD ROCm ecosystem. Engineers at AMD admit that GPUFORT currently does not have adequate performance at “analyzing which parts of the code can be downloaded and which cannot, … reorganizing the loops and the assignments to maximize the available parallelism,” and “does not implement the complete OpenACC standard.”
AMD published GPUFORT on coding site GitHub under MIT licensing.
