AdaptiveCpp Enables CUDA Dialect Support on Apple GPUs Through New Metal Backend

Summary

AdaptiveCpp, an open-source SYCL implementation project, is adding support for CUDA dialect APIs on Apple's Metal GPUs through a new pull request. The development, submitted by contributor resetius, aims to enable the 'pcuda' API on Metal backend, potentially allowing CUDA-based code to run on Apple Silicon GPUs. This represents a significant step in cross-platform GPU computing, as it could bridge the gap between NVIDIA's CUDA ecosystem and Apple's Metal framework.

AdaptiveCpp (formerly known as hipSYCL) is designed to provide a single-source C++ programming model for heterogeneous computing across different hardware vendors. The project supports multiple backends including CUDA, ROCm, Level Zero, and now increasingly robust Metal support for Apple platforms. By enabling CUDA dialect support on Metal, developers could potentially port CUDA applications to Apple Silicon without complete rewrites.

The pull request is currently open and awaiting review in the project's GitHub repository. While technical details of the implementation are limited in the public submission, the move aligns with broader industry efforts to improve GPU computing portability. Apple's Metal Performance Shaders and GPU capabilities have been growing, particularly with the M-series chips, making cross-compatibility with established frameworks like CUDA increasingly valuable for the scientific computing and machine learning communities.

• This advancement could benefit machine learning and scientific computing workloads on Apple Silicon by leveraging existing CUDA codebases