AdaptiveCpp 24.06 Released As "The Fastest Heterogeneous C++ Compiler" - Beats CUDA

AdaptiveCpp as the open-source compiler formerly known as hypSYCL and Open SYCL is out with a new feature release for this C++ heterogeneous compiler supporting all major CPUs and GPUs.

AdaptiveCpp's tag line is "the fastest heterogeneous C++ compiler - free from vendor politics" and its documentation for the new release goes on to note:

"This release increases performance even further, while also adding various new features. AdaptiveCpp 24.06 is now without a doubt one of the leading heterogeneous C++ compilers when it comes to performance. In many cases, it is faster than vendor-supported compiler stacks such as CUDA or oneAPI. At the same time, as a purely community-driven project, is is completely free from vendor politics, giving the community back control over their preferred programming models."

AdaptiveCpp 24.06 brings new JIT optimizations, C++ standard parallelism offloading using std::execution::par and experimental std::atomic support, runtime latency optimizations, better SYCL 2020 support, and their OpenCL back-end is also now more complete.

AdaptiveCpp 24.06 also brings support for dynamic functions, handling for the sycl::specialized extension, and other improvements.

Benchmarks coming from the AdaptiveCpp developers of this release show that this compiler can outperform CUDA 12.1 and Intel's oneAPI quite often. Results on AMD Radeon Pro hardware also show it typically competitive with AMD ROCm. Intel integrated graphics performance is also in good shape relative to oneAPI.

For the benchmarks and more details on the AdaptiveCpp 24.06 release along with downloads, visit the project's GitHub.