AMD AOCC 3.2 Compiler Released Along With AOCL 3.1 CPU Libraries
AMD AOCC 3.2 is the new version of their LLVM Clang downstream focused on providing the latest optimized compiler support for Zen-based processors whether it be EPYC, Ryzen, or Ryzen Threadripper processors. With AOCC 3.2 they have re-based to LLVM/Clang 13.0 as their compiler release. LLVM 13.0 was released this autumn and is the latest stable LLVM version. AMD now has all of their yet-to-be-upstreamed or unsuitable for upstream patches re-based on this new version.
AOCC 3.2 features a number of improvements around its Flang-based Fortran compiler front-end for LLVM, including various sanitizers working. The (NO)FREEFORM pragma is also now supported and there is improved debug support. OpenMP 4.5 is also now working for Fortran and better support for the Fortran 2008 language version.
AOCC 3.2 remains binary-only and that AMD official Zen code compiler can be downloaded from developer.amd.com. I'll be working on some fresh AOCC 3.2 comparison compiler benchmarks in the days ahead.
AMD also released AOCL 3.1 as their Optimizing CPU Libraries. AOCL 3.1 consists of their optimized versions of BLIS, libFLAME, FFTW, LibM, Sparse, ScaLAPCK, MUMPS, and other optimized libraries. These various libraries popular for different math purposes and used in numerical computing and HPC feature various optimizations aimed for enhancing the performance on AMD Zen-based processors. AOCL 3.1 is available for download as well from developer.amd.com.