LLVM/Clang 9.0 Merges Support For Intel "Cooperlake" CPU Target

The LLVM 9.0 compiler code in development along with the Clang 9.0 C/C++ front-end now have support for the -march=cooperlake target for optimizing the generated code for next-generation Intel Cooper Lake processors.

Cooper Lake is the successor to the recently launched Cascade Lake processors. Cooper Lake sticks with 14nm++ and is expected to be out in H1'2020 with support for eight memory channels per CPU, possible PCI Express 4.0, and other modest improvements over current-generation Xeon Scalable processors.

Merged into the LLVM/Clang compiler development code today is the handling for the -march=cooperlake option.

The new code does confirm Cooper Lake having AVX-512 BF16 (BFloat16) support, which was already merged into the GCC and LLVM/Clang compilers separately. BFloat16 is designed to help deep learning / machine learning workloads as this new optimized floating-point format. But beyond the BF16 support, there aren't any other new instruction set extensions enabled for Cooper Lake over existing Skylake Server processors.

As of writing, the GCC 10 compiler hasn't yet seen "-march=cooperlake" support merged but that is also expected to happen soon and will be well in time for the GCC 10.1 stable release due out next year. LLVM/Clang 9.0 meanwhile should debut around September allowing this Cooper Lake compiler support to become more widespread ahead of the processor's debut.