Intel Cascade Lake Target Added To LLVM / Clang Compiler
As of today the cascadelake target was added to LLVM and hooked up as well for Clang with the "-march=cascadelake" targeting for optimized code around this successor to Skylake-SP.
The addition is quite simple and reuses the existing Xeon Scalable (Skylake) optimizations/tables while adding in VNNI support. VNNI support was previously added to LLVM and GCC at the time with plans on it premiering in Icelake. Vector Neural Network Instructions is part of AVX-512 and is part of the deep learning improvements for Cascade Lake.
That's it on the compiler side with just adding AVX-512 VNNI on top of the existing Xeon Skylake Server parts. Outside of the compiler scope are also other improvements with Cascade Lake, mainly support for 3D XPoint memory modules and in-hardware mitigations for Spectre and Meltdown.
The Intel Cascade Lake support for LLVM/Clang will premiere with the LLVM 8.0 release due out early next year.