NVIDIA Carmel Support Finally Mainlined In LLVM/Clang
NVIDIA's Carmel design is based on an 8 core layout and based on ARMv8.2+FP16 with SIMD, VFP, and the other usual extensions. Carmel offers much greater CPU performance than earlier Tegra SoCs. At least until NVIDIA Orin hardware begins shipping at scale, the Tegra Xavier SoC with the Carmel CPU cores remains their latest and greatest within the DRIVE Xavier, Jetson AGX, Jetson Xavier, and other products.
With the support now added to LLVM/Clang is -mcpu=carmel and -mtune=carmel support for generating optimized binaries for the Tegra194 hardware. Making use of the Carmel target will enable the relevant supported AArch64 instructions for a more convenient build experience.
The Carmel support is merged for the LLVM 11.0 release due out around September.