Following this week's Intel Atom Silvermont / Bay Trail launch
and my notes about Intel Bay Trail / Silvermont Linux support
, improvements for the new Atom SoCs have been committed to the LLVM tree.
LLVM/Clang has already supported the "-march=slm" architecture flag for optimizing binaries to all of the CPU instruction set extensions provided by Silvermont while now implemented in the LLVM compiler infrastructure is the instruction scheduler latencies for Silvermont that were derived from the new Intel Silvermont Optimization Guide. There's also auto-detection now of Intel Silvermont hardware.
The work that was done by an Intel engineer for LLVM can be found with this Git commit