Better Fujitsu A64FX Support Arrives For GCC, LLVM Clang Compilers
The Fujitsu A64FX, which powers the "Fugaku" supercomputer among other accomplishments, has seen open-source compiler work going back a year while now the latest upstream GNU Compiler Collection (GCC) and LLVM Clang are seeing more complete support.
While GCC 11 and Clang 11 support -mcpu=a64fx for targeting the A64FX processor, only this month is the more complete support in place. LLVM 12 is seeing the complete scheduling model catering to the A64FX design. Similarly, GCC is seeing the cost table for the A64FX. This work by Fujitsu is based on their public documentation around the A64FX architecture.
These additions should allow the open-source compilers to make better code generation decisions when explicitly targeting the A64FX as part of builds -- for those lucky enough with hardware access to this high performance 64-bit ARM hardware.
Those curious about the A64FX design and wanting some fun technical weekend reading can check out Fujitsu's A64FX GitHub repository for the microarchitecture manual, A64FX HPC Extension specification, and more.