In 2010-2012, we developed the first complete ARM ports of the ATLAS 3.8.4 and FFTW 3.2.2 floating-point libraries, and distributed them under open source licenses. These ports are now obsolete, and we no longer support them.
ARM support is provided by newer versions of the original software packages, available at the links below. They will give you equal or better performance, and we recommend using them instead of our original ports.
ATLAS 3.10 is a stable release of the ATLAS distribution with ARM support. We contributed the NEON matrix multiplication routines from our version 3.8.4 ARM, and additional improvements and level-2 routines were added by ATLAS lead author Clint Whaley and his students.
FFTW 3.3 includes NEON support written by the original developers of FFTW. Performance is very similar to our NEON implementation.
For users with historical interest, or those who do not wish to change to the newer versions, our original ARM floating point web pages, including source code downloads, benchmark results, and build instructions, are available here.