Chris@42: Next: Generating your own code, Previous: Installation on non-Unix systems, Up: Installation and Customization [Contents][Index]
Chris@42:FFTW’s planner actually executes and times different possible FFT Chris@42: algorithms in order to pick the fastest plan for a given n. In Chris@42: order to do this in as short a time as possible, however, the timer must Chris@42: have a very high resolution, and to accomplish this we employ the Chris@42: hardware cycle counters that are available on most CPUs. Chris@42: Currently, FFTW supports the cycle counters on x86, PowerPC/POWER, Alpha, Chris@42: UltraSPARC (SPARC v9), IA64, PA-RISC, and MIPS processors. Chris@42:
Chris@42: Chris@42:Access to the cycle counters, unfortunately, is a compiler and/or
Chris@42: operating-system dependent task, often requiring inline assembly
Chris@42: language, and it may be that your compiler is not supported. If you are
Chris@42: not supported, FFTW will by default fall back on its estimator
Chris@42: (effectively using FFTW_ESTIMATE for all plans).
Chris@42:
Chris@42:
You can add support by editing the file kernel/cycle.h; normally,
Chris@42: this will involve adapting one of the examples already present in order
Chris@42: to use the inline-assembler syntax for your C compiler, and will only
Chris@42: require a couple of lines of code. Anyone adding support for a new
Chris@42: system to cycle.h is encouraged to email us at fftw@fftw.org.
Chris@42:
If a cycle counter is not available on your system (e.g. some embedded
Chris@42: processor), and you don’t want to use estimated plans, as a last resort
Chris@42: you can use the --with-slow-timer option to configure (on
Chris@42: Unix) or #define WITH_SLOW_TIMER in config.h (elsewhere).
Chris@42: This will use the much lower-resolution gettimeofday function, or even
Chris@42: clock if the former is unavailable, and planning will be
Chris@42: extremely slow.
Chris@42:
Chris@42: Next: Generating your own code, Previous: Installation on non-Unix systems, Up: Installation and Customization [Contents][Index]
Chris@42: