On 15 Feb 2008, lemming@henning-thielemann.de wrote:
On Fri, 15 Feb 2008, Jed Brown wrote:
On 15 Feb 2008, lemming@henning-thielemann.de wrote:
On Thu, 14 Feb 2008, Jed Brown wrote:
Hopefully these are mature enough to be generally useful. I would appreciate any comments regarding the interface. The FFTW API is not particularly nice for building a pure interface on.
because FFTW stores global data (called 'plan's)
Yes. In the Haskell interface, the existence of these plans is hidden From the user.
With locks for threaded use?
In FFTW, planning is not thread-safe, but execution of plans is. The Haskell library takes a lock during planning, but releases it before execution. Planning can be time-consuming if you want a high quality plan, and unfortunately this phase is single threaded. It would be nice if FFTW would expose slightly more of its internal interface, or take its own lock when modifying it's cache so that planning would be thread-safe. Once a plan for your transform size is created (or imported through the wisdom interface) the call to the planner will return immediately and everything else can run in parallel. In the Haskell interface, the default planner is 'estimate' which just employs heuristics rather than actually computing some transforms. If you have imported the correct wisdom, this will still do an optimized transform. If not, it will be less than optimal, but there will be no time-consuming planning phase. If you are concerned about getting the absolute best performance, you can use the advanced interface (which still hides locking, memory allocation, and internals of multi-dimensional transforms) for more control over planning. Jed