Clemens Heitzinger <cheitzin@...> writes:
> Would be great if anybody had an idea how to speed up something like
> (* z1 (log (- z0))), where z1 and z0 are complex double-floats. I get
> this compiler note. I am mostly interested in Linux/x86, if this make
> any difference.
> ; in: DEFUN ADD-TO-LOCAL-EXPANSION
> ; (LOG (- FAST-MULTIPOLE-METHOD::Z0))
> ; note: doing complex double float to pointer coercion (cost 13)
> ; compilation unit finished
> ; printed 1 note
This, I'm afraid, is somewhat of a late reply... (hm, only thirteen
(deftransform log ((x) ((complex double-float)))
'(complex (log (abs x)) (phase x)))
work for you? (Or, I guess, anyone who still cares...) Simple
testing says that functions of the form above go about six or seven
times faster with this transform enabled.