Re: Computing exp(z)
- From: gh14tq5@xxxxxxxxx
- Date: 19 May 2005 16:42:24 -0700
Richard E Maine wrote:
> One would sort of expect that to be the way of things.
> If a user can correctly duplicate the complete functionality of an
> intrinsic procedure and do so faster than the intrinsic, then either
> 1. The intrinsic was implemented pretty sloppily.
I don't know how the intrinsic is implemented.
> 2. The user is exceptionally talented and compiler vendors should
> consider hiring him.
2. This is definitely not the case. Half the time, I'm amazed my
programs even work at all.
> 3. It didn't actually correctly duplicate the complete functionality.
> Case 3 is probably the most likely, though the others are certainly
> possible and have probably happened at one time or other.
> Of course, there are times when the complete functionality isn't
> That's a different story. If you know the range of values that will
> used and don't have to fuss with edge cases, maybe this can turn into
> speedups. Likewise you might be able to trade off some accuracy for
> speed if the answer doesn't have to be accurate to the last bit.
I think this is the case. I do not expect my implementation to be
accurate for all values of z. But, in my program, the range of z
values is limited. In this range, the substitute seems to give
identical results to the intrinsic. My problem is basically an inverse
Fourier transform ( int(f(w)exp(-j*w*t)dw) ) that has been slighly
deformed into the complex plane for small values (in magnitude) of the
variable w to avoid some problem points. For large values of w, w is
limited to the real axis. t is real and and limited in magnitude.