One-line FFT, nice!

[Brad Roberts]

On 9/8/2012 6:01 PM, Timon Gehr wrote:
> On 09/09/2012 02:57 AM, Timon Gehr wrote:
>> On 09/09/2012 12:56 AM, Mehrdad wrote:
>>> I was pretty excited to figure out that a one-liner FFT is
>>> possible in D!
>>> creal[] dft(creal[] v) { return v.length > 1 ? (p => chain(map!(q
>>> => q[0] + q[1])(p), map!(q => q[0] -
>>> q[1])(p)))(zip(dft(v.stride(2).array()), map!(p => p[1] *
>>> expi(p[0] * -2 * PI / v.length))(zip(iota(v.length / 2),
>>> dft(v.drop(1).stride(2).array()))))).array() : v; }
>>>
>>> Of course, the Python version is still shorter:
>>> def dft(v): return (lambda e, o: map(add, e, o) + map(sub, e,
>>> o))(dft(v[0::2]), [o * rect(1, k * -2 * pi / len(v)) for k, o in
>>> enumerate(dft(v[1::2]))]) if len(v) > 1 else v
>>>
>>> but it's still cool (barring the unreadability, haha).
>>> Yay D!
>>
>> I usually get NaNs. Are you sure this is correct?
> 
> Works correctly in 32 bit mode. 64 bit code gen bug.

Bug number?