Sorry to be so long replying and thanking you for your help -- I've spent
most of the last week in hospital with very limited communications.
Thank you -- lots of interesting information.
I ended up for the moment with something like (retyping from memory)...
method double {
$!index += 8;
nativecast((num64), Blob.new(@!bytes[ $index - 8 ..^ $index ] );
}
which (summary to memory errors) is working fine, at least until I try to
run the code on a machine with big-endian floats, when it will require some
reversing of bytes, or a machine without IEEE floating point (do such
machines still exist?) in which case Marcel's code could be very useful. By
then perhaps there will have been consensus and implementation of a
pack/unpack native solution for Perl 6.
Thanks everybody.
On 21 April 2016 at 01:47, mt1957 <mt1...@gmail.com> wrote:
>
> On 19-04-16 10:21, Elizabeth Mattijsen wrote:
>
>> FWIW, I’ll take PR’s for the PackUnpack distribution to make ‘f’ and ‘d’
>> work :-)
>>
>
> Hi Elizabeth,
>
> For the PackUnpack distro this might come in handy... or might go in the
> examples corner?
>
> Done some experiments and looks well. Please check the attachment for
> encoding/decoding of doubles. decode-double has also an index in the buffer
> to pull out a specific spot in the buffer where the encoded double should
> be. The code is made from snippets from Timo Paulsen and
> David Warren.
>
> You might want to know how much faster it has become. Well it's not that
> much of an improvement, only encoding a double is about 4 times faster.
> The decoding only 1.13 times.
>
> emulated encoded
> 3000 runs total time = 7.211524 s, 0.002404 s per run, 416.000809 runs per
> s
>
> native encoded
> 3000 runs total time = 1.720918 s, 0.000574 s per run, 1743.256109 runs
> per s
>
> emulated decode
> 3000 runs total time = 1.038615 s, 0.000346 s per run, 2888.461538 runs
> per s
>
> native decode
> 3000 runs total time = 0.918133 s, 0.000306 s per run, 3267.498734 runs
> per s
>
>
> Before I pat myself on the back for the 'fast' emulated encode/decode
> already in place in the BSON package the above results might mean that the
> native routines can be done better.
>
> Regards,
> Marcel
>
>
>
