[Stk] bug in JCRev? (oHop too big)

[Stephen Sinclair]

One approach might be to develop under a kind of unit test approach.
Run the per-sample computations and block-based computations under the
same conditions and check that they produce the same results..

Steve

On Thu, Mar 3, 2011 at 4:56 PM, Morgan Packard <morgan at morganpackard.com> wrote:
> Thanks Gary,
> I definitely hear you w/regard to "rather spend my time on other issues"!
> Unfortunately, the "modern CPU's are so fast" statement isn't true for those
> of us developing software for mobile devices.
> If I wind up trying to improve the performance of one of those reverbs I'll
> share my code with any who want to take a look.
> -Morgan
>
> On Wed, Mar 2, 2011 at 8:54 PM, Gary Scavone <gary at ccrma.stanford.edu>
> wrote:
>>
>> Yes, in fact this is a bug in many or all of the Effects classes.  I fixed
>> it a while back in my own version with the hope that a new release would be
>> forthcoming shortly.  Needless to say, things have been delayed more than I
>> expect.  But hopefully, the new release will happen this Spring.
>>
>> Regarding block-based processing, I considered it but eventually didn't
>> bother because:
>>
>> - the possible performance improvement is not necessarily great (based on
>> a few basic tests)
>>
>> - there is the possibility of introducing yet more bugs
>>
>> - modern CPUs are so fast now that I'd rather spend my time on other
>> issues
>>
>> Best,
>>
>> --gary
>>
>> On 2011-03-01, at 7:10 PM, Morgan Packard wrote:
>>
>> > Hello all,
>> >
>> > FYI, I think I've found a bug in the multi-frame version of JCRev ::
>> > tick. The method was hopping three samples on every input frame instead of
>> > two samples.
>> > Perhaps I'm using it incorrectly, but it looks to me like no one has
>> > ever tested this method. Anyway, after a simple adjustment to oHop, it works
>> > fine for me. My fix is here:
>> >
>> > http://tinypaste.com/7b6860
>> >
>> > Also on the subject of JCRev, I noticed that it's calculating frames
>> > sample-by-sample rather than in blocks of StkFrames, which is not the most
>> > optimized way to do things. Is there a reverb in STK which has been written
>> > with more of an eye toward performance? I hesitate to offer to optimize this
>> > class since I'm basically a noob with regard to both DSP and C++, but I'd be
>> > happy to contribute if I can.
>> >
>> > thanks,
>> >
>> > -Morgan
>> >
>> > --
>> > ================================
>> > Web:
>> > http://www.morganpackard.com
>> >
>> > Music/Art:
>> > Latest album: Moment Again Elsewhere
>> > iOS app Thicket available on iTunes store.
>> > ================================
>> >
>> > _______________________________________________
>> > Stk mailing list
>> > Stk at ccrma.stanford.edu
>> > http://ccrma-mail.stanford.edu/mailman/listinfo/stk
>>
>
>
>
> --
> ================================
> Web:
> http://www.morganpackard.com
> Music/Art:
> Latest album: Moment Again Elsewhere
> iOS app Thicket available on iTunes store.
> ================================
>
> _______________________________________________
> Stk mailing list
> Stk at ccrma.stanford.edu
> http://ccrma-mail.stanford.edu/mailman/listinfo/stk
>
>