Thanks for the information. I seem to recall that you discussed this code somewhere, and I read your very clear words and thus assumed you had written the code.

At any rate, I do not see how this code could possibly be improved, but there may be instructions in some processors that could improve it. This should be researched.

In spite of your opinion and my opinion, the final arbiter should be the profiler.

Regards,
Mike

-----Original Message-----
>From: Richard Dobson 
>Sent: Apr 16, 2008 8:25 AM
>To: Developer discussions 
>Subject: Re: [Cs-dev] Profiling
>
>Michael Gogins wrote:
>> Array indexing, yes. Of course it is a lot of work.
>> 
>> Int casting, yes. We should have a macro or inline function for this.
>> 
>> Power of two, I'm not sure about that, it's one thing to profile. Dobson or 
>> someone made this modulo 2 wrap around for oscillator phase that is quite 
>> efficient, we may need to keep that.
>> 
>
>
>Not me, it is the original Vercoe code from way back when. The modulo 
>form results in  code with no conditional tests, from which much of the 
>efficiency is obtained. The difference (compared to "normal" table 
>wraparound) is not trivial.
>
>Richard Dobson
>
>
>-------------------------------------------------------------------------
>This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
>Don't miss this year's exciting event. There's still time to save $100. 
>Use priority code J8TL2D2. 
>http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
>_______________________________________________
>Csound-devel mailing list
>Csound-devel@lists.sourceforge.net
>https://lists.sourceforge.net/lists/listinfo/csound-devel




-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Csound-devel mailing list
Csound-devel@lists.sourceforge.net

Michael Gogins wrote:
> Thanks for the information. I seem to recall that you discussed this
> code somewhere, and I read your very clear words and thus assumed you
> had written the code.
> 
> At any rate, I do not see how this code could possibly be improved,
> but there may be instructions in some processors that could improve
> it. This should be researched.
> 
> In spite of your opinion and my opinion, the final arbiter should be
> the profiler.
> 

It was probably the standalone version I did for Rick''s book, comparing 
lookup methods, etc. With hardware f/p these days the cpu differences 
must be much less nowadays than they were. Profiling info will be very 
interesting to see; of course we want code that runs acceptably fast on 
a variety of platforms without too much (if any) machine-specific stuff. 
The modulo-2 lookup code is only some 10 lines or so, with a couple of 
MAC type operations, and AND and a shift (so, very nice for a DSP chip). 
Not sure if there is much scope for vector acceleration there; but there 
could well be for a composite object such as an oscillator bank or 
quadrature oscillator.

Richard Dobson


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Csound-devel mailing list
Csound-devel@lists.sourceforge.net