Adam

I have just written an MSP pitch detector object in java. It works perfectly but has a very high load on the CPU. In fact, if I use a signal vector size of more than 256 it causes clicking artefacts in the audio. The code isn't really that big and I am using a brand new intel mac book 2.2 GHz machine.

Is this because of some speed limitation of using mxj quickie to build java objects? How could I improve the speed?

I am normally a C programmer and have coded much much larger things that run in real-time as VST plug-ins for example - and they generally are not that much of a burden on the CPU. I know Java is slower than C but I didn't expect it to be quite so limiting, is there something I am doing wrong?

Alternatively, if anyone could point me to a good introduction to programming Max/MSP objects in C I would be very grateful (apologies as this question prob should go in the other forum but the first question should be here...).

I'd be very happy if someone could point me in the right direction,

how-efficient-is-coding-objects-using-java

Java is definitely slower than C but it is possible

with some care to get it running pretty efficiently.

I could take a look at your code if you want. Things to avoid

are unnecessary object allocation, method call overhead,

etc. Generally having a knowledge of what the JVM will do

I generally tend to think as if I am writing C code when writing

mxj~ stuff and stay as much as possible with the primitive constructs

On Oct 24, 2007, at 07:54 AM, Adam Stark wrote:

> I have just written an MSP pitch detector object in java. It works 

> perfectly but has a very high load on the CPU. In fact, if I use a 

> signal vector size of more than 256 it causes clicking artefacts in 

> the audio. The code isn't really that big and I am using a brand 

> Is this because of some speed limitation of using mxj quickie to 

> build java objects? How could I improve the speed?

> I am normally a C programmer and have coded much much larger things 

> that run in real-time as VST plug-ins for example - and they 

> generally are not that much of a burden on the CPU. I know Java is 

> slower than C but I didn't expect it to be quite so limiting, is 

> Alternatively, if anyone could point me to a good introduction to 

> programming Max/MSP objects in C I would be very grateful 

> (apologies as this question prob should go in the other forum but 

> I'd be very happy if someone could point me in the right direction,

Java is definitely slower than C but it is possible
with some care to get it running pretty efficiently.

I could take a look at your code if you want. Things to avoid
are unnecessary object allocation, method call overhead,
etc. Generally having a knowledge of what the JVM will do
when it executes certain instructions.

I generally tend to think as if I am writing C code when writing
mxj~ stuff and stay as much as possible with the primitive constructs
of the java language.

>
> Hi,
>
> I have just written an MSP pitch detector object in java. It works  
> perfectly but has a very high load on the CPU. In fact, if I use a  
> signal vector size of more than 256 it causes clicking artefacts in  
> the audio. The code isn't really that big and I am using a brand  
> new intel mac book 2.2 GHz machine.
>
> Is this because of some speed limitation of using mxj quickie to  
> build java objects? How could I improve the speed?
>
> I am normally a C programmer and have coded much much larger things  
> that run in real-time as VST plug-ins for example - and they  
> generally are not that much of a burden on the CPU. I know Java is  
> slower than C but I didn't expect it to be quite so limiting, is  
> there something I am doing wrong?
>
> Alternatively, if anyone could point me to a good introduction to  
> programming Max/MSP objects in C I would be very grateful  
> (apologies as this question prob should go in the other forum but  
> the first question should be here...).
>
> I'd be very happy if someone could point me in the right direction,
>
> Many thanks,
>
> Adam


Using the benchmark message allows you to quantify any optimizations 

and could be helpful if you are not aware of it. See the mxj~ 

Using the benchmark message allows you to quantify any optimizations  
that you might be making
and could be helpful if you are not aware of it. See the mxj~  
examples for use of the benchmark
message.

Thanks for your replies. It would be great if you could have a look at my code. I have attached the java file. The function causing the problems is the pitch_YIN function.

I also have one more question which is about C coding Max/MSP objects - so far, all the examples I have seen so far in C have been massively complex, uncommented and unexplained. The Java equivalent is so simple. Is there a simple C example somewhere or a good guide? Or is it generally complicated?

I am about to get on a plane here but will check the code in the next 

The c external api is covered in writing max/msp externals by david 

It is not overly complicated. Just a little different.

I agree that there could be a higher level c api for externs. I was 

working on a similar object to mxj called cfunk that addresses these 

issues by allowing you to write in a similar way but in c instead of 

java .I have stopped development of cfunk but perhaps I will finish it 

up someday or someone else is interested in taking it over.

Hola
I am about to get on a plane here but will check the code in the next 
day or so.

The c external api is covered in writing max/msp externals by david 
zicarelli.

I agree that there could be a higher level c api for externs. I was 
working on a similar object to mxj called cfunk that addresses these 
issues by allowing you to write in a similar way but in c instead of 
java .I have stopped development of cfunk but perhaps I will finish it 
up someday or someone else is interested in taking it over.


the reason why your code is slow is not due to the language (at least 

not entirely), but to the fact that you try to calculate

the square difference function in the time domain. in the nested for- 

loops you are doing a lot of calculations...

but using FFT you can easily calculate the auto-correlation much 

i believe that's the trick other yin-pitchtrackers use.

On 25 Oct 2007, at 15:19, Adam Stark wrote:

> Thanks for your replies. It would be great if you could have a look 

> at my code. I have attached the java file. The function causing the 

> I also have one more question which is about C coding Max/MSP 

> objects - so far, all the examples I have seen so far in C have 

> been massively complex, uncommented and unexplained. The Java 

> equivalent is so simple. Is there a simple C example somewhere or a 

> good guide? Or is it generally complicated?

> I'm very very appreciative of your help,

hi adam,
the reason why your code is slow is not due to the language (at least  
not entirely), but to the fact that you try to calculate
the square difference function in the time domain. in the nested for- 
loops you are doing a lot of calculations...
this wouldn't be efficient in C either.
but using FFT you can easily calculate the auto-correlation much  
quicker.
i believe that's the trick other yin-pitchtrackers use.

> Hi Topher,
>
> Thanks for your replies. It would be great if you could have a look  
> at my code. I have attached the java file. The function causing the  
> problems is the pitch_YIN function.
>
> I also have one more question which is about C coding Max/MSP  
> objects - so far, all the examples I have seen so far in C have  
> been massively complex, uncommented and unexplained. The Java  
> equivalent is so simple. Is there a simple C example somewhere or a  
> good guide? Or is it generally complicated?
>
> I'm very very appreciative of your help,
>
> Thanks
>
> Adam
>


Yes, that is a good suggestion and would make it run much faster. I also found another problem. This is maybe very obvious to more experienced Java users but I am new to using Java for signal processing: I found that, in my loop, when I calculated the square of a number using:

my function took 19ms to do it's calculations. However, when I used:

the processing time was reduced to just 3ms.

Anyway, thank you everyone for your help,

I still haven't found a spot to view your attachment but that makes 

sense. Method call overhead in tight loops adds up and the java Math 

library is typicaly slow on top of that. In the mxj~ noise examples you 

wil see I opted to write my own random fucntion instead of using 

Math.nextfloat. It seems like you are able to make use of the 

I still haven't found a spot to view your attachment but that makes 
sense. Method call overhead in tight loops adds up and the java Math 
library is typicaly slow on top of that. In the mxj~ noise examples you 
wil see I opted to write my own random fucntion instead of using 
Math.nextfloat. It seems like you are able to make use of the 
benchmarking stuff as well. Kudos.


How efficient is coding objects using Java?