Using emlc to generate C code - function within a function problem

Hello fellow Matlab users,

Ive been trying to convert a massive code from matlab to c and implement it for real time signal processing on a DSP processor.

So far ive been succesfull and emlc is an amazing function! The problem i have reached now is that I need to compile a code that has the function "butter" inside (gives out butterworth filter coefficients.)

The problem is that the compilation report complains about butter`s input arguments are not constant.

To make things a bit more clear:

C_generating_script.m => Function_1.m => butter.m

Now my question is:

How can I define/declare the inputs of the butter function as constants which is within another function and allow it to compile?

Answer 1

[Kaustubha Govind]

Have you tried specifying constant inputs using emlcoder.egc: http://www.mathworks.com/help/toolbox/eml/ug/bq2wkmb-47.html#brl11_g-1

[Andreas Prokopiou]

yes but im not quite sure where i need to make this declaration.

The function that complains about constant inputs (butter.m) is within another function (ParamGen.m) and ParamGen.m is the one I want to compile to C

[Kaustubha Govind]

Could you give a simple example of what you are doing? I tried the following, and it compiled without using emlcoder.egc:

-------- mytest.m ----------

function [b, a] = mytest

%#codegen

n = 4; Wn = 0.5;

a = zeros(1,5);

b = zeros(1,5);

[b a] = subfun(n,Wn);

-------- subfun.m ----------

function [b a] = subfun(n, Wn)

%#codegen

[b a] = butter(n, Wn);

Answer 2

[Andreas Prokopiou]

I have changed a bit the organization of the code but I still get the same problem. Here`s the section of the code that doesn`t get converted to C:

The function call is:

parameters = MAP_ParamGen(dt,fasterdt,CF_of_channel,audiotimeofblock);

takes in as input 4 double floats and gives out a struct with around 150 entries

This is the c-generating code im using:

emlc -eg {dt,fasterdt,CF_of_channel,audiotimeofblock }  MAP_ParamGen.m  -c -T RTW -report

the error comes from:

function [PARAMS] = MAP_ParamGen(dt,fasterdt,CF_of_channel,audiotimeofblock) %#eml  
.
.
.
cutOffFrequency = 100;
order = 1;
sampleRate=1/dt;
Nyquist=sampleRate/2;

[PARAMS.OME_b_low,PARAMS.OME_a_low] = butter(order, cutOffFrequency/Nyquist, 'low');
.
.
.

[Kaustubha Govind]

You should use:

emlc -eg {emlcoder.egc(dt),fasterdt,CF_of_channel,audiotimeofblock } MAP_ParamGen.m -c -T RTW -report

To specify that 'dt' as a constant, since this determines one of the inputs to butter.

Answer 3

[Andreas Prokopiou]

It works! Thank you! I finally understand how the constant declaration works.

I have another problem now though:

.
.
.
OMEfilters =  [0  900     1;2   800  2;0  900     3;2   800  4];

sampleRate=1/dt;
Nyquist=sampleRate/2;

[nFilters nValues]=size(OMEfilters);

for i=1:nFilters
    gaindB(i)=OMEfilters(i,1);
    lowCutOff=OMEfilters(i,2);
    filterOrder(i)=OMEfilters(i,3);
    if nValues>3
        highCutOff=filters(i,4);
    else
        highCutOff=sampleRate/3;
    end

      [b_high,a_high] = butter(filterOrder(i), [lowCutOff highCutOff] /Nyquist);
      for j = 1:filterOrder(i)*2+1
          OME_b_high(i,j) = b_high(j);
          OME_a_high(i,j) = a_high(j);
      end
  end
  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

All of the sizes of arrays are well defined and there is no problem there. Again it complains about constant inputs. Is it because the input value to the function changes per for loop iteration? if yes,Is there a way to resolve this without unravelling the for loop?

[Kaustubha Govind]

Did you try by specifying all relevant inputs: dt and OMEfilters as constant inputs using emlcoder.egc?

[Andreas Prokopiou]

Yes, In fact OMEfilters is contained within the MAP_ParamGen.m file which im compiling so emlc should be able to "see" that that parameter will not change.

As far as I can tell is exactly the same situation as before with the only difference the for loop and the if statement.

I tried fixing the highCutOff so it doesnt go through the if statement thus not allowing it two execution paths but there is still an error.

The error seems to be because the inputs to the 'butter' function change at every for loop iteration even though they are defined within the same function as a constant value. Is this something that is not possible to convert to C?

Is there a way to define within the compiled function that a certain parameter might change value but not type?

Answer 4

[Fred Smith]

Try using eml.unroll on the for-loop. That will cause the loop to be unrolled in the generated code.

This may or may not be acceptable depending on how many iterations the loop has. If it is not acceptable then you need a butter function that supports non-constant inputs. That capability is not available today.

If that is what you need, could you explain your use case and we'll create an enhancement request to add this capability to the product.