Request for Replace Custom Function with Optimized Lookup Table example

Dear All,

I'm trying to implement " Replace Custom Function with Optimized Lookup Table " example.

But I came across the below error message every time.

============= Step3: Generate Fixed Point Code ==============

### Generating Fixed Point MATLAB Code mlhdlc_replacement_exp_fixpt using Proposed Types
### Generating Fixed Point MATLAB Design Wrapper mlhdlc_replacement_exp_wrapper_fixpt
### Generating Mex file for ' mlhdlc_replacement_exp_wrapper_fixpt '
??? Function 'log' is not defined for values of class 'embedded.fi'.

Error in ==> mlhdlc_replacement_exp_fixpt Line: 15 Column: 8
Code generation failed: View Error Report
Build failed.
Use help codegen for more information on using this command.
Function name 'mlhdlc_replacement_exp' specified more than once.
Use help codegen for more information on using this command.
Error using codegen

Actually, I want to make a custom hdl code by using "log" function.

Would you please help me how to to that example ?

Or Would you please give me example code?

I used with the script

clear
clc

clear design_name testbench_name fxpCfg hdlcfg interp_degree

    design_name = 'mlhdlc_replacement_exp';
    testbench_name = 'mlhdlc_replacement_exp_tb';
    interp_degree = 0;

fixed point converter config

    fxpCfg = coder.config('fixpt');
    fxpCfg.TestBenchName = 'mlhdlc_replacement_exp_tb';
    fxpCfg.TestNumerics = true;
    %    specify this - for optimized HDL
    fxpCfg.DefaultWordLength = 40;
    fxpCfg.ProposeWordLengthsForDefaultFractionLength = true;
    fxpCfg.DefaultFractionLength = 30;
             
  %  %%   exp - replacement config
  %  mathFcnGenCfg = coder.approximation('log');
  %  %    generally use to increase accuracy; specify this as power of 2 for optimized HDL
  %  mathFcnGenCfg.NumberOfPoints = 1024;
  %  mathFcnGenCfg.InterpolationDegree = interp_degree; %         can be 0,1,2, or 3
  %  fxpCfg.addApproximation(mathFcnGenCfg);

    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
  %   %saturateExp = @(x) 1/(1+exp(-x));   
  %   saturateExp = @(y) log(u)*0.4342944819;
  %   mathFcnGenCfg = coder.approximation('Function','saturateExp',...
  %   'CandidateFunction', saturateExp,...
  %   'NumberOfPoints',50,'InputRange',[0,10]); 
  %   fxpCfg.addApproximation(mathFcnGenCfg);

saturateExp = @(x) 1/(1+exp(-x));   
custAppx = coder.approximation('Function','saturateExp',...
'CandidateFunction', saturateExp,...
'NumberOfPoints',50,'InputRange',[0,10]); 

   fxpCfg.addApproximation(custAppx);

    %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

HDL config object

    hdlcfg = coder.config('hdl');
    hdlcfg.TargetLanguage = 'Verilog';
    hdlcfg.DesignFunctionName = design_name;
    hdlcfg.TestBenchName = testbench_name;
    hdlcfg.GenerateHDLTestBench=true;

    hdlcfg.ConstantMultiplierOptimization = 'auto'; %optimize out any multipliers from interpolation
    hdlcfg.PipelineVariables = 'y u idx_bot x x_idx';%
    hdlcfg.InputPipeline = 2;
    hdlcfg.OutputPipeline = 2;
    hdlcfg.RegisterInputs = true;
    hdlcfg.RegisterOutputs = true;

    codegen('-float2fixed',fxpCfg,'-config',hdlcfg,'mlhdlc_replacement_exp')

Answer 1

[Walter Roberson]

See https://www.mathworks.com/help/fixedpoint/ug/what-to-do-when-a-function-is-not-supported-for-fixed-point-conversion.html

[satish guruma]

Thanks. But, I came across the below error when I follow your recommends.

Would you please help me what am I supposed to do to resolve this problem?

In fi(var1,var2,...varN) if var1 is not a constant then var2 to varN must be or specify a complete numerictype.

[Walter Roberson]

I meant more to follow up to the parts about writing replacement functions. You probably do not have a double precision core available for your FPGA, so you are going to have to provide your own log function.