Generating Modular Code for Functions
This example shows how to generate modular HDL code from MATLAB® code containing functions.
Introduction
HDL Coder™ by default inlines the body of all MATLAB functions called in the top level design and generates HDL code in a single file. However turning on the advanced coding option 'Generate instantiable code for functions' helps you generate module code in multiple files.
MATLAB Design
The MATLAB design used in the example is an implementation of an LMS filter. The LMS filter is used to identify an FIR filter signal embedded in the noise.
design_name = 'mlhdlc_lms_fcn'; testbench_name = 'mlhdlc_lms_fir_id_tb';
Let us take a look at the MATLAB design,
dbtype(design_name);
1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2 % MATLAB Design: Adaptive Noise Canceler algorithm using Least Mean Square
3 % (LMS) filter implemented in MATLAB
4 %
5 % Key Design pattern covered in this example:
6 % (1) Use of function calls
7 % (2) Function inlining vs instantiation knobs available in the coder
8 % (3) Use of system objects in the testbench to stream test vectors into the design
9 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
10
11 %#codegen
12
13 function [filtered_signal, y, fc] = mlhdlc_lms_fcn(input, desired, step_size, reset_weights)
14 %
15 % 'input' : The signal from Exterior Mic which records the ambient noise.
16 % 'desired': The signal from Pilot's Mic which includes
17 % original music signal and the noise signal
18 % 'err_sig': The difference between the 'desired' and the filtered 'input'
19 % It represents the estimated music signal (output of this block)
20 %
21 % The LMS filter is trying to retrieve the original music signal ('err_sig')
22 % from Pilot's Mic by filtering the Exterior Mic's signal and using it to
23 % cancel the noise in Pilot's Mic. The coefficients/weights of the filter
24 % are updated(adapted) in real-time based on 'input' and 'err_sig'.
25
26 % register filter coefficients
27 persistent filter_coeff;
28 if isempty(filter_coeff)
29 filter_coeff = zeros(1, 40);
30 end
31
32 % Variable Filter
33 % call 'tapped_delay_fcn' function on path to create 40-step tapped delay
34 delayed_signal = mtapped_delay_fcn(input);
35
36 % apply filter coefficients
37 weight_applied = delayed_signal .* filter_coeff;
38
39 % call 'treesum' function on matlab path to sum up the results
40 filtered_signal = mtreesum_fcn(weight_applied);
41
42 % Output estimated Original Signal
43 td = desired;
44 tf = filtered_signal;
45
46 esig = td - tf;
47 y = esig;
48
49 % Update Weights
50 % call 'update_weight_fcn' function on matlab path to
51 % calculate the new weights
52 updated_weight = update_weight_fcn(step_size, esig, delayed_signal, ...
53 filter_coeff, reset_weights);
54
55 % update filter coefficients register
56 filter_coeff = updated_weight;
57
58 fc = filter_coeff;
59
60
61 function y = mtreesum_fcn(u)
62 %Implement the 'sum' function without a for-loop
63 % y = sum(u);
64
65 % The loop based implementation of 'sum' function is not ideal for
66 % HDL generation and results in a longer critical path.
67 % A tree is more efficient as it results in
68 % delay of log2(N) instead of a delay of N delay
69
70 % This implementation shows how to explicitly implement the vector sum in
71 % a tree shape to enable hardware optimizations.
72
73 % The ideal way to code this generically for any length of 'u' is to use
74 % recursion but it is not currently supported by MATLAB Coder
75
76
77 % NOTE: To instruct MATLAB Coder to compile an external function,
78 % add the following compilation directive or pragma to the function code
79 %#codegen
80
81 % This implementation is hardwired for a 40tap filter.
82
83 level1 = vsum(u);
84 level2 = vsum(level1);
85 level3 = vsum(level2);
86 level4 = vsum(level3);
87 level5 = vsum(level4);
88 level6 = vsum(level5);
89 y = level6;
90
91
92
93 function output = vsum(input)
94
95 coder.inline('always');
96
97 vt = input(1:2:end);
98
99 for i = int32(1:numel(input)/2)
100 k = int32(i*2);
101 vt(i) = vt(i) + input(k);
102 end
103
104 output = vt;
105
106 function tap_delay = mtapped_delay_fcn(input)
107 % The Tapped Delay function delays its input by the specified number
108 % of sample periods, and outputs all the delayed versions in a vector
109 % form. The output includes current input
110
111 % NOTE: To instruct MATLAB Coder to compile an external function,
112 % add the following compilation directive or pragma to the function code
113 %#codegen
114
115 persistent u_d;
116 if isempty(u_d)
117 u_d = zeros(1,40);
118 end
119
120
121 u_d = [u_d(2:40), input];
122
123 tap_delay = u_d;
124
125
126 function weights = update_weight_fcn(step_size, err_sig, delayed_signal, filter_coeff, reset_weights)
127 % This function updates the adaptive filter weights based on LMS algorithm
128
129 % Copyright 2007-2015 The MathWorks, Inc.
130
131 % NOTE: To instruct MATLAB Coder to compile an external function,
132 % add the following compilation directive or pragma to the function code
133 %#codegen
134
135 step_sig = step_size .* err_sig;
136 correction_factor = delayed_signal .* step_sig;
137 updated_weight = correction_factor + filter_coeff;
138
139 if reset_weights
140 weights = zeros(1,40);
141 else
142 weights = updated_weight;
143 end
MATLAB test bench,
Let us take a look at the MATLAB test bench,
dbtype(testbench_name);
1
2 % Copyright 2011-2015 The MathWorks, Inc.
3
4 clear ('mlhdlc_lms_fcn');
5 % returns an adaptive FIR filter System object,
6 % HLMS, that computes the filtered output, filter error and the filter
7 % weights for a given input and desired signal using the Least Mean
8 % Squares (LMS) algorithm.
9
10 stepSize = 0.01;
11 reset_weights =false;
12
13 hfilt = dsp.FIRFilter; % System to be identified
14 hfilt.Numerator = fir1(10, .25);
15
16 rng('default'); % always default to known state
17 x = randn(1000,1); % input signal
18 d = step(hfilt, x) + 0.01*randn(1000,1); % desired signal
19
20 hSrc = dsp.SignalSource(x);
21 hDesiredSrc = dsp.SignalSource(d);
22
23 hOut = dsp.SignalSink;
24 hErr = dsp.SignalSink;
25 %%%%%%%%%%%%%%%%%%%%%%%%%%%%
26 %Call to the design
27 %%%%%%%%%%%%%%%%%%%%%%%%%%%%
28 while (~isDone(hSrc))
29 [y, e, w] = mlhdlc_lms_fcn(step(hSrc), step(hDesiredSrc), stepSize, reset_weights);
30 step(hOut, y);
31 step(hErr, e);
32 end
33
34 figure('Name', [mfilename, '_plot']);
35 subplot(2,1,1), plot(1:1000, [d,hOut.Buffer,hErr.Buffer]);
36 title('System Identification of an FIR filter');
37 legend('Desired', 'Output', 'Error');
38 xlabel('time index'); ylabel('signal value');
39 subplot(2,1,2); stem([hfilt.Numerator.', w(end-10:end).']);
40 legend('Actual','Estimated');
41 xlabel('coefficient #'); ylabel('coefficient value');
Create a New Folder and Copy Relevant Files
Execute the following lines of code to copy the necessary example files into a temporary folder.
mlhdlc_demo_dir = fullfile(matlabroot, 'toolbox', 'hdlcoder', 'hdlcoderdemos', 'matlabhdlcoderdemos'); mlhdlc_temp_dir = [tempdir 'mlhdlc_fcn_partition']; % create a temporary folder and copy the MATLAB files. cd(tempdir); [~, ~, ~] = rmdir(mlhdlc_temp_dir, 's'); mkdir(mlhdlc_temp_dir); cd(mlhdlc_temp_dir); copyfile(fullfile(mlhdlc_demo_dir, [design_name,'.m*']), mlhdlc_temp_dir); copyfile(fullfile(mlhdlc_demo_dir, [testbench_name,'.m*']), mlhdlc_temp_dir); % At the end of this step you should see the design files being copied to % a temporary folder.
Simulate the Design
It is always a good practice to simulate the design with the testbench prior to code generation to make sure there are no runtime errors.
mlhdlc_lms_fir_id_tb
Create a New HDL Coder™ Project
coder -hdlcoder -new mlhdlc_fcn_partition
Next, add the file 'mlhdlc_lms_fcn.m' to the project as the MATLAB Function and 'mlhdlc_lms_fir_id_tb.m' as the MATLAB Test Bench.
You can refer to Getting Started with MATLAB to HDL Workflow tutorial for a more complete tutorial on creating and populating MATLAB HDL Coder projects.
Run Fixed-Point Conversion and HDL Code Generation
Launch the Workflow Advisor from the Build tab and right click on the 'Code Generation' step and choose the option 'Run to selected task' to run all the steps from the beginning through the HDL code generation.
Examine the generated HDL code by clicking on the hyperlinks in the Code Generation Log window. Notice that the HDL code for all the functions in the design is inlined into a single file 'mlhdlc_lms_fcn_FixPt.vhd'
Generate Instantiable Code
To generate modular code for all the functions in separate files please turn on this advanced coding option 'Generate instantiable code for functions'
Rerun Code Generation Step
Now reset the code generation task and rerun the HDL code generation step to see how code is generated in multiple files for the functions
Control Inlining Per Function
Sometimes it is preferable to inline code for helpers and utilities and instantiate them. To locally control inlining of such functions use the 'coder.inline' pragma in the MATLAB code.
coder.inline('always') coder.inline('never') coder.inline('default')
Type the following command to see how to use coder.inline pragma
help coder.inline
Known Limitations
The following limitations exist when instantiating HDL code from functions.
Function calls inside conditional expressions and for loops get inlined and do not get instantiated.
Functions with states get inlined.
Clean up the Generated Files
You can run the following commands to clean up the temporary project folder.
mlhdlc_demo_dir = fullfile(matlabroot, 'toolbox', 'hdlcoder', 'hdlcoderdemos', 'matlabhdlcoderdemos'); mlhdlc_temp_dir = [tempdir 'mlhdlc_fcn_partition']; clear mex; cd (mlhdlc_demo_dir); rmdir(mlhdlc_temp_dir, 's');