Serial fread is too slow

I'm using MATLAB R2017a & Linux.

    version -java
    'Java 1.7.0_60-b19 with Oracle Corporation Java HotSpot(TM) 64-Bit Server VM mixed mode'

The program below, 75% of time spends onto fread(). How can it be speed-up?

        s = serial(serial_port,'BaudRate',baud_rate,'DataBits',8,'InputBufferSize',buffer_size); 
        fopen(s);   

        size_to_read=s.BytesAvailable;
        if (size_to_read>10)
            out = fread(s,size_to_read,'uint16')
            // some code //
        end

Probably direct access for Java will be faster, but I dont know how.

Another solution is use this MEX https://nl.mathworks.com/matlabcentral/fileexchange/62545-mex-c-serial-interface This output is 8bit digits, but I need 16 bit, can it be changed?

[Rik]

Are you sure the bottleneck is in Matlab and not in the thing the port is connecting to?

[Jan]

As soon as "// some code //" takes more computing time, the percentage used in fread will decrease. So what is your actual problem? What about reading more than 10 UINT16?

[Nikolay Konstantinov]

[UPD] Serial speed is 4Mbit/s, I've got around 400Kb/s. If I save the flow to txt and then read it from txt, not from serial port, the program will be finished much more faster. Also, it can be seen from profiler, that fread() takes many time.

[Walter Roberson]

You should use a Bytesavailablefnc callback to avoid polling the available size.

[Nikolay Konstantinov]

I've tried, but it becomes more slower.. no idea why. I will try again. Actually, polling takes only 14%, 70% used by fread() (com.mathworks.toolbox.instrument.SerialComm)

[Nikolay Konstantinov]

Probably I understand why it was slow, since I use nested function. But event function cant be nested, right? I need, that function can changed variables and than use it in next iteration, like this:

          function test_var = READY_TO_READ(obj,event,test_var)
              // some manipulation with test_var

but not working..

Only global var are working.

[UPD] it can be nested.

[Walter Roberson]

There is no problem using nested functions for callbacks or using shared objects with serial ports, and doing so is often a good structuring mechanism. Shared variables are one of the faster forms of variable access.

I notice that you are using // for comments. Those are not supported for MATLAB, which suggests you are possibly using Octave.

[Tillmann Spellauge]

I have a similar problem.

I am using Matlab R2018b and linux. My program is spending 90+% of its runtime in "com.mathworks.toolbox.instrument.SerialComm". More specifically in the igetfield() function used in line 150 of the fscanf() function. It also makes no difference using fread(), fgets() or fgetl().

Did you eventually find a solution?