long execution time for simple lines like end and vector value addition

I wanted to make my code more time efficient so looked in the profiler and saw that the most time consuming lines are 'end' and another quite simple line. it is a long loop but other lines in it runs much faster. Does anyone know why they take so long or have an idea how ti improve this loop?

Thanks!

[Jan]

Under Matlab R2016b/Win10_64/i7 I get

sec   calls
0.08  2399800   Efoot(n)=Efoot(n-1)+dt*(qin(n-1)+qm_foot-qsk(n-1)); 
0.08  2399800   Esk(n)=Esk(n-1)+dt*(qsk(n-1)+qm_sk-qout(n-1));

In your example the input data are 4 times bigger, but ths code needs 240 times longer. This is surprising. The main work is done in findpeaks.

Which Matlab version are you using?

Answer 1

[Walter Roberson]

>> 19.30/10799100
ans =
      1.78718596920114e-06

The iterations are not taking long; you are just doing a lot of them.

You are doing 2 additions, 4 subtractions, and 1 multiplication on the line that is taking the most time.

You could potentially reduce the time by assigning n-1 to a variable, but I would not be surprised if the Just In Time compiler is effectively already doing that internally.

[Oded Scharf]

and why is the 'end' taking relativly long?

[Walter Roberson]

When MATLAB detects a code sequence that can be handled by the high performance libraries it calls into them. The code pattern might cross multiple lines so the required time cannot always be allocated to one line. MATLAB allocates it to the end statement instead.

Answer 2

[Jan]

A very important part of the code is missing: Is Efoot and Esk pre-allocated properly? Otherwise the arrays grow iteratively, which is very expensive.

Analysing the run time behavior of a code, which is shown as screenshot and partially only, is a very bold task. I cannot guess e.g., if "qin" is an array or a function. Please post the code as text, such that it can be used by copy&paste, and add some meaningful test data - either by attaching them as MAT file or by some rand() commands with relevant sizes.

[Oded Scharf]

I posted the code. thank you for your help.

Answer 3

[Oded Scharf]

Hey, All of the vectors are pre-allocated and there are no inner functions in this function. Im attaching the code and the relavent parameters (one as an array and the other as a script to be run before the function) Thank you for your help! code:

function:

    function [ Tsk_f, Tfoot_f,ssTsk,ssTfoot,skinrisetime, footrisetime,skp2p ] =...
        newTempcalc(dt, Qfoot,roblood,Cblood,Cfoot,qm_foot,Rsk,Csk,qm_sk,Rh)

    cycles=200;
    %%setting start conditions

    %cycle vectors

    RQ=(roblood*Cblood*Qfoot).^-1;
    Rout=ones(size(Qfoot))*Rh;
    Tfoot=zeros(size(Qfoot));
    Tsk=zeros(size(Qfoot));
    Ta=ones(size(Qfoot))*37;

    Troom=ones(size(Qfoot))*24;
    Efoot=ones(size(Qfoot))*Tfoot(1)*Cfoot;
    Esk=ones(size(Qfoot))*Tsk(1)*Csk;
    qin=zeros(size(Qfoot));
    qsk=zeros(size(Qfoot));
    qout=zeros(size(Qfoot));

    %full vectors

    Efoot_f=zeros(1,cycles*length(Qfoot));
    Esk_f=zeros(1,cycles*length(Qfoot));
    Tfoot_f=zeros(1,cycles*length(Qfoot));
    Tsk_f=zeros(1,cycles*length(Qfoot));
    qin_f=zeros(1,cycles*length(Qfoot));
    qsk_f=zeros(1,cycles*length(Qfoot));
    qout_f=zeros(1,cycles*length(Qfoot));

    %%iterations

    for cyc=1:cycles %loop for every cycle

        for n=2:length(Qfoot) % lop for every time step
            %updating enrgy stored in capacitors
            Efoot(n)=Efoot(n-1)+dt*(qin(n-1)+qm_foot-qsk(n-1));
            Esk(n)=Esk(n-1)+dt*(qsk(n-1)+qm_sk-qout(n-1));

            %updating tissues temp
            Tfoot(n)=Efoot(n)/Cfoot;
            Tsk(n)=Esk(n)/Csk;

            %updating heat transfer through resistors
            qin(n)=(Ta(n)-Tfoot(n))/RQ(n);
            qsk(n)=(Tfoot(n)-Tsk(n))/Rsk;
            qout(n)=(Tsk(n)-Troom(n))/Rout(n);
        end
        % cycling variables    
        start=(cyc-1)*length(Qfoot)+1;
        stop=cyc*length(Qfoot);
        Efoot_f(start:stop)=Efoot;
        Esk_f(start:stop)=Esk;
        Tfoot_f(start:stop)=Tfoot;
        Tsk_f(start:stop)=Tsk;
        qin_f(start:stop)=qin;
        qsk_f(start:stop)=qsk;
        qout_f(start:stop)=qout;

        Efoot(1)=Efoot(length(Qfoot));
        Esk(1)=Esk(length(Qfoot));
        Tfoot(1)=Tfoot(length(Qfoot));
        Tsk(1)=Tsk(length(Qfoot));
        qin(1)=qin(length(Qfoot));
        qsk(1)=qsk(length(Qfoot));
        qout(1)=qout(length(Qfoot));
    end
    ssTsk=mean(Tsk_f(length(Tsk_f)-10000:end));
    ssTfoot=mean(Tfoot_f(length(Tfoot_f)-10000:end));
    [~,skinrisetime] =min(abs(Tsk_f-(0.63*(ssTsk-Tsk_f(1))+Tsk_f(1))));
    [~,footrisetime] =min(abs(Tfoot_f-(0.63*(ssTfoot-Tfoot_f(1))+Tfoot_f(1))));
    skinrisetime=skinrisetime*dt;
    footrisetime=footrisetime*dt;
    %risetimeval=Tsk_f(risetime);
    [~, maxindex]=findpeaks(Tsk_f);
    [~, minindex]=findpeaks(-Tsk_f);
    if isempty(maxindex) || isempty(minindex)
        skp2p=0;
    else
        skp2p=Tsk_f(maxindex(end))-Tsk_f(minindex(end));
    end
    end

script:

params
[ Tsk_f, Tfoot_f,ssTsk,ssTfoot,skinrisetime, footrisetime,skp2p ] =...
 newTempcalc(dt, Qfoot,roblood,Cblood,Cfoot,qm_foot,Rsk,Csk,qm_sk,Rh)

[Jan]

The run-time of the "end" is not meaningful. It might be an effect of tzhe JIT acceleration (although this is disables as far as I know during profiling?!). To find out more, e.g. if the code can be accelerated, ir would be required to run it, but the posted version does not run:

 Undefined function or variable 'Qfoot'.
 Error in asd (line 3)
 [ Tsk_f, Tfoot_f,ssTsk,ssTfoot,skinrisetime, footrisetime,skp2p ] =...

A proper code formatting would be useful also: Mark the code and press the "{} Code" button.

[Oded Scharf]

I uploaded the array Qfoot seperatly. I didn't understand what is the JIT acceleration.

thank you.

[Jan]

For the JIT see e.g. https://blogs.mathworks.com/loren/2016/02/12/run-code-faster-with-the-new-matlab-execution-engine/ .