How do I convert my nested for loop into a nested parfor loop?

I have a 36x34x8812 hourly current data called tpres3d. I used the matlab toolbox t_tide from Rich Pawlowicz toolbox to separate tidal currents. Since my data is a 3D array. I used a nested for loop that works but is taking too long to finish. I use the parallel computing toolbox and incorporate a parfor loop to speed up the process. My problem is I don't know how to properly write a parfor loop from a normal for loop. When I use this code:

tpres3d = tpres_U + tpres_V*sqrt(-1) %36x34x8812
[a,b,c] = size(tpres3d)
Ttides_hfdr = NaN*ones(a,b,c)

parfor i = 1:a
    for j = 1:b
        t0=squeeze(tpres3d(i,j,:));
        I=find(~isnan(t0));
        if ~isempty(I)
            try
                [~,~,~,tide_hfdr]=t_tide(t0, ...
                    'interval',1, ...
                    'start time',ttime(1), ...
                    'latitude',y, ...
                    'rayleigh',constituents, ...
                    'synthesis',0);
            catch
            end
        end
    end
    Ttides_hfdr(i,j,:)=tide_hfdr
end

It gives a warning:

Warning: The temporary variable 'tide_hfdr' will be cleared at the beginning of each iteration of the

parfor-loop. If 'tide_hfdr' is used before it is set, a runtime error will occur. For more information,

see Parallel for Loops in MATLAB, "Uninitialized Temporaries".

and an error:

Error: When indexing the sliced variable 'Ttides_hfdr' with the nested for-loop variable 'j', the sliced

variable must be inside a for-loop that defines the range of the for-loop variable. For more information,

see Parallel for Loops in MATLAB, "Nested for-Loops with Sliced Variables".

Trust me, I read the guides and tried it out but I can't seem to grasp how to fix my code. Please help. Thank you.

Answer 1

[Matt J]

One way:

tpres3d = tpres_U + tpres_V*sqrt(-1); %36x34x8812
[a,b,c] = size(tpres3d);

T0=reshape(tpres3D,[],c).';
Ttides_hfdr = nan(size(T0));
ttime1=ttime(1);

parfor k=1:width(T0)
     
        t0=T0(:,k);

        if any(~isnan(t0))
            try
                [~,~,~,tide_hfdr]=t_tide(t0, ...
                    'interval',1, ...
                    'start time',ttime1, ...
                    'latitude',y, ...
                    'rayleigh',constituents, ...
                    'synthesis',0);
            catch
            end
        end
  
    Ttides_hfdr(:,k)=tide_hfdr;
 end
 
 Ttides_hfdr=reshape(Ttides_hfdr.',[a,b,c]);

[Edcel]

Thank you for replying. It still gives the warning:

Warning: The temporary variable 'tide_hfdr' will be cleared at the beginning of each

iteration of the parfor-loop. If 'tide_hfdr' is used before it is set, a runtime error

will occur. For more information, see Parallel for Loops in MATLAB, "Uninitialized

Temporaries".

And a new error:

Error using script_name

The source code

(C:\...\script_name.m)

for the parfor-loop that is trying to execute on the worker could not be found.

Caused by:

Unrecognized function or variable 'tide_hfdr'.

Error using parallel.internal.parfor.cppRemoteParallelFunction

Worker unable to find file.

Unrecognized function or variable 'tide_hfdr'.

[Walter Roberson]

You do not set tide_hdfr in your catch block. If this were a serial loop then in this iteration if catch occurred then tide_hdfr would have been set by a previous iteration. But in parfor the previous iteration might or might not exist or might be for a completely unrelated iteration.

You need to set tide_hdfr in your catch block.

[Edcel]

Can you give me an example on how do I set it up in my catch block?

[Edcel]

Is this what you mean on setting it up inside the catch block? I got no errors but Ttides_hfdr gave me NaNs as result. I'm starting to think that parallel computing does not support t_tide at all.

tpres3d = tpres_U + tpres_V*sqrt(-1); %36x34x8812
[a,b,c] = size(tpres3d);

T0=reshape(tpres3d,[],c).';
Ttides_hfdr = nan(size(T0));
ttime1=ttime(1);

parfor k=1:width(T0)
    
    t0=T0(:,k);
    
    if any(~isnan(t0))
        try
            [~,~,~,tide_hfdr]=t_tide(t0, ...
                'interval',1, ...
                'start time',ttime1, ...
                'latitude',y, ...
                'rayleigh',constituents, ...
                'synthesis',0);
            Ttides_hfdr(:,k)=tide_hfdr;
        catch
        end
    end
    
end

Ttides_hfdr=reshape(Ttides_hfdr.',[a,b,c]);

[Matt J]

See if the t_tide() is ever getting executed, e.g., using,

tpres3d = tpres_U + tpres_V*sqrt(-1); %36x34x8812
[a,b,c] = size(tpres3d);

T0=reshape(tpres3d,[],c).';
Ttides_hfdr = nan(size(T0));
ttime1=ttime(1);

fails=0;

parfor k=1:width(T0)
    
    t0=T0(:,k);
    
    if any(~isnan(t0))
        try
            [~,~,~,tide_hfdr]=t_tide(t0, ...
                'interval',1, ...
                'start time',ttime1, ...
                'latitude',y, ...
                'rayleigh',constituents, ...
                'synthesis',0);
            Ttides_hfdr(:,k)=tide_hfdr;
        catch
            fails=fails+1;
        end
    else
        fails=fails+1;
    end
    
end
Ttides_hfdr=reshape(Ttides_hfdr.',[a,b,c]);

disp("Failures occurred "+fails/a/b*100+" percent")

[Edcel]

Failures occurred 26.7974 percent

[Matt J]

If so, you shouldn't be getting Ttides_hfdr=nan all the time. That would seem to imply that t_tide() is always returning NaNs.

[Edcel]

I agree. Maybe tide_hfdr gets deleted right away each iteration as Ttides_hfdr seems to fail to capture any result at all. I just ran a normal for loop, and it's still ongoing after 24 hours!

Thanks for trying to help, Matt. I appreciate it.

[Matt J]

Just run your original loop until a combination of i,j giving a non-trivial tide_hfdr is found. Then it will be easier -- we just need to compare and troubleshoot the two versions for that particular (i,j),

for i = 1:a
    for j = 1:b
        
        t0=squeeze(tpres3d(i,j,:));

        if ~all(isnan(t0))
            try
                [~,~,~,tide_hfdr]=t_tide(t0, ...
                    'interval',1, ...
                    'start time',ttime(1), ...
                    'latitude',y, ...
                    'rayleigh',constituents, ...
                    'synthesis',0);
                
                if any(isfinite(tide_hfdr)); 
                    disp 'Located'
                    i,j
                    break; 
                end
            catch
            end
        end
    end

end

[Edcel]

I just ran the script you provided. I should store each (i, j) right?

[Matt J]

There should only be one (i,j). You should also store the t0 and tide_hfdr that correspond to it.