Data store memory(DSM)

Hello

I am using Data store memory block in my simulink model to initialize a variable(array) to use in a matlab function block.Its a big model with tight tolerances. When I run I see that the DSM block looks into the full array(variable is size 50) at every time step so basically the simulation is EXTREMELY SLOW. How can I get rid of this.

Thank you

James

Answer 1

[Nirmal Gunaseelan]

Hi James,

Looking through all the warnings, it appears from the diagnostics that the MATLAB Fcn block is running at minor times steps (the model has continuous states or you have chosen a continuous state solver) and obviously the block is storing values at all time steps (including minor time steps). I don't think the DSM block has anything to do with this apart from providing the diagnostics. These write after write diagnostics are to make sure that this is what you intended to do and are not an error by themselves. Currently, the way you have modeled is making the block write to elements at all minor time steps. Could you change the solver type (say to discrete if your model doesnt have continuous states) and see if minor steps go away? This will also have a significant improvement in performance, which was your original issue.

-Nirmal

[James]

My model has continuous states, hence the solver. But is there a possibility to run only the MATLAB Fcn block with larger time steps. In the mean time, I will try changing the solver type

-James

[James]

I tried using a sample time of 0.1 and yes it look into all 50 elements at 0.1,0.2,0.3...etc

As i mentioned, I have a counter to fill in this array, so if my counter is at 10, why doesn't it try to read only 10th element instead of looking into all elements?

-James

[Nirmal Gunaseelan]

I'm going to try out a few things, but in the mean time could you also provide a snippet of your access code in the MATLAB Fcn block? I just want to verify that element wise access is actually happening.

[James]

Oh, the function is a very simple one

function x_LP = dist(u, j)

%#codegen

global distance;

if j == 50

distance=shiftu(distance,1,1);

end

distance(j) = u;

x_LP=distance;

I have 2 functions(exactly same). 'j' is the counter and 'distance' is the variable in DSM. So after 50, I shift the whole array by one row and add a new one.

Thanks a lot for the help!

Answer 2

[Nirmal Gunaseelan]

Hi James,

Could you give more details on what blocks are being used with the DSM block? Are you using a Data Store Write block as well?

Is the problem that the DSM block is accessing at every time step, or is it that you want only certain parts to be accessed? Have you tried Element wise assignment/selection?

-Nirmal

[James]

Basically I am using a dynamic look up table. So I have assigned an array with 100 elements in the DSM block. Initially I used only the DSM block but then I tried it with the write and read blocks too. So I have two arrays(initialized globally using DSM blocks). These arrays are used in MATLAB function blocks. The values in the array are being updated while the simulation runs.

So as you said, I get the warning that DSM block is being accessed every time step. Basically I want to use the elements in the array sequentially regardless the timestep.

I hope I have made myself more clear!

James

[Nirmal Gunaseelan]

Could you also post the error message here? I'm wondering if the message you're getting is about the DSM blocks not being in sync (read before write) or actually about DSM initializing many times.

As I understand, you are using DSM to initialize a chunk of memory and after that, that array is used in a MATLAB Fcn block. And you are suspecting that the DSM block is accessing the memory again (initializing again?) or the Data Store Read/Write block is accessing at every time step? (this seems normal to me).

[James]

Ok, I feel I am failing to convey my real situation. the set of warnings are similar to this:

The block 'tractor_semitrailer_control_DSM/semitrailer/MATLAB Function4' is writing to

element 1 of the data store 'tractor_semitrailer_control_DSM/semitrailer/Data Store Memory2' but

the block 'tractor_semitrailer_control_DSM/semitrailer/MATLAB Function4' has already written to

this memory in a minor time step at time 0.017329639746273648

Warning: The block 'tractor_semitrailer_control_DSM/semitrailer/MATLAB Function4' is reading

from element 2 of the data store 'tractor_semitrailer_control_DSM/semitrailer/Data Store

Memory2' before any blocks have written to this memory in a minor time step at time

0.017329639746273648

Warning: The block 'tractor_semitrailer_control_DSM/semitrailer/MATLAB Function4' is reading

from element 3 of the data store 'tractor_semitrailer_control_DSM/semitrailer/Data Store

Memory2' before any blocks have written to this memory in a minor time step at time

0.017329639746273648

So what I understand is that DSM is looking into every cell of the array(size 50) in each timestep(1e-4). I want the variable stored in the DSM to behave like a normal array so that i can use it in the matlab funtion block along with a counter to add new elements to consecutive cells(while the simulation is running) and simultaneously feed this array to a dynamic lookup table.

[James]

this is without using the read/write blocks!

Answer 3

[James]

Hi Nirmal,

The simulation runs now and, I get good results. But since I increased my sample time in the MATLAB Fcn block, it a bit faster. But I still have the warnings where DSM looks into every cell.my simulation time is 60 seconds. My simulation runs for 30 min, unlike my previous attempts where it took 45 min. I believe if DSM looks into ony one particular cell, it might be faster. Do you suggest any other way to make this process faster!?

-James

[Nirmal Gunaseelan]

Hi James,

I wanted to try out other ways of achieving your functionality without using shiftu/if conditions. I am not sure if we are preventing full access by using those lines. I'll get back to you.

-Nirmal

[James]

Oh, okay! I was also thinking that shiftu/if conditions are causing this.

Thanks for your help and let me know if you end up with some effective results!

-James