I'd like to find some dynamic lib files(.so format) required to run .mexa64 format file.

I'm trying to inferface matlab with Cadence Spectre (a type of Spice used for electronic ciricuit simulation). There is a SPECTRE/RF Matlab Toolbox that enables interfacing between MATLAB and Spectre using a bundle of function.

So I copied and pasted that toolbox file into the working directory where my main.m file is located, and ran 'main.m' after adding the path of the toolbox file, got the following error.

as far I understood, it said cds_srr called cds_innersrr and cds_innersrr.mexa64 called the dynamic lib named 'libsrrjni.so' which turned out to be non-existing. so I typed the 'ldd: ./cds_innersrr.mex64' to find out the the list of dynamic library files(.so format) required to run cds_innersrr.mexa64, and the result was the following

it reported that some .so format lib file were not found, which I assumed to be the cause of the original error.

my question is, where can I get those lib files? should I contact the provider of 'SPECTRE/RF Matlab Toolbox' from Cadence?

I left the main.m below in case it is necessary to solve this issue. it requires additional code and Spectre simulator, and SPECTRE/RF Matlab Toolbot in order to be executed. I ran Matlab under Linux environment.

% Matlab script for technology characterization
% Boris Murmann, Stanford University
% Tested with MMSIM12.11.134
% September 12, 2017

clearvars; 
close all;

% Load configuration
%c = techsweep_config_psp_65_spectre;
c = techsweep_config_bsim4_28_spectre;

% Write sweep info
nch.INFO   = c.modelinfo;
nch.CORNER = c.corner;
nch.TEMP   = c.temp;
nch.NFING  = c.NFING;
nch.L      = c.LENGTH';
nch.W      = c.WIDTH;
nch.VGS    = c.VGS';
nch.VDS    = c.VDS';
nch.VSB    = c.VSB';
%
pch.INFO   = c.modelinfo;
pch.CORNER = c.corner;
pch.TEMP   = c.temp;
pch.NFING  = c.NFING;
pch.L      = c.LENGTH';
pch.W      = c.WIDTH;
pch.VGS    = c.VGS';
pch.VDS    = c.VDS';
pch.VSB    = c.VSB';

% Simulation loop
for i = 1:length(c.LENGTH)
    str=sprintf('L = %2.3f', c.LENGTH(i));
    disp(str);
    for j = 1:length(c.VSB)
        % Write simulation parameters
        fid=fopen('techsweep_params.scs', 'w');
        fprintf(fid,'parameters length = %d\n', c.LENGTH(i));
        fprintf(fid,'parameters sb = %d\n', c.VSB(j));
        fclose(fid);
        
        pause(5)
        
        % Run simulator
        [status,result] = system(c.simcmd);
        if(status)
            disp('Simulation did not run properly. Check techsweep.out.')
            return;
        end    
        
        % Initialize data blocks
        for m = 1:length(c.outvars)
            nch.(c.outvars{m})(i,:,:,j) = zeros(length(c.VGS), length(c.VDS));
            pch.(c.outvars{m})(i,:,:,j) = zeros(length(c.VGS), length(c.VDS));
        end
        
        % Read and store results
        for k = 1:length(c.n)
            params_n = c.n{k};
            struct_n = cds_srr(c.outfile, c.sweep, params_n{1});
            values_n = struct_n.(params_n{2});
            params_p = c.p{k};
            struct_p = cds_srr(c.outfile, c.sweep, params_p{1});
            values_p = struct_p.(params_p{2});
            for m = 1:length(c.outvars)
                nch.(c.outvars{m})(i,:,:,j)  = squeeze(nch.(c.outvars{m})(i,:,:,j)) + values_n*params_n{3}(m);
                pch.(c.outvars{m})(i,:,:,j)  = squeeze(pch.(c.outvars{m})(i,:,:,j)) + values_p*params_p{3}(m);
            end
        end
        % Noise results
        for k = 1:length(c.n_noise)
            params_n = c.n_noise{k};
            % note: using cds_innersrr, since cds_srr is buggy for noise
            struct_n = cds_innersrr(c.outfile, c.sweep_noise, params_n{1},0);
            field_names = fieldnames(struct_n);
            values_n = struct_n.(field_names{4});
            params_p = c.p_noise{k};
            % note: using cds_innersrr, since cds_srr is buggy for noise
            struct_p = cds_innersrr(c.outfile, c.sweep_noise, params_p{1},0);
            field_names = fieldnames(struct_p);
            values_p = struct_p.(field_names{4});
            nch.(c.outvars_noise{k})(i,:,:,j) = squeeze(values_n);
            pch.(c.outvars_noise{k})(i,:,:,j) = squeeze(values_p);
        end
    end
end

save(c.savefilen, 'nch');
save(c.savefilep, 'pch');

[Walter Roberson]

those libraries do appear to be from Cadence https://community.cadence.com/cadence_technology_forums/f/custom-ic-design/15682/ic613-shared-libraries

I do not find libssrjni but I do find reference to libssrjava which seems likely to be a different implementation of the same function.

[Jongsung Park]

@Walter Roberson

after posting this video, I found the required libraries in the install directory of Spectre.

actually, the manual that guide interfacing Matlab with Specter explained where the required libraries are, and how to link them, but I failed to spot it because the exact directory where those libray exist isn't the same as the one manual indicated. More importanlty, I hadn't realized those required libraries were the ones that I faild to find.

and now I'm trying to do as the manual says, and it says

Check the dynamic library path. Check the environment variable LD_LIBRARY_PATH. Make sure that <instdir>/tools/dfII/lib and <instdir>/tools/lib are included. For c shell user, you can use command like this:

the problem is, in my Centos cshell environment, LD_LIBRARY_PATH is undefined variable, and even after I defined it as the manual guided, Matlab still failed to find the requried libraries.

do you think it would work if I just put the several libraries that is required to run the cds_srr inside the working directory? if not, should I find another way to let matab know where those libraries are located?

[dpb]

"...put the several libraries that is required to run the cds_srr inside the working directory..."

That's not the clean install way, but the app probably will find them; can't hurt anything to try.

"...LD_LIBRARY_PATH is undefined variable, and even after I defined it as the manual guided, Matlab still failed to find the requried libraries."

The environment variable will have to be set in the process which spawns the MATLAB session in which you're running; possibly the environment is cached on startup and then not rescanned after you started your session and tried to set it there...

The install should set the needed environment variable(s) globally; Windows is bad about having to reboot to make such things show up; I dunno much (as in anything, really) about Linux behavior in that regards; I'm sure Walter knows much more in that area...

[Jongsung Park]

@dpb

Thanks for leaving a comment. as you said it,

"The environment variable will have to be set in the process which spawns the MATLAB session in which you're running; possibly the environment is cached on startup and then not rescanned after you started your session and tried to set it there..."

If I understood it correctly, you meant I made a mistake of setting a environmental variable while Matlab was running, which I shoud've ran after setting was done. So, I exit the Matlab, set the LD_LIBRARY_PATH to the designated directory, and ran Matlab again.

Now in the command window, Matlab seems to find the required libraries, but it reproted another error :

Invalid MEX-file '/~~~/MATLAB Add-Ons/spectre_toolbox/64bit/cds_innersrr.mexa64': libsrrjni.so: wrong ELF class:

ELFCLASS32

I don't know what this error means. Besides, there is another error I haven't mentioned, which reported instantly whenver I ran the main.m code.

ERROR: ld.so: object '~~~/Downloads/matlab/bin/glnxa64/glibc-2.17_shim.so' from LD_PRELOAD cannot be preloaded: ignored

I wonder whether this could be also fatal issue.

[Jongsung Park]

@dpb

I googled the error, and found out that I set LD_LIBRARY_PATH to the 32bit library files.

So I ran the code again after having edited it, but this time Matlab crashed, and in the crash log

it reported the segmentation error.

[Jongsung Park]

I posted another question here.

[dpb]

"... this time Matlab crashed, and in the crash log it reported the segmentation error."

That's now probably outside the realm of being a MATLAB-itself Q? but indicative of one of two things (besides the app not doing a real good job of verifying inputs to avoid "going boom in the dark" instead of trapping user errors)

1. You passed a wrong set of arguments to the function that caused a nasty behavior failure, or
2. The app itself is buggy.

It's quite easy to create seg errors with mex code; a bum pointer (as in NULL) is a sure-fire way to do so. A mex file should be able to catch most user input errors and trap them and return a useful error message to the user, but less-than-expert code developers may either not even try and put the onus on the user to ensure never passes anything but perfect input, or may be incomplete to all possible user errors (not terribly uncommon, and can be difficult to be 100% bullet proof).

I'd venture it may well be time to contact the vendor or their user support group; this now is outside MATLAB itself except for the small population that may have the same toolbox downloaded.