Trouble inlining C-Mex S-Functions

Hello everyone!

After spending quite a long time fighting with this issue I think I’m at the point of having to ask for some help to clear up my confussion. I consulted the documentation, examples and discussions, and thought I understood, but my failure to get it working proves the contrary.

So: I want to inline a quite heavy iterative algorithm that was given to me in form of a C-Mex S-Function. I’m working with code generation for embedded microcontroller targets, compiling the code in the corresponding IDE (TI CCSv3.3 and CCSv5). Everything was working fine until I decided to implement this S-Function and to inline it, mainly to avoid the S-Function API and improve the efficiency on target.

Writing a wohle new TLC is not a reasonable option for me. So I thought I could use the Legacy Code Tool. Am I wrong here and is this just thought for “pure” C code? Maybe I still could use it later on for registry operations, but first I need to implement the algorithm.

All the legacy_code commands seemed to work (sfcn_cmex_generate, slblock_generate, sfcn_tlc_generate…) except the ‘compile’ which failed with docens of messages like: “sfun_orig.obj : error LNK2005: _ssSetStateAbsTol already defined in ex_sfun_out.obj”

I’m therefore assuming I must use a wrapper. I understand this requires writing a small TLC to embodie the calling function. This shouldn’t be a big deal, as the algorithm has 2 inputs, 1 output, 1 discrite inherited sample time and just needs a couple of values from the previous steps... but I still didn’t manage it. Using it as a common level-2 C-Mex S-Functions works fine.

I’m even starting to think of “extracting” the c routines of the s-function and create a regular .c to work with, but I’m not sure if it’s a good solution... and maybe it’s an unnecessary mess.

Hopefully someone will give me a clue… so that I know in which way to focus my efforts.. .

Regards, and thanks a lot in advance!

Martin

[Kaustubha Govind]

Are you using functions like ssSetStateAbsTol in your C code? If yes, that may be the issue. The code that you give to the Legacy Code Tool is supposed to be something that can run independently (including on an embedded target).

[Martin Dendaluce]

Thanks for your answer!... and sorry for my huge reaction time, I was unexpectedly overloaded in another project.

I do have the functions ssSetNumSFcnParams, ssSetNumContStates, ssSetNumDiscStates, ssSetNumInputPorts, ssSetInputPortWidth, ssSetInputPortRequiredContiguous, ssSetInputPortDataType, ssSetInputPortDirectFeedThrough, ssSetNumOutputPorts, ssSetOutputPortWidth, ssSetOutputPortDataType, ssSetNumSampleTimes, ssSetNumRWork, ssSetNumIWork, ssSetNumPWork, ssSetNumModes, ssSetNumNonsampledZCs, ssSetOptions, ssSetSampleTime, ssSetOffsetTime inside the functions static void mdlInitializeSizes(SimStruct *S) and static void mdlInitializeSampleTimes(SimStruct *S).

Where you also refferring to those??

I have made a first attempt to adapt it into a s-function builder block to generate a Wrapper TLC, but I'm not sure if this makes sense.

Could I use the legacy code tool for this? Maybe I'm using it wrong. My aim is to make the code execution more efficient, avoiding the s-function API.

Regards and thanks again!

Martin

Answer 1

[Kaustubha Govind]

I think you might be using the Legacy Code Tool incorrectly - the code that you supply to the tool is supposed to be standalone C code (probably some legacy algorithms that you inherited which you want to run as part of a Simulink model), and the tool generates the "wrapper" S-function interface like mdlInitializeSizes, mdlOutputs, etc. along with TLC code which basically just asks Simulink Coder to generates calls into your standalone C code. I wonder if it might be easier if you manually create the TLC file, unless there is an easy way to separate out the algorithm part of your S-function into standalone C code, and then generate a new S-function and TLC using Legacy Code Tool or S-function Builder.

[Martin Dendaluce]

Thanks again! Then yes, I was using it incorrectly, which actually is good news :-) I will try to separate the pure C code from the s-function, as I'm not very sure about manually creating a tlc. I understand this tlc would just be the skeleton to call the function, similar to what an empty s-function-builder block creates? (code at the end)

I think the legacy code tool might also come in handy for me in future cases. Am I wrong assuming it also is the most efficient way?

I understand that this docmentation should should keep being my major reference point.

Regards!

Martin

%%File : sbuild_dummy.tlc
%%Created: Thu Mar 14 13:47:28 2013
%%Description:

Simulink Coder wrapper functions interface generated for

S-function "sbuild_dummy.c".

File generated by S-function Builder Block

For more information on using the Target Language with the

Simulink Coder, see the Target Language Compiler manual

(under Simulink Coder) in the "Inlining S-Functions"

chapter under the section and subsection:

"Writing Block Target Files to Inline S-Functions",

"Function-Based or Wrappered Code".

%implements  sbuild_dummy "C"
%%Function: BlockTypeSetup ===================================================
%%Purpose:

Set up external references for wrapper functions in the

generated code.

%function BlockTypeSetup(block, system) Output
  %openfile externs

    extern void sbuild_dummy_Outputs_wrapper(const real_T *u0,
                            real_T *y0);
    %closefile externs
    %<LibCacheExtern(externs)>
  %endfunction

%%Function: Outputs ==========================================================
%%Purpose:

Code generation rules for mdlOutputs function.

%function Outputs(block, system) Output
  %assign pu0 = LibBlockInputSignalAddr(0, "", "", 0)
  %assign py0 = LibBlockOutputSignalAddr(0, "", "", 0)
  %assign py_width = LibBlockOutputSignalWidth(0)
  %assign pu_width = LibBlockInputSignalWidth(0)
  sbuild_dummy_Outputs_wrapper(%<pu0>, %<py0> );
%endfunction

%%[EOF] sbuild_dummy.tlc

[Kaustubha Govind]

Yes, if you separate out your algorithm into generic C, then your TLC will just be a skeleton that calls into your algorithm entry points.

[Martin Dendaluce]

Is there a significant performance difference between using the Legacy Tool and a S-Function Builder which generates a Wrapper TLC?

It's just that I'm not very confident regarding to manually implementing the TLC with the entry points and the data mapping.

Thanks!

[Kaustubha Govind]

Martin: I don't know of a significant difference in the generated TLC for performance - I think you can use whatever you find more convenient.