Documentation
Last update : 28/09/2007

sci_files - How to write conversion functions

Description

To convert calls to Matlab functions, mfile2sci uses a function called sci_<Matlab_function_name> . All these functions are defined in sci_files in directory SCI/macros/m2sci/sci_files/. The set of sci_files given in Scilab distribution does not allow to convert calls to all Matlab functions yet. However, a Scilab user can add sci_files (for Matlab functions or for user defined functions) to Scilab using the following tips.

In M2SCI, a function call is considered as a "tree" (it is also the case for the instructions of the file to convert), represented in Scilab by a tlist with following fields:

  • name : Matlab function name
  • lhsnb : number of Matlab function output parameters
  • lhs : list of Matlab function output parameters
  • rhs : list of Matlab function input parameters
  • A sci_function has one input called tree which is also the output of the function. A sci_function has to convert this incoming "tree" so that it is compatible with Scilab by changing name, lhsnb, lhs and/or rhs. The other task that has to be done by this function is inference. Incoming tree contains inference data in its lhs that have to be updated with what can be infered for the outputs of this function.

    Some useful functions have been written to help to create M2SCI tlists while writing this conversion function:

  • Funcall : create a tree representing a function call
  • Operation : create a tree representing an operation
  • Variable : create a tree representing a variable
  • Cste : create a tree representing a constante value
  • Infer : create a tree representing inference data
  • Type : create a tree representing type for inference
  • Equal : create a tree representing an instruction
  • Some other functions have been designed to get properties of operands/inputs. Considering A is tlist used in macro tree, you can use the following functions:

    Function returns %T if...
    is_empty(A) all dimensions of A are 0
    not_empty(A) all dimensions of A are known and at least one dimension of A is not 0
    is_a_scalar(A) all dimensions of A are 1
    not_a_scalar(A) all dimensions of A are known and at least one dimension of A is not 1
    is_a_vector(A) all dimensions of A are known and all dimensions of A but one are equal to 1
    not_a_vector(A) all dimensions of A are known and at least two dimensions of A are greater than one
    is_real(A) A is real
    is_complex(A) A is complex
    isdefinedvar(A) A is a variable already created in M-file currently converted
    allunknown(A) all dimensions of A are unknown

    Some other functions have been written for specific needs while writing conversion files:

  • first_non_singleton : is an equivalent to firstnonsingleton for an M2SCI tlist. Calling sequence: dim = first_non_singleton(A)
  • gettempvar : generates a temporary variable having a name which does not already exist. Calling sequence: v = gettempvar()
  • insert : allows to insert instructions. Calling sequence: insert(Equal(...),opt) with opt~=1 to insert before current instruction and opt=1 to insert after it.
  • getoperands : can be used to get each operand as a variable. Calling sequence: [A,B] = getoperands(operation_tlist)
  • getrhs : can be used to get each parameter as a variable. Calling sequence: [A,...] = getrhs(funcall_tlist)
  • convert2double : change type of input when this type is not implemented for a particular function is Scilab. Calling sequence: A = convert2double(A)
  • To have more informations about how to write such files, refer to directory SCI/macros/m2sci/sci_files/ which gives many examples from very simple ones (e.g. sci_abs.sci) to very complex ones (e.g. sci_zeros.sci).

    See Also

    m2scideclare ,   Funcall ,   Operation ,   Variable ,   Cste ,   Infer ,   Type ,   Equal ,  

    Authors

    V.C.