COMPILER TECHNIQUES FOR MATLAB PROGRAMS ... - CiteSeerX

architectures and the use of the semantic information in order to facilitate the work of a
parallelizing compiler.
As described previously, due to its interactive nature, its extensive set of functions and
higher-level operators, and its lack of requirements for specication of intrinsic type and
dimensions of variables, MATLAB is a very powerful programming tool for prototyping algorithms
and applications. This lack of \declarations", however, is one of the major challenges
for the translation. The fact that the language is not typed simplies the programmer's job
because every function and operator is polymorphic and, therefore, type casting or function
cloning to accept dierent input and output types is not necessary.
Consider, for example, the simple computation of a square root function ( p k). It may
result in an integer, real, orcomplex output, depending on the value of k. In MAT-
LAB, however, the user does not need to be concerned about the intrinsic type of the input
variable to compute this function furthermore, the same statement works for any variable,
independent of its intrinsic type, number of dimensions (rank), or size of each dimension
(shape). Moreover, the intrinsic type, rank, and shape of the output variable are not important
for the correct execution of the program. On the other hand, for a typed language such
as Fortran and C, the user is required to know these properties for all the variables in order
to select the correct method to compute the square root. Therefore, an inference mechanism
is necessary to generate the declarations for a typed language.
The inference mechanism developed for our MATLAB compiler combines static and dynamic
inference methods, and is enhanced with symbolic and value propagation analyses.
The goal of our inference system is to determine in a MATLAB program the following variable
properties:
intrinsic type: (e.g, complex or real)
rank: (e.g., vector, matrix, scalar)
shape: (i.e., size of each dimension) and
5