Does MATLAB auto parallelize

I recently started becoming familiar with pfor and the parallel programming toolbox. In discussing this with a colleague it was mentioned that MATLAB may do some parallelizing under the hood.

Is this true?

The reason its important is that we are evaluating performance of prototype algorithms in matlab that are to be converted to C++. Past experience has show that expertly written matlab may come out 2 times faster in C/C++ but matlab written by mortals (not fully vectorized) can gain a factor of 10 or more in speed when rewritten in C/C++.

Thanks for the information.

Answer 1

[Walter Roberson]

MATLAB does do some parallelizing under the hood. In particular, for some of the more common array operations, once the array gets "big enough", MATLAB invokes LAPACK or BLAS to do the work. Those two libraries are both optimized and threaded.

Answer 2

[David Bergman]

Walter, thanks. Can you elaborate on LAPACK and BLAS libs being threaded? does matlab use FORTRAN or C versions of these libs? Do you expect threaded behavior from these libs in C/C++ implementations? Lastly, can one turn this off with some flag?

I guess what I'm really after is a way of verifying algorithm performance honestly. For the applications I'm used to developing we can match performance to a hand count of flops fairly well.