MATLAB COMPILER RELEASE NOTES User's Guide download pdf (Page 173)

Optimizing Arrays

6-5

Optimizing Arrays

Scalar Arrays

(fold_scalar_mxarrays) When this optimization is enabled, all constant,

scalar-valued array operations are folded at compile time and are stored in a

constant pool that is created once at program initialization time. Folding

reduces the number of computations that are performed at runtime, thus

improving runtime performance.

Scalar folding can dramatically improv e the performance of code that is

manipulating scalar arrays, but it makes the code less readable. For e xample,

function y = foo(x)

y = 2*pi*x;

If you compile this with the -O none option, you get

...

mlfAssign(&y, mclMtimes(mlfScalar(6.283185307179586),

mclVa(x, "x")));

...

Compiling with -O none -O fold_scalar_mxarrays:on,gives

...

mlfAssign(&y, mclMtimes(_mxarray0_, mclVa(x, "x")));

...

In the optimized case, this code uses _mxarray0_, which is initialized at

program start-up to hold the correct value. All constants with the s ame value

use the same

mxArray variable in the constant pool.

Nonscalar Arrays

(fold_non_scalar_mxarrays) This optimization is very similar to

fold_scalar_mxarrays. It folds nonscalar mxArray values into compile-time

arrays that are initialized at program start-up. This can have a large

performance impact if you are constructing arrays that use

[] or {} within a

loop. This optimization makes the code less readable. For example,

function y = test

y = [ 1 0; 0 1] * [ pi pi/2; -pi -pi/2 ];

1 2 ... 168 169 170 171 172 173 174 175 176 177 178 ... 263 264

No comments

MATLAB COMPILER RELEASE NOTES User's Guide Page 173