How to set Matlab precision to Single

my code requires a precision 32 bit float. Matlab uses double by default, is there a way to set it to single by default so that when ever my program executes , all my variables are in single

Answer 1

[Walter Roberson]

This information can be found in http://undocumentedmatlab.com. My memory is that it is

system_specific('precision', '24')

but you should check.

However, it should definitely not be trusted that all internal calculations will be made in single precision even if all the variables involved are single precision.

[Star Strider]

I couldn't find a ‘system_specific’ command mentioned when I looked on that site just now. There were a few mentions on a Web search (none later than 2010), but nothing else about it related to MATLAB. When I typed system_specific at the command line (Windows 7, 2012a), I got:

Undefined function or variable 'system_specific'.

It would be nice if a few such customization functions existed, though.

[Walter Roberson]

Ah, system_dependent(). Or feature(). See http://undocumentedmatlab.com/blog/undocumented-feature-function/

[Star Strider]

Thank you! (Bookmarked for later reference.)

An experimental system_dependent() call generated no errors (or to the best of my knowledge, anything else), and a feature() call generated:

Error using feature
Not enough input arguments.

I was hoping that both calls would generate some documentation themselves, but no such luck. I don't surf Undocumented MATLAB much because it seems to be mostly Java. I may need to visit it more often.

[Jan]

The command is: feature('SetPrecision', 24). The deafult value is 53, which means standard 64 bit DOUBLE precision, while 64 enables the storing of intermediate values in the 80 bit registers of Intel processors.

But this command enables rounding to single precision. This is useful to check the numerical precision of an algorithm, but it neither saves time nor memory. Therefore I assume, that this is not useful for you. Please explain, what you are wanting to achieve.

Answer 2

[Azzi Abdelmalek]

 single(x)

[Pratheek]

i want all my variable to work automatically with single precision, not by casting

[Azzi Abdelmalek]

if your data are single , all your variables will be single

 a=single(2)% single
 b=14       % double
 c=a*b      % single

Answer 3

[Star Strider]

The best you can likely do is described in this thread ‘Single Precision by Default? ... lots of auxiliary variables to cast’ last year.

[Yash]

change format