right shift of integer variable

Hi,

I want to do a right shift to a give signed integer variable. Can any help me with the syntax please.

My C-code looks like below.

"a = b>>12;"

where the variables 'a' and 'b' are signed integer.

I tried several methods like 'bitsra', 'bitshift' etc. But I was not succesful. Kindly help me please.

Answer 1

[NDKA]

Thanks for your reply Richard. My data is random, sometimes it is positive and sometimes it can be negative i.e. int16.

So, you mean, there is no direct command in matlab to do right/left shift of an signed integer variable (int16)?

Please advise. Thank You

[Richard Brown]

No there isn't.

p.s. when replying to answers, use comments (like this one) rather than writing new answers

[NDKA]

Its sad, but I'm glad I now I have to find another way to solve my issue. Thank You Richard.

Answer 2

[Richard Brown]

bitshift is for unsigned integers. What exactly do you want your signed right shift to do?

Answer 3

[NDKA]

Just like in C-language. For example, if the value of 'b' is '-8192', then if I execute the below line, I expect the value of 'a' as '2'.

" a = b>>12 "

I want the equalant matlab code for the above. I know I can just use divide operator "a = b/4096", but I'd like to use a shift operator here.

Thanks a lot Richard.

[Walter Roberson]

In int16, (-8192) >> 12 is *not* 2: it is either -2 (sign fill) or 6 (no sign fill).

[Jan]

"b / 4096" is useful. Do you have a good reason to avoid this?

Answer 4

[Richard Brown]

the C language doesn't officially specify what happens on a right shift of a negative signed integer, it's up to any given compiler to decide. I guess you need to decide what sign you want, and then right shift an unsigned version of it. Or you could create a mex file that calls the C >> operator on your variables

Answer 5

[Geoff]

MatLab is pretty infuriating sometimes.

For example, if you cast a float as an int, it 'helpfully' rounds the value instead of truncating.

And if you cast a negative integer as unsigned, it returns zero.

Don't get me started on MatLab's fscanf function!

If you want to do a lot of shifting and don't want to multiply or divide by 2^shift, AND don't want to implement a bitshift wrapper to 'cure' the negative case, then I guess you'll need to write it in C and use MEX.

Remember, it's doubtful whether you will get any useful speed advantage using MatLab's bitshift, if that's what you want. Call me cynical =) I haven't actually tested that theory.

[NDKA]

I hope Matlab will consider to include this option in future. As of now, I'll try to find an alternate option. Thanks for your comments Geoff, it really helped me.