Archives Discussions

[drstrip]

I have some code where I would like to perform a circular shift of the contents of an int4 by one word.  Can this be done with some sort of shift operator, or something that is more efficient than 5 assignments?

int4 a; int4 b; b.x = a.x; a.x = a.y; a.y = a.z; a.z = a.w; a.w = b.x;

[ryta1203]

The above should only take 1 cycle, assuming the T ALU can do a MOV.

EDIT: The other assumption is, of course, that the Brook+ compiler is smart enough to make the IL that the CAL compiler can produce the good corresponding ISA.

You could always code this fairly simply in CAL/IL or CAL/ISA (of course, if AMD would ever get this code path working).

[genaganna]

following one could be more efficient.

a=a.wxyz

[ryta1203]

genaganna, you code does this:

a.x = a.w

a.y = a.x

a.z = a.y

a.w = a.z

Which I don't think is what he wanted, for starters you are overwriting one of the values before you have a chance to write it back, correct?

But like I said above, either way this should only take 1 cycle if the compilers are good enough to handle it.

[genaganna]

Originally posted by: ryta1203 genaganna, you code does this:

a.x = a.w

a.y = a.x

a.z = a.y

a.w = a.z

Which I don't think is what he wanted, for starters you are overwriting one of the values before you have a chance to write it back, correct?

      it works fine. compiler generates appropriate code for such statements.

But like I said above, either way this should only take 1 cycle if the compilers are good enough to handle it.

a=a.yzwx

sorry previous one is lelf-cricular shift