Talk:XCore Architecture
Latest comment: 12 years ago by Chris Chittleborough in topic Encoding three numbers from 0 to 11
 | This article was nominated for deletion on 24 October 2020. The result of the discussion was redirect. |
 | This redirect does not require a rating on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | |||||||
| ||||||||
Encoding three numbers from 0 to 11 edit
Some readers will be curious, as I was, about how XMOS encode three numbers from 0 to 11 inclusive (eg., general purpose register numbers) in 16-bit instructions. Note that 12×12×12 = 1728 = 2**10.75, so you need 11 bits.
Let x, y, and z be the numbers being encoded, all ≥ 0 and ≤ 11.
- Use 5 bits to hold (x/4)×9 + (y/4)×3 + (z/4)
which can never exceed 2×9+2×3+2=26 - Use 2 bits to hold (x & 0x3)
- Use 2 bits to hold (y & 0x3)
- Use 2 bits to hold (z & 0x3)
(I guess the 5-bit field is small enough to decode directly into (x/4), (y/4) and (z/4), without using any division or modulo operations.)
This allows XMOS to use values 26-31 in that 5 bit field for other purposes. Neat, hey? CWC 07:01, 8 January 2012 (UTC)