[Libre-soc-bugs] [Bug 230] Video opcode development and discussion

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=230

--- Comment #35 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to cand from comment #34)
> That's great and definitely useful. I'm just not sure what it means if
> additional functionality is desired for such ops, is that allowed or a no-no?

well... maybe.  saturate definitely.

as long as it is 1-in 1-out (not trying to zip or unzip) because rhat really is
a new op, ordinary 1in 1out ops should actually fit, but not need new opcodes.

by that i mean, it is perfectly fine for an FU to do double duty as long as the
regfile (operand) profile matches.

there is a bugreport outstanding to add OR, AND and XOR etc. to the ALU
pipeline for example, but not the other way round because Logical does not have
XER carry.

in this case: logical not, exts*, FP2INT, FPCVT, fabs, those all "fit" because
they are 1in 1out and not a lot in the way of gates, each.

however anything that is arithmetic i am wary of due to the effect it would
have on data routing.  arithmetic which takes 2 inputs and *reduces* that to 1
out is very different from something that takes 2 inputs and splices them to
create the same number of bits in the output, if you know what i mean.

i am also tempted but also wary of anything that would take large immediates,
although in theory it should be fine.

was there anything you had in mind?

-- 
You are receiving this mail because:
You are on the CC list for the bug.