[Libre-soc-dev] v3.1B prefix

Cole Poirier colepoirier at gmail.com
Sat Dec 5 17:06:27 GMT 2020


On Saturday, December 5, 2020, Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

> carrying on from late last night: some recovery of prior work done can
> be achieved in the following way:
>
> 32bit SV instructions can be invented by:
>
>   - selecting 1 or 2 major opcodes
>   - using the older SV-P48 11 bit format
>   - taking the v1 16bit Compressed form
>   - joining the 3 together
>

This answers some of my questions from last night. Important lesson, it’s
better to wait and see what your subsequent emails say.


> following on from that:
>
>   - given that these are SV prefixed
>   - given that they are v3.0B 32bit
>   - logically it follows they fit v3.1 64 pref
>   - *and the 24 bits are entirely free*
>
> this i feel is how we should do src1/src2 swizzled operations, for
> example.  there will be limitations on both the range of registers and
> also the range of operations (because C16 is so small) however it is
> better than a kick in the teeth.
>

Yes, better to try to do a little with a little than a lot with nothing.


> and, if the C bank-selection idea is kept, even the v3.1 prefixed ops
> can be retargetted to specialist tasks when needed most.
>
> in particular: the 2 bits N and M of the 16bit form, by not doing
> "state" any more, may be allocated to alternative purposes.  the
> highest priority of these i feel would be one for selecting between CR
> or INT predication.  the other for extending the ops that can be
> selected.
>
> although to be honest with the focus being entirely different, a full
> redesign of C16 to only be SV prefixed is probably on the cards.
>
> argh.
>
> we have to invent an entire sub-ISA which is EXACTLY what i did not
> want us to be doing.
>

My mental model is stuck on this: is the C bank switching done through the
VLE page table bit? Following from this, is the invention of an entire sub
ISA behind this protection/isolation VLE-like page?

A simple yes or no to both questions is sufficient for me to progress in
developing my mental model.

Cole


More information about the Libre-soc-dev mailing list