[Libre-soc-bugs] [Bug 838] sync or at least statically check fields.text, power_decoder, trans/svp64, CSVs between each other

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=838

--- Comment #41 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Dmitry Selyutin from comment #40)
> (In reply to Luke Kenneth Casson Leighton from comment #39)
> > yes TX is the extra bit that turns PackedSIMD (VSX) reg
> > numbers from 5-bit (32 regs) to 6-bit (64 regs).  they
> > probably put the one extra bit elsewhere, leaving the
> > 5 bits RA RB etc. etc. where they are to avoid the MUXes
> > just for PackedSIMD [a common hardware decode tactic]
> 
> Well we might have it as an alias to Form.X. What do you think?

No doesn't work that way.  It is a hierarchy. Form.X.TX
Form.whatever.field

All of the aliases you can see in PowerFields should never have been added and
under no circumstances should they be used. They were part of microwatt.

NO pipelines use self.forms.RA

ALL pipelines use self.forms.FormX.RA.

so no, under no circumstances should any "aliases" be added.

The operands are accessible ONLY via the 2 level hierarchy or not at all

Now we come to a painful bit.  IBM former custodians of Power ISA had the
attitude that the Forms are not for machine readability and neither is the ISA.
this is stupid as it costs time and money which apparently they must clearly
have plenty of.

If you look closely there are very unfortunately MULTIPLE fields named "L" even
WITHIN the same Form.

this because clearly anyone wishing to do a machine readable spec must be
stupid or "well we never intended it to be, we never did it that way, so tough"
was the response.

Not impressed with that I *renamed* L to L1 L2 L3 to avoid conflicts and this
is just how it has to be.

> > ehn? how did i miss those, well-spotted :)
> 
> I think these should rather appear _before_ our custom forms. I guess
> enumeration values do not value much for Form enum, just a logical thought.

I had only added the instructions needing them to the
Simulator ISACaller, it is the HDL that would need them
and that hasn't been done yet. ISACaller does not use
Power enum Forms.


> Well I cleaned these up. Perhaps I should drop a note to them.

Yes there will soon be a external RFC process available
Although I really want to keep track of things, they have
the perspective that all and any even slightest minor
revisions and corrections shall remain secretive and
confidential. Which makes sense for certain scenarios
but not all.


> I think we should contact them to tell about this; any email to reach them?

No there will be an RFC web page up shortly, which will
even have a question "do you want to attach a diff patch"

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