[Libre-soc-bugs] [Bug 713] PartitionedSignal enhancement to add partition-context-aware lengths

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Thu Oct 7 20:51:55 BST 2021


https://bugs.libre-soc.org/show_bug.cgi?id=713

--- Comment #29 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #28)
> (In reply to Luke Kenneth Casson Leighton from comment #27)
> > 2) we make the tough decision to break the rule "no changes to any
> >    pipeline code" and AVOID the need entirely.
> 
> Well, we'll need to break that rule anyway for FP pipelines, which isn't
> much of an additional problem since they still all need modification anyway
> to add F16/BF16

FP16 was already done, it was possible to do full coverage using it, and
found bugs which, through parameterisation, meant that FP32 and FP64 didn't
have them either.

BF16 is a one-liner in the dictionary i mentioned (and can't remember where
it is).  unit tests are 

> and to handle flags/exceptions.

that's orthogonal and a separate series of patches that have no bearing
on this.

> Idea: have a XLEN global that is an instance of a new class SimdMap (a
> generalization of SimdLayout's dict handling code to arbitrary values 

jacob: again, that's *another* patch - a seriously big one.  it can
be considered.... *AFTER* the completion of the first (minimalist)
set of changes.

the other idea would mean that the ALUs become critically dependent
on PartitionedSignal, which may *or may not* have the full capabilities
of Signal (Array), and has *not* been designed to be a "scalar Signal
replacement".

it always creates sub-modules for example.  none of those modules
have been tested with zero-length (no) partitioning.

if used as a single-scalar it would result in significantly extra gates
and may not actually work with zero partitions (which is what a scalar
is).

that would therefore be extra work and, to remind you for about the
fiftieth time, we are under time and budgetary pressure.

please, please: *focus*.

the more time is spent discussing possible future designs and enhancements
instead of doing the work that's needed, the more i have to cut drastically
back on what can be done.

we

are

running

out 

of

time.

please.... *focus* on the task that's needed, and listen to the time-saving
risk-reducing techniques that took me over 20 years to learn, at considerable
cost, in my career.

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


More information about the libre-soc-bugs mailing list