[Libre-soc-bugs] [Bug 708] PartitionedSignal API design

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Thu Sep 23 23:12:45 BST 2021


https://bugs.libre-soc.org/show_bug.cgi?id=708

--- Comment #4 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #0)
> exponent_field_layout = { # probably in a FP layout class
>     FPULayoutEnum.F16x4: 5,
>     FPULayoutEnum.F32x2: 8,
>     FPULayoutEnum.F64x1: 11
> }

yes, this would solve the tricky issue of irregular sizes,
that change depending on partition size.  over-allocation
will be required, in effect, where the total bits allocated
must be more than *might* be used for certain operation
widths, and the class ends up *dynamically* selecting
the appropriate sub-parts depending on the partition
points.

really quite mind-bending.

unfortunately, like ArrayProxy, i feel it is just too much
to tackle in a "first cut" of the API. i would much rather
that we:

a) complete the work as planned (esp. given that the budget's
   already spent)

b) go through the whole rigmarole of workarounds, documenting
   them clearly, to deal with the varying sizes explicitly
   (using the existing PartitionedSignal)

one of the primary reasons for (b) is because there has been
such extensive testing and comprehensive Formal Correctness
Proofs done.  we *know* that __eq__ etc work.

if we get into serious trouble or get completely fried and
brain-melted with how exasperating PartitionedSignal turns
out to be, that's easy to justify an extra Grant.

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


More information about the libre-soc-bugs mailing list