[Libre-soc-bugs] [Bug 419] MUL pipeline formal proof needed

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=419

--- Comment #3 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Samuel A. Falvo II from comment #2)

> I don't understand how copying the code in that class would help though; can
> you clarify?  Thanks.

apologies, we need to speed things up, a lot.  i will explain why at an
appropriate time.

https://git.libre-soc.org/?p=soc.git;a=commitdiff;h=d3deda12916c450f05f7d03610f0ef79c9cbdafb

i have set up a triple chain using StageChain which you can see was being
called from nmutil PipeModBase.  i had to work out how to use it because the
last time i called StageChain was 18 months ago.

a dummy object is set up with a dut.i and dut.o so that the remaining
cut/pasted code does not need significant mods.

you can go ahead and use dut.i and dut.o as if there was a single combinatorial
block between them, from a single module.

stage 1 abs's its input
stage 2 does the mul
stage 3 works out which bits of result to use

however with all 3 chained together you can ignore that entirely and implement
the  OP_MUL* microcode operations in any way you see fit that is preferably
simple and clear.

however please do not be surprised to find that symbiyosys MUL runs as slow as
molasses flowing downhill in arctic winter.  there is not a lot we can do about
that.

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