[Libre-soc-dev] daily kan-ban update 19nov2021
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Fri Nov 19 08:23:36 GMT 2021
yesterday spent half a day asleep and the evening sorting out forms for
NGI POINTER fundingbox.
i've managed to get a Finite State Machine running in core which can
(optionally) be requested to respect Read-after-Write hazard tracking.
it didn't work... because the decoded instruction, broadcast to multiple
Function Unts into "Satellite" decoders, only contains the immediates
and other operands *not* the register numbers that need to be written
the reason why this did not matter previously is because TestIssuer
would "hold" the register read/write numbers asserted on the Regfile
Ports, waiting for that one instruction to complete.
now that there are multiple instructions in-flight, having a *single*
location "hold" those read/write numbers is completely inadequate:
the reg nums will have to be captured in independent per-Function-Unit
funnily enough this is precisely and exactly one of the jobs of a
Dependency Matrix FU-Regs Table, aka the 6600 "Q-Table" in the
original 6600 patents.
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
More information about the Libre-soc-dev