[libre-riscv-dev] [Bug 270] New: investigate nmigen clock gating

[libre-riscv-dev] [Bug 270] New: investigate nmigen clock gating

[bugzilla-daemon]

http://bugs.libre-riscv.org/show_bug.cgi?id=270

            Bug ID: 270
           Summary: investigate nmigen clock gating
           Product: Libre-SOC's first SoC
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: CONFIRMED
          Severity: enhancement
          Priority: ---
         Component: Source Code
          Assignee: lkcl@lkcl.net
          Reporter: lkcl@lkcl.net
                CC: libre-riscv-dev@lists.libre-riscv.org
   NLnet milestone: ---

see
http://lists.libre-riscv.org/pipermail/libre-riscv-dev/2020-March/005482.html

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev

[libre-riscv-dev] [Bug 270] investigate nmigen clock gating

[bugzilla-daemon]

http://bugs.libre-riscv.org/show_bug.cgi?id=270

--- Comment #1 from Luke Kenneth Casson Leighton <lkcl@lkcl.net> ---
> The principle is that you save power by not clocking the parts of the circuit
> that don't have to do any computing. I think this could be a more
> general way to only enable the stages in your pipeline who actually 
> are doing computation.

ok so if i understand this correctly:

* the clock still runs at 1600mhz
* the clock runs a cyclic shift-register of length equal to the
  number of stages, at 1600 mhz.
* only every *alternate* one of those elements in the shift register
  is enabled (or, if you want full speed, all of them). 
* through EnableInserter each stage is clocked by a *different* bit
  in the shifted-register

> That said I think this feature does not fit in the MVP scope of the October
> prototype so that chip should IMO not use clock gating nor the pass-through
> register feature from the original discussion. 

no, i agree, and, more to the point, we don't need it for the 180nm ASIC
(except perhaps to test the concept).

one thing that we have is, the use of OO python has the entirety of the
stages themselves *completely* separated firmly behind a general-purpose
API, where the construction of pipelines, from those stages, using entirely
different pipeline techniques, is *literally* a one-line change.

so we could conceivably do the *entire* suite of pipelines - convert them
to use this clock gating technique - *literally* in well under a day,
after first experimenting with EnableInserter and a quick and simple unit
test.

re-running the IEEE754 FP unit tests on the other hand... *sigh* :)

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev

[libre-riscv-dev] [Bug 270] investigate nmigen clock gating

[bugzilla-daemon]

http://bugs.libre-riscv.org/show_bug.cgi?id=270

Staf Verhaegen <staf@fibraservi.eu> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |staf@fibraservi.eu

--- Comment #2 from Staf Verhaegen <staf@fibraservi.eu> ---
(In reply to Luke Kenneth Casson Leighton from comment #1)
> > The principle is that you save power by not clocking the parts of the circuit
> > that don't have to do any computing. I think this could be a more
> > general way to only enable the stages in your pipeline who actually 
> > are doing computation.
> 
> ok so if i understand this correctly:
> 
> * the clock still runs at 1600mhz
> * the clock runs a cyclic shift-register of length equal to the
>   number of stages, at 1600 mhz.
> * only every *alternate* one of those elements in the shift register
>   is enabled (or, if you want full speed, all of them). 
> * through EnableInserter each stage is clocked by a *different* bit
>   in the shifted-register

Correct, the clock is the pipeline clock. In theory other parts of the CPU
could for example run at half the clock frequency. This will then naturally
automatically only committing a new operation every other cycle at maximum.

I did not test it but EnableInserter should work in simulation and FPGA.
Depending on FPGA you likely won't see the full power improvements as I think
that the enabling is implemented as an enable input to each FF and not with
gating parts of the clock tree. It will still guarantee that the output of FFs
don't change.
As said implementing clock gating for ASICs will not be a simple task.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev