[libre-riscv-dev] go die/cancel signals in fsm

[libre-riscv-dev] go die/cancel signals in fsm

[Jacob Lifshay]

I looked through the vcd generated for soc.simple.test.test_core and
was unable to locate the signals used to tell the pipeline stages that
the instruction they are currently operating on is canceled. This is
needed to not have to wait 63(!) more clock cycles for a canceled div
instruction to finish computing before another div instruction can be
started.

Help wanted,
Jacob

_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev

Re: [libre-riscv-dev] go die/cancel signals in fsm

[Luke Kenneth Casson Leighton]

On Wed, Jul 22, 2020 at 3:01 AM Jacob Lifshay <programmerjake@gmail.com> wrote:
>
> I looked through the vcd generated for soc.simple.test.test_core and
> was unable to locate the signals used to tell the pipeline stages that
> the instruction they are currently operating on is canceled.

that's because we're using SimpleHandshakeRedir rather than
MaskCancellable, and are not going to do cancellation in the 180nm
core unless we get more people, and i can focus in the remaining very
short amount of time on doing absolutely nothing but add in the
Dependency Matrices.

three months is already far too short to even consider doing that, so
it is extremely unlikely to be added.

> This is
> needed to not have to wait 63(!) more clock cycles for a canceled div
> instruction to finish computing before another div instruction can be
> started.

indeed.

ok so i added some code comments explaining what goes into CommonPipeSpec:
https://git.libre-soc.org/?p=soc.git;a=commitdiff;h=71af4fadd8a431079df13fc5812d745c133074be

i'll also update the README...

https://git.libre-soc.org/?p=soc.git;a=commitdiff;h=593c41881b455aeeb19fe98071647d7b4ffca15c

l.

_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev

Re: [libre-riscv-dev] go die/cancel signals in fsm

[Luke Kenneth Casson Leighton]

On Wed, Jul 22, 2020 at 10:51 AM Luke Kenneth Casson Leighton
<lkcl@lkcl.net> wrote:

> > This is
> > needed to not have to wait 63(!) more clock cycles for a canceled div
> > instruction to finish computing before another div instruction can be
> > started.
>
> indeed.

for now, you can see that issuer.py is a FSM:
https://git.libre-soc.org/?p=soc.git;a=blob;f=src/soc/simple/issuer.py;hb=HEAD#l124

so there's not going to be any cancellation.  instructions are done
one at a time, total freeze until the instruction is completed.

l.

_______________________________________________
libre-riscv-dev mailing list
libre-riscv-dev@lists.libre-riscv.org
http://lists.libre-riscv.org/mailman/listinfo/libre-riscv-dev