[libre-riscv-dev] Introduction

Luke Kenneth Casson Leighton lkcl at lkcl.net
Fri May 3 01:37:36 BST 2019


---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68

On Thu, May 2, 2019 at 7:29 PM Samuel Falvo II <sam.falvo at gmail.com> wrote:
>
> On Thu, May 2, 2019 at 10:39 AM Luke Kenneth Casson Leighton <lkcl at lkcl.net>
> wrote:
>
> > the project has a charter, we do make an effort to honour it :)
> > https://libre-riscv.org/charter/discussion/
>
>
> Thanks for this; it seems pretty common sense and is quite agreeable.

 logical, rational, and also designed on some of the best resources i
could find on goal-focussed organisations / community.

> I feel I'm perhaps most useful with respect to 53000-related concerns; I'm
> willingly going to abstain from libre-riscv issues for now, at least until
> I learn more about its history and rationale.

 sounds reasonable.

 http://chiselapp.com/user/kc5tja/repository/kestrel-3/artifact/96ea88450c1a3fc6

 ha! you incorporated the conversion from verilog-formal to
nmigen-formal, already, ha! :)

 hey... ah...
 http://chiselapp.com/user/kc5tja/repository/kestrel-3/artifact/5c470620924b19d7

 is that... are you storing *only* the modifiable bits, to save gates?
 if so, cool!


 btw:
    for i in range(0, self.o_dat.nbits):

 that can be just
   for i in range(self.o_dat.nbits):

and:
self.bits = Signal(len(self.retained_bits))

            j = 0
            for i in range(0, self.o_dat.nbits):
                if i in self.retained_bits:
                    m.d.sync += self.bits[j].eq(self.i_dat[i])
                    j = j + 1

can be arranged as:

def elaborate(...)
   i_bits = []
   for i in self.retained_bits:
       bits.append(self.i_dat[i])
   i_bits = Cat(*bits)
   with m.If(self.o_valid & self.i_we):
        m.d.sync += self.bits.eq(bits)

(1) the key there is, you don't need to walk the length of o_dat.nbits
to get an index which you then check to see if it's in
self.retained_bits...  *just iterate the retained_bits*

 :)

(2) Cat is awesome, and the graphviz is (yosys "show") is way *way*
cleaner than doing a sequence of eqs on individual bits.


you can do the same trick with the other loop, however yeah j still
has to stick around.

        j = 0
        o_bits = []
        for i in range(self.o_dat.nbits):
            if i in self.retained_bits:
                o_bits.append(self.bits[j])
                j = j + 1
            else:
                o_bits.append(C(0,1))
            m.d.comb += self.o_dat.eq(Cat(*o_bits))

there's proooobably.... there's probably a way to get that down
somewhat (getting rid of j) by using "yield from", moving it to a
function...

       def something_like_this():
         for i in range(self.o_dat.nbits):
           if i in self.retained_bits:
                yield from self.bits
            else:
                yield C(0,1)
        o_bits = Cat(*list(something_like_this()))
        m.d.comb += self.o_dat.eq(o_bits)

maybe you can get those last two on one line, maybe don't need the
conversion to a list, maybe make the function name shorter...

       def bits_gen():
         for i in range(self.o_dat.nbits):
           if i in self.retained_bits:
                yield from self.bits
            else:
                yield C(0,1)
        m.d.comb += self.o_dat.eq(Cat(*bits_gen())

which starts to look a lot cleaner.  attached file.

btw... try this before and after, then do yosys read_ilang memcsr2.il
and "show MemCSR"
python3 mod_common_csr.py MemCSR generate -t il > memcsr2.il

and compare it to what you get without the above modifications - the
difference is quite shocking.

l.


More information about the libre-riscv-dev mailing list