[libre-riscv-dev] [Bug 329] coriolis2 experiment layout for Dependency Matrices

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Wed Jun 3 23:18:28 BST 2020


https://bugs.libre-soc.org/show_bug.cgi?id=329

--- Comment #10 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jean-Paul.Chaput from comment #9)
> (In reply to Luke Kenneth Casson Leighton from comment #8)
> > after a little thought, see the diagram in attachment "fu-fu matrix"
> > https://bugs.libre-soc.org/attachment.cgi?id=58
> 
> Hello Luke,
> 
> I'm starting to have automated matrix placement working for the FU-FU
> matrix.

superb.  do commit it, we can take a look.

> But the first comparison I can make between matrix placement
> and automated placement still got the later winning (by a little).

that's veery unlikely to be the case at 32x12, 32x25 or, when we go
to GPU DMs, 48x30 or even 128x30.   (this is FU-REGs, we are currently
investigating FU-FU, which will still potentially be be up to 30x30)


> I have two little remarks about the FU-FU example you gave me:
> 
> 1. It seems to me that the current example has 4 unit and 3 regs
>    (inverted in comment #5).

probably :)  actually for FU-FU it should be a square matrix.

> 2. Nothing is connected to gowr3_i, seems strange to me.

because if it's not a square matrix, that's what happens.

> I did try to generate a bigger FU-FU, but got stuck here:
> 
> Traceback (most recent call last):
>   File "fu_fu_matrix.py", line 169, in <module>
>     test_fu_fu_matrix()
>   File "fu_fu_matrix.py", line 166, in test_fu_fu_matrix
>     run_simulation(dut, d_matrix_sim(dut), vcd_name='test_fu_fu_matrix.vcd')
>   File "/usr/lib/python3.6/site-packages/nmigen/compat/sim/__init__.py",
> line 22, in run_simulation
>     fragment.domains += ClockDomain("sync")
> AttributeError: 'FUFUDepMatrix' object has no attribute 'domains'
> 
> I should have the latest nmigen, soc & nmutil repositry versions.

you need the version nmigen from whitequark, not from m-labs.
check the repo url in .git/config, it should be
https://github.com/nmigen/nmigen/

and you can ignore that, anyway - the ilang file-generation should
have happened before that.


> Would it be possible for you to generate a FU-FU matrix as close as possible
> to the one intended in terms of number of FU and registers (the idea is to
> have a representative amount of wires and gates).

let's make it 30x30 - that's pretty damn big.  for 180nm we'll likely go
with 1/2 that size.

> Best regards,
> 
> PS1: I use the approach of "not getting to the bottom of things but start
> like
>      a copying monkey", but always still felt a little bit like cheating.
> Now, I will
>      do it proudly...

haha very funny :)

> PS2: Maybe you have seen this... (personally, I'm at 110 characters)
>     
> https://www.phoronix.com/scan.php?page=news_item&px=Linux-Kernel-Deprecates-
> 80-Col

sigh there's some severe problems with internet connectivity (much more
sophisticated than a DDOS attack), i still cannot get access to the
majority of news websites.  slashdot, theregister, phoronix, zoom, they're
all timing out (!)

if you can see this page https://slashdot.org/~lkcl then you'll see my
comments on this one.  summary: it's a serious mistake.  this screenshot
shows why: https://libre-soc.org/HDL_workflow/640x-2020-01-24_11-56.png

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


More information about the libre-riscv-dev mailing list