00:05 <rjo> using optree for multiplication it is a bit uncommon but i did like the tree implementation. it a makes the synthesizing/balancing job quite a bit easier and is very idiomatic. maybe reduce_tree(...) makes the name more pythonic.

00:21 <sb0> rjo, do you have a strong need/opinion in favor of optree?

00:21 <sb0> nothing in migen/misoc/artiq uses it for multiplication

01:25 <rjo> i would keep it. maybe change the documentation and/or the name to contrast/compare it to functools.reduce()

05:08 <GitHub32> [migen] sbourdeauducq pushed 5 new commits to new: http://git.io/vZEZF

05:08 <GitHub32> migen/new 714ae43 Sebastien Bourdeauducq: fhdl: remove features new simulator won't use

05:08 <GitHub32> migen/new 49ef182 Sebastien Bourdeauducq: fhdl/tools: add input lister

05:08 <GitHub32> migen/new f9849fb Sebastien Bourdeauducq: style

05:09 <travis-ci> m-labs/migen#79 (new - fd98621 : Sebastien Bourdeauducq): The build has errored.

05:09 <travis-ci> Build details : https://travis-ci.org/m-labs/migen/builds/79789240

07:00 <cr1901_modern> Finally found the cominatorial loop in my Migen code... only took me nearly two weeks (on and off, of course). Funny, viewing waveforms didn't help this time. Simulator's display_run=True option saved the day here. So thanks whoever implemented that lol

09:51 <terpstra> ysionneau, how much effort would it be for you if your netbsd port had to use Sv32 <http://riscv.org/spec/riscv-privileged-spec-v1.7.pdf> instead of software refill, even on an LM32?

09:52 <ysionneau> well, let me process that :)

09:52 <terpstra> it's section 4.5

09:52 <ysionneau> ah thx

09:58 <ysionneau> I'll get back to you on this, do you have an email address where I can contact you?

09:59 <terpstra> wesley@terpstra.ca

09:59 <terpstra> (not my work email address, please :)

10:00 <terpstra> it's definitely a bit quirky. it has 34-bit physical addresses from 32-bit virtual.. and 4MB + 4KB page mappings

10:56 <ysionneau> I would say it would simplify the low level part of the code, you don't need to refill tlb anymore, you just need to update the page table in the OS and give the pointer to the HW

10:56 <ysionneau> but one would need to add support for Sv32 to qemu for instance

10:57 <ysionneau> and also to actually add support for such an mmu to lm32 :)

10:57 <ysionneau> I wonder what those 2 extra physical bits are for

10:59 <ysionneau> hmm I don't see any asid field in the pte

11:03 <ysionneau> I don't understand how the mmu knows is some pte is "active" or not for current asid/address space

11:03 <ysionneau> -is some+if some

11:18 <terpstra> i intend to use one of the high physical bits to distinguish between cacheable and uncacheable memory

11:19 <terpstra> well, i wasn't planning on backporting Sv32 to the LM32, but rather supporting LM32 on my processor which implements Sv32 :)

11:19 <terpstra> there is a CSR which says the root of the page table

11:20 <terpstra> so, you can tag the loaded PTE entries in the TLB to refer to that particular CSR value

11:20 <terpstra> ie: you don't need to actually blow away all the TLB entries, when someone updates the CSR indicating the root of the page table

11:23 <ysionneau> oh, ok it's like this

11:25 <ysionneau> with asid you could just evict tlb entries when needed, and not just save the life of global ones

11:26 <terpstra> you can still do this

11:26 <ysionneau> oh, or maybe the csr value is like the asid ?

11:26 <terpstra> yes

11:27 <terpstra> when someone writes to the CSR, assign a small tag that means that value

11:27 <terpstra> use that tag on the TLB entries

11:27 <ysionneau> I see, this is totally implementation specific, right? Programmer don't have to know this

11:27 <terpstra> maybe even index the TLB as hash(tag,virtual bits) for a bit better utilization

11:27 <terpstra> exactly

11:28 <terpstra> programmers just fill in the PTE table

11:28 <terpstra> then write to the CSR when they want a context switch

11:28 <ysionneau> yep got it

11:28 <ysionneau> sounds nice!

11:28 <terpstra> how/if the processor optimizes TLB eviction is an implementation detail

19:45 <cr1901_modern> software refill? I THINK it's self explanatory...

21:37 <cr1901_modern> whitequark: re https://twitter.com/whitequark/status/642188969426427904, Urbit may be something similar to what you're looking for from a technical point of view. >>

21:37 <cr1901_modern> Personally, any technical merit the project has IMHO is marred by the creators political views, but YMMV.