04:37 <cr1901_modern> sb0: I wish to add little-endian CSRs. I think the easiest way to do this is to add a new member to _CSRBase that allows a user to specify the endianness. >>

04:38 <cr1901_modern> Anything that subclasses _CompoundCSR should do something like: if self.endianness = LITTLE, for i in range(nwords) else for i in reversed(range(nwords))

04:38 <cr1901_modern> Does that sound good?

04:58 <cr1901_modern> Scratch part of that. I shouldn't need to add anything to _CSRBase, since CSR regs are 8 bits.

05:02 <sb0> cr1901_modern, what do you need LE for?

05:02 <sb0> this whole LE/BE problem is a waste of time

05:06 <cr1901_modern> I want to use Migen for other projects besides LM32/Milkymist. If I can almost fully replace Verilog for writing synthesizable code, that would make my life easier

05:07 <cr1901_modern> Why do you think it's a waste of time?

05:08 <cr1901_modern> Yes, of course I can just make my own custom register bus if you don't want me to screw with CSRs. Just though I'd try avoiding reinventing the wheel

05:21 <cr1901_modern> sb0: I think I get where you're coming from. CSR is meant to be primarily used for MiGen/MiSoc-related projects only, correct?

06:23 <sb0> no, it's because every feature added is a liability

06:24 <sb0> have to maintain it, makes the code complex and harder to read, may break when other features are added, provides hiding space for bugs, etc.

06:24 <cr1901_modern> Okay, then. I'll use LE for my own personal projects then. Not a big deal.

06:25 <cr1901_modern> (i.e. I'll just make my own personal module for it)

06:25 <sb0> why not have the software functions access the CSR bus in BE?

06:26 <sb0> CSR is slow anyway, the LE->BE conversion won't actually make any difference since the CPU itself is splitting the transfers into 8-bit chunks

06:26 <cr1901_modern> B/c the CPU in question is one of the only CPUs I can think of that's still produced with an 8-bit external data bus AND supports 16-bit xfers

06:27 <cr1901_modern> I didn't think of that

06:27 <sb0> you can also use a 16-bit CSR bus. or wishbone.

06:28 <cr1901_modern> "CPU itself is splitting the transfers into 8-bit chunks" <== Again, I didn't think of that. My fault.

06:29 <sb0> maybe the C code of the access functions could even be the same for LE and BE CPUs....

06:29 <cr1901_modern> It's not difficult to write endian-independent code

06:31 <cr1901_modern> I do a lot of vintage work- for fun/hobby. So let's get that out of the way before ppl make fun of me. >>

06:32 <cr1901_modern> The CPU in question that I wanted to interface is a 6502 variant that can do 16-bit xfers. 6502 is a little-endian processor, 8-bit external bus. So that was the use case

06:33 <cr1901_modern> I'm not trying to justify what I was trying to do- just explaining my thought process. :P

08:12 <GitHub63> [artiq] sbourdeauducq pushed 2 new commits to master: http://git.io/vTTva

08:12 <GitHub63> artiq/master b74b8d5 Sebastien Bourdeauducq: Scheduling TNG

08:12 <GitHub63> artiq/master d8917fd Sebastien Bourdeauducq: Merge branch 'master' of github.com:m-labs/artiq

09:19 <sb0> oh, pyqtgraph comes with docking that does pretty much what i want

09:37 <cr1901_modern> Automated data slicing... that's cool

11:53 <sb0> sturmflut2, you said your emmc working correctly with the android kernel. is that with or without debug_quirks2?

11:54 <sb0> also, did you have the wifi enabled? I suspect some "intel storage cluster" bug that could manifest itself when both the emmc and wifi are in use...

11:54 <sb0> kernels that don't have the emmc bug coincidentally have a sdio bug that makes the wifi driver not recognize the card

11:57 <sb0> and I had the emmc bug kick in once right when I was inserting a sd card. could be a coincidence, but who knows....

11:58 <sb0> those overengineered intel systems are a royal pita

13:24 <sb0> https://github.com/pyqtgraph/pyqtgraph/blob/develop/pyqtgraph/Qt.py#L25

13:24 <sb0> https://github.com/harvimt/quamash/blob/master/quamash/__init__.py#L30

13:24 <sb0> oh, i love this sort of things

13:28 <sb0> and yes, my program is crashing right now because each of those libs is importing a different qt binding

21:29 <whitequark> wheee, a thermocouple feedthrough for $25 and free shipping

21:29 <whitequark> those ebay alerts really pay off