IRC channel logs

<stikonas>fossy: maybe you want to rebase my guile branch while I'm busy with risc-v work?

<stikonas>if not, I can do it at some point later...

<fossy>stikonas: happily

<stikonas>it needs /usr change...

<stikonas>and I guess has to be moved to sysc

<stikonas> https://github.com/stikonas/live-bootstrap/tree/guile

<stikonas>other than that I don't expect significant difficulties...

<stikonas>a bit annoying though that we couldn't bootstrap autogen (which is what I intended to use guile for)...

<fossy>how is riscv going btw

<stikonas>fossy: writing (I guess) first ever riscv C compiler in riscv assembly

<stikonas>I guess about 20% done

<stikonas>well, just finished M0 recently

<fossy>thats good progress!

<stikonas>so quite a bit nicer now

<stikonas>no more hex encoding

<stikonas>which gets tedious really quickly

<stikonas>well, at the moment I'm just porting cc_amd64 to riscv

<stikonas>there isn't that much difference between different arches

<fossy>as a starting point i guess?

<stikonas>maybe 30 or so lines

<stikonas>yeah

<stikonas>then I can just adjust which text I emit into M1 file

<stikonas>anyway, I've got tokenization working now and I think global variables are detected and emitted into labels

<stikonas>still need to implement all other C features

<stikonas>but riscv binaries are quite fast in qemu

<stikonas>since I found a good workaround

<stikonas>hex'es in other arches should probably be fixed too..

<stikonas>I tried running aarch64 bootstrap in qemu, and it took maybe 15 minutes

<stikonas>most of it was running hex2

<fossy>what was the workaround

<stikonas>basically avoid using global variables

<stikonas>(memory locations after program .text)

<stikonas>and I think without ELF section header those cause mprotect syscalls in qemu

<stikonas>moving variables to either stack or CPU registers (risc has plenty of them) seems to run significantly faster

<stikonas>I get 8x speedups or so

<stikonas>and hex2 suffers most from this problem

<stikonas>since everything is read and written byte by byte

<stikonas>this workaround also let me significantly reduce size of hex0

<stikonas>it went from something like 48x bytes to 392

<xentrac>yeah, that was amazing!

<xentrac>it inspired me to write dernocua/text/term-rewriting-micro-interpreter.md. I'm currently trying to debug it, but currently the interpreter is over a kilobyte

<fossy>oh thats smart, stikonas

<ekaitz>i'm probably late to the party but did you guys see this?

<ekaitz> https://openpowerfoundation.org/libre-soc-180nm-power-isa-asic-submitted-to-imec-for-fabrication/

<xentrac>cute

<oriansj>ekaitz: yeah I posted about https://libre-soc.org/ earlier this year. I look forward to it and the release of its design files. As I like having more options for improving control over our own computing.

<oriansj>when they add MMU support and become able to boot a Linux, it'll become much more useful.

IRC channel logs

2021-09-15.log