| ghostmansd[m] | lkcl, let's speak in terms of tests or something reproducible | 00:20 |
|---|---|---|
| ghostmansd[m] | Please post the command to launch so that I can check | 00:20 |
| ghostmansd[m] | "this breaks files in media/" vs "this breaks test X launched by command Y" makes a huge difference :-) | 00:22 |
| ghostmansd[m] | I'll take a look today. The changes are inevitable anyway, because otherwise SVP64.py is acting inconsistently and, I'd say, somewhat surprisingly. | 00:23 |
| ghostmansd[m] | Also, I'd like the test to be as simple as `cmdX A.s B0.o` vs `cmdY A.s B1.o` incantations. Yes sure A.s can (and most likely should) be generated. | 00:26 |
| ghostmansd[m] | And, even before this, I'll add binutils tests. | 00:26 |
| lkcl | ghostmansd[m], https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=media/README.txt;hb=HEAD | 00:58 |
| lkcl | you should find that the audio/mp3/mp3_0.sh tests work, the mp3_1.sh markos is working on | 00:58 |
| lkcl | the trail should be followable from the README.txt | 00:59 |
| lkcl | when converting this | 01:00 |
| lkcl | https://git.libre-soc.org/?p=openpower-isa.git;a=blob;f=media/audio/mp3/mp3_0_apply_window_float_basicsv.s;hb=HEAD | 01:00 |
| lkcl | lines 63 64 66 67 72 73 75 76 96-98 101-104 115-117 120-123 and 143 144 146 and 147 will all need the "# " removed in front of them | 01:02 |
| lkcl | if it's not entirely obvious let me know ok? | 01:11 |
| ghostmansd[m] | lkcl, star | 06:49 |
| programmerjake | lkcl, offered to pay jix: https://github.com/YosysHQ/yosys/pull/3319#issuecomment-1148844920 | 16:52 |
| programmerjake | i'm thinking maybe I could pay him out of the amount I'd get from nlnet | 16:53 |
| programmerjake | rather than jix having to go through all the nlnet obligations of being part of libre-soc | 16:54 |
| programmerjake | was wondering if you think we should offer him more than eur 200... | 17:23 |
| ghostmansd | Each and every time I want to re-create the environment it's a f*cking pain. All these pinmux errors or some missing crap like sphinx addons or anything else. Any ideas on making it convenient? | 18:01 |
| ghostmansd | Seriously, with binutils, I never had any issues, it's sweet and simple. With _any_ attempts to have a fresh and sweet copy of say openpower-isa it's a pain, even if I clone all repositories. | 18:02 |
| ghostmansd | The actual reason why I had to clone the repo again is stupid, but, still, it's a real f*cking pain compared to binutils. | 18:03 |
| ghostmansd | lkcl I know your opinion on docker, but perhaps there are alternatives? I simply would like to have an option to take an environment and work with it. | 18:04 |
| programmerjake | once we finish getting CI up and running, we could have a git repo with submodules set to the latest known-working commits...that might help | 18:04 |
| ghostmansd | That'd be awesome. Cannot wait! | 18:05 |
| ghostmansd | Thanks programmerjake! | 18:05 |
| ghostmansd | Seriously, for people who don't use venv on everyday basis it's difficult. Also, relying on bash scripts without even error handling... I'd say it's not the sanest option. | 18:06 |
| ghostmansd | All this to bit shit out of some shell scripts working on mp3 files. Madness. | 18:07 |
| lkcl | programmerjake, i saw, no that's a good idea (to pay directly). more than EUR 200 is perfectly fine with me | 18:10 |
| lkcl | ghostmansd, the dev-env-setup scripts are supposed to "just work" | 18:11 |
| lkcl | plus you should only need openpower-isa and really nothing else. | 18:11 |
| lkcl | not soc, not ieee754fp, not c4m-jtag, nothing | 18:11 |
| ghostmansd | yehyeh, except that you have to keep in mind that there's venv version for talos1 | 18:12 |
| programmerjake | lkcl, was asking because I don't have a good idea how much is appropriate... | 18:12 |
| lkcl | .... ermm.... ermermerm.. | 18:12 |
| ghostmansd | and that you did `python setup.py develop` | 18:12 |
| lkcl | programmerjake, i'd say likely up to EUR 500 would be reasonable? | 18:12 |
| ghostmansd | and any other crap inbetween :-) | 18:12 |
| lkcl | ghostmansd, the schroot script should still work on the talos1 workstation | 18:13 |
| ghostmansd | venv works I think | 18:13 |
| lkcl | someone tested the schroot dev-env-setup scripts a while back and... | 18:13 |
| lkcl | except nobody i know of uses venv | 18:13 |
| ghostmansd | one of the parts that confused me a lot is when I launched `make wget` in the root (openpower-isa) | 18:14 |
| lkcl | right | 18:14 |
| lkcl | ok | 18:14 |
| ghostmansd | ...and not in openpower-isa/media | 18:14 |
| lkcl | so that downloads a large file containing MPEG audio that is completely inappropriate to put into a git repository | 18:14 |
| ghostmansd | then sphinx started beating the shit out of me | 18:14 |
| lkcl | it's binary-formatted data and comes from ffmpeg | 18:14 |
| lkcl | is this in the "main" (non-chrooted) repos? | 18:15 |
| ghostmansd | yeah now that I do it in media/ it seems to work | 18:15 |
| programmerjake | i always use a venv for libre-soc stuff | 18:16 |
| ghostmansd | hm, on talos I think I don't have chroot | 18:16 |
| lkcl | ghostmansd: sudo bash | 18:17 |
| lkcl | no password required | 18:17 |
| ghostmansd | > All Libre-SOC dev dependenices should now be installed. | 18:17 |
| ghostmansd | OK should work on debian | 18:18 |
| lkcl | btw, please *under no circumstances* install arbitrary software "apt-get install of anything" on the talos1 workstation | 18:18 |
| lkcl | it had to be set up using a bare-minimum debian/testing netboot.iso loaded to its BMC over the internet | 18:18 |
| lkcl | you really, *really* need to use schroots | 18:18 |
| lkcl | hmmmm.... i need to do some base installs first, for you | 18:19 |
| lkcl | ok all good. debootstrap and schroot installed | 18:20 |
| lkcl | so now it's safe for you to first run mk-deb-chroot {some_dev_name_of_some_kind} | 18:22 |
| lkcl | then "schroot -c {the_name_you_gave_the_schroot}" | 18:23 |
| lkcl | then *re*-clone the dev-env-setup (or cp -aux it in there manually, i do that all the time) | 18:23 |
| lkcl | and then run whatever you like. | 18:23 |
| ghostmansd | ok launching mk-deb-chroot | 18:24 |
| ghostmansd | $(make tests) is running in media/ at my debian VM | 18:24 |
| ghostmansd | takes a while | 18:24 |
| lkcl | yes, it will :) | 18:25 |
| lkcl | on a basic laptop, ISACaller can only do about 250 instructions per second | 18:25 |
| lkcl | on my laptop which is 4.8 ghz with an NVMe SSD it's more like 2,000 | 18:25 |
| ghostmansd | iirc we had a task on new simulator, eh? | 18:26 |
| ghostmansd | in C | 18:26 |
| lkcl | uhhuhn | 18:27 |
| lkcl | which should be an astounding 300,000 instructions per second (about 1/4 of native CPU speed) | 18:27 |
| lkcl | but is limited to user-space execution only | 18:27 |
| lkcl | btw now you've got that trundling along, and it's not producing any errors, there is a short-cut | 18:29 |
| lkcl | run the .sh command manually | 18:29 |
| lkcl | 1 sec... | 18:29 |
| lkcl | cd media | 18:30 |
| lkcl | ./audio/mp3/mp3_0.sh 0 /tmp/out0 | 18:30 |
| lkcl | and if that works you're golden | 18:30 |
| lkcl | programmerjake, there's quite a lot available still EUR 6550 unallocated on the parent, so do let jix know that there's plenty more, if interested to help more https://bugs.libre-soc.org/show_bug.cgi?id=196 | 18:37 |
| lkcl | ghostmansd, btw feel free to use the "wall" command to annoy me back, on talos1 :) | 18:39 |
| ghostmansd | lkcl, ok :-) | 18:40 |
| ghostmansd | lkcl, I don't really get what I should do with $(./audio/mp3/mp3_0.sh 0 /tmp/out0) | 18:57 |
| ghostmansd | $(cmp /tmp/out0 data/audio/mp3/mp3_0_data/samples0) yields 0 as exit code | 18:57 |
| ghostmansd | aha OK, at least mp3_1.sh output differ, at least something to work with | 18:59 |
| ghostmansd | I re-read what you wrote, mp3_1.sh is a known issue, markos is working on it | 19:04 |
| ghostmansd | sorry, I'm way too slow today :-) | 19:04 |
| ghostmansd | OK, if I got it correctly, you think that after my changes we now have bad media/audio/mp3/mp3_0_apply_window_float_basicsv.s.sv file, right? | 19:12 |
| ghostmansd | yeah I see wildcard target which produce .bin files | 19:14 |
| ghostmansd | (if only everyone knew how I hate wildcards and other non-portable GNU make extensions) | 19:15 |
| ghostmansd | OK I'll check what's going on here | 19:15 |
| ghostmansd | lkcl, I don't observe differences upon running this in media/: $(rm -f audio/mp3/*.s.* audio/mp3/*.bin && make audio/mp3/mp3_0_apply_window_float.bin) | 19:22 |
| ghostmansd | https://pastebin.com/qvHDPYjr | 19:24 |
| ghostmansd | that's how I checked it | 19:24 |
| ghostmansd | could you, please, go into more details? | 19:24 |
| *** kylel1 is now known as kylel | 20:03 | |
| lkcl | ghostmansd, what that script does is it runs the c-version of the ffmpeg_apply_window_float.c program | 21:36 |
| lkcl | producing some output | 21:36 |
| lkcl | and it then compiles the *svp64-assembler* variant of the *exact* same program | 21:37 |
| lkcl | producing some output | 21:37 |
| lkcl | like i said, you have to "unpack" this somewhat | 21:37 |
| lkcl | examine the contents of the script | 21:38 |
| lkcl | are you familiar with Makefiles? | 21:40 |
| lkcl | basically by running "make tests" the four binaries have been constructed | 21:41 |
| lkcl | (the two we care about are for mp3_0) | 21:41 |
| lkcl | one binary is the c version | 21:41 |
| lkcl | one binary is the assembler version | 21:41 |
| lkcl | now look in audio/mp3/Makefile | 21:42 |
| lkcl | you see | 21:42 |
| lkcl | mp3_1_test: mp3_1_main.c imdct36_standalone.c | 21:42 |
| lkcl | gcc -o mp3_1_test mp3_1_main.c imdct36_standalone.c | 21:42 |
| lkcl | that's the c version | 21:42 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!