| programmerjake | oh, x86 also has a matrix extension: AMX | 00:04 |
|---|---|---|
| programmerjake | https://en.wikipedia.org/wiki/Advanced_Matrix_Extensions | 00:04 |
| programmerjake | that needs to be added to the isa comparison table | 00:05 |
| lkcl | ngggh ok :) | 02:34 |
| lkcl | it's a highly-specialist "tiling" unit. | 02:37 |
| lkcl | the more this table is filled out the more i am having a "surreal" moment that the entire world's mainstream ISA community, led primarily by ARM and Intel, is engaging in some sort of mass-insanity | 02:41 |
| programmerjake | you might be interested to know that the Rust Foundation is planning a full rewrite of their trademark policy, to happen around Q3: https://foundation.rust-lang.org/news/2022-07-27-what-has-the-foundation-been-up-to/ | 03:48 |
| lkcl | about fucking time | 04:26 |
| lkcl | i've had to have some extremely distressing conversations with a lot of people | 04:26 |
| lkcl | their Trademark policy basically is about the biggest fuck-up of any Trademark License in FOSS history | 04:28 |
| lkcl | worse than even that which caused debian to create iceweasel | 04:29 |
| lkcl | discussions are underway within the GNU Project to begin the process of removing the words "rust" and "cargo" from distributions | 04:30 |
| programmerjake | they are taking public input -- hopefully they fix their policy satisfactorily | 04:44 |
| programmerjake | lkcl, imho we should use markdown footnotes for the isa comparison table, rather than the DIY approach you've taken...they're supported by both the pdf conversion and by ikiwiki | 05:26 |
| programmerjake | they allow you to use names when writing markdown, and they auto-number all the footnotes, so you won't have to manually update all the numbers | 05:27 |
| programmerjake | see the Footnotes section in https://www.markdownguide.org/extended-syntax/ | 05:28 |
| markos | lkcl, I knew many of those restrictions, sadly there are more | 05:30 |
| markos | I find myself disliking SVE2 more and more | 05:30 |
| programmerjake | I changed the comparison table to use footnotes and changed texmunge.py to make the tex work with multiple references to the same footnote. I managed to cram it all on one page by reducing the font size of the footnotes somewhat... | 10:30 |
| programmerjake | https://git.libre-soc.org/?p=libreriscv.git;a=commitdiff;h=93b706bca64112a29a75cbb8e0f1a3fea80e5e2a | 10:31 |
| lkcl | yeah the table is massive, i have an xterm open at 300x65 and it's not fitting any more | 12:11 |
| programmerjake | imho it should be split into 2 tables --- isa and half the columns in 1 table, isa and the other columns in the other | 12:13 |
| lkcl | we were specifically asked to give only 2 pages. one of those is the executive summary, the other is the comparison table. | 12:16 |
| programmerjake | ah, ok. though you can always have 2 tables on 1 page... | 12:18 |
| programmerjake | also imho the footnotes are waay to verbose...2/3 page of footnotes, 1/3 page table | 12:19 |
| programmerjake | the 128-bit column can be dropped, it's inaccurate anyway, since most simd isas don't really support vectors of 128-bit elements, there may be a few instructions operating on 128-bit elements but it's mostly just a 128-bit vector with only <= 64-bit elements | 12:22 |
| programmerjake | ppc and rvv are the only ones i'm aware of with full 128-bit element support | 12:24 |
| programmerjake | iirc arm and x86 don't even have a u128xN add instruction | 12:25 |
| markos | the only thing I'm aware of for arm is expand-to-128-bit, but that doesn't really count as 128-bit support | 12:26 |
| programmerjake | similarly x86 has some 128-bit shifts iirc | 12:26 |
| programmerjake | but no add, sub, mul, compare, etc | 12:27 |
| lkcl | ok got it got it down to under 250 characters. was looking for a way to do "proper" footnotes, thank you for finding it. | 12:34 |
| lkcl | urrr really, no 128-bit? dang | 12:34 |
| lkcl | okaaay | 12:34 |
| lkcl | that's actually good | 12:34 |
| lkcl | nicely done with the footnote-munging | 12:41 |
| programmerjake | all the columns where svp64 is yes and everything else is no can be merged -- name the column additional features or something, leaving the full list to the footnote | 12:41 |
| programmerjake | yeah, texmunge took like 6hr to get working, turns out tex footnotes are kinda fragile and i'm not that familiar with them | 12:42 |
| lkcl | pandoc filters are pretty damn obtuse, and latex is as well | 12:48 |
| lkcl | whyy are we doing so many obtuse things all at once? :) | 12:48 |
| programmerjake | well, just be glad we don't have to write a ttf parser/renderer too...bezier curves anyone? | 12:49 |
| * lkcl starts shaking at the thought | 12:49 | |
| lkcl | the story behind tex is pretty hilarious. | 12:50 |
| programmerjake | though i have researched font rendering quite a bit... | 12:50 |
| programmerjake | still trying to figure out text shaping -- harfbuzz | 12:51 |
| programmerjake | nearly 5am here -- gn | 12:53 |
| lkcl | night. thanks for your help | 12:55 |
| programmerjake | do note adding more footnotes or rows will likely force it to take 2 pages in the pdf, it barely fit on 1 when I tried https://git.libre-soc.org/?p=libreriscv.git;a=commit;h=13c7963d8ee00b4b448add28d4d9fee9924c63ff | 13:07 |
| lkcl | i made space by ensuring that some of the table entries fitted onto 2 lines instead of 3 or 1 instead of 2. | 15:14 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!