| 17:54.48 | *** join/#debianppc meowray (znc@sjtug/member/MaskRay) |
| 17:56.36 | meowray | i have access to a machine which only has ipv6. how can i update packages? `sudo apt-get -o Acquire::ForceIPv6=true update` => Err:1 http://ftp.us.debian.org/debian buster InRelease |
| 17:56.46 | meowray | <PROTECTED> |
| 17:57.14 | meowray | 100.91.* is in a private network. i don't understand why ftp.us.debian.org is resolved to the address |
| 17:57.32 | meowray | from the output of `host ftp.us.debian.org`, it looks normal |
| 18:04.34 | psydroid | is there even a buster release? and for which architecture? |
| 18:17.12 | meowray | ppc64el |
| 18:35.28 | scientes | meowray, did you never get minicloud access? |
| 18:35.37 | scientes | are are trying to use Integricloud? |
| 18:36.10 | scientes | meowray, I met a woman working on lld at LLVM Dev Mtg that didn't know you |
| 18:36.23 | scientes | *for Google |
| 18:48.47 | meowray | scientes: Amy Huang on lld/COFF? i hadn't seen her before dev mtg |
| 18:49.34 | meowray | i don't know Integricloud.. |
| 18:51.23 | scientes | just wondering why you only have ipv6 |
| 18:51.45 | meowray | i was not in the US before 10/21/2019, ... so |
| 18:51.57 | scientes | meowray, yes, pretty sure it was Amy. I gave her your IRC nick |
| 18:52.12 | scientes | Oh, where are you from? |
| 18:53.15 | meowray | China, and stuck in China for 10 months due to administrative processing |
| 18:53.47 | scientes | on the chinese side or US side? |
| 18:54.29 | scientes | *or the google side |
| 18:54.38 | meowray | definitely us side.. |
| 18:54.54 | scientes | I ask because many chinese cannot get passports |
| 18:56.00 | meowray | i met Rui Ueyama, Michael Spencer, Sean Silva at llvm dev mtg. i think most other lld devs don't know me |
| 18:56.17 | scientes | which? |
| 18:56.20 | scientes | llvm euro? |
| 18:56.45 | meowray | the San Jose one |
| 18:56.56 | scientes | 2018? |
| 18:57.07 | meowray | 10/22/2019 |
| 18:57.12 | scientes | I went in 2013 in San Francisco, and it has gotten MUCH bigger |
| 18:57.14 | scientes | well shit |
| 18:57.16 | scientes | I was there too |
| 18:57.22 | scientes | I gave a talk |
| 18:57.54 | meowray | It was in the evening on the second day and many ppl (including me) had left .. |
| 18:58.54 | scientes | well, it wasn't too exciting either |
| 18:59.06 | scientes | but I guess being at the end doesn't help |
| 19:00.40 | meowray | bad time slot you got |
| 19:01.28 | scientes | ahhh ok |
| 19:01.47 | scientes | I was thinking about doing a overlay level in llvm so that I can put relative pointers into constant data |
| 19:01.51 | scientes | its kinda hacky |
| 19:02.05 | scientes | based on a reference to a callbr instruction |
| 19:02.27 | meowray | "overlay level"? |
| 19:02.29 | scientes | I don't think it would make sense to use a linker |
| 19:02.38 | scientes | well, you can't resolve the values until codegen |
| 19:02.46 | scientes | but they have memory locations |
| 19:02.53 | scientes | so again, you kinda have to hack this in |
| 19:03.31 | scientes | so it would be a constant array, with certain parts of memory being resolved at codegen time |
| 19:04.28 | meowray | how can that part be resolved at codegen time? doesn't that need relocations? |
| 19:04.48 | scientes | <PROTECTED> |
| 19:04.57 | scientes | cause you don't have the location of the jump instruction until link time |
| 19:05.03 | scientes | relative to the data |
| 19:05.12 | scientes | well no, its all in the same function |
| 19:05.16 | scientes | so you don't need a relocation |
| 19:05.36 | meowray | PC relative values |
| 19:05.44 | scientes | yes |
| 19:06.45 | scientes | the data structure I was thinking about was a binary search of variable length keys, with pc relative values, so its a bisection jump table |
| 19:06.55 | scientes | but this would also allow all the jump table stuff to be put in the middle-end |
| 19:07.14 | meowray | i decide to learn some ldc for codegen stuff |
| 19:07.45 | scientes | unless you ever do a function dispatch jump table with absolute jumps |
| 19:07.52 | scientes | but that would be easier to implement this way |
| 19:09.28 | meowray | but how can you decide they are constants before lowering to the MC layer |
| 19:09.35 | scientes | switch statements would still reach the backend because the backend handles branch weighing |
| 19:11.18 | meowray | did you attend the talk about Propeller (post-link optimization) |
| 19:11.46 | scientes | you have a byte array, and that would have a reference to a call instruction, and then each lazy value overlayed on the array would have a reference to the label being branced to by that callbr instruction |
| 19:12.07 | scientes | meowray, yes I did, patching machine code like that is pretty hard-core |
| 19:12.48 | meowray | your "lazy value" sounds like GNU ifunc |
| 19:13.24 | scientes | its resolved at codegen time |
| 19:16.16 | scientes | I think my idea is pretty ugly, but I don't see another way to do it, and it is best in the middle-end |
| 19:16.50 | scientes | as PC-relative instructions tend to look the same |
| 19:17.04 | scientes | *relative jumps |
| 19:18.39 | meowray | why do you want to do it in the middle-end? optimize expressions that use the lazy value? |
| 19:34.23 | scientes | ahh, the target-independent part of the back-end would probably be better |