Twitter | Pretraživanje | |
Giovanni Petrantoni
I like to research, create and sometimes destroy. From Italy (Sicily), I live in Japan (10 years+).
1.142
Tweetovi
1.012
Pratim
446
Osobe koje vas prate
Tweetovi
Giovanni Petrantoni 20 h
Btw, I commit a lot on / and I don’t care! It's not a show off, it's just being proactive by using continuous integration done right to find issues for me ahead of time!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 4. velj
Sounds like you guys need stuff like ! If I only had some support to keep developing it! But everyone seems stuck to bad tools and bad pipelines toasting their productivity...
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 1. velj
Odgovor korisniku/ci @trav_downs
Well, I'm not surprised at all, doing any syscall is basically like entering the unknown in any environment. But threadpools, custom user-land schedulers and coroutines are there for the rescue!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni proslijedio/la je tweet
will knight 30. sij
Wow.... and while using Windows 2000 by the looks of it.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 31. sij
Odgovor korisniku/ci @pcwalton
How about just stop rewriting and start writing new cool stuff! I love c++, rust, nim, lisp, etc. Each of them have uses and each of them is the right tool for certain jobs... seriously stop treating languages like football teams in terms of adoration! Don't be possessed!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 31. sij
This really shows that users experience and ease of use is indeed above all even in this field. Lots of lessons learned for new frameworks too! Also shows that users really don’t care about technical details and internal stuff (where I see other frameworks honestly superior).
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 28. sij
maybe this can help with the “fast_template”? Wonder if they could coexist. Will check if I have time one day!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 28. sij
Wow, I was looking for a, even local, hackmd like replacement! Had no idea you were working on this as well!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 24. sij
If you don’t know it already comby by is probably the best tool for code refactor you need. It saved me many hours of regex/(x paid tool) etc...
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 23. sij
Odgovor korisniku/ci @furan
I completely understand that, I have not so much verilog experience but with synthesis you have quite long "synthesis" time to wait to catch results it seemed to me. Where a c++ compiler can be extremely fast to iterate a small hot path many times.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 23. sij
Odgovor korisniku/ci @furan
That's also true, and in that case probably plain old style C is superior or Rust too. But when I dig into assembly to optimize some hot path iteratively control is really all I need due to the brute force approach. But yeah sounds almost like the last resort :)
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 23. sij
Odgovor korisniku/ci @furan
True but at same time this is probably the reason I like c/c++ , because you still have control over it or rather you can still workaround. Many new languages are nice and all but offer little control over those things, and that really matters when you try to optimize for perf.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 23. sij
Odgovor korisniku/ci @trav_downs
Ah! I changed a lot of hot path code lately to be idiomatic, but now I feel like a idiot... I swear it had to do with some measuring I did in valgrind and for some reason the memset version was always jumping somewhere else and I didn't like that. I think the code was "-O3" too.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 23. sij
Odgovor korisniku/ci @Jeffinatorator @trav_downs
This is probably the biggest piece of knowledge from this investigation btw! Had no idea either that it helped the compiler.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 22. sij
Odgovor korisniku/ci @science_dot @trav_downs @nickreal03
Well in that case the expansion might actually take a lot of lines/text storage :) lot of them are repetitions left and right. With the flag I posted the compiler will actually try to inline anything it can unless explicitly marked "no inline" very aggressively.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 22. sij
Odgovor korisniku/ci @trav_downs @science_dot @nickreal03
Lol try add "-mllvm -inline-threshold=100000" as well :) That said I use it when I don't care about footprint and indeed makes things fast! Got plenty of memory nowadays most of times!
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 21. sij
Odgovor korisniku/ci @F_Vaggi @rockikamil
Indeed! One of the reasons I don't like using those libraries is the over aggressive bad scheduling ( for my use cases at least, works maybe somehow for python due to GIL.. ). I know it's needed but delegating your CPU completely to a library is bad design and very limiting.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni 21. sij
Odgovor korisniku/ci @trav_downs
Interesting investigation as usual! Recently I replaced many “memset” with simple “= {}” in some codebase, I did actually valgrind and looked better… This article makes me wonder again tho. Anyway I end up reading your stuff always when I’m out drinking… gonna forget about it.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni proslijedio/la je tweet
Kamil Rocki 20. sij
No BLAS? No problem. This is the first post of the series implementing various algorithms in ANSI C with 0 dependencies for the DYI people. Stay tuned. Here' s #1: A 2-layer MLP getting beating OpenBLAS' performance and getting over 95 % in under 5s.
Reply Retweet Označi sa "sviđa mi se"
Giovanni Petrantoni proslijedio/la je tweet
Kamil Rocki 21. sij
One of the most important debugging technique of my life - numerical gradient check. This is a short intro and working piece of some for two models sharing the same grad-check scaffolding: a classifier and an Autoencoder. Pure Numpy!
Reply Retweet Označi sa "sviđa mi se"