Edward Z. Yang

@ezyang

FAAIRy and part-time adjunct faculty at NYU. Ask me about Haskell, PyTorch, or programming languages.

Vrijeme pridruživanja: svibanj 2008.

Tweetovi

Blokirali ste korisnika/cu @ezyang

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @ezyang

  1. proslijedio/la je Tweet
    prije 5 sati

    And here is its fancy type of matrix multiplication.

    Prikaži ovu nit
    Poništi
  2. proslijedio/la je Tweet
    prije 5 sati

    Here is a fancy type of matrices. (x <- xs is the type of pointers into a list.)

    Poništi
  3. proslijedio/la je Tweet
    prije 23 sata

    I wrote some thoughts on software performance:

    Poništi
  4. 29. sij

    JAX-style vmap written in less than a dozen lines of Haskell

    Poništi
  5. proslijedio/la je Tweet
    15. sij
    Odgovor korisniku/ci

    we're hiring for a QA eng at actually

    Poništi
  6. 9. sij

    If I want to do NaNoWriMo with GPT-2 how would I set it up

    Poništi
  7. 6. sij

    Running some stats and feeling about hack'slashing the SQL queries. Sooo much copy pasting.

    Poništi
  8. 6. sij

    That feeling when you're working on perf and you're like, "Wait, that was THIS half?!"

    Poništi
  9. 4. sij

    In as little time as possible, you want to extract the digits in an image like into machine readable form. What would you do?

    Poništi
  10. proslijedio/la je Tweet
    12. pro 2019.
    Odgovor korisnicima

    There are many opaque ways local bindings can mess with C++ programs that rely on symbol merging for correctness. RTTI is just one case of a pattern people use in regular code.

    Poništi
  11. proslijedio/la je Tweet
    12. pro 2019.
    Odgovor korisniku/ci

    I believe that's a known issue with ubsan, and the correctness depends on the stdlibc++'s implementation of type comparison. Some do pointer comparison, other look at a mangled string.

    Poništi
  12. 12. pro 2019.

    What about if your program does not use RTTI at all?

    Prikaži ovu nit
    Poništi
  13. 12. pro 2019.

    Fuck weak symbols

    Poništi
  14. 12. pro 2019.

    UBSAN is complaining that an object passed between two dynamic libraries is not the same because you didn't load with RTLD_GLOBAL and so they got distinct typeinfo

    Prikaži ovu nit
    Poništi
  15. proslijedio/la je Tweet
    12. pro 2019.
    Odgovor korisniku/ci

    You can maybe use LD_AUDIT facility () if it doesn’t interfere with what you want to measure. The example in the man page looks to include printing resolved symbol addresses.

    Poništi
  16. 11. pro 2019.

    I'm trying to debug an RTLD_LOCAL related UBSAN problem. Is there a way to get LD_DEBUG to print the addresses of symbols after it binds them?

    Poništi
  17. proslijedio/la je Tweet
    11. pro 2019.

    How do we solve the problem of communication breakdown between management and engineering? In business loans are a predictable tool, so "technical debt" sounds OK to business people. Technical debt is a poor metaphor. Technical debt is escalating risk - at

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    11. pro 2019.

    I missed this post on modern GCs from Mike Hearn. A must-read to understand the state of the art in low-latency GC:

    Poništi
  19. proslijedio/la je Tweet
    11. pro 2019.
    Odgovor korisniku/ci

    "Transformation", or perhaps "partial transformation" if the purpose of the sentinels is to indicate elements excluded from the domain.

    Poništi
  20. proslijedio/la je Tweet
    10. pro 2019.

    Helping my students to see beyond equations.

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·