damageboy

@damageboy

Antarctica
Vrijeme pridruživanja: svibanj 2008.

Medijski sadržaj

  1. 1. velj
    Odgovor korisnicima

    I think his is as vague without being misleading as I need it to be?

  2. 1. velj
  3. 31. sij
  4. 27. sij
    Odgovor korisniku/ci

    I'll make VxSort public later today, I'll link there directly but basically:

  5. 16. sij
    Odgovor korisnicima

    I dunno. Ole' grandma had so little growing up. All she could pass us, her beloved grandchildren, were those used up Cray V/A/S/T registers...

  6. 14. sij

    Last road block to publish removed. I'm now much more confident that I have a code generator for bitonic sort. I'll start squashing and pushing nugets soon. 3 first posts this weekend sound good?

  7. 9. sij

    Kind of bummed this didn't happen together with `out var`:

  8. 9. sij

    More tweaks to BitonicSort: 1. Had an (all-over) inlined function with a copy-pasted, useless AVX2 Max+Min 2. Tweaked the code to help the JIT allocate less registers 3. Removed one register dep Noting huge, but better. There's a credible chance now for 120 sized bitonic sort:

  9. 8. sij

    This probably a good time as any to reveal that the working title for my blog post series about this whole mess is: "This Goes to Eleven"... :)

  10. 8. sij

    Time to change on constant in the code and see where this takes us...:

    Prikaži ovu nit
  11. 8. sij

    This, in turn, means my Bitonic sort is now more effcient than vectorized partitioning all the way up to 112 elements. Notice the "Time/N" column is completely stable all the way till 112 elements... (It used to be like that "only" till 96).

    Prikaži ovu nit
  12. 8. sij

    I was trying to clean the BitonicSort before moving to T4 code-gen. Ended up looking at it again for the first time after 3 months... This is my last commit message. I managed to delete 1000 LOC. Somehow, by merging these blocks, I circumvented a bottleneck I was unaware of..

    Prikaži ovu nit
  13. 7. sij
    Odgovor korisniku/ci

    Come on... There is a notion of what is reasonably expected from a viewer: "The gladiators were highly prized fighters who got the most advanced training"...

  14. 7. sij

    Waiting for me to publish sorting stuff? It's coming... Rewriting two key parts for the PR: 1. Partitioning: to decrease code-size (also improves perf, as a side-effect). 2. Bitonic Sort: Going for code-generation, to make review process simpler (It's 1500 LOC). For now: 10.55x!

  15. 6. sij

    Having a hard time breathing here.

  16. 6. sij

    You even notice how there's always an idiot on twitter trying to start a fire by taking a clip and literally omitting the 2 previous sentences that provide context? Full relevant clip: Sick of these liars.

  17. 1. sij

    What the.... I had NO Idea... from: So... AMD exists because of IBM's insistence of a second supplier for the 8086? Mind.Blown.

  18. 28. pro 2019.

    Wow, my father is such a hipster. Can't believe he paid good money for 8 physical 3d prints of a "Save" icon.

  19. 26. pro 2019.

    Reminded me that I read 's anti-debugging and anti-anti-debugging TXTs back at the same era. I guess I couldn't come to terms that an Israeli wrote it. Sounded so crazy reading it that I imagined you are a really smart Indian hacker (!) Then I met you 5 years later.

  20. 23. pro 2019.

    My talk about going from "Array.Sort()" and ending with my initial vectorized/intrinsics based version in C# running 2.7x faster is online!

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:

    ·