Announcement

Collapse
No announcement yet.

Mesa Preps For Landing The On-Disk Shader Cache

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • treba
    replied
    Originally posted by starshipeleven View Post
    It works at driver level, so yeah it will work with that too, dunno if it will help tho.
    Well if it improves game loading times, it would be of great help. Hope it really does.

    Leave a comment:


  • trek
    replied
    can anyone explain to me if the radeonsi in-memory shader cache can be easily integrated to this on-disk shader cache?

    it seems to me the main difference is that the radeonsi code extrapolates prolog and epilog from the shader before saving it to the cache

    the implementation for r600 should be more difficult as there is no cache at all and it is different from radeonsi that uses llvm and intel that do not uses gallium

    Leave a comment:


  • dungeon
    replied
    Originally posted by tarceri View Post
    Silly comments like these are why I don't comment on these forums much anymore.
    Well i am talking about users here (not me exactly) as users are sort of silly (if you are not aware of that), as users don't wanna know your code better then you

    Leave a comment:


  • tarceri
    replied
    Originally posted by treba View Post
    And another question (no edit button, grr): does anybody know if the cache uses any compression, like SMOL-V does for SPIR shaders?
    Or do I get something wrong and the compression on SPIR-V is only so effective because it's an intermediate representation and therefor not as optimized as the shaders in the cache here?
    There is no compression the cache objects are already very small, compression would slow things down for no good reason. Also we don't cache any IR just the binary's and a small amount of OpenGL related meta-data so compression would not be very effective.

    Leave a comment:


  • tarceri
    replied
    Originally posted by dungeon View Post

    Whatever, best implementation should be if user does not need to think about it... enable/disable and that is it.
    Silly comments like these are why I don't comment on these forums much anymore. If your running Mesa from git your already thinking about it, Mesa is intended to be run via released versions in which case you don't need to think about it.

    Originally posted by dungeon View Post
    Which sound like - I plan bugs to have
    No it's just doing things the most simple way to start with and improving things if there is a real need later on.

    Leave a comment:


  • dungeon
    replied
    Originally posted by tarceri View Post

    Even those upgrading often should see benefits, for example skipping shader compiles between levels etc.
    Whatever, best implementation should be if user does not need to think about it... enable/disable and that is it.

    Also we could make it so we only invalidate if we really need to i.e if the cached objects are really incompatible.
    Which sound like - I plan bugs to have

    Leave a comment:


  • starshipeleven
    replied
    Originally posted by treba View Post
    Will this also work with wine games? That would be great!
    It works at driver level, so yeah it will work with that too, dunno if it will help tho.

    Leave a comment:


  • pal666
    replied
    will it help borderlands?

    Leave a comment:


  • treba
    replied
    And another question (no edit button, grr): does anybody know if the cache uses any compression, like SMOL-V does for SPIR shaders?
    Or do I get something wrong and the compression on SPIR-V is only so effective because it's an intermediate representation and therefor not as optimized as the shaders in the cache here?
    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux Hardware, Linux benchmarking, Desktop Linux, GNU/Linux benchmarks, Open Source AMD, Linux How To, X.Org drivers, Ubuntu hardware, Phoronix Test Suite

    Leave a comment:


  • treba
    replied
    Will this also work with wine games? That would be great!

    Leave a comment:

Working...
X