More optimized PT versions, please!

  • fabrizziop
    30th Jul 2010 Member 0 Permalink
    I'm asking gently, anyone who knows how to compile, to make SSE3 version, and beyond if possible(SSSE3,SSE4.1, blah blah...).

    I think SSE3 have a very good perfomance improvement. If it does, Simon, please include it in your releases!.

    Other thing: Is it possble to make a DX9 or DX10 or DX11 GPU accelerated powder toy?. We know DX9 have very programmable shaders, DX10(some) have DXcompute 4, and stream precoessors, and DX11 have OpenCL, DXcompute 5.

    I know(because of some demos) that DX9 GPUs can accelerate fluids(see X800 fluid demo)
  • Felix
    30th Jul 2010 Member 0 Permalink
    Since this game is made to be cross-platform OpenCL would be a better choice. And, it is mostly un-optimized algorithms that gives it so bad performance(electrodes to mention a thing) but you should realise that it would be better to fix the performance issues and make it multithreaded because not everyone even has a gfx card capable of those stuff.

    and, as a side note. Download the source and run make powder-sse3, doing that on linux 64bit is broken btw. I think it works on 32bit. But you can compile normally on 64bit.
  • Simon
    30th Jul 2010 Administrator 0 Permalink
    Felix
    There's already an SSE3 version for Linux, There's almost no performance increase from SSE2 to SSE3 though, and Microsoft Compiler doesn't do SSE3 optimisations.
  • Felix
    30th Jul 2010 Member 0 Permalink
    Simon:
    Felix
    There's already an SSE3 version for Linux, There's almost no performance increase from SSE2 to SSE3 though, and Microsoft Compiler doesn't do SSE3 optimisations.

    I wrote that it didn't work for 64bit, for me atleast. It need something called gnash-32, which sounds like a 32bit library. Strange since it does not appear without any SSE.
  • fabrizziop
    30th Jul 2010 Member 0 Permalink
    well, I think even a low-end DX11 GFX(5470 , fermi mini BIC lol) have more flops than a high-end CPU.

    Anyway, I'm no programmer, so it's your choice.

    I want to know how can I compile my own powder toy?, and some things like in what language it's written and so...
  • Felix
    30th Jul 2010 Member 0 Permalink
    fabrizziop:
    well, I think even a low-end DX11 GFX(5470 , fermi mini BIC lol) have more flops than a high-end CPU.

    Anyway, I'm no programmer, so it's your choice.

    I want to know how can I compile my own powder toy?, and some things like in what language it's written and so...

    So? Not everyone uses windows.

    Download the source, it is all in there.