VideoHelp Forum
+ Reply to Thread
Results 1 to 7 of 7
Thread
  1. Member
    Join Date: Nov 2005
    Location: United States
    Search Comp PM
    for those of you that don't know, the lead developer of x264 goes by the online name dark shikari. he recently started his own forum, named doom10, after he was banned by the mod from doom9 for disagreeing with him. in an ironic twist, i was banned from doom10 for challenging some of the idiotic claims made by his darkness.

    anyway, a poster over at doom10 asked if x264 will support the new avx instruction set, here is what he said:

    http://freeproxyserver.net/index.php?q=aHR0cDovL2Rvb20xMC5vcmcvaW5kZXgucGhwP3RvcGljPTUxNC4w

    What about 256bit AVX? Each module can process only one at time (AFAIK) and x264 will surely support these new instructions (I hope eheh)


    Float-only, thus a useless pile of tripe.
    Quote Quote  
  2. I haven't looked at the x264 source but video encoding is predominantly done in integer, not floating point.
    Last edited by jagabo; 7th Nov 2010 at 19:51.
    Quote Quote  
  3. Member
    Join Date: Nov 2005
    Location: United States
    Search Comp PM
    Originally Posted by jagabo View Post
    I haven't looked at the x264 source but video encoding is predominantly done in integer, not floating point.
    the reason video encoding is done in integer is due to historical reasons, it has it's origins back in the days when the fpu was known as the math co-processor and was a separate chip. i remember my teachers back then telling us to structure our code so that the math was done by the alu. this was true with basic, fortran, pascal and c.

    it's trivial to modify the source so that the math operations switch from integer to floating point, simply re-declare int as float and you have now instructed the compiler to swap in the assembler instructions required to carry out the math on the fpu instead of the integer unit.

    a similar approach can be used change simd ints to simd floats.

    his darkness seems to have some bizarre aversion to a number of technologies, including anything related to gpu acceleration and he seems to have a serious grudge against intel. over at the doom9 forums an intel developer that had worked on sse4 offered to assist with getting x264 to use sandy bridge's video transcoding engine.

    shikari responded by ridiculing the guys work, ripping into intel and basically dismissing SB without actually knowing shit about it. what's worse is that he has a chorus of supporters who joined in on the fun. then when the intel developer told him to forget about it if that was going to be his attitude darkness tried to back peddle and claimed that the developer was always welcome to help.

    a similar occurrence takes place any time anyone dares question any of his idiotic assertions concerning gpu's and the best part is that he ripped into the theora code base claiming it was a mess that was poorly commented, a hilariously ironic statement when you consider who it came from and the code he and his cohorts have produced.
    Quote Quote  
  4. Member
    Join Date: Sep 2007
    Location: Canada
    Search Comp PM
    You mean at D10 ? You should read the entire thread, because they are buddy-buddy now as things got patched things up over IRC

    I saw the "deadrats" thread at D9 ... funny stuff . Did you get banned for spamming or just a suspension in the penalty box ?
    Quote Quote  
  5. You don't need to explain computer programming to me. I've been doing it for 40 years.

    Since conversion of x264 to floats is so trivial why don't you just do it and blow us all away with how fast your version is?
    Last edited by jagabo; 7th Nov 2010 at 21:17.
    Quote Quote  
  6. Member
    Join Date: Nov 2005
    Location: United States
    Search Comp PM
    Originally Posted by jagabo View Post
    Since conversion of x264 to floats is so trivial why don't you just do it and blow us all away with how fast your version is?
    a few reasons:

    1) the fpu on current cpu's still lags behind the alu, in fact microsoft and intel have been suggesting for years that programmers avoid using regular floats in favor of ssefp for years now. furthermore amd had taken the same stance for a while. with bulldozer 256 bit fpu per module, it seems that amd is reversing it's stance.

    2) when bulldozer comes out i do plan on working on an x264/x262 fork that features greatly simplified code and relies entirely on 32 bit floating point operations.

    3) if you have been programming for 40 years then i'm assuming you know that simply re-declaring int as float causes the math operations to be calculated on the fpu instead of the alu.

    that is neither here nor there, it's absurd for his darkness to dismiss modifying the code base to take advantage of avx, why would you use 128 bit sse (that can only handle four 32 bit calculations per cycle) when you can use 256 bit avx that can handle eight 32 bit calculations per cycle?
    Quote Quote  
  7. Member
    Join Date: Nov 2005
    Location: United States
    Search Comp PM
    Originally Posted by poisondeathray View Post
    You mean at D10 ? You should read the entire thread, because they are buddy-buddy now as things got patched things up over IRC

    I saw the "deadrats" thread at D9 ... funny stuff . Did you get banned for spamming or just a suspension in the penalty box ?
    i didn't get banned from doom9, i got banned from doom10, permanently, for posting a rather in depth rebuttal to the various claims his darkness has made with regards to gpu encoding, with a conservative sprinkling of some mocking.

    also, he banned my ip, i can't access doom10, i can use a proxy server, but i really don't see the need.
    Quote Quote  



Similar Threads