Oh this is some great stuff too! This will help me with another problem I have (CUDA and OpenCL), Thanks.
On Fri, 2011-07-01 at 16:40 -0700, Reverend Chip wrote: > On 7/1/2011 2:26 PM, chromatic wrote: > > On Friday, July 01, 2011 at 10:58 am, Chip Salzenberg wrote: > > > >> On 7/1/2011 7:24 AM, Kartik Thakore wrote: > >>> chip during my talk you mentioned some technology for doing OpenGL > >>> context/display lists better. Can you mention it again chip? I seem to > >>> have forgotten it. > >> Hi! I was talking about tcc, the > >> http://en.wikipedia.org/wiki/Tiny_C_Compiler . It's x86-specific but > >> otherwise it seems like a convenient and really fast way to create > >> structures defined by C. > > Are you suggesting an approach like Python's Weave or Cinpy? > > > > http://scipy.org/Weave > > http://www.cs.tut.fi/~ask/cinpy/ > > Yes, quite. > > > The approach I had in mind was to build an XS file full of Perl <-> C > > thunks > > and then use the closure-over-dlfunc-pointer trick of P5NCI to avoid > > writing > > (and paying the cost of) hundreds of otherwise-identical XS wrappers. > > Ah, well, I had perhaps an incomplete understanding of the problem. I > thought we needed to create complex data structures for some newer > OpenGL calls, so I wanted to ease the difficulty of making complex C > data structures from Perl. If, OTOH, the primary problem is exposing a > very broad API from C to Perl, then yes, reusing XS wrappers is very > good (reduced memory use, better cache coherency, blah blah). > > > > This is > > even one spot where a little AUTOLOAD magic would help memory usage even > > more. > > > > -- c -- Kartik Thakore <thakore.kar...@gmail.com>