|
I sign up to the high performance matrix math interest group. :)Currently, my work involves parallel processing of big multidimensional matrices in C++. There is interest in expanding the libraries to be usable from Lua, and it's sure interesting to read where people have been banging their heads until now.
In the long run, it would be meaningful to have a base matrix math library akin to what LuaSocket is for networking. LuaMatrix could be a good name. Intel Threading Building Blocks is an interesting piece of API to make use of multiple cores doing matrix calculations.
-asko Chris kirjoitti 20.2.2008 kello 3:57:
On Feb 19, 2008 8:37 PM, Jean-Claude Wippler <jcw@equi4.com> wrote:The following can be made to work: vertex_vector(20, 'x').value = 2.5 by having the call return a table or userdata with a suitable __newindex.This is exactly the problem I'm trying to avoid. The problem is when you call this millions of times you get millions of temporary objects. It's slow (lots of memory allocation) and causes garbage collector thrashing. Technically with that same approach it can look "proper": vertex_vector[20].x = 2.5 It's the same problem though with needing temporary objects. CR