>>>>> "Chisheng" == Chisheng Huang <[EMAIL PROTECTED]> writes:
Chisheng> I'm calling a C function from CMUCL to do hierarchical clustering
on a
Chisheng> computer with 2 GB's of RAM. This C function allocates memory to
store
Chisheng> distances between the observations being clustered. When I
cluster a
Chisheng> data set of 18500 observations, this C function fails to allocate
the
Chisheng> required 1306 MB's of memory when my computer still has around
1800 MB's
Chisheng> of free memory. However, I can allocate 1306 MB's of memory with
plain
Chisheng> C code without any problem.
There is a issue in where CMUCL has it heap located and where C starts
allocating memory. The C malloc area starts somewhere above the C
code and grows up (I think). It grows until to runs into the memory
allocated for CMUCL's heap. This is about 256 MB. (It used to be 8
MB).
I can only see 2 ways around this problem:
o Move the heap (and possibly other spaces) to higher addresses. Will
require a cross-compile. You get to maintain your own custom
version of CMUCL.
o Try to allocate the necessary space in the Lisp heap and have C use
that instead. There's some sample code in the User's manual showing
how this can be done.
I think option 2 is the easiest way, and is how Matlisp can allocate
huge arrays. But if you can't do that, I think option 1 is your only
choice.
Ray
