-
Notifications
You must be signed in to change notification settings - Fork 1
Home
Welcome to the RHMC-on-GPU wiki! (under development)
The typical compilation follows the pattern:
./configure && make
this will compile a code to run in a single core CPU.
In order to enable CUDA routines type the following command (also look at the
variour options in configure --help
)
./configure --enable-cuda=ARCH
where ARCH should be substituted by one of the following strings "sm_XX" in order to choose the target NVIDIA architecture
- sm_10 ISA_1, Basic features
- sm_11 atomic memory operations on global memory
- sm_12 atomic memory operations on shared memory, vote instructions
- sm_13 double precision floating point support (Tesla Cards)
- sm_20 Fermi architecture support
please refer to the original NVIDIA documentation for more details on which cards support the several architectures.
Some newer compilers, like GNU gcc 4.5
, are incompatible with the NVIDIA nvcc
compiler (CUDA 4.0). In order to solve this problem you should install a different
version of the compiler, e.g. gcc 4.4
, in the directory PATH_TO_COMPILER
and then
configure with the following command
./configure --enable-cuda=ARCH --with-cuda_comp=PATH_TO_COMPILER CXX=PATH_TO_COMPILER/g++ CC=PATH_TO_COMPILER/gcc
Also, if the mandatory library MPFR (Multiple Precision) is installed in a
non standard location (not /usr/lib
) use the flag --with-mpfr=PATH
where PATH
is the location
of the library in your system
Please refer to the wiki page: Running the code