GMP and CUDA kernels

Allan Menezes amenezes007 at sympatico.ca
Thu Aug 6 08:03:03 CEST 2009


Hi AnyOne,
   I am sorry if I seem un-enlightened about GMP... but is there generic 
C code alternative instead of assembler for CPUs that GMP does not 
support at the assembler level?
This must be so in GMP for the mpn routines. But can someone tell me if 
all of mpn assembly for each processor is mirrored in C for unsupported 
processors and where in the untarred version of GMP would one find ALL 
the necessary
routines to make GMP work?
I ask this because for a CUDA port for GMP feasibility study I would 
have to study the C code instead of assembly as it would be far easier 
to read and a lesser learning curve plus C --> cuda .cu would be a 
viable alternative making the
otherwise assembler level mpn routines run on a NVIDIA GPU CUDA enabled 
device writing cuda kernels for these instead of assembler and leaving 
the rest of the higher level C and C++ code intact in their own 
respective directories.
Only the mpn  code would have to converted to .cu or cuda  kernels to be 
executed in parallel on a CUDA GPU.
Please share your insights into the above ideas and questions.
Please Google(nvidia cuda) if you have not heard of CUDA.
Thank you very much for GMP! I love it!
Regards,
Allan MeneZes


More information about the gmp-discuss mailing list