Hash table with nvcc

Accelerated Computing CUDA CUDA Programming and Performance
1

I am writing some code to solve support vector machines and I am handling some “cache” on the host of the kernel matrix.

When I try to compile using nvcc and include the <tr1/unordered_map> I get an cfee++ error.

I thought it was possible to compile the kernel using nvcc and then use gcc to compile the rest of the code, but I think the gcc on the server I am using is too old. I have no admin control of the server. What options do I have to use a hast table?

nvcc and gcc versions:

-bash-3.2$ nvcc --version
nvcc: NVIDIA ® Cuda compiler driver
Copyright © 2005-2007 NVIDIA Corporation
Built on Thu_Jun_19_03:38:28_PDT_2008
Cuda compilation tools, release 2.0, V0.2.1221

-bash-3.2$ gcc --version
gcc (GCC) 4.1.2 20071124 (Red Hat 4.1.2-42)
Copyright © 2006 Free Software Foundation, Inc.

2

get your own machine :) the gtx 260 isn’t that expensive.

it could also be that you can’t use existing host code on the GPU. Try writing the hash table yourself. Often simpler algorithms will work better on the GPU so long as they’re sufficiently parallel.