Cuda::Entry function uses too much local data

Question

I am trying to run one algorithm for Polygon Intersection on Tesla C2075.But when I try to compile this algorithm, the compiler is giving me error :"Entry Function 'Polygon_Intersection'uses too much local data(0x40e8 bytes,0x4000).

My algorithm is little bit complex and it make use of lots of local variable inside the kernel. So my question is ,Is there any way to solve this problem like Can I use Global memory for local variable declaration for each kernel or somehow use the shared memory?

Answer 1

Compile with -arch=sm_20 to enable device-side malloc so that you can use heap memory for thread-local data.