help__can not find shared memory on C2050

Accelerated Computing CUDA CUDA Programming and Performance
#1

I am a nvidia C2050 GPU user, I found a problem when I compile my code.

the message shows as follows:

ptxas error : Entry function ‘_Z20cudgseis_source_termiPiPdS0_iS0_iiS0_S0_S0_iS0_S0_S0_S0_iS_S_S0_S0_S0_S0_S0_S0_S0_S0_S0_diiS0_i’ uses too much shared data (0x5990 bytes + 0x10 bytes system, 0x4000 max)

I think this means there are 16384 bytes shared memory on board, but there should be 49152 bytes memory there, would anyone help me to solve this problem?

#2

Add the --arch=sm_20 option to your nvcc command. The compiler has to know you are targeting the newer architecture before it will let you use more than 16 kB of shared memory.

#3

Thank you so much!