Problem accessing data accurately from shared memory in kernel

[bit_mapper]

Hi I have a weired problem accessing data accurately from shared memory on kernel.

Basically I have a data array(A[500]) in shared memory, if I copy them to another array B[500] and move it back to host side for print, it looks exactly what I expect to see.

shared char A[500];

for (int i=0; i<500; i++)
B[i]=A[i];

But if I further randomly pick up some data from the array A[500] after the above copy like below

shared char A[500];

for (int i=0; i<500; i++)
B[i]=A[i]

B[500]=A[38];
B[501]=A[179];
B[502]=A[444];
B[503]=A[321];
…

When I move B to the host side and print everything out, B[0]~B[499] are always correct, but the data after that(B[500]~B[503]) are partly correctly and partly just 0…So weired, Is that the problem of possible bank conflict? synchronize seems not that helpful to this problem. Any idea?

Thanks.

[tera]

Bank conflicts never lead to wrong results, they only slow down execution.

Can you post a self contained example (kernel and host code calling it) that demonstrates your problem? There are numerous things that could have gone wrong here as shared memory is a per-block resource and you don’t write anything about your thread and block configuration. When you post the code, please enclose in in [code] [/code] tags for better readability.