Hi,
When I compile a very complex model on GPU using Ansor. It shows the Compilation error:
ptxas error : Entry function ‘tvmgen_default_fused_nn_conv2d_add_nn_relu_17_kernel0’ uses too much shared data (0x10900 bytes, 0xc000 max)
It seems that I use too much data which are out of the shared memory? How can I control it explicitly?
Thanks a lot!