I’ve encountered various errors for `invalid device pointers’, but the program seems to be running well w/o errors. Is there any clue in which wrong ways have I used the pointers? Out of bound accesses, or any other causes? Thanks!
Almost always using a host pointer as a device pointer.
It seems that the errors persist with CUDA 2.3, but disappears with CUDA 3.0 beta. The 2.3 version is running with a Tesla C1060. The 3.0 version is running with a GTX285. Both are running 64-bit linux. Codes and datasets are also aligned. Dunno why…