CUDA Fortran Error: copyin Symbol Memcpy FAILED: 74< misaliged address>

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hello, guys！

I encountered some serious problems in CUDA Fortran and need your help urgently.

I have two desktops named PC1 and PC2 respectively, which are configured as follows：
PC1: Win xp 32bit OS,VS2010, CUDA5.0, PVF2010, GeForce GTX680
PC2: Win7 64bit OS, VS2015，CUDA8.0, PVF15, GeForce GTX750Ti

I write a CUDA FORTRAN program on PC1, which can be compiled and run without any problem. However, when I take these codes to PC2, although they can also be compiled successfully , the following error is reported during the running.

0：copyin Symbol Memcpy < dev=0x0000000140572D20, host=0x0000000021FD0AA0, size=5121600, offset=506140424> FAILED : 74<misaligned address>

I’ve been debugging for many days. Unfortunately, no solution has been found. So, sincerely hope that you can help me for some suggestions and methods.

Thank you very much.

ZJQ007

2

Hi ZJQ007,

Are you able to provide a reproducing example that we could look at? This is a somewhat generic error so it’s difficult to give guidance without an example.

Since PVF2015 still supported 32-bits, you might try compiling to 32-bits to see if the problem is with porting your code to 64-bits. Though there’s many other changes here, different compiler version, different OS, newer CUDA version, and different device. Any one of these could be the difference.

-Mat

3

Ok，Thanks