Stepping into OpenACC region with cuda-gdb

Hi,
How can I step into my OpenACC code region with cuda-gdb. Setting a breakpoint at the line number does not seem to work. E.g.,

[Launch of CUDA Kernel 6 (calc_force_des_150_gpu<<<(18471,1,1),(256,1,1)>>>) on Device 0]
[Launch of CUDA Kernel 7 (calc_force_des_180_gpu_red<<<(1,1,1),(256,1,1)>>>) on Device 0]
[Termination of CUDA Kernel 6 (calc_force_des_150_gpu<<<(18471,1,1),(256,1,1)>>>) on Device 0]

Breakpoint 1, calc_force_des () at calc_force_des.f:408
408     !$acc data copy(tmp_ax(dimn)) 
(cuda-gdb) 
(cuda-gdb) n
[Launch of CUDA Kernel 8 (calc_force_des_408_gpu<<<(18471,1,1),(1,1,1)>>>) on Device 0]
[Launch of CUDA Kernel 9 (calc_force_des_444_gpu_red<<<(4,1,1),(256,1,1)>>>) on Device 0]
[Termination of CUDA Kernel 8 (calc_force_des_408_gpu<<<(18471,1,1),(1,1,1)>>>) on Device 0]
667
(cuda-gdb) n
0x00000000007a9e40 in pgf90io_src_info03 ()
(cuda-gdb) n
Single stepping until exit from function pgf90io_src_info03,
which has no line number information.
0x00000000007a8130 in pgf90io_encode_fmt ()
(cuda-gdb) n
Single stepping until exit from function pgf90io_encode_fmt,
which has no line number information.
0x00000000007afa40 in pgf90io_fmtw_init ()
(cuda-gdb) n
Single stepping until exit from function pgf90io_fmtw_init,
which has no line number information.
0x00000000007b2c20 in pgf90io_sc_i_fmt_write ()
...
...

Seems like above I have already exited from the kernel I want to examine, which is calc_force_des_408_gpu.


I also tried setting a breakpoint at the start of cuda kernels, but then I get messages like

(cuda-gdb) break calc_force_des_408_gpu
Function "calc_force_des_408_gpu" not defined.
Make breakpoint pending on future shared library load? (y or [n])

I went throught the cuda-gdb manual from http://www.pgroup.com/userforum/posting.php?mode=newtopic&f=12, but it does not have very detailed instructions.

Thanks
Anirban

Sorry the cuda-gdb manual I went through is at
http://developer.download.nvidia.com/compute/cuda/2_1/cudagdb/CUDA_GDB_User_Manual.pdf

Hi Anirban,

Sorry but we don’t support on device debugging yet. It’s in process as we shift to using an LLVM back-end where can add DWARF information. Should be available in an early 14.x compiler.

  • Mat

Thanks much Mat. Will look forward to v14.