CUDA Toolkit same Error with the cuda-gdb as in Version 12.5 (original) (raw)

August 2, 2024, 12:17pm 1

Hi
But now I’m really angry, they announced that this error has been fixed in the new version, nothing, nothing at all happened, the same error again, apparently AMD GPU toolkits are the better solution

exec: /usr/local/cuda-12.6/bin/cuda-gdb-python3.12-tui --version
/usr/local/cuda-12.6/bin/cuda-gdb-python3.12-tui: symbol lookup error: /usr/local/cuda-12.6/bin/cuda-gdb-python3.12-tui: undefined symbol: curscr

mmason August 2, 2024, 5:23pm 2

Thank you for raising this issue. To investigate, we need a couple of pieces of additional information:

The Linux distribution you’re using.
The output of ldd /usr/local/cuda-12.5/bin/cuda-gdb-python3.12-tui

In the meantime you can use cuda-gdb-minimal - it’s lacking Python and TUI integration, but should run on any Linux distribution that’s supported by the CUDA Toolkit.

this is the ldd output :
franz@tumbleweed:~> ldd /usr/local/cuda-12.6/bin/cuda-gdb-python3.12-tui
linux-vdso.so.1 (0x00007f3e30210000)
libgmp.so.10 => /lib64/libgmp.so.10 (0x00007f3e30116000)
libncurses.so.6 => /lib64/libncurses.so.6 (0x00007f3e300eb000)
libpython3.12.so.1.0 => /lib64/libpython3.12.so.1.0 (0x00007f3e2fa00000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f3e300e6000)
libutil.so.1 => /lib64/libutil.so.1 (0x00007f3e300e1000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007f3e300da000)
libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f3e2f600000)
libm.so.6 => /lib64/libm.so.6 (0x00007f3e2f919000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f3e300ad000)
libc.so.6 => /lib64/libc.so.6 (0x00007f3e2f200000)
/lib64/ld-linux-x86-64.so.2 (0x00007f3e30212000)
libtinfo.so.6 => /lib64/libtinfo.so.6 (0x00007f3e30072000)
the same error as in the version before
cuda-gdb-minimal is not a practicable solution, why is the error not fixed in the new 12.6 Version
as announced ?