Using libcu++ in programs compiled to OptiX-IR.

[to-mas-kral]

I'm trying to compile a program which uses libcu++ to the OptiX-IR format.

The compilation fails with:

  __int128_t __convert_to_integral(__int128_t __val) { return __val; }
             ^

/opt/cuda/bin/../targets/x86_64-linux/include/cuda/std/detail/libcxx/include/__utility/convert_to_integral.h(56): error: "__uint128_t (__uint128_t)" contains a 128-bit integer, which is not supported in device code
  __uint128_t __convert_to_integral(__uint128_t __val) { return __val; }

I want to ask if this is something that's supposed to work or if it's just unsupported. If this is a bug, should I report it to the libcu++ team or the OptiX team ?

System information:
nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2023 NVIDIA Corporation Built on Fri_Nov__3_17:16:49_PDT_2023 Cuda compilation tools, release 12.3, V12.3.103 Build cuda_12.3.r12.3/compiler.33492891_0

/opt/cuda/bin/g++ --version
g++ (GCC) 12.3.0 Copyright (C) 2022 Free Software Foundation, Inc.

[miscco]

Hi @to-mas-kral I believe this is a bug on our side.

We did pull 128 bit integer support in from libc++ but should ensure that it is not used on device. I have opened a bug report with your reproducer

[to-mas-kral]

Alright, thanks !