How do I download the CUDA12.8 NVCC toolchain for the DRIVE Thor platform?

Autonomous Vehicles DRIVE AGX Thor DRIVE AGX Thor General
1

DRIVE OS Version: SDK 7.0.3.0

Issue Description: How can I obtain the CUDA12.8 nvcc toolchain compatible with the DRIVE Thor platform?

3

Dear @zaiwei ,
DRIVE OS 7.0.3 docker has environment to cross compile the DOS samples already. Could you check cross compile?

4

Thanks for your quick reply.
We would like to download a standalone nvcc toolchain (for example, Jetson Jetpack 6.2 cuda-nvcc-12-6_12.6.68-1_amd64 and DRIVE Orin cuda-nvcc-11-4_11.4.160-1_amd64) so that we can integrate it into our application compilation pipeline.

7 
root@7.0.3.0-0010-linux-nsr:~# /usr/local/cuda-12.8/bin/nvcc -v --cuda -ccbin /usr/bin/clang-17 '-std=c++17' '-gencode=arch=compute_101,code=sm_101' -Xcompiler '-std=c++17,--target=aarch64-linux-gnu' -o test.cu.cpp.ii test.cu
#$ _NVVM_BRANCH_=nvvm
#$ _SPACE_= 
#$ _CUDART_=cudart
#$ _HERE_=/usr/local/cuda-12.8/bin
#$ _THERE_=/usr/local/cuda-12.8/bin
#$ _TARGET_SIZE_=
#$ _TARGET_DIR_=
#$ _TARGET_DIR_=targets/x86_64-linux
#$ TOP=/usr/local/cuda-12.8/bin/..
#$ CICC_PATH=/usr/local/cuda-12.8/bin/../nvvm/bin
#$ NVVMIR_LIBRARY_DIR=/usr/local/cuda-12.8/bin/../nvvm/libdevice
#$ LD_LIBRARY_PATH=/usr/local/cuda-12.8/bin/../lib:
#$ PATH=/usr/local/cuda-12.8/bin/../nvvm/bin:/usr/local/cuda-12.8/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
#$ INCLUDES="-I/usr/local/cuda-12.8/bin/../targets/x86_64-linux/include"  
#$ LIBRARIES=  "-L/usr/local/cuda-12.8/bin/../targets/x86_64-linux/lib/stubs" "-L/usr/local/cuda-12.8/bin/../targets/x86_64-linux/lib"
#$ CUDAFE_FLAGS=
#$ PTXAS_FLAGS=
#$ "/usr/bin"/clang-17 -std=c++17 -D__CUDA_ARCH_LIST__=1010 -D__NV_LEGACY_LAUNCH -E -x c++ -D__CUDACC__ -D__NVCC__  -std=c++17 --target=aarch64-linux-gnu "-I/usr/local/cuda-12.8/bin/../targets/x86_64-linux/include"    -D__CUDACC_VER_MAJOR__=12 -D__CUDACC_VER_MINOR__=8 -D__CUDACC_VER_BUILD__=93 -D__CUDA_API_VER_MAJOR__=12 -D__CUDA_API_VER_MINOR__=8 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include "cuda_runtime.h" -m64 "test.cu" -o "/tmp/tmpxft_0000024d_00000000-5_test.cpp4.ii" 
#$ cudafe++ --c++17 --clang --clang_version=170000 --display_error_number --orig_src_file_name "test.cu" --orig_src_path_name "/root/test.cu" --unicode_source_kind=UTF-8 --allow_managed  --m64 --parse_templates --gen_c_file_name "/tmp/tmpxft_0000024d_00000000-6_test.cudafe1.cpp" --stub_file_name "tmpxft_0000024d_00000000-6_test.cudafe1.stub.c" --gen_module_id_file --module_id_file_name "/tmp/tmpxft_0000024d_00000000-4_test.module_id" "/tmp/tmpxft_0000024d_00000000-5_test.cpp4.ii" 
/usr/bin/../lib/gcc-cross/aarch64-linux-gnu/13/../../../../aarch64-linux-gnu/include/bits/math-vector.h(106): error: identifier "__SVFloat32_t" is undefined
  typedef __SVFloat32_t __sv_f32_t;
          ^
/usr/bin/../lib/gcc-cross/aarch64-linux-gnu/13/../../../../aarch64-linux-gnu/include/bits/math-vector.h(107): error: identifier "__SVFloat64_t" is undefined
  typedef __SVFloat64_t __sv_f64_t;
          ^
/usr/bin/../lib/gcc-cross/aarch64-linux-gnu/13/../../../../aarch64-linux-gnu/include/bits/math-vector.h(108): error: identifier "__SVBool_t" is undefined
  typedef __SVBool_t __sv_bool_t;
          ^
3 errors detected in the compilation of "test.cu".
# --error 0x2 --

Compiling using the nvcc toolchain within Docker also resulted in an error. How can I resolve this error?
Does it seem to be a compatibility issue between the NVIDIA CUDA Toolkit version and the GNU C Library (glibc) version?

8

Dear @zaiwei ,
Did you try CUDA sample compilation steps from GitHub - NVIDIA/cuda-samples: Samples for CUDA Developers which demonstrates features in CUDA Toolkit . Please use CUDA 12.8 samples. Once it works, check adding your code as CUDA sample . Let us know observations.