Encountered NVIDIA support issues while compiling spconv-cpp

[Mr-MoNET]

Environment
device : PC
platform : X86(amd64)
system : Ubuntu 22.04
GPU : RTX 4070
CUDA : 12.4
cmake : 4.2.0

Problem Description
I recently encountered some problems while compiling the autoware source code, which depends on third-party libraries such as spconv and cumm. So I first tried to compile and install spconv and cumm, but when I compiled spconv, I encountered the following error:

(spconv) zhsd@zhsd-B760-GAMING-X-GEN5:~/Desktop/spconv_cpp/spconv/build-amd64$ make -j8 [ 0%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 0%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 0%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 0%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 1%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 1%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o [ 2%] Building CUDA object src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:18976：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m128n128k8m32n64k8A1_200_S1_GemmKernel.dir/all] error 2 make[1]: *** 正在等待未完成的任务.... nvcc fatal : Unsupported gpu architecture 'compute_100' nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:18912：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/all] error 2 make[1]: *** [CMakeFiles/Makefile2:18944：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n512k8m32n64k8A1_200_S1_GemmKernel.dir/all] error 2 [ 2%] Building CUDA object src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/StaticAllocator_full_int.cu.o nvcc fatal : Unsupported gpu architecture 'compute_100' nvcc fatal : Unsupported gpu architecture 'compute_100' nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 [ 2%] Building CUDA object src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/StaticAllocator_full_float.cu.o nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:18816：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/all] error 2 make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:18880：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32tnt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/all] error 2 make[1]: *** [CMakeFiles/Makefile2:18784：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m64n256k8m32n64k8A1_200_S1_GemmKernel.dir/all] error 2 make[2]: *** [src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/build.make:80：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/GemmKernel_gemm_kernel.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:18848：src/spconvlib/cumm/gemm/main/Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1/GemmKernel/CMakeFiles/spconvlib_cumm_gemm_main_Simt_f32f32f32f32f32ttt_m32n128k16m32n32k8A1_200_S1_GemmKernel.dir/all] error 2 nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/build.make:150：src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/StaticAllocator_full_int.cu.o] error 1 make[2]: *** Waiting for unfinished tasks.... nvcc fatal : Unsupported gpu architecture 'compute_100' make[2]: *** [src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/build.make:165：src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/StaticAllocator_full_float.cu.o] error 1 make[1]: *** [CMakeFiles/Makefile2:43616：src/spconvlib/spconv/csrc/sparse/alloc/StaticAllocator/CMakeFiles/spconvlib_spconv_csrc_sparse_alloc_StaticAllocator.dir/all] error 2 make: *** [Makefile:156：all] error 2

Wooooooo! According to autowarefoundation/spconv-cpp, spconv-cpp supports CUDA version 12.8. This confuses me, as amd64.env in Autoware explicitly states cuda_version=12.4, which seems contradictory. So, should I use CUDA 12.4 or CUDA 12.8?

[urusaidog]

Hi @Mr-MoNET
This may be caused due to a recent PR in which blackwell support was added. Since you have 12.4 you can not use this version I think. Can you not install the previous release? The code it is the same but compatible with < 12.8