Unable to compile cumm (fails with "error: ambiguous overload for ‘operator=’")

[mawright]

I'm on Ubuntu 22.04 in WSL and trying to install cumm in a conda environment. Building the wheels fails with the attached error log.

log.txt

The text of the error log is too long to paste into the issue window. Here's a snippet from the full log of what I believe is the main error and the stacktrace

Building wheels for collected packages: cumm-cu120
  Building wheel for cumm-cu120 (pyproject.toml): started
  Building wheel for cumm-cu120 (pyproject.toml): finished with status 'error'
  error: subprocess-exited-with-error

  × Building wheel for cumm-cu120 (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [675 lines of output]
      running bdist_wheel
      running build
      running build_py
      copying cumm/__version__.py -> build/lib.linux-x86_64-cpython-311/cumm
      running egg_info
      writing cumm_cu120.egg-info/PKG-INFO
      writing dependency_links to cumm_cu120.egg-info/dependency_links.txt
      writing requirements to cumm_cu120.egg-info/requires.txt
      writing top-level names to cumm_cu120.egg-info/top_level.txt
      reading manifest file 'cumm_cu120.egg-info/SOURCES.txt'
      reading manifest template 'MANIFEST.in'
      adding license file 'LICENSE'
      writing manifest file 'cumm_cu120.egg-info/SOURCES.txt'
      running build_ext
      [1/39] [GCC][c++]/home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/core_cc_pybind_main/core_cc_pybind_main.cc.o
      [2/39] [GCC][c++]/home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc.o
      FAILED: /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc.o
      g++ -MMD -MT /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc.o -MF /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc.o.d -I "/home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include" -I "/tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include" -I "/home/matt/miniconda3/envs/test/include/python3.11" -I "/usr/local/cuda/include" -I "/home/matt/Code/cumm/include" -O3 -std=c++14 -fPIC -DTV_CUDA -c /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc -o /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc.o
      In file included from /usr/include/c++/11/bits/forward_list.h:38,
                       from /usr/include/c++/11/forward_list:38,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../detail/common.h:306,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../attr.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/class.h:12,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/pybind11.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/stl.h:12,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/PyBindTensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:1:
      /usr/include/c++/11/bits/stl_algobase.h: In instantiation of ‘static _OI std::__copy_move<false, false, std::random_access_iterator_tag>::__copy_m(_II, _II, _OI) [with _II = const long int*; _OI = __half*]’:
      /usr/include/c++/11/bits/stl_algobase.h:495:30:   required from ‘_OI std::__copy_move_a2(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:522:42:   required from ‘_OI std::__copy_move_a1(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:530:31:   required from ‘_OI std::__copy_move_a(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:620:7:   required from ‘_OI std::copy(_II, _II, _OI) [with _II = const long int*; _OI = __half*]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1643:24:   required from ‘tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>::<lambda(auto:28)>::<lambda(auto:29)> [with auto:29 = tv::detail::_identity]’
      /home/matt/Code/cumm/include/tensorview/core/cc17.h:124:47:   [ skipping 12 instantiation contexts, use -ftemplate-backtrace-limit=0 to disable ]
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:106:26:   required from ‘constexpr F tv::detail::mp_for_each_impl(tv::mp_list<Ts ...>, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:497:34:   required from ‘constexpr F tv::mp_for_each(F&&) [with L = tv::mp_list<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16>; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:372:3:   required from ‘bool tv::dispatch_noexcept(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:384:32:   required from ‘void tv::dispatch(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:526:29:   required from ‘void tv::Dispatch<T<Args ...> >::operator()(tv::DType, F&&) [with F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>; T = std::tuple; Args = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1614:43:   required from here
      /usr/include/c++/11/bits/stl_algobase.h:385:25: error: ambiguous overload for ‘operator=’ (operand types are ‘__half’ and ‘const long int’)
        385 |               *__result = *__first;
            |               ~~~~~~~~~~^~~~~~~~~~
      In file included from /usr/local/cuda/include/cuda_fp16.h:4748,
                       from /home/matt/Code/cumm/include/tensorview/dtypes.h:18,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_fp16.hpp:149:50: note: candidate: ‘__half& __half::operator=(float)’
        149 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const float f) { __x = __float2half(f).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:150:50: note: candidate: ‘__half& __half::operator=(double)’
        150 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const double f) { __x = __double2half(f).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:257:50: note: candidate: ‘__half& __half::operator=(short int)’
        257 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const short val) { __x = __short2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:258:50: note: candidate: ‘__half& __half::operator=(short unsigned int)’
        258 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned short val) { __x = __ushort2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:259:50: note: candidate: ‘__half& __half::operator=(int)’
        259 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const int val) { __x = __int2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:260:50: note: candidate: ‘__half& __half::operator=(unsigned int)’
        260 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned int val) { __x = __uint2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:261:50: note: candidate: ‘__half& __half::operator=(long long int)’
        261 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const long long val) { __x = __ll2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:262:50: note: candidate: ‘__half& __half::operator=(long long unsigned int)’
        262 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned long long val) { __x = __ull2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      In file included from /home/matt/Code/cumm/include/tensorview/dtypes.h:18,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_fp16.h:4210:26: note: candidate: ‘constexpr __half& __half::operator=(const __half&)’
       4210 | struct __CUDA_ALIGN__(2) __half {
            |                          ^~~~~~
      /usr/local/cuda/include/cuda_fp16.h:4210:26: note: candidate: ‘constexpr __half& __half::operator=(__half&&)’
      In file included from /usr/include/c++/11/bits/forward_list.h:38,
                       from /usr/include/c++/11/forward_list:38,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../detail/common.h:306,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../attr.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/class.h:12,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/pybind11.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/stl.h:12,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/PyBindTensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:1:
      /usr/include/c++/11/bits/stl_algobase.h: In instantiation of ‘static _OI std::__copy_move<false, false, std::random_access_iterator_tag>::__copy_m(_II, _II, _OI) [with _II = const long unsigned int*; _OI = __half*]’:
      /usr/include/c++/11/bits/stl_algobase.h:495:30:   required from ‘_OI std::__copy_move_a2(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:522:42:   required from ‘_OI std::__copy_move_a1(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:530:31:   required from ‘_OI std::__copy_move_a(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __half*]’
      /usr/include/c++/11/bits/stl_algobase.h:620:7:   required from ‘_OI std::copy(_II, _II, _OI) [with _II = const long unsigned int*; _OI = __half*]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1643:24:   required from ‘tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>::<lambda(auto:28)>::<lambda(auto:29)> [with auto:29 = tv::detail::_identity]’
      /home/matt/Code/cumm/include/tensorview/core/cc17.h:124:47:   [ skipping 12 instantiation contexts, use -ftemplate-backtrace-limit=0 to disable ]
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:106:26:   required from ‘constexpr F tv::detail::mp_for_each_impl(tv::mp_list<Ts ...>, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:497:34:   required from ‘constexpr F tv::mp_for_each(F&&) [with L = tv::mp_list<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16>; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:372:3:   required from ‘bool tv::dispatch_noexcept(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:384:32:   required from ‘void tv::dispatch(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:526:29:   required from ‘void tv::Dispatch<T<Args ...> >::operator()(tv::DType, F&&) [with F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>; T = std::tuple; Args = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1614:43:   required from here
      /usr/include/c++/11/bits/stl_algobase.h:385:25: error: ambiguous overload for ‘operator=’ (operand types are ‘__half’ and ‘const long unsigned int’)
        385 |               *__result = *__first;
            |               ~~~~~~~~~~^~~~~~~~~~
      In file included from /usr/local/cuda/include/cuda_fp16.h:4748,
                       from /home/matt/Code/cumm/include/tensorview/dtypes.h:18,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_fp16.hpp:149:50: note: candidate: ‘__half& __half::operator=(float)’
        149 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const float f) { __x = __float2half(f).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:150:50: note: candidate: ‘__half& __half::operator=(double)’
        150 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const double f) { __x = __double2half(f).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:257:50: note: candidate: ‘__half& __half::operator=(short int)’
        257 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const short val) { __x = __short2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:258:50: note: candidate: ‘__half& __half::operator=(short unsigned int)’
        258 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned short val) { __x = __ushort2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:259:50: note: candidate: ‘__half& __half::operator=(int)’
        259 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const int val) { __x = __int2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:260:50: note: candidate: ‘__half& __half::operator=(unsigned int)’
        260 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned int val) { __x = __uint2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:261:50: note: candidate: ‘__half& __half::operator=(long long int)’
        261 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const long long val) { __x = __ll2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      /usr/local/cuda/include/cuda_fp16.hpp:262:50: note: candidate: ‘__half& __half::operator=(long long unsigned int)’
        262 | __CUDA_HOSTDEVICE__ __CUDA_FP16_INLINE__ __half &__half::operator=(const unsigned long long val) { __x = __ull2half_rn(val).__x; return *this; }
            |                                                  ^~~~~~
      In file included from /home/matt/Code/cumm/include/tensorview/dtypes.h:18,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_fp16.h:4210:26: note: candidate: ‘constexpr __half& __half::operator=(const __half&)’
       4210 | struct __CUDA_ALIGN__(2) __half {
            |                          ^~~~~~
      /usr/local/cuda/include/cuda_fp16.h:4210:26: note: candidate: ‘constexpr __half& __half::operator=(__half&&)’
      In file included from /usr/include/c++/11/bits/forward_list.h:38,
                       from /usr/include/c++/11/forward_list:38,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../detail/common.h:306,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../attr.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/class.h:12,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/pybind11.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/stl.h:12,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/PyBindTensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:1:
      /usr/include/c++/11/bits/stl_algobase.h: In instantiation of ‘static _OI std::__copy_move<false, false, std::random_access_iterator_tag>::__copy_m(_II, _II, _OI) [with _II = const long int*; _OI = __nv_bfloat16*]’:
      /usr/include/c++/11/bits/stl_algobase.h:495:30:   required from ‘_OI std::__copy_move_a2(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:522:42:   required from ‘_OI std::__copy_move_a1(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:530:31:   required from ‘_OI std::__copy_move_a(_II, _II, _OI) [with bool _IsMove = false; _II = const long int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:620:7:   required from ‘_OI std::copy(_II, _II, _OI) [with _II = const long int*; _OI = __nv_bfloat16*]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1643:24:   required from ‘tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>::<lambda(auto:28)>::<lambda(auto:29)> [with auto:29 = tv::detail::_identity]’
      /home/matt/Code/cumm/include/tensorview/core/cc17.h:124:47:   [ skipping 12 instantiation contexts, use -ftemplate-backtrace-limit=0 to disable ]
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:106:26:   required from ‘constexpr F tv::detail::mp_for_each_impl(tv::mp_list<Ts ...>, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:497:34:   required from ‘constexpr F tv::mp_for_each(F&&) [with L = tv::mp_list<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16>; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:372:3:   required from ‘bool tv::dispatch_noexcept(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:384:32:   required from ‘void tv::dispatch(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:526:29:   required from ‘void tv::Dispatch<T<Args ...> >::operator()(tv::DType, F&&) [with F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>; T = std::tuple; Args = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1614:43:   required from here
      /usr/include/c++/11/bits/stl_algobase.h:385:25: error: ambiguous overload for ‘operator=’ (operand types are ‘__nv_bfloat16’ and ‘const long int’)
        385 |               *__result = *__first;
            |               ~~~~~~~~~~^~~~~~~~~~
      In file included from /usr/local/cuda/include/cuda_bf16.h:4756,
                       from /home/matt/Code/cumm/include/tensorview/dtypes.h:21,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_bf16.hpp:123:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(float)’
        123 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(const float f) { __x = __float2bfloat16(f).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:124:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(double)’
        124 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(const double f) { __x = __double2bfloat16(f).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:238:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(short int)’
        238 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(short val) { __x = __short2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:239:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(short unsigned int)’
        239 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned short val) { __x = __ushort2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:240:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(int)’
        240 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(int val) { __x = __int2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:241:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(unsigned int)’
        241 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned int val) { __x = __uint2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:242:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(long long int)’
        242 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(long long val) { __x = __ll2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:243:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(long long unsigned int)’
        243 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned long long val) { __x = __ull2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      In file included from /home/matt/Code/cumm/include/tensorview/dtypes.h:21,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_bf16.h:4215:26: note: candidate: ‘constexpr __nv_bfloat16& __nv_bfloat16::operator=(const __nv_bfloat16&)’
       4215 | struct __CUDA_ALIGN__(2) __nv_bfloat16 {
            |                          ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.h:4215:26: note: candidate: ‘constexpr __nv_bfloat16& __nv_bfloat16::operator=(__nv_bfloat16&&)’
      In file included from /usr/include/c++/11/bits/forward_list.h:38,
                       from /usr/include/c++/11/forward_list:38,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../detail/common.h:306,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/../attr.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/detail/class.h:12,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/pybind11.h:13,
                       from /tmp/pip-build-env-sh7fzxa7/overlay/lib/python3.11/site-packages/pybind11/include/pybind11/stl.h:12,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/PyBindTensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:1:
      /usr/include/c++/11/bits/stl_algobase.h: In instantiation of ‘static _OI std::__copy_move<false, false, std::random_access_iterator_tag>::__copy_m(_II, _II, _OI) [with _II = const long unsigned int*; _OI = __nv_bfloat16*]’:
      /usr/include/c++/11/bits/stl_algobase.h:495:30:   required from ‘_OI std::__copy_move_a2(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:522:42:   required from ‘_OI std::__copy_move_a1(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:530:31:   required from ‘_OI std::__copy_move_a(_II, _II, _OI) [with bool _IsMove = false; _II = const long unsigned int*; _OI = __nv_bfloat16*]’
      /usr/include/c++/11/bits/stl_algobase.h:620:7:   required from ‘_OI std::copy(_II, _II, _OI) [with _II = const long unsigned int*; _OI = __nv_bfloat16*]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1643:24:   required from ‘tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>::<lambda(auto:28)>::<lambda(auto:29)> [with auto:29 = tv::detail::_identity]’
      /home/matt/Code/cumm/include/tensorview/core/cc17.h:124:47:   [ skipping 12 instantiation contexts, use -ftemplate-backtrace-limit=0 to disable ]
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:106:26:   required from ‘constexpr F tv::detail::mp_for_each_impl(tv::mp_list<Ts ...>, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/core/mp_helper.h:497:34:   required from ‘constexpr F tv::mp_for_each(F&&) [with L = tv::mp_list<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16>; F = tv::dispatch_noexcept<float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)> >(tv::DType, tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>&&)::<lambda(auto:3)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:372:3:   required from ‘bool tv::dispatch_noexcept(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:384:32:   required from ‘void tv::dispatch(tv::DType, F&&) [with Ts = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}; F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:526:29:   required from ‘void tv::Dispatch<T<Args ...> >::operator()(tv::DType, F&&) [with F = tv::Tensor::astype(tv::DType, bool) const::<lambda(auto:27)>; T = std::tuple; Args = {float, double, signed char, short int, int, long int, unsigned char, short unsigned int, unsigned int, long unsigned int, bool, __half, __nv_bfloat16}]’
      /home/matt/Code/cumm/include/tensorview/tensor.h:1614:43:   required from here
      /usr/include/c++/11/bits/stl_algobase.h:385:25: error: ambiguous overload for ‘operator=’ (operand types are ‘__nv_bfloat16’ and ‘const long unsigned int’)
        385 |               *__result = *__first;
            |               ~~~~~~~~~~^~~~~~~~~~
      In file included from /usr/local/cuda/include/cuda_bf16.h:4756,
                       from /home/matt/Code/cumm/include/tensorview/dtypes.h:21,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_bf16.hpp:123:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(float)’
        123 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(const float f) { __x = __float2bfloat16(f).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:124:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(double)’
        124 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(const double f) { __x = __double2bfloat16(f).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:238:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(short int)’
        238 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(short val) { __x = __short2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:239:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(short unsigned int)’
        239 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned short val) { __x = __ushort2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:240:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(int)’
        240 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(int val) { __x = __int2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:241:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(unsigned int)’
        241 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned int val) { __x = __uint2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:242:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(long long int)’
        242 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(long long val) { __x = __ll2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.hpp:243:61: note: candidate: ‘__nv_bfloat16& __nv_bfloat16::operator=(long long unsigned int)’
        243 |     __CUDA_HOSTDEVICE__ __CUDA_BF16_INLINE__ __nv_bfloat16 &__nv_bfloat16::operator=(unsigned long long val) { __x = __ull2bfloat16_rn(val).__x; return *this; }
            |                                                             ^~~~~~~~~~~~~
      In file included from /home/matt/Code/cumm/include/tensorview/dtypes.h:21,
                       from /home/matt/Code/cumm/include/tensorview/tensorview.h:28,
                       from /home/matt/Code/cumm/include/tensorview/cuda/driverops.h:16,
                       from /home/matt/Code/cumm/include/tensorview/contexts/core.h:20,
                       from /home/matt/Code/cumm/include/tensorview/context.h:2,
                       from /home/matt/Code/cumm/include/tensorview/tensor.h:94,
                       from /home/matt/Code/cumm/include/tensorview/pybind.h:16,
                       from /home/matt/Code/cumm/include/tensorview/pybind_utils.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/include/tensorview_bind/TensorViewBind.h:2,
                       from /home/matt/Code/cumm/build/temp.linux-x86_64-cpython-311/cumm/build/core_cc/src/tensorview_bind/PyBindTensorViewBind/PyBindTensorViewBind_bind_TensorViewBind.cc:2:
      /usr/local/cuda/include/cuda_bf16.h:4215:26: note: candidate: ‘constexpr __nv_bfloat16& __nv_bfloat16::operator=(const __nv_bfloat16&)’
       4215 | struct __CUDA_ALIGN__(2) __nv_bfloat16 {
            |                          ^~~~~~~~~~~~~
      /usr/local/cuda/include/cuda_bf16.h:4215:26: note: candidate: ‘constexpr __nv_bfloat16& __nv_bfloat16::operator=(__nv_bfloat16&&)’

[mawright]

I just posted this but I think I solved it. The issue seems to be the hardcoded path usr/local/cuda in https://github.com/FindDefinition/cumm/blob/92a3dfa7e3ad4f0d185fbe14a38c79adfb63d313/cumm/common.py#L249-L251, which made the compiler look there instead of the Cuda that was part of the conda environment. Changing that path to the conda environment directories solved this error. However, I'm now encountering a similar error when trying to compile spconv but haven't found a similar hardcoded directory in the code there yet.

[rem821]

I'm experiencing the same issue. Trying to build on Jetson Orin AGX with JetPack 6.0DP, CUMM_CUDA_VERSION = 12.0 has anyone been able to resolve this issue consistently? Changing the path sadly doesn't work

[rem821]

What worked for me in the end was to build the CPU only version by unsetting the CUMM_CUDA_VERSION variable

[l1351868270]

1.install cuda env

conda install "cuda<12.2" -c nvidia
do not install, check the cuda-* version is or not suitable
if cuda-* large than 12.1
conda install "cuda-*<12.2"

2.copy the headers necessary to the include

cp /root/miniconda3/envs/spconv_cu121/targets/x86_64-linux/include/* /root/miniconda3/envs/spconv_cu121/include/

3.modify the _CACHED_CUDA_INCLUDE_LIB in https://github.com/FindDefinition/cumm/blob/main/cumm/common.py#L231

_CACHED_CUDA_INCLUDE_LIB = ("/root/miniconda3/envs/spconv_cu121/include", "/root/miniconda3/envs/spconv_cu121/lib")

4. install the cumm

   pip install -e .
   python -c "import cumm"