Failed to build on latest pytorch master.

[xkszltl]

Bug

torchvision failed to build with latest pytorch master.
Same build script was working weeks ago, indicating this can be a recent regression.

  Compiling objects...                                                                                                                                                                                                                                        
  Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)                                                                                                                                           
  [1/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build/t
emp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-option
s ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60 -gencode=
arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14                                                                             
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.o                                                                                                                           
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build/temp.lin
ux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14                                                                                    
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu(160): warning: variable "error_unused" was declared but never referenced                                                                                                                
                                                                                                                                                                                                                                                              
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu(106): warning: variable "device_guard" was declared but never referenced                                                                                                                
                                                                                                                                                                                                                                                              
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu(169): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu(169): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, double)>>, &vision::ops::<unnamed>::nms_ke
rnel>)
              object type is: torch::Library

  2 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/nms_kernel.cu".
  [2/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu -o /tmp/pip-req-build-tzug7t
tm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constex
pr --compiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,
code=sm_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.o
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu -o /tmp/pip-req-build-tzug7ttm/buil
d/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --co
mpiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm
_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(644): warning: variable "guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(806): warning: variable "guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(938): warning: variable "guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(1210): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(1209): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, const at::Tensor &, const at::Tensor &, co
nst at::Tensor &, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, __nv_bool)>>, &vision::ops::<unnamed>::deform_conv2d_forward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(1213): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu(1212): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here                 
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here                                  
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<std::tuple<at::Tensor, at::Tensor, at::Tensor, at::Tensor, at::Tensor> (const at::Tensor &, c$
nst at::Tensor &, const at::Tensor &, const at::Tensor &, const at::Tensor &, const at::Tensor &, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, __nv_bool)>>, &vision::ops::<unnamed>::deform_conv2d_backward_kernel>)              
              object type is: torch::Library                                                                                                                                                                                                                  
                                                                                                                                                                                                                                                              
  4 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/deform_conv2d_kernel.cu".                                                       
  [3/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu -o /tmp/pip-req-build-tzug7ttm
/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr -
-compiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code
=sm_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.o
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build/
temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compil
er-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60
-gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(177): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(205): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(158): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(248): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(272): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(233): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(280): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(279): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<std::tuple<at::Tensor, at::Tensor> (const at::Tensor &, const at::Tensor &, double, int64_t, i
nt64_t)>>, &vision::ops::<unnamed>::ps_roi_pool_forward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(283): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(282): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, const at::Tensor &, double, int64_t, int64
_t, int64_t, int64_t, int64_t, int64_t)>>, &vision::ops::<unnamed>::ps_roi_pool_backward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu(85): warning: parameter "num_rois" was declared but never referenced

  4 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_pool_kernel.cu".
  [4/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu -o /tmp/pip-req-build-tzug7tt
m/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr
 --compiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,co
de=sm_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.o
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build
/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --comp
iler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_6
0 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(337): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(366): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(318): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(411): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(436): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(396): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(444): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(443): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<std::tuple<at::Tensor, at::Tensor> (const at::Tensor &, const at::Tensor &, double, int64_t, i
nt64_t, int64_t)>>, &vision::ops::<unnamed>::ps_roi_align_forward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(447): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(446): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, const at::Tensor &, double, int64_t, int64
_t, int64_t, int64_t, int64_t, int64_t, int64_t)>>, &vision::ops::<unnamed>::ps_roi_align_backward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(21): warning: parameter "index" was declared but never referenced
            detected during instantiation of "void vision::ops::<unnamed>::ps_roi_align_forward_kernel_impl(int, const T *, T, int, int, int, int, int, int, const T *, int, T *, int *) [with T=double]"
  (349): here

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(206): warning: parameter "num_rois" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu(156): warning: parameter "index" was declared but never referenced
            detected during instantiation of "void vision::ops::<unnamed>::ps_roi_align_backward_kernel_impl(int, const T *, const int *, int, T, int, int, int, int, int, int, int, T *, const T *) [with T=double]"
  (418): here

  4 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/ps_roi_align_kernel.cu".
  [5/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu -o /tmp/pip-req-build-tzug7ttm/bu
ild/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compi
ler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60
 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.o
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build/tem
p.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-opt
ions ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60 -genco
de=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(163): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(183): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(141): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(226): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(256): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(210): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(264): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(263): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<std::tuple<at::Tensor, at::Tensor> (const at::Tensor &, const at::Tensor &, double, int64_t, i
nt64_t)>>, &vision::ops::<unnamed>::roi_pool_forward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(267): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(266): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, const at::Tensor &, double, int64_t, int64
_t, int64_t, int64_t, int64_t, int64_t)>>, &vision::ops::<unnamed>::roi_pool_backward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(85): warning: parameter "num_rois" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu(86): warning: parameter "spatial_scale" was declared but never referenced

  4 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_pool_kernel.cu".
  [6/32] /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/
dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu -o /tmp/pip-req-build-tzug7ttm/b
uild/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --com
piler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_
60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  FAILED: /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.o
  /usr/local/cuda/bin/nvcc  -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-pa
ckages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu -o /tmp/pip-req-build-tzug7ttm/build/te
mp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-o
ptions ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_60,code=sm_60 -gen
code=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -ccbin /usr/bin/gcc-8 -std=c++14
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(355): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(376): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(336): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(415): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(445): warning: variable "error_unused" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(401): warning: variable "device_guard" was declared but never referenced

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(453): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(452): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, double, int64_t, int64_t, int64_t, __nv_bo
ol)>>, &vision::ops::<unnamed>::roi_align_forward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(456): error: expression must have a constant value
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
  /usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_allowlist.h(62): note: called from:

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(455): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(606): here
              function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
  /usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(616): here
              argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, double, int64_t, int64_t, int64_t, int64_t
, int64_t, int64_t, int64_t, __nv_bool)>>, &vision::ops::<unnamed>::roi_align_backward_kernel>)
              object type is: torch::Library

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(21): warning: parameter "index" was declared but never referenced
            detected during instantiation of "void vision::ops::<unnamed>::roi_align_forward_kernel_impl(int, const T *, T, int, int, int, int, int, int, __nv_bool, const T *, T *) [with T=double]"
  (360): here

  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu(159): warning: parameter "index" was declared but never referenced
            detected during instantiation of "void vision::ops::<unnamed>::roi_align_backward_kernel_impl(int, const T *, T, int, int, int, int, int, int, __nv_bool, T *, const T *, int, int, int, int) [with T=double]"
  (425): here

  4 errors detected in the compilation of "/tmp/pip-req-build-tzug7ttm/torchvision/csrc/ops/cuda/roi_align_kernel.cu".
  [7/32] /usr/bin/g++-8 -MMD -MF /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip-req-build-tzug7ttm/torchvision/csrc/vision.o.d -DNDEBUG -g -fwrapv -O2 -Wall -g -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORT
IFY_SOURCE=2 -fPIC -DWITH_CUDA -I/tmp/pip-req-build-tzug7ttm/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-packages/to
rch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-tzug7ttm/torchvision/csrc/vision.cpp -o /tmp/pip-req-build-tzug7ttm/build/temp.linux-x86_64-3.6/tmp/pip
-req-build-tzug7ttm/torchvision/csrc/vision.o -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1013"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=1 -std=c++14
  In file included from /tmp/pip-req-build-tzug7ttm/torchvision/csrc/vision.cpp:1:
  /tmp/pip-req-build-tzug7ttm/torchvision/csrc/vision.h:10:40: warning: ‘_register_ops’ initialized and declared ‘extern’
   extern "C" VISION_INLINE_VARIABLE auto _register_ops = &cuda_version;
                                          ^~~~~~~~~~~~~

Environment

• PyTorch / torchvision Version (e.g., 1.0 / 0.4.0): master
• OS (e.g., Linux): Ubuntu 18.04
• How you installed PyTorch / torchvision (conda, pip, source): source
• Build command you used (if compiling from source): pip install from git repo
• Python version: 3.6
• CUDA/cuDNN version: 11.3

[trougnouf]

Likely the same as #3695

[fmassa]

Is this issue still happening today? Our CI seems to be working fine.

[xkszltl]

The last log in our side for pytorch+torchvision (Fri Apr 30 02:27:17 UTC 2021) still shows the issue, checking again...

[xkszltl]

@fmassa I don't think it's fixed.

[xkszltl]

Also I noticed our pure cmake build of torchvision works fine, but pip build from git repo doesn't work.

[xkszltl]

Let me grab some log for you about the same file nms_kernel.cu:

CMake build:

ccache /usr/local/cuda/bin/nvcc -forward-unknown-to-host-compiler -DGFLAGS_IS_A_DLL=0 -DGOOGLE_GLOG_DLL_DECL="" -DGOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS="" -DWITH_CUDA -D__CUDA_NO_HALF_OPERATORS__ -Dtorchvision_EXPORTS -I../torchvision/csrc -isystem=/usr/local/include/torch/csrc/api/include -isystem=/usr/local/cuda/include -isystem=/usr/include/python3.6m --expt-relaxed-constexpr -O3 -DNDEBUG -Xcompiler=-fPIC -D_GLIBCXX_USE_CXX11_ABI=1 -std=c++14 -MD -MT CMakeFiles/torchvision.dir/torchvision/csrc/ops/cuda/nms_kernel.cu.o -MF CMakeFiles/torchvision.dir/torchvision/csrc/ops/cuda/nms_kernel.cu.o.d -x cu -c ../torchvision/csrc/ops/cuda/nms_kernel.cu -o CMakeFiles/torchvision.dir/torchvision/csrc/ops/cuda/nms_kernel.cu.o
../torchvision/csrc/ops/cuda/nms_kernel.cu(106): warning: variable "device_guard" was declared but never referenced

pip build:

/usr/local/cuda/bin/nvcc --generate-dependencies-with-compile --dependency-output /tmp/pip-req-build-9e2kvinu/build/temp.linux-x86_64-3.6/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.o.d -DWITH_CUDA -I/tmp/pip-req-build-9e2kvinu/torchvision/csrc -I/usr/local/lib/python3.6/dist-packages/torch/include -I/usr/local/lib/python3.6/dist-packages/torch/include/torch/csrc/api/include -I/usr/local/lib/python3.6/dist-packages/torch/include/TH -I/usr/local/lib/python3.6/dist-packages/torch/include/THC -I/usr/local/cuda/include -I/usr/include/python3.6m -c -c /tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.cu -o /tmp/pip-req-build-9e2kvinu/build/temp.linux-x86_64-3.6/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options ''"'"'-fPIC'"'"'' -DTORCH_API_INCLUDE_EXTENSION_H '-DPYBIND11_COMPILER_TYPE="_gcc"' '-DPYBIND11_STDLIB="_libstdcpp"' '-DPYBIND11_BUILD_ABI="_cxxabi1011"' -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_61,code=compute_61 -gencode=arch=compute_61,code=sm_61 -gencode=arch=compute_70,code=compute_70 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=compute_75 -gencode=arch=compute_75,code=sm_75 -ccbin /usr/bin/gcc-8 -std=c++14
/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.cu(160): warning: variable "error_unused" was declared but never referenced

/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.cu(106): warning: variable "device_guard" was declared but never referenced

/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.cu(169): error: expression must have a constant value
/usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(585): note: constexpr memory comparison is only supported for top-level integer or array-of-integer objects
/usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(340): note: called from:
/usr/local/lib/python3.6/dist-packages/torch/include/c10/util/string_view.h(369): note: called from:
/usr/local/lib/python3.6/dist-packages/torch/include/ATen/core/op_registration/op_whitelist.h(62): note: called from:

/tmp/pip-req-build-9e2kvinu/torchvision/csrc/ops/cuda/nms_kernel.cu(169): error: more than one instance of overloaded function "torch::Library::impl" matches the argument list:
function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<false>, Func &&) &"
/usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(593): here
function template "torch::Library &torch::Library::impl(torch::detail::SelectiveStr<true>, Func &&) &"
/usr/local/lib/python3.6/dist-packages/torch/include/torch/library.h(603): here
argument types are: (torch::detail::SelectiveStr<<error-constant>>, c10::CompileTimeFunctionPointer<std::remove_pointer_t<std::remove_reference_t<at::Tensor (const at::Tensor &, const at::Tensor &, double)>>, &vision::ops::<unnamed>::nms_kernel>)
object type is: torch::Library

A clear difference is the simple -isystem=/usr/local/include/torch/csrc/api/include vs. many -I for pytorch.
Wondering if there's any include with the same file name that can lead to the difference.

[fmassa]

This seems to be an issue with PyTorch registration mechanism, or in the CppExtensions we are using from PyTorch.

@ezyang is this a known issue in PyTorch?

[datumbox]

I noticed that you set -D_GLIBCXX_USE_CXX11_ABI=1. Could you try adding -D_GLIBCXX_USE_CXX11_ABI=0 in your build flags?

[ezyang]

I attempted to build a minimal repro for this but failed:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Mar_21_19:15:46_PDT_2021
Cuda compilation tools, release 11.3, V11.3.58
Build cuda_11.3.r11.3/compiler.29745058_0
root@40738805190e:~/pytorch# cat test.cu
#include <c10/util/string_view.h>
#include <cassert>

namespace c10 {

constexpr bool f(c10::string_view op_name) {
        assert(op_name.find("::") != string_view::npos);
        return true;
}

  template <bool enabled>
  class SelectiveStr {
  public:
    constexpr SelectiveStr(const char* name) : name_(name) {}
    constexpr operator const char*() { return name_; }
  private:
    const char* name_;
  };


#define TORCH_SELECTIVE_NAME(n) SelectiveStr<f(n)>(n)

void impl(SelectiveStr<true>) {}
void impl(SelectiveStr<false>) {}

void f() {
        impl(TORCH_SELECTIVE_NAME("foo::bar"));
}

}
root@40738805190e:~/pytorch# nvcc -std=c++14 --expt-relaxed-constexpr -c -I. test.cu

I'm OK with commenting out the assert (or ifdefing it so it doesn't run on nvcc) in pytorch core. Can someone send the patch?

[ezyang]

This was on nvidia/cuda:11.3.0-cudnn8-devel-ubuntu20.04 Docker image.

[malfet]

This should have been fixed by pytorch/pytorch#57322, @xkszltl can you please try the build that includes this PR?

[xkszltl]

Confirmed the fix, thanks!