pytorch_CUDA10.patch

diff --git a/aten/src/THC/THCAtomics.cuh b/aten/src/THC/THCAtomics.cuh
index bdb7859d9..a5e6ce321 100644
--- a/aten/src/THC/THCAtomics.cuh
+++ b/aten/src/THC/THCAtomics.cuh
@@ -4,8 +4,7 @@
 #include "THC.h"
 #include "THCHalf.h"
 #include "THCNumerics.cuh"
-
-namespace at { struct Half; }
+#include "ATen/ATen.h"
 
 template <typename T, size_t n>
 struct AtomicAddIntegerImpl;
@@ -96,7 +95,7 @@ static inline __device__ void atomicAdd(int64_t *address, int64_t val) {
   AtomicAddIntegerImpl<int64_t, sizeof(int64_t)>()(address, val);
 }
 
-#ifdef CUDA_HALF_TENSOR
+#if ((CUDA_VERSION < 10000) || (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 700)))
 static inline  __device__ void atomicAdd(half *address, half val) {
   unsigned int * address_as_ui =
     (unsigned int *) ((char *)address - ((size_t)address & 2));
@@ -119,10 +118,10 @@ static inline  __device__ void atomicAdd(half *address, half val) {
     old = atomicCAS(address_as_ui, assumed, old);
   } while (assumed != old);
 }
-static inline __device__ void atomicAdd(at::Half *address, half val) {
-  return atomicAdd(reinterpret_cast<half*>(address), val);
-}
 #endif
+static inline __device__ void atomicAdd(at::Half *address, at::Half val) {
+  atomicAdd(reinterpret_cast<half*>(address), val);
+}
 
 #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 600 || CUDA_VERSION < 8000)
 // from CUDA C Programmic Guide
diff --git a/caffe2/utils/GpuDefs.cuh b/caffe2/utils/GpuDefs.cuh
index cf54f9e85..c69fcd38c 100644
--- a/caffe2/utils/GpuDefs.cuh
+++ b/caffe2/utils/GpuDefs.cuh
@@ -8,7 +8,7 @@ namespace caffe2 {
 // Static definition of GPU warp size for unrolling and code generation
 
 #ifdef __CUDA_ARCH__
-#if __CUDA_ARCH__ <= 700
+#if __CUDA_ARCH__ <= 730
 constexpr int kWarpSize = 32;
 #else
 #error Unknown __CUDA_ARCH__; please define parameters for compute capability
diff --git a/cmake/Modules_CUDA_fix/upstream/FindCUDA.cmake b/cmake/Modules_CUDA_fix/upstream/FindCUDA.cmake
index 78df03e89..28d5a64c8 100644
--- a/cmake/Modules_CUDA_fix/upstream/FindCUDA.cmake
+++ b/cmake/Modules_CUDA_fix/upstream/FindCUDA.cmake
@@ -1,5 +1,4 @@
-#.rst:
-# FindCUDA
+
 # --------
 #
 # .. note::
@@ -630,7 +629,10 @@ macro(cuda_unset_include_and_libraries)
   unset(CUDA_cudart_static_LIBRARY CACHE)
   unset(CUDA_cudadevrt_LIBRARY CACHE)
   unset(CUDA_cublas_LIBRARY CACHE)
-  unset(CUDA_cublas_device_LIBRARY CACHE)
+  if(CUDA_VERSION VERSION_LESS "10.0")
+    # cublas_device is removed in CUDA 10"
+    unset(CUDA_cublas_device_LIBRARY CACHE)
+  endif()
   unset(CUDA_cublasemu_LIBRARY CACHE)
   unset(CUDA_cufft_LIBRARY CACHE)
   unset(CUDA_cufftemu_LIBRARY CACHE)
@@ -981,7 +983,7 @@ if(NOT CUDA_VERSION VERSION_LESS "3.2")
     find_cuda_helper_libs(nvcuvid)
   endif()
 endif()
-if(CUDA_VERSION VERSION_GREATER "5.0")
+if(CUDA_VERSION VERSION_GREATER "5.0" AND CUDA_VERSION VERSION_LESS "10.0")
   find_cuda_helper_libs(cublas_device)
 endif()
 
@@ -1019,7 +1021,11 @@ if (CUDA_BUILD_EMULATION)
   set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublasemu_LIBRARY})
 else()
   set(CUDA_CUFFT_LIBRARIES ${CUDA_cufft_LIBRARY})
-  set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublas_LIBRARY} ${CUDA_cublas_device_LIBRARY})
+  if(CUDA_VERSION VERSION_LESS "10.0")
+    set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublas_LIBRARY} ${CUDA_cublas_device_LIBRARY})
+  else()
+    set(CUDA_CUBLAS_LIBRARIES ${CUDA_cublas_LIBRARY})
+  endif()
 endif()
 
 ########################
@@ -1787,26 +1793,50 @@ function(CUDA_LINK_SEPARABLE_COMPILATION_OBJECTS output_file cuda_target options
     endif()
 
     if (do_obj_build_rule)
-      add_custom_command(
-        OUTPUT ${output_file}
-        DEPENDS ${object_files}
-        COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} -dlink ${object_files} ${CUDA_cublas_device_LIBRARY} -o ${output_file}
-        ${flags}
-        COMMENT "Building NVCC intermediate link file ${output_file_relative_path}"
-        COMMAND_EXPAND_LISTS
-        ${_verbatim}
-        )
+      if(CUDA_VERSION VERSION_LESS "10.0")
+        add_custom_command(
+          OUTPUT ${output_file}
+          DEPENDS ${object_files}
+          COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} -dlink ${object_files} ${CUDA_cublas_device_LIBRARY} -o ${output_file}
+          ${flags}
+          COMMENT "Building NVCC intermediate link file ${output_file_relative_path}"
+          COMMAND_EXPAND_LISTS
+          ${_verbatim}
+          )
+      else()
+        add_custom_command(
+          OUTPUT ${output_file}
+          DEPENDS ${object_files}
+          COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} -dlink ${object_files} -o ${output_file}
+          ${flags}
+          COMMENT "Building NVCC intermediate link file ${output_file_relative_path}"
+          COMMAND_EXPAND_LISTS
+          ${_verbatim}
+          )
+      endif()
     else()
       get_filename_component(output_file_dir "${output_file}" DIRECTORY)
-      add_custom_command(
-        TARGET ${cuda_target}
-        PRE_LINK
-        COMMAND ${CMAKE_COMMAND} -E echo "Building NVCC intermediate link file ${output_file_relative_path}"
-        COMMAND ${CMAKE_COMMAND} -E make_directory "${output_file_dir}"
-        COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} ${flags} -dlink ${object_files} ${CUDA_cublas_device_LIBRARY} -o "${output_file}"
-        COMMAND_EXPAND_LISTS
-        ${_verbatim}
-        )
+      if(CUDA_VERSION VERSION_LESS "10.0")
+        add_custom_command(
+          TARGET ${cuda_target}
+          PRE_LINK
+          COMMAND ${CMAKE_COMMAND} -E echo "Building NVCC intermediate link file ${output_file_relative_path}"
+          COMMAND ${CMAKE_COMMAND} -E make_directory "${output_file_dir}"
+          COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} ${flags} -dlink ${object_files} ${CUDA_cublas_device_LIBRARY} -o "${output_file}"
+          COMMAND_EXPAND_LISTS
+          ${_verbatim}
+          )
+      else()
+        add_custom_command(
+          TARGET ${cuda_target}
+          PRE_LINK
+          COMMAND ${CMAKE_COMMAND} -E echo "Building NVCC intermediate link file ${output_file_relative_path}"
+          COMMAND ${CMAKE_COMMAND} -E make_directory "${output_file_dir}"
+          COMMAND ${CUDA_NVCC_EXECUTABLE} ${nvcc_flags} ${flags} -dlink ${object_files} -o "${output_file}"
+          COMMAND_EXPAND_LISTS
+          ${_verbatim}
+          )
+      endif()
     endif()
  endif()
 endfunction()
@@ -1995,7 +2025,11 @@ macro(CUDA_ADD_CUBLAS_TO_TARGET target)
   if (CUDA_BUILD_EMULATION)
     target_link_libraries(${target} ${CUDA_LINK_LIBRARIES_KEYWORD} ${CUDA_cublasemu_LIBRARY})
   else()
-    target_link_libraries(${target} ${CUDA_LINK_LIBRARIES_KEYWORD} ${CUDA_cublas_LIBRARY} ${CUDA_cublas_device_LIBRARY})
+    if(CUDA_VERSION VERSION_LESS "10.0")
+      target_link_libraries(${target} ${CUDA_LINK_LIBRARIES_KEYWORD} ${CUDA_cublas_LIBRARY} ${CUDA_cublas_device_LIBRARY})
+    else()
+      target_link_libraries(${target} ${CUDA_LINK_LIBRARIES_KEYWORD} ${CUDA_cublas_LIBRARY})
+    endif()
   endif()
 endmacro()
 
@@ -2021,3 +2055,4 @@ macro(CUDA_BUILD_CLEAN_TARGET)
   # have been removed.
   set(CUDA_ADDITIONAL_CLEAN_FILES "" CACHE INTERNAL "List of intermediate files that are part of the cuda dependency scanning.")
 endmacro()
+