fix #226: workaround for cuda 9.0/9.1

src/cuhash/CMakeLists.txt

 if(WIN32)
     add_library(cuhash SHARED hash_functions.cu hash_table.cpp hash_table.cu hash_functions.cpp)
 else()
-    add_library(cuhash STATIC hash_functions.cu hash_table.cpp hash_table.cu hash_functions.cpp)
+    add_library(cuhash SHARED hash_functions.cu hash_table.cpp hash_table.cu hash_functions.cpp)
 endif()
 target_include_directories(cuhash PRIVATE ${ALL_INCLUDE} )
 set_property(TARGET cuhash PROPERTY CUDA_STANDARD 14)

src/spconv/cublas_gemm.cc

@@ -46,8 +46,5 @@ cublasStatus_t cublasTgemm(cublasHandle_t handle, cublasOperation_t transa,
                      beta, C, ldc);
 }
 
-template <> inline __half constant_scalar(float data) {
-  return __float2half(data);
-}
 
 } // namespace spconv
\ No newline at end of file

src/spconv/reordering.cu

@@ -29,12 +29,17 @@ namespace spconv {
 
 using float_types_t = tv::mp_list<float, double, at::Half>;
 using int_types_t = tv::mp_list<int32_t, int64_t>;
+
+
 template <typename T>
-using half_vec_t =
-    std::conditional_t<std::is_same<T, at::Half>::value, int4, int4>;
+struct half_vec{
+  using type = typename std::conditional_t<std::is_same<T, at::Half>::value, int4, int4>;
+};
+
 template <typename T>
-using half_vec_sadd_t =
-    std::conditional_t<std::is_same<T, at::Half>::value, int4, int4>;
+struct half_vec_sadd{
+  using type = typename std::conditional_t<std::is_same<T, at::Half>::value, int4, int4>;
+};
 using kernel_block_t = tv::mp_list_c<int, 64, 32, 16>;
 
 void sparse_gather_cuda(torch::Tensor buffer, torch::Tensor features,
@@ -47,7 +52,7 @@ void sparse_gather_cuda(torch::Tensor buffer, torch::Tensor features,
   auto inds_dtype = indices.scalar_type();
   tv::DispatchTorch<float_types_t>()(dtype, [&](auto TValue) {
     using T = decltype(TValue);
-    using vecload_type_t = half_vec_t<T>;
+    using vecload_type_t = typename half_vec_sadd<T>::type;
     tv::DispatchTorch<int_types_t>()(inds_dtype, [&](auto IndexValue) {
       using Index = decltype(IndexValue);
       bool notFound = true;
@@ -136,7 +141,7 @@ void sparse_scatter_add_cuda(torch::Tensor buffer, torch::Tensor outFeatures,
 
   tv::DispatchTorch<float_types_t>()(dtype, [&](auto TValue) {
     using T = decltype(TValue);
-    using vecload_type_t = half_vec_sadd_t<T>;
+    using vecload_type_t = typename half_vec_sadd<T>::type;
     tv::DispatchTorch<int_types_t>()(inds_dtype, [&](auto IndexValue) {
       using Index = decltype(IndexValue);
       bool notFound = true;
@@ -231,7 +236,7 @@ void batch_sparse_gather_cuda(torch::Tensor buffer, torch::Tensor features,
   int feature_stride = buffer.size(1);
   tv::DispatchTorch<float_types_t>()(dtype, [&](auto TValue) {
     using T = decltype(TValue);
-    using vecload_type_t = half_vec_t<T>;
+    using vecload_type_t = typename half_vec<T>::type;
     tv::DispatchTorch<int_types_t>()(inds_dtype, [&](auto IndexValue) {
       using Index = decltype(IndexValue);
       bool notFound = true;
@@ -304,7 +309,7 @@ void batch_sparse_scatter_add_cuda(torch::Tensor buffer,
 
   tv::DispatchTorch<float_types_t>()(dtype, [&](auto TValue) {
     using T = decltype(TValue);
-    using vecload_type_t = half_vec_sadd_t<T>;
+    using vecload_type_t = typename half_vec_sadd<T>::type;
     tv::DispatchTorch<int_types_t>()(inds_dtype, [&](auto IndexValue) {
       using Index = decltype(IndexValue);
       bool notFound = true;