Fix cudaAPI wrapper errors for CUDA_MALLOC_ASYNC (kokkos#6346)

* Fix errors relating to KOKKOS_ENABLE_IMPL_CUDA_MALLOC_ASYNC

- default arg for cupda api wrappers fix
- input stream name for wrappers used with typo

* add to CI build

* cast int -> size_t to avoid comparison warning

* Change def of memory_threashold_g to size_t

.jenkins

@@ -458,6 +458,7 @@ pipeline {
                                 -DKokkos_ENABLE_CUDA_LAMBDA=ON \
                                 -DKokkos_ENABLE_LIBDL=OFF \
                                 -DKokkos_ENABLE_IMPL_MDSPAN=ON \
+                                -DKokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC=ON \
                               .. && \
                               make -j8 && ctest --verbose && \
                               cd ../example/build_cmake_in_tree && \

core/src/Cuda/Kokkos_CudaSpace.cpp

@@ -141,7 +141,7 @@ CudaUVMSpace::CudaUVMSpace() : m_device(Kokkos::Cuda().cuda_device()) {}
 
 CudaHostPinnedSpace::CudaHostPinnedSpace() {}
 
-int memory_threshold_g = 40000;  // 40 kB
+size_t memory_threshold_g = 40000;  // 40 kB
 
 //==============================================================================
 // <editor-fold desc="allocate()"> {{{1

core/src/Cuda/Kokkos_Cuda_Instance.hpp

@@ -487,16 +487,16 @@ class CudaInternal {
 #if (defined(KOKKOS_ENABLE_IMPL_CUDA_MALLOC_ASYNC) && CUDART_VERSION >= 11020)
   template <bool setCudaDevice = true>
   cudaError_t cuda_malloc_async_wrapper(void** devPtr, size_t size,
-                                        cudaStream_t hStream == nullptr) const {
+                                        cudaStream_t hStream = nullptr) const {
     if constexpr (setCudaDevice) set_cuda_device();
-    return cudaMallocAsync(devPtr, size, get_input_stream(stream));
+    return cudaMallocAsync(devPtr, size, get_input_stream(hStream));
   }
 
   template <bool setCudaDevice = true>
   cudaError_t cuda_free_async_wrapper(void* devPtr,
-                                      cudaStream_t hStream == nullptr) const {
+                                      cudaStream_t hStream = nullptr) const {
     if constexpr (setCudaDevice) set_cuda_device();
-    return cudaFreeAsync(devPtr, get_input_stream(stream));
+    return cudaFreeAsync(devPtr, get_input_stream(hStream));
   }
 #endif