From f11272516a85c4345e0ff644ebb2d3d0628729e4 Mon Sep 17 00:00:00 2001
From: cliffburdick <cburdick@nvidia.com>
Date: Thu, 8 Aug 2024 16:21:34 -0700
Subject: [PATCH] Fixed issues with host compiler with C++17 and C++20 modes

---
 include/matx/core/half.h            | 12 ++++++++++--
 include/matx/core/storage.h         |  8 ++++----
 include/matx/operators/scalar_ops.h |  4 ++++
 3 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/include/matx/core/half.h b/include/matx/core/half.h
index 9fbd9e03a..0aba9a93c 100644
--- a/include/matx/core/half.h
+++ b/include/matx/core/half.h
@@ -701,7 +701,11 @@ __MATX_HOST__ __MATX_DEVICE__ __MATX_INLINE__ matxHalf<T> rsqrt(const matxHalf<T
 #ifdef __CUDA_ARCH__
   return hrsqrt(x.x);
 #else
-  return static_cast<T>(::rsqrt(static_cast<float>(x.x)));
+  #ifdef __CUDACC__
+    return static_cast<__nv_bfloat16>(::rsqrt(static_cast<float>(x.x)));
+  #else
+    return static_cast<__nv_bfloat16>(1.f / static_cast<float>(x.x));
+  #endif
 #endif
 }
 
@@ -719,7 +723,11 @@ rsqrt(const matxHalf<__nv_bfloat16> &x)
 #if __CUDA_ARCH__ >= 800
   return hrsqrt(x.x);
 #else
-  return static_cast<__nv_bfloat16>(::rsqrt(static_cast<float>(x.x)));
+  #ifdef __CUDACC__
+    return static_cast<__nv_bfloat16>(::rsqrt(static_cast<float>(x.x)));
+  #else
+    return static_cast<__nv_bfloat16>(1.f / static_cast<float>(x.x));
+  #endif
 #endif
 }
 
diff --git a/include/matx/core/storage.h b/include/matx/core/storage.h
index 112b5d8fe..06e6f3c6c 100644
--- a/include/matx/core/storage.h
+++ b/include/matx/core/storage.h
@@ -282,7 +282,7 @@ namespace matx
      * @brief Default construct a smart_pointer_buffer. This should only be used when temporarily
      * creating an empty tensor for construction later.
     */
-    smart_pointer_buffer<T>() {};
+    smart_pointer_buffer() {};
 
     /**
      * @brief Construct a new smart pointer buffer from an existing object
@@ -290,7 +290,7 @@ namespace matx
      * @param ptr Smart poiner object
      * @param size Size of allocation
      */
-    smart_pointer_buffer<T>(T &&ptr, size_t size) : data_(std::forward<T>(ptr)), size_(size) {
+    smart_pointer_buffer(T &&ptr, size_t size) : data_(std::forward<T>(ptr)), size_(size) {
       static_assert(is_smart_ptr_v<T>);
     }
 
@@ -328,7 +328,7 @@ namespace matx
      * 
      * @param rhs Object to move from
      */
-    smart_pointer_buffer<T>(smart_pointer_buffer<T> &&rhs) {
+    smart_pointer_buffer(smart_pointer_buffer<T> &&rhs) {
       size_ = rhs.size_; 
       data_ = std::move(rhs.data_);
     }    
@@ -337,7 +337,7 @@ namespace matx
      * @brief Default destructor
      * 
      */  
-    ~smart_pointer_buffer<T>() = default;
+    ~smart_pointer_buffer() = default;
 
     /**
      * @brief Get underlying data pointer
diff --git a/include/matx/operators/scalar_ops.h b/include/matx/operators/scalar_ops.h
index fa7e27168..661aa73b7 100644
--- a/include/matx/operators/scalar_ops.h
+++ b/include/matx/operators/scalar_ops.h
@@ -162,7 +162,11 @@ static __MATX_INLINE__ __MATX_HOST__ __MATX_DEVICE__ auto _internal_rsqrt(T v1)
     return rsqrt(v1);
   }
   else {
+#ifdef __CUDACC__
     return ::rsqrt(v1);
+#else
+    return static_cast<T>(1) / v1;
+#endif
   }
 }
 template <typename T> struct RSqrtF {