3.6.0 update (#2005)

* 3.6.0 update

* doc and swap stuff

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>

include/cute/pointer_base.hpp

@@ -30,9 +30,10 @@
  **************************************************************************************************/
 #pragma once
 
-#include <cute/config.hpp>                 // CUTE_HOST_DEVICE
-#include <cute/numeric/numeric_types.hpp>  // cute::sizeof_bits
-#include <cute/util/type_traits.hpp>       // cute::declval, cute::void_t, etc
+#include <cute/config.hpp>                     // CUTE_HOST_DEVICE
+#include <cute/numeric/numeric_types.hpp>      // cute::sizeof_bits
+#include <cute/numeric/integral_constant.hpp>  // Int<0>
+#include <cute/util/type_traits.hpp>           // cute::declval, cute::void_t, etc
 
 namespace cute
 {
@@ -115,6 +116,14 @@ raw_pointer_cast(T* ptr) {
   return ptr;
 }
 
+// The statically-known alignment of a dynamic pointer is unknown
+template <class T>
+CUTE_HOST_DEVICE constexpr
+Int<0>
+max_alignment(T*) {
+  return {};
+}
+
 //
 // A very simplified iterator adaptor.
 // Derived classed may override methods, but be careful to reproduce interfaces exactly.
@@ -169,6 +178,13 @@ raw_pointer_cast(iter_adaptor<I,D> const& x) {
   return raw_pointer_cast(x.ptr_);
 }
 
+template <class I, class D>
+CUTE_HOST_DEVICE constexpr
+auto
+max_alignment(iter_adaptor<I,D> const& x) {
+  return max_alignment(x.ptr_);
+}
+
 //
 // counting iterator -- quick and dirty
 //