Merge branch 'develop' into amd-develop

include/ck_tile/core/tensor/null_tile_window.hpp

@@ -35,6 +35,8 @@ struct null_tile_window
 
     CK_TILE_DEVICE constexpr auto get_window_origin() const { return BottomTensorIndex{}; }
 
+    CK_TILE_DEVICE void init_raw() {}
+
     WindowLengths window_lengths_;
 };
 

include/ck_tile/core/tensor/tensor_view.hpp

@@ -36,6 +36,8 @@ struct tensor_view
     {
     }
 
+    CK_TILE_HOST_DEVICE void init_raw() { buf_.init_raw(); }
+
     CK_TILE_HOST_DEVICE constexpr auto& get_tensor_descriptor() const { return desc_; }
 
     CK_TILE_HOST_DEVICE static constexpr index_t get_num_of_dimension()
@@ -85,30 +87,34 @@ struct tensor_view
     // "coord" is coordinate of DataType, not X. "coord" should be aligned to X
     template <typename X,
               bool oob_conditional_check = true,
+              bool pre_nop               = false,
               typename std::enable_if<
                   std::is_same_v<typename vector_traits<remove_cvref_t<X>>::scalar_type,
                                  typename vector_traits<remove_cvref_t<DataType>>::scalar_type>,
                   bool>::type = false>
-    CK_TILE_HOST_DEVICE void
-    get_vectorized_elements_raw(remove_cvref_t<X>& dst,
-                                const TensorCoord& coord,
-                                bool_constant<oob_conditional_check> = {}) const
+    CK_TILE_HOST_DEVICE void get_vectorized_elements_raw(remove_cvref_t<X>& dst,
+                                                         const TensorCoord& coord,
+                                                         bool_constant<oob_conditional_check> = {},
+                                                         bool_constant<pre_nop> = {}) const
     {
-        return buf_.template get_raw<X, oob_conditional_check>(
+        return buf_.template get_raw<X, oob_conditional_check, pre_nop>(
             dst,
             coord.get_offset(),
-            coordinate_has_valid_offset_assuming_top_index_is_valid(desc_, coord));
+            coordinate_has_valid_offset_assuming_top_index_is_valid(desc_, coord),
+            bool_constant<pre_nop>{});
     }
 
     template <typename X,
+              bool pre_nop = false,
               typename std::enable_if<
                   std::is_same_v<typename vector_traits<remove_cvref_t<X>>::scalar_type,
                                  typename vector_traits<remove_cvref_t<DataType>>::scalar_type>,
                   bool>::type = false>
-    CK_TILE_HOST_DEVICE constexpr void async_get_vectorized_elements(remove_cvref_t<DataType>* smem,
-                                                                     const TensorCoord& coord) const
+    CK_TILE_HOST_DEVICE constexpr void async_get_vectorized_elements_raw(
+        remove_cvref_t<DataType>* smem, const TensorCoord& coord, bool_constant<pre_nop> = {}) const
     {
-        return buf_.template async_get<X>(smem, coord.get_offset(), true /*not used*/);
+        return buf_.template async_get_raw<X>(
+            smem, coord.get_offset(), true /*not used*/, bool_constant<pre_nop>{});
     }
 
     // X is vector of DataType.

library/include/ck/library/utility/host_tensor.hpp

@@ -43,7 +43,15 @@ std::ostream& LogRangeAsType(std::ostream& os, Range&& range, std::string delim)
             first = false;
         else
             os << delim;
-        os << static_cast<T>(v);
+
+        if constexpr(std::is_same_v<T, ck::f8_t> || std::is_same_v<T, ck::bf8_t>)
+        {
+            os << ck::type_convert<float>(v);
+        }
+        else
+        {
+            os << static_cast<T>(v);
+        }
     }
     return os;
 }