CUTLASS 2.6 (#298)

CUTLASS 2.6

include/cutlass/transform/pitch_linear_thread_map.h

@@ -517,6 +517,9 @@ struct TransposePitchLinearThreadMap {
       layout::PitchLinearShape<ThreadMap::Iterations::kStrided,
                                ThreadMap::Iterations::kContiguous>;
 
+  static_assert(Iterations::kContiguous == 1,
+    "Contiguous iteration has to be one to reuse the same shared store function with those that don't need transpose");
+
   static_assert(Iterations::kCount, "Number of iterations must be non-zero");
 
   ///< Delta betweeen accesses (units of elements, concept: PitchLinearShape)
@@ -595,6 +598,9 @@ struct TransposePitchLinearThreadMapSimt {
 
     static_assert(Iterations::kCount, "Number of iterations must be non-zero");
 
+    static_assert(Iterations::kStrided == 1,
+      "Strided iteration has to be one to reuse the same shared store function with those that don't need transpose");
+
     /// Shape of access by each thread
     using ThreadAccessShape = typename ThreadMap::ThreadAccessShape;