CUTLASS 2.6 (#298)

CUTLASS 2.6
2021-07-23 00:40:53 -04:00
parent 6c29fe20ba
commit e5d51840e8
308 changed files with 32408 additions and 4722 deletions
@@ -51,6 +51,8 @@ namespace thread {
 template <
  typename ElementOutput_,                             ///< Data type used to load and store tensors
  int Count,                                           ///< Number of elements computed per operation
+                                                       ///< Usually it is 128/sizeof_bits<ElementOutput_>,
+                                                       ///< but we use 64 or 32 sometimes when there are not enough data to store
  typename ElementAccumulator_ = ElementOutput_,       ///< Accumulator data type
  typename ElementCompute_ = ElementOutput_,           ///< Data type used to compute linear combination
  FloatRoundStyle Round = FloatRoundStyle::round_to_nearest