48   typename InterleavedTileShape,  
    72     InterleavedTileShape::kM / InstructionShape::kM,
    73     InterleavedTileShape::kN / InstructionShape::kN
    78     WarpShape::kM / InterleavedTileShape::kM,
    79     WarpShape::kN / InterleavedTileShape::kN
    83   static int const kElementsPerMma = 8;
    84   static int const kRowsPerIteration = 16;
    91   static int const kElementsPerAccess = 4;
    94   static int const kAccessesPerInterleavedTile = 4;
    97   static int const kIterations = TileIterations::kRow * 2;
   109     kElementsPerAccess * kAccessesPerInterleavedTile * TileIterations::kColumn>;
   114     TileIterations::kCount * MmaIterations::kCount * kElementsPerMma>;
   127   using ElementC = float;
   135     InterleavedTileShape::kM / InstructionShape::kM,
   136     InterleavedTileShape::kN / InstructionShape::kN
   141     WarpShape::kM / InterleavedTileShape::kM,
   142     WarpShape::kN / InterleavedTileShape::kN
   146   static int const kElementsPerMma = 8;
   147   static int const kRowsPerIteration = 16;
   154   static int const kElementsPerAccess = 2;
   157   static int const kAccessesPerInterleavedTile = 8;
   160   static int const kRowsPerMmaTile = 2;
   163   static int const kIterations = TileIterations::kRow * MmaIterations::kRow;
   175     kElementsPerAccess * kAccessesPerInterleavedTile * TileIterations::kColumn>;
   180     TileIterations::kCount * MmaIterations::kCount * kElementsPerMma>;
 Describes the size of a matrix tile. 
Definition: matrix_shape.h:42
Definition: aligned_buffer.h:35
Array< ElementC, kElementsPerAccess *kAccessesPerInterleavedTile *TileIterations::kColumn > Fragment
This is the fragment size produced by one access of the iterator. 
Definition: volta_tensor_op_policy.h:175
Aligned array type. 
Definition: array.h:511
IEEE half-precision floating-point type. 
Definition: half.h:126
Defines common types used for all GEMM-like operators. 
Defines a Shape template for matrix tiles. 
WarpShape_ WarpShape
Definition: volta_tensor_op_policy.h:125
Shape of a matrix multiply-add operation. 
Definition: include/cutlass/gemm/gemm.h:57
Mapping function for row-major matrices. 
Definition: layout/matrix.h:50
Array< ElementC, kElementsPerAccess *kAccessesPerInterleavedTile *TileIterations::kColumn > Fragment
This is the fragment size produced by one access of the iterator. 
Definition: volta_tensor_op_policy.h:109
Defines layout functions used by TensorRef and derived classes. 
Policy details related to the epilogue. 
Definition: volta_tensor_op_policy.h:52
Array< ElementC, TileIterations::kCount *MmaIterations::kCount *kElementsPerMma > AccumulatorTile
This is the complete warp-level accumulator tile. 
Definition: volta_tensor_op_policy.h:180
Array< ElementC, TileIterations::kCount *MmaIterations::kCount *kElementsPerMma > AccumulatorTile
This is the complete warp-level accumulator tile. 
Definition: volta_tensor_op_policy.h:114
WarpShape_ WarpShape
Definition: volta_tensor_op_policy.h:62