GlobalParameters:
  EnqueuesPerSync: 1
  LibraryPrintDebug: 0
  NumElementsToValidate: 1000
  KernelTime: True
  SleepPercent: 0
  DataInitTypeBeta : 0
  DataInitTypeA: 3
  DataInitTypeB: 3
  DataInitTypeC: 4
  DataInitTypeAlpha: 2
  DataInitTypeBeta: 0
  PrintSolutionRejectionReason: 0
  PrintWinnersOnly: 0
  PrintTensorA: 0
  PrintTensorB: 0
  PrintTensorD: 0

BenchmarkProblems:
  ########################################
  # NT - standard
  ########################################
  -
    - # ProblemType
      OperationType: TensorContraction
      DataType: s
      UseBeta: False
      NumIndicesC: 2
      IndexAssignmentsA: [ 3, 0, 2]
      IndexAssignmentsB: [ 3, 1, 2]
      # 3 is Cin, 2 is batch index converted to a second summation index
      #
      # Standard batched NT is :
      #  IndicesFree:  [0, 1]  # 0 is H*W, 1 is Cout
      #  IndicesBatch: [2]  # N, but convert to summation in the config above
      #  IndicesSum:   [3]
      #  IndexAssignmentsA:   [3, 0, 2]
      #  IndexAssignmentsB:   [3, 1, 2]

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - KernelLanguage: ["Assembly"]
        - PrefetchLocalRead: [False, True]
        - PrefetchGlobalRead: [False, True]
        - ThreadTile:
          - [ 2, 2 ]
          - [ 8, 8 ]
        - WorkGroup:
          - [ 8, 8, 1 ]
          - [ 16, 16, 1 ]
        - StaggerU: [0,32]
        - DepthU: [ 4,8]
        - VectorWidth: [1,-1]
        - FractionalLoad: [0,1]
      BenchmarkForkParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [127,1,129], 0, [1,5], [4,4,32]]
          - Range: [ [16], [16], [1,1,4], [32] ]
          - Range: [ [16,1,128], [16], [1,1,4], [32] ]
          - Exact: [ 32, 32, 2, 4]
          - Exact: [192, 1024, 64, 192]