GlobalParameters:
  MinimumRequiredVersion: 4.0.1
  PrintLevel: 1
  ForceRedoBenchmarkProblems: True
  ForceRedoLibraryLogic: True
  ForceRedoLibraryClient: True
  CMakeBuildType: Release
  EnqueuesPerSync: 1
  SyncsPerBenchmark: 1
  LibraryPrintDebug: False
  DebugKernel: False
  NumElementsToValidate: 1000
  ValidationMaxToPrint: 100
  ValidationPrintValids: False
  ShortNames: False
  MergeFiles: True
  Platform: 0
  Device: 0
  KernelTime: True
  PinClocks: True
  SleepPercent: 200
  PrintSolutionRejectionReason: True
  DataInitTypeA: 5           # 0=0, 1=1, 2=serial, 3=rand, 4=NaN
  DataInitTypeB: 5           # 0=0, 1=1, 2=serial, 3=rand, 4=NaN
  DataInitTypeBeta: 0        # 0=0, 1=1, 2=serial, 3=rand, 4=NaN
  DataInitTypeC : 0          # 0=0, 1=1, 2=serial, 3=rand, 4=NaN
  PrintTensorA : 0
  PrintTensorB : 0
  CodeFromFiles: 1

BenchmarkProblems:
  ########################################
  # NT - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: 1
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - LoopTail: [True]
        - PrefetchLocalRead: [False]
      ForkParameters:
        - FractionalLoad: [1]
        - PrefetchGlobalRead: [False]
        - ThreadTile:
          - [ 4, 4 ]
        - WorkGroup:
          - [ 8, 8, 1 ]
        - WorkGroupMapping: [8]
        - LocalDotLayout: [2]
        - InnerUnroll: [2]
        - GlobalSplitU: [1]
        - DepthU: [ 4 ]
        - GlobalReadVectorWidth: [-1]
        - VectorWidth: [4]
        - AssertSummationElementMultiple: [2]
      BenchmarkForkParameters:
      JoinParameters:
        - MacroTile
        - GlobalSplitU
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          #- Range: [ [16], [16], [1], [16] ]
          #- Range: [ [32], [32], [1], [16] ]
          - Range: [ [64], [64], [1], [64] ]

  ########################################