# A config to generate a fast kernel for large DGEMMs
GlobalParameters:
  EnqueuesPerSync: 1
  LibraryPrintDebug: False
  NumElementsToValidate: 0
  KernelTime: True
  #PinClocks: True
  SleepPercent: 0
  DataInitTypeBeta : 0
  PrintSolutionRejectionReason: 0
  PrintWinnersOnly: 1

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

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - PrefetchLocalRead: [True]
        - PrefetchGlobalRead: [True]
        - ThreadTile:
          - [ 8, 8 ]
        - WorkGroup:
          - [ 16, 32, 1 ]
          - [ 32, 16, 1 ]
          - [ 16, 16, 1 ]
        - WorkGroupMapping: [1,8,64]
        - StaggerU: [0,32]
        - StaggerUMapping: [0,1]
        - DepthU: [ 8,16 ]
        - VectorWidth: [4]
        - FractionalLoad: [1]
        - AssertSummationElementMultiple: [4]
        - AssertFree0ElementMultiple: [4]
        - AssertFree1ElementMultiple: [4]
      BenchmarkForkParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # A couple classic large matrix problems:
          - Range: [ [4096], [4096], [1], [4096] ]
          - Range: [ [5760], [5760], [1], [4096] ]
          - Range: [ [5760], [5760], [1], [5760] ]
          - Range: [ [7744], [7744], [1], [7744] ]

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