GlobalParameters:
  MinimumRequiredVersion: 4.4.0
  PrintLevel: 1
  ForceRedoBenchmarkProblems: True
  ForceRedoLibraryLogic: True
  ForceRedoLibraryClient: True
  CMakeBuildType: Release
  EnqueuesPerSync: 1
  SyncsPerBenchmark: 2
  LibraryPrintDebug: False
  NumElementsToValidate: 128
  ValidationMaxToPrint: 4
  ValidationPrintValids: False
  ShortNames: False
  MergeFiles: True
  Platform: 0
  Device: 0
  KernelTime: True
  DataInitTypeBeta : 0

BenchmarkProblems:
  - # cgemm NN
    - # ProblemType
      OperationType: GEMM
      DataType: c
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - ProblemSizes:
          - Exact: [ 2880, 2880, 2, 1536 ]
        - EdgeType: ["ShiftPtr"]
        - LoopTail: [True]
      ForkParameters:
        - ThreadTile:
          - [ 4, 8 ]
          - [ 4, 4 ]
          - [ 2, 4 ]
        - WorkGroup:
          - [  4,  4, 16 ]
          - [  4,  8,  8 ]
          - [  8,  8,  4 ]
          - [  8, 16,  2 ]
          - [ 16, 16,  1 ]
        - DepthU: [ 4, 8, 16, 32 ]
        - PrefetchGlobalRead: [False, True]
        - PrefetchLocalRead: [False, True]
      BenchmarkForkParameters:
      JoinParameters:
        - MacroTile
      BenchmarkJoinParameters:
        - WorkGroupMapping: [1, 2, 4, 8, 16, 32, 64]
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [32, 32, 32, 400], [32, 32, 32, 400], [2], [1536] ]

  - # cgemm NT
    - # ProblemType
      OperationType: GEMM
      DataType: c
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - ProblemSizes:
          - Exact: [ 2880, 2880, 2, 1536 ]
        - EdgeType: ["ShiftPtr"]
        - LoopTail: [True]
      ForkParameters:
        - ThreadTile:
          - [ 4, 8 ]
          - [ 4, 4 ]
          - [ 2, 4 ]
        - WorkGroup:
          - [  4,  4, 16 ]
          - [  4,  8,  8 ]
          - [  8,  8,  4 ]
          - [  8, 16,  2 ]
          - [ 16, 16,  1 ]
        - PrefetchGlobalRead: [False, True]
        - PrefetchLocalRead: [False, True]
        - DepthU: [ 4, 8, 16, 32 ]
      BenchmarkForkParameters:
      JoinParameters:
        - MacroTile
      BenchmarkJoinParameters:
        - WorkGroupMapping: [1, 2, 4, 8, 16, 32, 64]
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [32, 32, 32, 400], [32, 32, 32, 400], [2], [1536] ]

  - # cgemm TN
    - # ProblemType
      OperationType: GEMM
      DataType: c
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - ProblemSizes:
          - Exact: [ 2880, 2880, 2, 1536 ]
        - EdgeType: ["ShiftPtr"]
        - LoopTail: [True]
      ForkParameters:
        - ThreadTile:
          - [ 4, 8 ]
          - [ 4, 4 ]
          - [ 2, 4 ]
        - WorkGroup:
          - [  4,  4, 16 ]
          - [  4,  8,  8 ]
          - [  8,  8,  4 ]
          - [  8, 16,  2 ]
          - [ 16, 16,  1 ]
        - PrefetchGlobalRead: [False, True]
        - PrefetchLocalRead: [False, True]
        - DepthU: [ 4, 8, 16, 32 ]
      BenchmarkForkParameters:
      JoinParameters:
        - MacroTile
      BenchmarkJoinParameters:
        - WorkGroupMapping: [1, 2, 4, 8, 16, 32, 64]
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [32, 32, 32, 400], [32, 32, 32, 400], [2], [1536] ]

  - # cgemm TT
    - # ProblemType
      OperationType: GEMM
      DataType: c
      TransposeA: True
      TransposeB: True
      UseBeta: True
      Batched: True

    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
        - WorkGroupMapping: [1]
      BenchmarkCommonParameters:
        - ProblemSizes:
          - Exact: [ 2880, 2880, 2, 1536 ]
        - EdgeType: ["ShiftPtr"]
        - LoopTail: [True]
      ForkParameters:
        - ThreadTile:
          - [ 4, 8 ]
          - [ 4, 4 ]
          - [ 2, 4 ]
        - WorkGroup:
          - [  4,  4, 16 ]
          - [  4,  8,  8 ]
          - [  8,  8,  4 ]
          - [  8, 16,  2 ]
          - [ 16, 16,  1 ]
        - PrefetchGlobalRead: [False, True]
        - PrefetchLocalRead: [False, True]
        - DepthU: [ 4, 8, 16, 32 ]
      BenchmarkForkParameters:
      JoinParameters:
        - MacroTile
      BenchmarkJoinParameters:
        - WorkGroupMapping: [1, 2, 4, 8, 16, 32, 64]
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [32, 32, 32, 400], [32, 32, 32, 400], [2], [1536] ]

LibraryLogic:
    ScheduleName: "vega10"
    DeviceNames: ["Device 6863", "Device 6862", "Device 687f", "Device 6860", "Device 6861", "Vega 10 XTX [Radeon Vega Frontier Edition]", "Vega [Radeon RX Vega]",
        Vega, Device 6864, Device 686c]
    ArchitectureName: "gfx900"

#   ScheduleName: "mi25"
#   DeviceNames: ["Device 6860"]
#   ArchitectureName: "gfx900"

#   ScheduleName: "r9nano"
#   DeviceNames: ["Device 7300"]
#   ArchitectureName: "gfx803"

#   ScheduleName: "hip"
#   DeviceNames: ["Device 0000"]
#   ArchitectureName: "fallback"

LibraryClient: