TestParameters:
  marks: [skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx950, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201, skip-gfx1250] # not supported by arch

GlobalParameters:
  SyncsPerBenchmark: 0
  MinimumRequiredVersion: 5.0.0
  SleepPercent: 50
  NumElementsToValidate: -1
  DataInitTypeBias: 0
  DataInitTypeBeta: 1
  DataInitTypeAlpha: 1
  DataInitTypeScaleA: 2
  DataInitTypeScaleB: 3
  CSVExportWinner: true
  CSVMergeSameProblemID: true
  Device: 0
  MaxWorkspaceSize: 3355443200

BenchmarkProblems:
  ########################################
  # TN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8B8N
      DestDataType: S
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
      UseBias: 1
      BiasDataTypeList: [s,b]
      UseScaleAB: "Scalar"
      Activation:    True
      ActivationType: hipblaslt_all
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16,16,32, 1, 1, 1,1, 1,1]  # 16x16
          - [16,16,32, 1, 1, 1,1, 2,2]  # 32x32
          - [16,16,32, 1, 1, 2,2, 1,1]  # 32x32
          - [16,16,32, 1, 1, 1,3, 4,1]  # 64x48
          - [16,16,32, 1, 1, 2,2, 2,2]  # 64x64
          - [32,32, 16, 1, 1, 1,1, 1,1]  # 32x32
          - [32,32, 16, 1, 1, 1,1, 2,2]  # 64x64
          - [32,32, 16, 1, 1, 2,2, 1,1]  # 64x64
          - [32,32, 16, 1, 1, 2,2, 2,2]  # 128x128
        - DepthU: [ 32, 64 ]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - VectorWidthA: [1]
        - VectorWidthB: [1]
        - GlobalReadVectorWidthA: [-1]
        - GlobalReadVectorWidthB: [1,-1]
        - LocalReadVectorWidth: [8]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1]
        - ExpandPointerSwap: [true]
        - TransposeLDS: [1]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - WaveSeparateGlobalReadB: [1]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1, 2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0,2]
        - StorePriorityOpt: [false, true]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [4608, 1335, 1, 640]
          - Exact: [4608, 1408, 1, 640]
          - Exact: [128,   128, 1, 640]
        - BiasTypeArgs: ['s', 'b']
        - ActivationArgs:
          - [Enum: Relu]

  ########################################
  # NT - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8B8N
      DestDataType: S
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
      UseBias: 1
      BiasDataTypeList: [s,b]
      UseScaleAB: "Scalar"
      Activation:    True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16,16,32, 1, 1, 1,1, 1,1]  # 16x16
          - [16,16,32, 1, 1, 1,1, 2,2]  # 32x32
          - [16,16,32, 1, 1, 2,2, 1,1]  # 32x32
          - [16,16,32, 1, 1, 1,3, 4,1]  # 64x48
          - [16,16,32, 1, 1, 2,2, 2,2]  # 64x64
          - [32,32, 16, 1, 1, 1,1, 1,1]  # 32x32
          - [32,32, 16, 1, 1, 1,1, 2,2]  # 64x64
          - [32,32, 16, 1, 1, 2,2, 1,1]  # 64x64
          - [32,32, 16, 1, 1, 2,2, 2,2]  # 128x128
        - DepthU: [ 32, 64 ]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - GlobalReadVectorWidthA: [-1]
        - GlobalReadVectorWidthB: [1,-1]
        - VectorWidthA: [1]
        - VectorWidthB: [1]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1]
        - ExpandPointerSwap: [true]
        - TransposeLDS: [0]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - WaveSeparateGlobalReadB: [1]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1, 2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0,2]
        - StorePriorityOpt: [false, true]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [4608, 1335, 1, 640]
          - Exact: [4608, 1408, 1, 640]
          - Exact: [128,   128, 1, 640]
        - BiasTypeArgs: ['s', 'b']
        - ActivationArgs:
          - [Enum: none]