TestParameters:
  marks: [skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, 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
  NumElementsToValidate: -1
  MinimumRequiredVersion: 5.0.0
  PrintLevel: 1
  PrintSolutionRejectionReason: True
  Device: 0
  CMakeBuildType: Release
  KernelTime: True
  MaxWorkspaceSize: 13421772800
  DataInitTypeAlpha: 1
  DataInitTypeBeta: 0
  BoundsCheck: 2

BenchmarkProblems:
  ########################################
  # TN BBS - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: b
      DestDataType: b
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      ActivationType: all
      UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16,16, 1,  1,   4, 4,  2,2 ] # 128x128
          - [16, 16,16, 1,  1,   4, 2,  2,2 ] # 128x128
          - [16, 16,16, 1,  1,   2, 2,  2,2 ] # 64x64
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - GlobalReadVectorWidthA: [-1,1,2,4]
        - GlobalReadVectorWidthB: [-1,1,2,4]
        - LocalReadVectorWidth: [-1]
        - TransposeLDS: [1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - StaggerU: [0,16]
        - StaggerUStride: [-1]
        - WorkGroupMapping: [8]
        - StaggerUMapping: [0]
        - WaveSeparateGlobalReadA: [0,1,2]
        - WaveSeparateGlobalReadB: [0,1,2]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1,2]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [16]
        - ClusterLocalRead: [1]
        - UnrollLoopSwapGlobalReadOrder: [1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [1,     127, 1, 127]
          - Exact: [2,     127, 1, 127]
          - Exact: [3,     127, 1, 127]
          - Exact: [1,     1,   1, 127]
          - Exact: [127,   1,   1, 127]
          - Exact: [127,   2,   1, 127]
          - Exact: [127,   3,   1, 127]
          - Exact: [127,   127, 1, 127]
          - Exact: [128,   128, 1, 128]
          - Exact: [129,   129, 1, 129]
          - Exact: [127,   127, 1, 128]
          - Exact: [127,   127, 1, 129]
          - Exact: [127,   128, 1, 576]
          - Exact: [129,   128, 1, 640]

        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

  ########################################
  # NN BBS - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: b
      DestDataType: b
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      ActivationType: all
      UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16,16, 1,  1,   4, 4,  2,2 ] # 128x128
          - [16, 16,16, 1,  1,   2, 4,  2,2 ] # 64x128
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - DepthU: [128]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - GlobalReadVectorWidthA: [-1,1,2,4]
        - GlobalReadVectorWidthB: [-1,1,2,4]
        - LocalReadVectorWidth: [-1]
        - TransposeLDS: [1]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - StaggerU: [0]
        - StaggerUStride: [-1]
        - WorkGroupMapping: [8]
        - StaggerUMapping: [0]
        - WaveSeparateGlobalReadA: [2]
        - WaveSeparateGlobalReadB: [2]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0]
        - ClusterLocalRead: [1]
        - UnrollLoopSwapGlobalReadOrder: [1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [1,     127, 1, 127]
          - Exact: [2,     127, 1, 127]
          - Exact: [3,     127, 1, 127]
          - Exact: [1,     1,   1, 127]
          - Exact: [127,   1,   1, 127]
          - Exact: [127,   2,   1, 127]
          - Exact: [127,   3,   1, 127]
          - Exact: [127,   127, 1, 127]
          - Exact: [128,   128, 1, 128]
          - Exact: [129,   129, 1, 129]
          - Exact: [127,   127, 1, 128]
          - Exact: [127,   127, 1, 129]
          - Exact: [127,   128, 1, 576]
          - Exact: [129,   128, 1, 640]

        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

  ########################################
  # TT BBS - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: b
      DestDataType: b
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: True
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      ActivationType: all
      UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16,16, 1,  1,   4, 2,  2, 2 ]
        - AssertFree0ElementMultiple: [1]
        - AssertFree1ElementMultiple: [1]
        - AssertSummationElementMultiple: [1]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - DepthU: [32]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - GlobalReadVectorWidthA: [1]
        - GlobalReadVectorWidthB: [1]
        - LocalReadVectorWidth: [-1]
        - TransposeLDS: [1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - StaggerU: [16]
        - StaggerUStride: [256]
        - StaggerUMapping: [1]
        - WorkGroupMapping: [1]
        - WorkGroupMappingXCC: [8]
        - WorkGroupMappingXCCGroup: [304]
        - 1LDSBuffer: [-1]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [0.5]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [8]
        - ClusterLocalRead: [1]
        - NonTemporalD: [0]
        - UnrollLoopSwapGlobalReadOrder: [1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [128,  128, 1, 1024  ]  
          - Exact: [128,  128, 1, 2048  ]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

  ########################################
  # NT BBS - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: b
      DestDataType: b
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      ActivationType: all
      UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16,16, 1,  1,   3, 4,  1, 4 ]
          - [16, 16,16, 1,  1,   4, 2,  2, 2 ]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - DepthU: [32]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - GlobalReadVectorWidthA: [-1]
        - GlobalReadVectorWidthB: [-1]
        - LocalReadVectorWidth: [-1]
        - TransposeLDS: [0]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - StaggerU: [0]
        - StaggerUStride: [-1]
        - StaggerUMapping: [0]
        - WorkGroupMapping: [1]
        - WorkGroupMappingXCC: [1]
        - WorkGroupMappingXCCGroup: [304]
        - 1LDSBuffer: [-1]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [0.5]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [16]
        - ClusterLocalRead: [1]
        - NonTemporalD: [0]
        - NonTemporalB: [4]
        - UnrollLoopSwapGlobalReadOrder: [1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [128,  128, 1, 1024  ]  
          - Exact: [128,  128, 1, 2048  ]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]