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
  NumElementsToValidate: -1
  MinimumRequiredVersion: 5.0.0
  PrintLevel: 1
  #PrintSolutionRejectionReason: True
  Device: 0
  CMakeBuildType: Release
  KernelTime: True
  MaxWorkspaceSize: 13421772800
  DataInitTypeA: 13
  DataInitTypeB: 12
  DataInitTypeAlpha: 1
  DataInitTypeBeta: 1
  DataInitTypeBias: 13
  DataInitTypeScaleAlphaVec: 12

BenchmarkProblems:
  ########################################
  # HHS NN DTVA pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 2,  4,1 ]
        #  - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   8, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  4,1 ]
        #  - [32, 32,  8, 1,  1,   2, 2,  4,1 ]
        #  - [32, 32,  8, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16,  4, 4,  4,   1, 4,  4,1 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  4,1 ]
        #  - [32, 32,  4, 2,  2,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [2,4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - StreamK: [0,3]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   1, 4,  4,1 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [32, 32,  4, 2,  2,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS NT DTVA pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   8, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   2, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  4,1 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [2,4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,32,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - StreamK: [0,3]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 8,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 16,  4,1 ]

      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS NT DTVB pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 8,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 2,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  1,4 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [2,4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [2,4,8]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- SourceSwap: [0, 1]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - StreamK: [0]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 8,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  1,4 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS TN DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   8, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   2, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  4,1 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - StreamK: [0,3]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS TN DTVA + BIAS + Activation
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      UseScaleAlphaVec: 1
      UseScaleAB: "Scalar"
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 16, 1,  1,   1, 1,  4,1 ] # MT = 64x16. Case to check kernel writer works
          - [16, 16, 16, 1,  1,   2, 1,  4,1 ] # MT = 128x16. Check kernel works without vgprValuA
          - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
          - [16, 16, 16, 1,  1,   8, 1,  4,1 ] # MT = 512x16. Case to check LDS usage of Epilogues
          - [16, 16, 16, 1,  1,   5, 1,  2,1 ] # MT = 160x16. Case to check kernel writer works
          - [32, 32,  8, 1,  1,   2, 1,  4,1 ]
          - [32, 32,  8, 1,  1,   4, 1,  4,1 ]

          - [16, 16, 16, 1,  1,  8,1,   1,4 ] # 128x64
          - [16, 16, 16, 1,  1,  8,1,   2,2 ] # 256x32
          - [16, 16, 16, 1,  1,  16,1,  2,2 ] # 512x32
          - [16, 16, 16, 1,  1,  10,1,  2,2 ] # 160x32
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [2,4,8]
        - GlobalReadVectorWidthB: [2,4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [32,64,128]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false, true]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [2,4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - ActivationFuncCall: [false, true]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [127,   128, 1, 640]
          - Exact: [129,   128, 1, 640]
          - Exact: [256,   255, 1, 1101]
          - Exact: [256,   257, 1, 1101]
          - Exact: [512,   128, 1, 256]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: relu]

  ########################################
  # HHS TN DTVB
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 8,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 2,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16,  4, 4,  4,   4, 1,  1,4 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [4,8]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 8,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  1,4 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS TN DTVB + BIAS + Activation
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
      UseBias: 1
      Activation: True
      UseScaleAlphaVec: 1
      UseScaleAB: "Scalar"
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 16, 1,  1,   1, 1,  1,4 ] # MT = 16x64. Case to check kernel writer works
          - [16, 16, 16, 1,  1,   1, 2,  1,4 ] # MT = 16x128. Check kernel works without vgprValuA
          - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
          - [16, 16, 16, 1,  1,   1, 8,  1,4 ] # MT = 16x512. Case to check LDS usage of Epilogues
          - [16, 16, 16, 1,  1,   1, 5,  1,2 ] # MT = 16x160. Case to check kernel writer works
          - [32, 32,  8, 1,  1,   1, 2,  1,4 ]
          - [32, 32,  8, 1,  1,   1, 4,  1,4 ]

          - [16, 16, 16, 1,  1,  1,8,   4,1 ] # 64x128
          - [16, 16, 16, 1,  1,  1,8,   2,2 ] # 32x256
          - [16, 16, 16, 1,  1,  1,16,  2,2 ] # 32x512
          - [16, 16, 16, 1,  1,  1,10,  2,2 ] # 32x160
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [2,4,8]
        - GlobalReadVectorWidthB: [2,4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [32,64,128]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false, true]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false]
        - TransposeLDS: [1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - TransposeLDS: [0,1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [2,4,8]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - ActivationFuncCall: [false, true]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [127,   128, 1, 640]
          - Exact: [129,   128, 1, 640]
          - Exact: [256,   255, 1, 1101]
          - Exact: [256,   257, 1, 1101]
          - Exact: [128,   512, 1, 256]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: relu]

  ########################################
  # HHS TT DTVB pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 2,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 8,  1,4 ]
        #  - [32, 32,  8, 1,  1,   2, 2,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  1,4 ]
        #  - [ 4,  4,  4,16,  1,   2, 2,  1,4 ]
        #  - [16, 16,  4, 4,  1,   2, 2,  1,4 ]
        #  - [32, 32,  4, 2,  1,   2, 2,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [2,4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [2,4,8]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- SourceSwap: [0, 1]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - StreamK: [0]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 8,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 4,  4,   2, 2,  1,4 ]
            - MatrixInstruction: [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,1 ]
              WorkGroup: [16,4,4]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 256]

  ########################################
  # SGEMM NT DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      DestDataType: s
      ComputeDataType: s
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   4, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   2, 1,  4,1 ]
        #  - [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2,4]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - StreamK: [0]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # SGEMM NT DTVB
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      DestDataType: s
      ComputeDataType: s
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  1,4 ]
        #  - [16, 16,  4, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16,  4, 1,  1,   1, 4,  1,4 ]
        #  - [32, 32,  2, 1,  1,   1, 1,  1,4 ]
        #  - [32, 32,  2, 1,  1,   1, 2,  1,4 ]
        #  - [ 4,  4,  1,16,  1,   8, 1,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2]
        - GlobalReadVectorWidthB: [1,2,4]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        #- StaggerUStride: [256]
        #- StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - TransposeLDS: [0]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - StreamK: [0,3]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [ 4,  4,  1,16,  1,   8, 1,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # SGEMM NN DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      DestDataType: s
      ComputeDataType: s
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   4, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   2, 1,  4,1 ]
        #  - [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
        #  - [32, 32,  1, 2,  2,   1, 1,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2,4]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1,2]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [true]#[0, 1]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - StreamK: [0]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
            - MatrixInstruction: [32, 32,  1, 2,  2,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # SGEMM TN DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      DestDataType: s
      ComputeDataType: s
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   4, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  2, 1,  1,   2, 1,  4,1 ]
        #  - [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
        #  - [32, 32,  1, 2,  2,   1, 1,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2,4]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [ 4,  4,  1,16, 16,   1, 8,  4,1 ]
            - MatrixInstruction: [32, 32,  1, 2,  2,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # DGEMM NT DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: d
      DestDataType: d
      ComputeDataType: d
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   4, 1,  4,1 ]
        #  - [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # DGEMM NT DTVB
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: d
      DestDataType: d
      ComputeDataType: d
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  1,4 ]
        #  - [16, 16,  4, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16,  4, 1,  1,   1, 4,  1,4 ]
        #  - [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        #- StaggerUStride: [256]
        #- StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - TransposeLDS: [0]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 2,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # DGEMM NN DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: d
      DestDataType: d
      ComputeDataType: d
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16,  4, 1,  1,   1, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16,  4, 1,  1,   4, 1,  4,1 ]
        #  - [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1,2]
        - GlobalReadVectorWidthB: [1,2]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1,2]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [true]#[0, 1]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [ 4,  4,  4, 4,  4,   2, 2,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  17]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 128]

  ########################################
  # F8F8S NT DTVA pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: f8n
      DestDataType: f8n
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 32, 1,  1,   1, 1,  4,1 ]
          - [16, 16, 32, 1,  1,   2, 1,  4,1 ]
          - [16, 16, 32, 1,  1,   4, 1,  4,1 ]
          - [16, 16, 32, 1,  1,   8, 1,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [true]#[0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 32, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 32, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 32, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 32, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 32, 1,  1,   2, 1,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [16, 16, 32, 1,  1,   2, 1,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1, 193]
          - Exact: [255,   255, 1, 256]

  ########################################
  # F8F8S NT DTVB pack
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: f8n
      DestDataType: f8n
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 32, 1,  1,   1, 1,  1,4 ]
          - [16, 16, 32, 1,  1,   1, 2,  1,4 ]
          - [16, 16, 32, 1,  1,   1, 4,  1,4 ]
          - [16, 16, 32, 1,  1,   1, 8,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [2]#[1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [32,64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        - SourceSwap: [false, true]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1, 193]
          - Exact: [255,   255, 1, 256]

  ########################################
  # F8F8S TN DTVA
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: f8n
      DestDataType: f8n
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 32, 1,  1,   4, 1,  4,1 ]
          - [16, 16, 32, 1,  1,   8, 1,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,2] #[1,2]
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 256]

  ########################################
  # F8F8S TN DTVB
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: f8n
      DestDataType: f8n
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 32, 1,  1,   1, 4,  1,4 ]
          - [16, 16, 32, 1,  1,   1, 8,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [true]#[0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1, 193]
          - Exact: [255,   255, 1, 256]

  ########################################
  # F8H_HHS NT DTVA pack+input conversion
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DataTypeA: f8n
      DataTypeB: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   8, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   2, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  4,1 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [2,4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        #- LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - ConvertAfterDS: [true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HF8_HHS NT DTVB pack+input conversion
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DataTypeA: h
      DataTypeB: f8n
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 8,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 2,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  1,4 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [2,4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [2,4,8]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        - ExpandPointerSwap: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- SourceSwap: [0, 1]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - ConvertAfterDS: [true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 8,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # F8H_HHS TN DTVA+input conversion
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DataTypeA: f8n
      DataTypeB: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   2, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16, 16, 1,  1,   8, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   2, 1,  4,1 ]
        #  - [32, 32,  8, 1,  1,   4, 1,  4,1 ]
        #  - [16, 16,  4, 4,  4,   2, 2,  4,1 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  4,1 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [0,1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [4,8]
        - DirectToVgprA: [true]
        - UseSgprForGRO: [0,1]
        - TransposeLDS: [0,1]
        - ConvertAfterDS: [true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   8, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HF8_HHS TN DTVB
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DataTypeA: h
      DataTypeB: f8n
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard - All problem
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        #- MatrixInstruction:
        #  - [16, 16, 16, 1,  1,   1, 2,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16, 16, 1,  1,   1, 8,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 1,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 2,  1,4 ]
        #  - [32, 32,  8, 1,  1,   1, 4,  1,4 ]
        #  - [16, 16,  4, 4,  4,   4, 1,  1,4 ]
        #  - [ 4,  4,  4,16, 16,   2, 2,  1,4 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4,8]
        - GlobalReadVectorWidthB: [4,8]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [4]
        - StaggerUStride: [256]
        - StaggerUMapping: [2]
        - WorkGroupMapping: [1]
        - ScheduleIterAlg: [3]
        #- ExpandPointerSwap: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [false, true]
        - LocalReadVectorWidth: [4,8]
        - DirectToVgprB: [true]
        - UseSgprForGRO: [0,1]
        - ConvertAfterDS: [true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 8,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 2,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 4,  1,2 ]
              WorkGroup: [16,8,2]
            - MatrixInstruction: [32, 32,  8, 1,  1,   2, 2,  1,1 ]
              WorkGroup: [16,4,4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - Exact: [255,   255, 1,  47]
          - Exact: [255,   255, 1,  63]
          - Exact: [255,   255, 1,  79]
          - Exact: [255,   255, 1,  95]
          - Exact: [255,   255, 1, 111]
          - Exact: [255,   255, 1, 127]
          - Exact: [255,   255, 1, 191]
          - Exact: [255,   255, 1, 255]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS TT DTVA with/wo tail loop
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: True
      TransposeB: True
      UseBeta: True
      Batched: True
      # UseBias: 1
      # BiasDataTypeList: [s,h]
      # Activation: True
      # ActivationType: hipblaslt_all
      # UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 16, 1,  1,  1,8,   4,1 ] # 64x128
          - [16, 16, 16, 1,  1,  2,8,  4,1 ] # 128x128
          - [16, 16, 16, 1,  1,  4,16,  4,1 ] # 256x256

          - [16, 16, 16, 1,  1,  4,2,   1,4 ] # 64x128
          - [16, 16, 16, 1,  1,  8,2,   1,4 ] # 128x128
          - [16, 16, 16, 1,  1,  16,4,  1,4 ] # 256x256
        - AssertFree0ElementMultiple: [16]
        - AssertFree1ElementMultiple: [16]
        - AssertSummationElementMultiple: [32]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,2,4]
        - DepthU: [128] # Has tail
        - GlobalReadVectorWidthA: [-1,8] # force testing 8 (prepared for swizzleA)
        - GlobalReadVectorWidthB: [-1]
        - VectorWidthA: [-1,1] # force testing 1 (prepared for swizzleA)
        - VectorWidthB: [-1]
        - LocalReadVectorWidth: [-1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        # - StaggerU: [0] #
        # - StaggerUStride: [-1] #
        # - WorkGroupMapping: [1] #
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [0,1]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - NumElementsPerBatchStore: [0]
        - ClusterLocalRead: [1]
        - WorkGroupMappingXCC: [8]
        - UseSgprForGRO: [0]
        - DirectToVgprA: [true]

      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [960, 1280, 1, 192]
          - Exact: [960, 1280, 1, 256] # no tail
          - Exact: [1280, 1280, 1, 192]
          - Exact: [1600, 1280, 1, 192]

  ########################################
  # HHS NN DTVB with/wo tail loop
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: h
      DestDataType: h
      ComputeDataType: s
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
      # UseBias: 1
      # BiasDataTypeList: [s,h]
      # Activation: True
      # ActivationType: hipblaslt_all
      # UseScaleAlphaVec: 1
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 16, 1,  1,  4,2,   1,4 ] # 64x128
          - [16, 16, 16, 1,  1,  8,2,  1,4 ] # 128x128
          - [16, 16, 16, 1,  1,  16,4,  1,4 ] # 256x256

          - [16, 16, 16, 1,  1,  2,4,   4,1 ] # 128x64
          - [16, 16, 16, 1,  1,  2,8,  4,1 ] # 128x128
          - [16, 16, 16, 1,  1,  4,16,  4,1 ] # 256x256
        - AssertFree0ElementMultiple: [16]
        - AssertFree1ElementMultiple: [16]
        - AssertSummationElementMultiple: [32]
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,2,4]
        - DepthU: [128] # Has tail
        - GlobalReadVectorWidthA: [-1]
        - GlobalReadVectorWidthB: [-1,8] # force testing 8 (prepared for swizzleB)
        - VectorWidthA: [-1]
        - VectorWidthB: [-1,1] # force testing 1 (prepared for swizzleB)
        - LocalReadVectorWidth: [-1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        # - StaggerU: [0] #
        # - StaggerUStride: [-1] #
        # - WorkGroupMapping: [1] #
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [0,1]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreRemapVectorWidth: [0]
        - StoreVectorWidth: [-1]
        - NumElementsPerBatchStore: [0]
        - ClusterLocalRead: [1]
        - WorkGroupMappingXCC: [8]
        - UseSgprForGRO: [0]
        - DirectToVgprB: [true]

      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [960, 1280, 1, 192]
          - Exact: [960, 1280, 1, 256]
          - Exact: [1280, 1280, 1, 192]
          - Exact: [1600, 1280, 1, 192]