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
  DataInitTypeAlpha: 1
  DataInitTypeBeta: 1
  BoundsCheck: 2
  #KeepBuildTmp: True

BenchmarkProblems:
  ########################################
  # HHS NN DTL
  ########################################
  -
    - # 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]
        - GlobalReadVectorWidthB: [2,4]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,32]
        - 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]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - LocalReadVectorWidth: [4]
        - DirectToLds: [1]
        - AssertSummationElementMultiple: [1, 2]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  2,2 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  2,2 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  14]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  34]
          - Exact: [255,   255, 1,  62]
          - Exact: [255,   255, 1,  80]
          - Exact: [255,   255, 1,  94]
          - Exact: [255,   255, 1, 110]
          - Exact: [255,   255, 1, 126]
          - Exact: [255,   255, 1, 190]
          - Exact: [255,   255, 1, 256]

  ########################################
  # HHS NT DTL
  ########################################
  -
    - # 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, 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]
        - GlobalReadVectorWidthB: [2,4]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,32]
        - 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]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToLds: [1]
        - AssertSummationElementMultiple: [1, 2]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  2,2 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  2,2 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  15]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  33]
          - 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]

  ########################################
  # HHS TN DTL
  ########################################
  -
    - # 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, 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]
        - GlobalReadVectorWidthB: [2,4]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16,32]
        - 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]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToLds: [1]
        - AssertSummationElementMultiple: [1, 2]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  2,2 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  2,2 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  14]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  34]
          - Exact: [255,   255, 1,  62]
          - Exact: [255,   255, 1,  80]
          - Exact: [255,   255, 1,  94]
          - Exact: [255,   255, 1, 110]
          - Exact: [255,   255, 1, 126]
          - Exact: [255,   255, 1, 190]
          - Exact: [255,   255, 1, 256]
  ########################################
  # HHS TT DTL
  ########################################
  -
    - # 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,  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]
        - GlobalReadVectorWidthB: [2,4]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1,2]
        - 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: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        #- LocalReadVectorWidth: [4,8]
        - DirectToLds: [1]
        - AssertSummationElementMultiple: [1, 2]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16, 16, 1,  1,   2, 2,  2,2 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  2,2 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  8, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [16, 16, 16, 1,  1,   4, 2,  2,1 ]
              WorkGroup: [16,8,2]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1,  14]
          - Exact: [255,   255, 1,  32]
          - Exact: [255,   255, 1,  34]
          - Exact: [255,   255, 1,  62]
          - Exact: [255,   255, 1,  80]
          - Exact: [255,   255, 1,  94]
          - Exact: [255,   255, 1, 110]
          - Exact: [255,   255, 1, 126]
          - Exact: [255,   255, 1, 190]
          - Exact: [255,   255, 1, 256]

  ########################################
  # SGEMM NT DTL
  ########################################
  -
    - # 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]
        - GlobalReadVectorWidthB: [1]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [8,16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false]
        - LocalWritePerMfma: [-1]
        - StaggerU: [0]
        #- 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]
        - GlobalSplitU: [1,2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToLds: [1]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  1,4 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
      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 DTL
  ########################################
  -
    - # 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 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1]
        - GlobalReadVectorWidthB: [1]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [8,16]
        - 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]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToLds: [1]
        - LocalReadVectorWidth: [1]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
      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 DTL
  ########################################
  -
    - # 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 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [1]
        - GlobalReadVectorWidthB: [1]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [16]
        - VectorWidthA: [-1]
        - VectorWidthB: [-1]
        - MIArchVgpr: [false, true]
        - 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]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToLds: [1]
        - LocalReadVectorWidth: [1]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
      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 TT DTL
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      DestDataType: s
      ComputeDataType: s
      TransposeA: True
      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]
        - GlobalReadVectorWidthB: [1]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - 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: [false, true]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - StorePriorityOpt: [false]
        - VectorStore: [-1]
        - StoreSyncOpt: [0]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        #- SourceSwap: [0, 1]
        - DirectToLds: [1]
        - LocalReadVectorWidth: [1]
        - DirectToVgprA: [false, true]
        - DirectToVgprB: [false, true]
        - Groups:
          -
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   4, 1,  4,1 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   1, 4,  1,4 ]
            - MatrixInstruction: [32, 32,  2, 1,  1,   2, 1,  2,2 ]
            - MatrixInstruction: [16, 16,  4, 1,  1,   2, 2,  2,1 ]
              WorkGroup: [16,8,2]
      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 TN DTL
  ########################################
  -
    - # 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,   2, 2,  2,2 ]
          - [16, 16, 32, 1,  1,   4, 2,  2,2 ]
        - WorkGroup:
          - [16,16,1]
        - GlobalReadVectorWidthA: [4]
        - GlobalReadVectorWidthB: [4]
        - PrefetchGlobalRead: [1]#[1,2] # DTL needs PGR=1
        - PrefetchLocalRead: [1,2]
        - ClusterLocalRead: [1]
        - NumElementsPerBatchStore: [0]
        - DepthU: [64,128]
        - 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: [0]
        #- GlobalSplitU: [1,2]
        #- GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"]
        - SourceSwap: [true]#[0, 1]
        - LocalReadVectorWidth: [8]
        - DirectToLds: [1]
        - AssertSummationElementMultiple: [1, 4]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [255,   255, 1, 193]
          - Exact: [255,   255, 1, 256]