TheRock/rocm-libraries/shared/tensile/Tensile/Tests/emulation/mfma/1LDSB.yaml

TestParameters:
  marks: [skip-gfx900, skip-gfx906, skip-gfx1010, skip-gfx1011, emulation1] # not supported by arch

GlobalParameters:
  NumElementsToValidate: -1
  BoundsCheck: True
  KernelTime: True

BenchmarkProblems:
  ########################################
  # TN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      TransposeA: True
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - MatrixInstruction:
          - [32, 32, 1, 2]
          # - [32, 32, 2, 1]
          - [16, 16, 1, 4]
          # - [16, 16, 4, 1]
        - ThreadTile:
          - [ 1, 32 ]
          - [ 2, 64 ]
        - WorkGroup:
          - [ 64, 4, 1 ]
        - AssertSummationElementMultiple: [1]
        - AssertFree0ElementMultiple: [1]
        - AssertFree1ElementMultiple: [1]
        - VectorWidth: [1]
        - PrefetchGlobalRead: [1]
        - PrefetchLocalRead: [1,5,9,17]
        - DepthU: [16,32]
        - GlobalReadVectorWidth: [1]
        - LocalReadVectorWidth: [-1,1,2,4]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1,2,4]
        - TransposeLDS: [0,1]
        - LdsBlockSizePerPadA: [-1]
        - LdsBlockSizePerPadB: [-1]
        - LdsPadA: [-1]
        - LdsPadB: [-1]
        - 1LDSBuffer: [1]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Range: [ [128], [256], [1], [256,1,257] ]

  # -
  #   - # ProblemType
  #     OperationType: GEMM
  #     DataType: b
  #     DestDataType: B
  #     ComputeDataType: s
  #     HighPrecisionAccumulate: True
  #     TransposeA: True
  #     TransposeB: False
  #     UseBeta: True
  #     Batched: True
  #   - # BenchmarkProblemSizeGroup - Standard
  #     InitialSolutionParameters:
  #     BenchmarkCommonParameters:
  #       - KernelLanguage: ["Assembly"]
  #       - EdgeType: ["ShiftPtr"]
  #     ForkParameters:
  #       - MatrixInstruction:
  #         - [32, 32, 2, 2]
  #         - [32, 32, 4, 1]
  #         - [16, 16, 2, 4]
  #         - [16, 16, 8, 1]
  #       - ThreadTile:
  #         - [ 1, 32 ]
  #         - [ 2, 64 ]
  #       - WorkGroup:
  #         - [ 64, 4, 1 ]
  #       - AssertSummationElementMultiple: [1]
  #       - AssertFree0ElementMultiple: [1]
  #       - AssertFree1ElementMultiple: [1]
  #       - VectorWidth: [2]
  #       - PrefetchGlobalRead: [1]
  #       - PrefetchLocalRead: [1,5,9,17]
  #       - DepthU: [32,64]
  #       - GlobalReadVectorWidth: [2]
  #       - LocalReadVectorWidth: [-1,2,4,8]
  #       - ScheduleIterAlg: [3]
  #       - InnerUnroll: [1,2,4]
  #       - TransposeLDS: [0,1]
  #       - LdsBlockSizePerPadA: [-1]
  #       - LdsBlockSizePerPadB: [-1]
  #       - LdsPadA: [-1]
  #       - LdsPadB: [-1]
  #       - 1LDSBuffer: [1]
  #     BenchmarkJoinParameters:
  #     BenchmarkFinalParameters:
  #       - ProblemSizes:
  #         - Range: [ [128], [256], [1], [255,1,257] ]

  # -
  #   - # ProblemType
  #     OperationType: GEMM
  #     DataType: h
  #     ComputeDataType: s
  #     HighPrecisionAccumulate: True
  #     TransposeA: True
  #     TransposeB: False
  #     UseBeta: True
  #     Batched: True
  #   - # BenchmarkProblemSizeGroup - Standard
  #     InitialSolutionParameters:
  #     BenchmarkCommonParameters:
  #       - KernelLanguage: ["Assembly"]
  #       - EdgeType: ["ShiftPtr"]
  #     ForkParameters:
  #       - MatrixInstruction:
  #         - [32, 32, 4, 2]
  #         - [32, 32, 8, 1]
  #         - [16, 16, 4, 4]
  #         - [16, 16, 16, 1]
  #       - ThreadTile:
  #         - [ 1, 32 ]
  #         - [ 2, 64 ]
  #       - WorkGroup:
  #         - [ 64, 4, 1 ]
  #       - AssertSummationElementMultiple: [1]
  #       - AssertFree0ElementMultiple: [1]
  #       - AssertFree1ElementMultiple: [1]
  #       - VectorWidth: [2]
  #       - PrefetchGlobalRead: [1]
  #       - PrefetchLocalRead: [1,3,5,9]
  #       - DepthU: [32,64]
  #       - GlobalReadVectorWidth: [2,4]
  #       - LocalReadVectorWidth: [-1,2,4,8]
  #       - ScheduleIterAlg: [3]
  #       - InnerUnroll: [1,2]
  #       - TransposeLDS: [0,1]
  #       - LdsBlockSizePerPadA: [-1]
  #       - LdsBlockSizePerPadB: [-1]
  #       - LdsPadA: [-1]
  #       - LdsPadB: [-1]
  #       - 1LDSBuffer: [1]
  #     BenchmarkJoinParameters:
  #     BenchmarkFinalParameters:
  #       - ProblemSizes:
  #         - Range: [ [128], [256], [1], [255,1,257] ]

LibraryLogic:
    ScheduleName: "arcturus"
    DeviceNames: ["Device 7380", "Device 7388", "Device 738c", "Device 7390"]
    ArchitectureName: "gfx908"