TestParameters:
  marks: [skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx940, skip-gfx941, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201, skip-gfx1250] # not supported by arch

# benchmark assembly and source kernels
GlobalParameters:
#  MinimumRequiredVersion: 4.2.0
  CMakeBuildType: Release
  PrintLevel: 1
  ForceRedoBenchmarkProblems: True
  ForceRedoLibraryLogic: True
  ForceRedoLibraryClient: True
  EnqueuesPerSync: 1
  SyncsPerBenchmark: 0
  NumElementsToValidate: -1
  ValidationMaxToPrint: 4
  ValidationPrintValids: False
  # PrintTensorD: 0x3
  # PrintTensorC: 0x3
  # PrintTensorB: 0x3
  # PrintTensorA: 0x3 #0x1
  # PrintTensorRef: 0x3
  # ShortNames: False
  DataInitTypeA: 3 #12 #3
  DataInitTypeB: 3 #13 #3
  DataInitTypeC: 3 #3
  DataInitTypeAlpha: 1
  DataInitTypeBeta: 0
  KernelTime: True
  # PrintSolutionRejectionReason: True

BenchmarkProblems:

  - # zgemm NN
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
      ComplexConjugateA: False
      ComplexConjugateB: True

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        # - LoopTail: [True]
        # - EdgeType: ["ShiftPtr"]
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
        # - 1LDSBuffer: [-1,0,1]
        # - ThreadTile:
          ##- [ 3, 5 ]
          # - [ 2, 2 ]
          # - [ 8, 8 ]
        # - VectorWidth: [-1]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          # - Exact: [32, 32, 1, 8]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [64, 64, 1, 16]
          # - Exact: [64, 64, 1, 128]
          # - Exact: [4096, 4096, 1, 8192]

          # - Range: [[16,37, 512],[16,57,512],[1],[1,1,64]]
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm NT
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: False
      TransposeB: True
      ComplexConjugateA: False
      ComplexConjugateB: False

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm TN
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: False
      ComplexConjugateA: False
      ComplexConjugateB: False

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm TT
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: True
      ComplexConjugateA: False
      ComplexConjugateB: False

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm NC
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: False
      TransposeB: True
      ComplexConjugateA: False
      ComplexConjugateB: True

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm CN
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: False
      ComplexConjugateA: True
      ComplexConjugateB: False

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm TC
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: True
      ComplexConjugateA: False
      ComplexConjugateB: True

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm CT
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: True
      ComplexConjugateA: True
      ComplexConjugateB: False

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]

  - # zgemm CC
    - # ProblemType
      OperationType: GEMM
      DataType: Z
      DestDataType: Z
      ComputeDataType: Z
      UseBeta: True
      Batched: True      
      TransposeA: True
      TransposeB: True
      ComplexConjugateA: True
      ComplexConjugateB: True

    - # BenchmarkProblemSizeGroup - Assembly
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16, 16, 4, 1, 1, 1, 1, 1, 1] #16x16
          - [16, 16, 4, 1, 1, 4, 4, 2, 2] #128x128
          - [16, 16, 4, 1, 1, 2, 1, 2, 2] #64x32
          - [16, 16, 4, 1, 1, 1, 2, 2, 2] #32x64
          - [16, 16, 4, 1, 1, 1, 1, 2, 2] #32x32
          - [16, 16, 4, 1, 1, 2, 2, 2, 1]
          - [16, 16, 4, 1, 1, 2, 2, 1, 2]
        - ScheduleIterAlg: [1, 3]
        - MIArchVgpr: [false, true]
        - 1LDSBuffer: [0, 1]
        - GlobalSplitU: [1, 2, 4, 8]
        - GlobalSplitUAlgorithm: [MultipleBuffer]
        - PrefetchLocalRead: [1]
        - PrefetchGlobalRead: [0, 2]
        - 1LDSBuffer: [0, 1]
        - DepthU: [4, 8, 16, 32]
        - WorkGroup:
          - [16, 4, 1]
          - [16, 4, 2]
          - [16, 4, 4]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          # - Exact: [2, 2, 1, 2]
          # - Exact: [16, 16, 1, 2]
          # - Exact: [2048, 2048, 1, 512] #128x128(times16)
          # - Exact: [1024, 1024, 1, 64] #64x64(times16)
          # - Exact: [512, 512, 1, 32] #32x32
          - Range: [ [127,1,129], 0, [2], [63,1,64] ]