TestParameters:
  marks: [skip-gfx900, skip-gfx906, skip-gfx908, skip-gfx90a, skip-gfx1010, skip-gfx1011, skip-gfx1012, skip-gfx1030, skip-gfx1031, skip-gfx1032, skip-gfx1033, skip-gfx1034, skip-gfx1035, skip-gfx1036, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1103, skip-gfx1150, skip-gfx1151, skip-gfx1152, skip-gfx1153] # not supported by arch

GlobalParameters:
  MinimumRequiredVersion: 4.14.0
  KernelTime: True
  NumElementsToValidate: -1 # check this out.
  # DataInitTypeA: 0
  # DataInitTypeB: 0
  # DataInitTypeC: 0
  # DataInitTypeD: 3
  # DataInitTypeBeta: 0 # to be consistent with the rocblas yaml creator
  # DataInitTypeAlpha: 1 # to be consistent with the rocblas yaml creator
  NewClient: 2
  BoundsCheck: True
  #MaxFileName: 256
  #Device: 0

BenchmarkProblems:
  ########################################
  # F8B8HS NN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8B8
      DestDataType: H
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - MatrixInstruction:
          - [32, 32, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - TransposeLDS: [1]
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

  ########################################
  # B8F8HS NN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: B8F8
      DestDataType: H
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - MatrixInstruction:
          - [32, 32, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - TransposeLDS: [1]
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

  ########################################
  # F8B8HS NT - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8B8
      DestDataType: H
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - MatrixInstruction:
          - [32, 32, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

  ########################################
  # B8F8HS NT - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: B8F8
      DestDataType: H
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: True
      UseBeta: True
      Batched: True
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
        - EdgeType: ["ShiftPtr"]
      ForkParameters:
        - MatrixInstruction:
          - [32, 32, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

  ########################################
  # F8B8HS TN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8B8
      DestDataType: 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, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - TransposeLDS: [1]
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

  ########################################
  # B8F8HS TN - standard
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: B8F8
      DestDataType: 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, 16, 1, 1, 1, 1, 2, 2] # 64x64
          - [16, 16, 32, 1, 1, 2, 2, 2, 2] # 64x64

        - 1LDSBuffer: [0]
        - DepthU: [64,128]
        - GlobalReadVectorWidth: [8,16] #
        - PrefetchGlobalRead: [1,2]
        - PrefetchLocalRead: [1,3] #[3,5,9] #
        - SourceSwap: [True] #[True, False]  # default False
        - StoreRemapVectorWidth: [0] #[0,8]  # default 0
        - StoreVectorWidth: [-1] #[-1, 2, 4,8 ]  # default -1
        - TransposeLDS: [1]
        - VectorWidth: [1,4] #[1, 2,4] #[-1,2] #[1,2,4] #

      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [1216,768,1,1024]
          - Exact: [777,555,1,111]