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: 21
  DataInitTypeB: 21
  DataInitTypeC: 21
  DataInitTypeAlpha: 1
  DataInitTypeBeta: 1
  DataInitTypeBias: 21
  DataInitTypeScaleAlphaVec: 21
  BoundsCheck: 2

BenchmarkProblems:
  ########################################
  # FP8HS with ScaleABCD
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8N
      DestDataType: h
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
      Activation:    True
      ActivationType: hipblaslt_all
      UseScaleAB: "Scalar"
      UseScaleCD: True
      UseScaleAlphaVec: 1
      UseBias: 1
      BiasDataTypeList: [s]
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16,16,32, 1, 1, 1,1, 1,1]  # 16x16
        - DepthU: [ 64 ]
        - AssertFree0ElementMultiple: [1]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - VectorWidthA: [1]
        - VectorWidthB: [1]
        - GlobalReadVectorWidthA: [1, 2, 4, 8, 16]
        - GlobalReadVectorWidthB: [1, 2, 4, 8, 16]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1]
        - ExpandPointerSwap: [true]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - TransposeLDS: [0, 1]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - WaveSeparateGlobalReadB: [1]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1, 2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0]
        - StorePriorityOpt: [false]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [1,   128, 1, 640]
          - Exact: [2,   128, 1, 640]
          - Exact: [3,   128, 1, 640]
          - Exact: [1,   1, 1, 640]
          - Exact: [127,   127, 1, 640]
          - Exact: [129,   129, 1, 640]
          - Exact: [128,   128, 1, 127]
          - Exact: [128,   128, 1, 129]
          - Exact: [127,   128, 1, 640]
          - Exact: [128,   128, 1, 640]
          - Exact: [129,   128, 1, 640]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

  ########################################
  # FP8SS with ScaleABCD
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8N
      DestDataType: s
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
      Activation:    True
      UseScaleAB: "Scalar"
      UseScaleCD: True
      UseScaleAlphaVec: 1
      UseBias: 1
      BiasDataTypeList: [s]
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16,16,32, 1, 1, 1,1, 1,1]  # 16x16
        - DepthU: [ 64 ]
        - AssertFree0ElementMultiple: [1]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - VectorWidthA: [1]
        - VectorWidthB: [1]
        - GlobalReadVectorWidthA: [4]
        - GlobalReadVectorWidthB: [4]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1]
        - ExpandPointerSwap: [true]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - WaveSeparateGlobalReadB: [1]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1, 2]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0]
        - StorePriorityOpt: [false]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [127,   128, 1, 640]
          - Exact: [128,   128, 1, 640]
          - Exact: [129,   128, 1, 640]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

  ########################################
  # FP8FP8S
  ########################################
  -
    - # ProblemType
      OperationType: GEMM
      DataType: F8N
      ComputeDataType: S
      HighPrecisionAccumulate: True
      TransposeA: False
      TransposeB: False
      UseBeta: True
      Batched: True
      Activation:    True
      UseScaleAB: "Scalar"
      UseScaleCD: True
      UseScaleAlphaVec: 1
      UseBias: 1
      BiasDataTypeList: [s]
    - # BenchmarkProblemSizeGroup - Standard
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - KernelLanguage: ["Assembly"]
      ForkParameters:
        - MatrixInstruction:
          - [16,16,32, 1, 1, 1,1, 1,1]  # 16x16
          - [16,16,32, 1, 1, 2,2, 1,1]  # 16x16
        - DepthU: [ 64 ]
        - AssertFree0ElementMultiple: [1]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - ClusterLocalRead: [1]
        - VectorWidthA: [1,2]
        - VectorWidthB: [1,2]
        - GlobalReadVectorWidthA: [4]
        - GlobalReadVectorWidthB: [4]
        - ScheduleIterAlg: [3]
        - InnerUnroll: [1]
        - ExpandPointerSwap: [true]
        - LdsBlockSizePerPadA: [0]
        - LdsBlockSizePerPadB: [0]
        - LdsPadA: [0]
        - LdsPadB: [0]
        - WaveSeparateGlobalReadB: [1]
        - 1LDSBuffer: [-1]
        - GlobalSplitU: [1]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - GlobalReadPerMfma: [1.0]
        - LocalWritePerMfma: [-1]
        - StoreVectorWidth: [-1]
        - SourceSwap: [true]
        - NumElementsPerBatchStore: [0]
        - StorePriorityOpt: [false]
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
          - Exact: [127,   128, 1, 640]
          - Exact: [128,   128, 1, 640]
          - Exact: [129,   128, 1, 640]
        - BiasTypeArgs: ['s']
        - ActivationArgs:
          - [Enum: none]

#####################################################
# FP8HS + Bias S + ScaleAlphaVec + SVW=16 for bpl 64
#####################################################

  - 
    - # ProblemType
      OperationType: GEMM
      DataType: F8N
      DestDataType: H
      ComputeDataType: S
      HighPrecisionAccumulate: true
      TransposeA: True
      TransposeB: False
      UseBeta: True
      BiasDataTypeList: [s]
      UseBias: 1
      UseScaleAlphaVec: 1
      Batched: True
      #Activation: True
      #ActivationHPA: True
    - InitialSolutionParameters: 
      BenchmarkCommonParameters:
        - KernelLanguage: [Assembly]
      ForkParameters:
        - 1LDSBuffer: [1]
        - DepthU: [64, 128]
        - GlobalReadVectorWidthA: [4, 16]
        - GlobalReadVectorWidthB: [4, 16]
        - WaveSeparateGlobalReadA: [0]
        - WaveSeparateGlobalReadB: [0]
        - NumElementsPerBatchStore: [16]
        - NonTemporalA: [0]
        - PrefetchGlobalRead: [2]
        - PrefetchLocalRead: [1]
        - SourceSwap: [true]
        - StaggerU: [8]
        - StorePriorityOpt: [true]
        - StoreSyncOpt: [4]
        - WorkGroupMapping: [-48]
        - WorkGroupMappingXCC: [1]
        - MIArchVgpr: [true] #[0,1]
        - GlobalSplitUAlgorithm: ["MultipleBuffer"]
        - Groups:
          - - NonTemporalC: 4
              NonTemporalD: 4
          - 
            
            - MatrixInstruction: [16, 16, 32,  1,  1, 16,  3,  1,  4] # MT 256x192 - TT 16x48  - WG 16x16  - MIBlockM:  1 -- sizes [[(57344, 8192, 1, 8192)]]
              MIArchVgpr: false
       
      BenchmarkJoinParameters: 
      BenchmarkFinalParameters:
      - ProblemSizes:
        - Exact: [ 1024, 1024, 1, 1024 ]
      - BiasTypeArgs: ['s']