GlobalParameters:
  MinimumRequiredVersion: 4.0.0
  PrintLevel: 1
  ForceRedoBenchmarkProblems: True
  ForceRedoLibraryLogic: True
  ForceRedoLibraryClient: True
  CMakeBuildType: Release
  EnqueuesPerSync: 1
  SyncsPerBenchmark: 1
  LibraryPrintDebug: False
  NumElementsToValidate: 1000
  ValidationMaxToPrint: 10
  ValidationPrintValids: False
  ShortNames: False
  MergeFiles: True
  Platform: 0
  Device: 0
  KernelTime: True
  PinClocks: True
  SleepPercent: 200
  DataInitTypeAlpha : 1
  DataInitTypeA : 3
  DataInitTypeB : 3
  PrintSolutionRejectionReason : 0

BenchmarkProblems:
  -
    - # ProblemType
      OperationType: GEMM
      DataType: s
      TransposeA: True
      TransposeB: True
      UseBeta: True
      Batched: True

  ########################################
  # NN - Batch
  ########################################
    - # Benchmark Group - ResNet 1x1:
      InitialSolutionParameters:
      BenchmarkCommonParameters:
        - EdgeType: ["ShiftPtr"]
        - KernelLanguage: ["Assembly"]

      ForkParameters:
        - FractionalLoad: [1]
        - PrefetchGlobalRead: [0]
        - PrefetchLocalRead: [1]
        - ThreadTile:
          - [ 8, 6 ]
        - WorkGroup:
          - [ 32, 8, 1 ]
            #- [ 16, 16, 1 ]
        - WorkGroupMapping: [64]
          #  - DepthU: [3,5,7,9,16] # some bugs with odd unroll dims
        - DepthU: [16]
        - VectorWidth: [2]
        - GlobalReadVectorWidth: [2]
      BenchmarkForkParameters:
      JoinParameters:
      BenchmarkJoinParameters:
      BenchmarkFinalParameters:
        - ProblemSizes:
            #- Range: [ [127,1,127], 0, [1], [64,1,64] ]

# to repro, disable this in SolutionStructs.py
# 1506         reject(state, "Fractional currently requires PreciseBoundsCheck") # debug this later
# Symptom is GPUVM fault,