BenchmarkProblems:
- - DataType: s
    IndexAssignmentsA: [3, 1, 2]
    IndexAssignmentsB: [3, 0, 2]
    NumIndicesC: 3
    OperationType: TensorContraction
    UseBeta: false
    UseInitialStridesAB: false
  - BenchmarkCommonParameters:
    - EdgeType: [ShiftPtr]
    BenchmarkFinalParameters:
    - ProblemSizes:
      - Exact: [16, 32, 1, 8] # fails - multiple tiles.
    BenchmarkForkParameters: null
    BenchmarkJoinParameters: null
    ForkParameters:
    - PrefetchGlobalRead: [0]
    - KernelLanguage: [Source,Assembly]
    - ThreadTile:
      - [2, 2]
    - WorkGroup:
      - [8, 8, 1]
    - DepthU: [8]
    - GlobalReadVectorWidth: [1]
    - VectorWidth: [1]
    - FractionalLoad: [0]
    InitialSolutionParameters: null
    JoinParameters: null
GlobalParameters: {CMakeBuildType: Debug, EnqueuesPerSync: 1, ForceRedoBenchmarkProblems: true,
  ForceRedoLibraryClient: true, ForceRedoLibraryLogic: true, KernelTime: true, LibraryPrintDebug: false,
  MergeFiles: true, MinimumRequiredVersion: 4.2.0, NumElementsToValidate: 1000,
  ShortNames: false, SolutionSelectionAlg: 1, SyncsPerBenchmark: 1,
  ValidationMaxToPrint: 4, ValidationPrintValids: false,
  DataInitTypeAB: 1,
  LibraryPrintDebug: 1
      }

# Case has free index for A is 1 and for B is 0.
# So does interesting transpose in the output space.
# Bug appears to be in the store path with second work-group, set AB=1 still fails.
#   Check usage of WorkGroup0/1 and consistency of output space assignment in reads and stores.
#
#   Check cstrides0/1 written by host code - seems ok, likely WG-related problem.
#   Maybe need to swap wg0,wg1 in store code for this case
#
#   Diff code with slightly different mapping