# benchmark assembly and source kernels GlobalParameters: MinimumRequiredVersion: 4.2.0 NumElementsToValidate: -1 BoundsCheck: True CodeObjectVersion: V5 BenchmarkProblems: - # sgemm NT - # ProblemType OperationType: GEMM DataType: s DestDataType: s TransposeA: False TransposeB: True UseBeta: True Batched: True - # BenchmarkProblemSizeGroup - Assembly InitialSolutionParameters: BenchmarkCommonParameters: - LoopTail: [True] - EdgeType: ["ShiftPtr"] - KernelLanguage: ["Assembly"] ForkParameters: - GlobalSplitU: [1] - PrefetchLocalRead: [True] - PrefetchGlobalRead: [True] - ThreadTile: - [ 4, 8 ] - WorkGroup: - [ 16, 16, 1 ] - DepthU: [-3] - VectorWidth: [-1] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [ 128, 128, 1, 128 ]