TestParameters: marks: [skip-gfx950, skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201, skip-gfx1250] # not supported yet GlobalParameters: SyncsPerBenchmark: 0 NumElementsToValidate: -1 MinimumRequiredVersion: 5.0.0 PrintLevel: 1 PrintSolutionRejectionReason: True Device: 0 CMakeBuildType: Release KernelTime: True MaxWorkspaceSize: 13421772800 DataInitTypeAlpha: 1 DataInitTypeBeta: 0 DataInitTypeA: 1 DataInitTypeB: 1 BenchmarkProblems: ######################################## # NN - standard ######################################## - - # ProblemType OperationType: GEMM DataType: d TransposeA: False TransposeB: False UseBeta: True Batched: True Activation: False B2BGemm: False - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: ForkParameters: - KernelLanguage: ["Assembly"] - DepthU: [8, 16] - GlobalSplitU: [1] - ScheduleIterAlg: [3] - MatrixInstruction: - [16, 16, 4, 1, 1, 1, 1, 1, 1] - AssertFree0ElementMultiple: [1] - VectorWidthA: [1, 2] - VectorWidthB: [1, 2] - GlobalSplitU: [1,2] - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [false, true] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 1024] - Exact: [4608, 1335, 1, 640] ######################################## # TN - standard ######################################## - - # ProblemType OperationType: GEMM DataType: d TransposeA: True TransposeB: False UseBeta: True Batched: True Activation: False B2BGemm: False - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: ForkParameters: - KernelLanguage: ["Assembly"] - DepthU: [8, 16] - GlobalSplitU: [1] - ScheduleIterAlg: [3] - MatrixInstruction: - [16, 16, 4, 1, 1, 1, 1, 1, 1] - AssertFree0ElementMultiple: [1] - VectorWidthA: [1, 2] - VectorWidthB: [1, 2] - GlobalSplitU: [1,2] - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [false, true] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 1024] - Exact: [4608, 1335, 1, 640] ######################################## # NT - standard ######################################## - - # ProblemType OperationType: GEMM DataType: d TransposeA: False TransposeB: True UseBeta: True Batched: True Activation: False B2BGemm: False - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: ForkParameters: - KernelLanguage: ["Assembly"] - DepthU: [8, 16] - GlobalSplitU: [1] - ScheduleIterAlg: [3] - MatrixInstruction: - [16, 16, 4, 1, 1, 1, 1, 1, 1] - AssertFree0ElementMultiple: [1] - VectorWidthA: [1, 2] - VectorWidthB: [1, 2] - GlobalSplitU: [1,2] - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [false, true] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 1024] - Exact: [4608, 1335, 1, 640] ######################################## # TT - standard ######################################## - - # ProblemType OperationType: GEMM DataType: d TransposeA: True TransposeB: True UseBeta: True Batched: True Activation: False B2BGemm: False - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: ForkParameters: - KernelLanguage: ["Assembly"] - DepthU: [8, 16] - GlobalSplitU: [1] - ScheduleIterAlg: [3] - MatrixInstruction: - [16, 16, 4, 1, 1, 1, 1, 1, 1] - AssertFree0ElementMultiple: [1] - VectorWidthA: [1, 2] - VectorWidthB: [1, 2] - GlobalSplitU: [1,2] - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [false, true] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 1024] - Exact: [4608, 1335, 1, 640]