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 DataInitTypeAlpha: 1 DataInitTypeBeta: 0 BoundsCheck: 2 BenchmarkProblems: ######################################## # BBS TN + DTVA ######################################## - - # ProblemType OperationType: GEMM DataType: b DestDataType: b ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: False UseBeta: True Batched: True UseBias: 1 Activation: True ActivationType: hipblaslt_all UseScaleAlphaVec: 1 - # BenchmarkProblemSizeGroup - Standard - All problem InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [16, 16, 16, 1, 1, 5, 16, 4, 1] - [16, 16, 16, 1, 1, 4, 16, 4, 1] - [16, 16, 16, 1, 1, 4, 17, 4, 1] - [16, 16, 16, 1, 1, 4, 18, 4, 1] - [16, 16, 16, 1, 1, 4, 19, 4, 1] - [16, 16, 16, 1, 1, 4, 20, 4, 1] - [16, 16, 16, 1, 1, 4, 21, 4, 1] - AssertFree0ElementMultiple: [1] - AssertFree1ElementMultiple: [1] - AssertSummationElementMultiple: [1] - GlobalReadVectorWidthA: [4, 8] - GlobalReadVectorWidthB: [-1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - DirectToVgprA: [true] - VectorWidthA: [1] - VectorWidthB: [-1] - MIArchVgpr: [false] - LocalWritePerMfma: [-1] - StaggerU: [32] - StaggerUStride: [256] - StaggerUMapping: [0] - WorkGroupMapping: [6] - WorkGroupMappingXCC: [8] - ScheduleIterAlg: [3] - ExpandPointerSwap: [false] - TransposeLDS: [1] - LdsBlockSizePerPadA: [-1] - LdsBlockSizePerPadB: [-1] - StorePriorityOpt: [false] - VectorStore: [-1] - StoreSyncOpt: [0] - LdsPadA: [-1] - LdsPadB: [-1] - 1LDSBuffer: [1] - GlobalSplitU: [1] #[1,2] - GlobalSplitUAlgorithm: ["MultipleBuffer",] #["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [true] - NumElementsPerBatchStore: [16] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [1, 127, 1, 127] - Exact: [2, 127, 1, 127] - Exact: [3, 127, 1, 127] - Exact: [1, 1, 1, 127] - Exact: [127, 1, 1, 127] - Exact: [127, 2, 1, 127] - Exact: [127, 3, 1, 127] - Exact: [127, 127, 1, 127] - Exact: [128, 128, 1, 128] - Exact: [129, 129, 1, 129] - Exact: [127, 127, 1, 128] - Exact: [127, 127, 1, 129] - Exact: [127, 128, 1, 640] - Exact: [129, 128, 1, 640] - Exact: [512, 512, 1, 640] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] ######################################## # HHS TN ######################################## - - # ProblemType OperationType: GEMM DataType: h DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: False UseBeta: True Batched: True UseBias: 1 Activation: True ActivationType: hipblaslt_all UseScaleAlphaVec: 1 - # BenchmarkProblemSizeGroup - Standard - All problem InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [16, 16, 16, 1, 1, 8, 8, 2, 2] - [16, 16, 16, 1, 1, 8, 9, 2, 2] - [16, 16, 16, 1, 1, 8, 10, 2, 2] - [16, 16, 16, 1, 1, 8, 11, 2, 2] - [16, 16, 16, 1, 1, 8, 12, 2, 2] - [16, 16, 16, 1, 1, 9, 8, 2, 2] - [16, 16, 16, 1, 1, 10, 8, 2, 2] - [16, 16, 16, 1, 1, 11, 8, 2, 2] - [16, 16, 16, 1, 1, 12, 8, 2, 2] - [16, 16, 16, 1, 1, 9, 9, 2, 2] - [16, 16, 16, 1, 1, 9, 10, 2, 2] - [16, 16, 16, 1, 1, 10, 9, 2, 2] - [16, 16, 16, 1, 1, 21, 4, 1, 4] - [16, 16, 16, 1, 1, 4, 21, 4, 1] - AssertFree0ElementMultiple: [1] - AssertFree1ElementMultiple: [1] - AssertSummationElementMultiple: [1] - GlobalReadVectorWidthA: [-1] - GlobalReadVectorWidthB: [-1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [-1] - VectorWidthB: [-1] - MIArchVgpr: [false] - LocalWritePerMfma: [-1] - StaggerU: [32] - StaggerUStride: [256] - StaggerUMapping: [0] - WorkGroupMapping: [6] - WorkGroupMappingXCC: [8] - ScheduleIterAlg: [3] - ExpandPointerSwap: [false] - TransposeLDS: [1] - LdsBlockSizePerPadA: [-1] - LdsBlockSizePerPadB: [-1] - StorePriorityOpt: [false] - VectorStore: [-1] - StoreSyncOpt: [0] - LdsPadA: [-1] - LdsPadB: [-1] - 1LDSBuffer: [1] - SourceSwap: [true] - NumElementsPerBatchStore: [16] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [1, 127, 1, 127] - Exact: [2, 127, 1, 127] - Exact: [3, 127, 1, 127] - Exact: [1, 1, 1, 127] - Exact: [127, 1, 1, 127] - Exact: [127, 2, 1, 127] - Exact: [127, 3, 1, 127] - Exact: [127, 127, 1, 127] - Exact: [128, 128, 1, 128] - Exact: [129, 129, 1, 129] - Exact: [127, 127, 1, 128] - Exact: [127, 127, 1, 129] - Exact: [127, 128, 1, 640] - Exact: [129, 128, 1, 640] - Exact: [512, 512, 1, 640] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] ######################################## # BBS NT + VW ######################################## - - # ProblemType OperationType: GEMM DataType: b DestDataType: b ComputeDataType: s HighPrecisionAccumulate: True TransposeA: False TransposeB: True UseBeta: True Batched: True UseBias: 1 Activation: True ActivationType: hipblaslt_all UseScaleAlphaVec: 1 - # BenchmarkProblemSizeGroup - Standard - All problem InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [16, 16, 16, 1, 1, 9, 8, 2, 2] - [16, 16, 16, 1, 1, 8, 9, 2, 2] - [16, 16, 16, 1, 1, 10, 8, 2, 2] - [16, 16, 16, 1, 1, 8, 10, 2, 2] - [16, 16, 16, 1, 1, 4, 17, 4, 1] - [16, 16, 16, 1, 1, 17, 4, 1, 4] - AssertFree0ElementMultiple: [1] - AssertFree1ElementMultiple: [1] - AssertSummationElementMultiple: [1] - GlobalReadVectorWidthA: [1,2,-1] - GlobalReadVectorWidthB: [1,2,-1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1,-1] - VectorWidthB: [-1] - MIArchVgpr: [false] - LocalWritePerMfma: [-1] - StaggerU: [32] - StaggerUStride: [256] - StaggerUMapping: [0] - WorkGroupMapping: [6] - WorkGroupMappingXCC: [8] - ScheduleIterAlg: [3] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [-1] - LdsBlockSizePerPadB: [-1] - StorePriorityOpt: [false] - VectorStore: [-1] - StoreSyncOpt: [0] - LdsPadA: [-1] - LdsPadB: [-1] - 1LDSBuffer: [1] - GlobalSplitU: [1,2,9] - GlobalSplitUAlgorithm: ["MultipleBuffer", "MultipleBufferSingleKernel"] - SourceSwap: [true] - NumElementsPerBatchStore: [16] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [1, 127, 1, 127] - Exact: [2, 127, 1, 127] - Exact: [3, 127, 1, 127] - Exact: [1, 1, 1, 127] - Exact: [127, 1, 1, 127] - Exact: [127, 2, 1, 127] - Exact: [127, 3, 1, 127] - Exact: [127, 127, 1, 127] - Exact: [128, 128, 1, 128] - Exact: [129, 129, 1, 129] - Exact: [127, 127, 1, 128] - Exact: [127, 127, 1, 129] - Exact: [127, 128, 1, 640] - Exact: [129, 128, 1, 640] - Exact: [512, 512, 1, 640] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] ######################################## # F8NBS TN ######################################## - - # ProblemType OperationType: GEMM DataType: f8n DestDataType: b ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: False UseBeta: True Batched: True UseBias: 1 Activation: True ActivationType: hipblaslt_all UseScaleAlphaVec: 1 BiasDataTypeList: [s, b] - # BenchmarkProblemSizeGroup - Standard - All problem InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [16, 16, 32, 1, 1, 8, 8, 2, 2] - [16, 16, 32, 1, 1, 8, 9, 2, 2] - [16, 16, 32, 1, 1, 8, 10, 2, 2] - [16, 16, 32, 1, 1, 8, 11, 2, 2] - [16, 16, 32, 1, 1, 8, 12, 2, 2] - [16, 16, 32, 1, 1, 8, 13, 2, 2] - [16, 16, 32, 1, 1, 11, 9, 2, 2] - [16, 16, 32, 1, 1, 11, 10, 2, 2] - [16, 16, 32, 1, 1, 9, 11, 2, 2] - [16, 16, 32, 1, 1, 10, 11, 2, 2] - [16, 16, 32, 1, 1, 10, 10, 2, 2] - [16, 16, 32, 1, 1, 9, 8, 2, 2] - [16, 16, 32, 1, 1, 10, 8, 2, 2] - [16, 16, 32, 1, 1, 11, 8, 2, 2] - [16, 16, 32, 1, 1, 12, 8, 2, 2] - [16, 16, 32, 1, 1, 13, 8, 2, 2] - [16, 16, 32, 1, 1, 9, 9, 2, 2] - [16, 16, 32, 1, 1, 9, 10, 2, 2] - [16, 16, 32, 1, 1, 10, 9, 2, 2] - AssertFree0ElementMultiple: [1] - AssertFree1ElementMultiple: [1] - AssertSummationElementMultiple: [1] - GlobalReadVectorWidthA: [-1] - GlobalReadVectorWidthB: [-1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [-1] - VectorWidthB: [-1] - MIArchVgpr: [false] - LocalWritePerMfma: [-1] - StaggerU: [32] - StaggerUStride: [256] - StaggerUMapping: [0] - WorkGroupMapping: [6] - WorkGroupMappingXCC: [8] - ScheduleIterAlg: [3] - ExpandPointerSwap: [false] - TransposeLDS: [1] - LdsBlockSizePerPadA: [-1] - LdsBlockSizePerPadB: [-1] - StorePriorityOpt: [false] - VectorStore: [-1] - StoreSyncOpt: [0] - LdsPadA: [-1] - LdsPadB: [-1] - 1LDSBuffer: [1] - SourceSwap: [true] - NumElementsPerBatchStore: [16] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [1, 127, 1, 127] - Exact: [2, 127, 1, 127] - Exact: [3, 127, 1, 127] - Exact: [1, 1, 1, 127] - Exact: [127, 1, 1, 127] - Exact: [127, 2, 1, 127] - Exact: [127, 3, 1, 127] - Exact: [127, 127, 1, 127] - Exact: [128, 128, 1, 128] - Exact: [129, 129, 1, 129] - Exact: [127, 127, 1, 128] - Exact: [127, 127, 1, 129] - Exact: [127, 128, 1, 640] - Exact: [129, 128, 1, 640] - Exact: [512, 512, 1, 640] - BiasTypeArgs: ['s', 'b'] - ActivationArgs: - [Enum: none]