TestParameters: marks: [skip-gfx1100, skip-gfx1101, skip-gfx1102, skip-gfx1200, skip-gfx1201, skip-gfx1250] # not supported yet GlobalParameters: SyncsPerBenchmark: 0 MinimumRequiredVersion: 5.0.0 PrintLevel: 1 PrintSolutionRejectionReason: True Device: 0 CMakeBuildType: Release KernelTime: True MaxWorkspaceSize: 13421772800 DataInitTypeAlpha: 1 DataInitTypeBeta: 0 # DataInitTypeA: 1 # DataInitTypeB: 1 UseUserArgs: False NumElementsToValidate: 128 BoundsCheck: 2 BenchmarkProblems: ######################################### # SHHS ######################################### - - # ProblemType OperationType: GEMM DataType: h DataTypeA: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: False TransposeB: False UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeA: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: False UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeA: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: False TransposeB: True UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeA: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: True UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] ######################################### # HSHS ######################################### - - # ProblemType OperationType: GEMM DataType: h DataTypeB: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: False TransposeB: False UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeB: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: False TransposeB: True UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeB: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: False UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none] - - # ProblemType OperationType: GEMM DataType: h DataTypeB: s DestDataType: h ComputeDataType: s HighPrecisionAccumulate: True TransposeA: True TransposeB: True UseBeta: True Batched: True UseBias: 1 - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] ForkParameters: - MatrixInstruction: - [32, 32, 8, 1, 1, 1, 1, 2,2 ] - AssertFree0ElementMultiple: [1] - PrefetchGlobalRead: [2] - PrefetchLocalRead: [1] - ClusterLocalRead: [1] - DepthU: [32] - VectorWidthA: [1] - VectorWidthB: [1] - ScheduleIterAlg: [3] - InnerUnroll: [1] - ExpandPointerSwap: [false] - TransposeLDS: [0] - LdsBlockSizePerPadA: [0] - LdsBlockSizePerPadB: [0] - LdsPadA: [0] - LdsPadB: [0] - WaveSeparateGlobalReadB: [1] - 1LDSBuffer: [-1] - GlobalSplitU: [1] - GlobalSplitUAlgorithm: ["MultipleBuffer"] - GlobalReadPerMfma: [1.0] - LocalWritePerMfma: [-1] - StoreVectorWidth: [-1] - SourceSwap: [true] - NumElementsPerBatchStore: [2] - StorePriorityOpt: [true] BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Exact: [128, 128, 1, 128] # - Exact: [] - BiasTypeArgs: ['s'] - ActivationArgs: - [Enum: none]