TestParameters: marks: [skip-gfx900, skip-gfx906, skip-gfx1010, skip-gfx1011, emulation2] # not supported by arch GlobalParameters: NumElementsToValidate: 65536 BoundsCheck: True KernelTime: True BenchmarkProblems: ######################################## # NT - standard ######################################## - # sgemm TN - # ProblemType OperationType: GEMM DataType: s TransposeA: True TransposeB: False UseBeta: True Batched: True - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] - EdgeType: ["ShiftPtr"] - PrefetchLocalRead: [True] ForkParameters: - MatrixInstruction: - [32, 32, 1, 2] - PrefetchGlobalRead: [True] - ThreadTile: - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # - [ 4, 64 ] - WorkGroup: # - [ 16, 16, 1 ] - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - [ 128, 4, 1 ] - WorkGroupMapping: [8] - GlobalSplitU: [1] - DepthU: [ 8 ] - VectorWidth: [1] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 2, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 1, 4] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # # - [ 4, 16 ] # # - [ 8, 16 ] # # - [ 2, 32 ] # # - [ 4, 32 ] # # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # # - [ 32, 8, 1 ] # # - [ 64, 4, 1 ] # # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 4, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [ 4, 4, 1, 16] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 8 ] # - [ 2, 16 ] # - [ 4, 8 ] # - WorkGroup: # - [ 8, 32, 1 ] # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 16 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] - # sgemm NN - # ProblemType OperationType: GEMM DataType: s TransposeA: False TransposeB: False UseBeta: True Batched: True # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 1, 2] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # - [ 4, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 2, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # # - [ 1, 32 ] # - [ 2, 32 ] # # - [ 4, 32 ] # # - [ 1, 64 ] # # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # # - [ 32, 8, 1 ] # # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 1, 4] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] - EdgeType: ["ShiftPtr"] - PrefetchLocalRead: [True] ForkParameters: - MatrixInstruction: - [16, 16, 4, 1] - PrefetchGlobalRead: [True] - ThreadTile: # - [ 2, 16 ] - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] - WorkGroup: - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] - WorkGroupMapping: [8] - GlobalSplitU: [1] - DepthU: [ 8 ] - VectorWidth: [1] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [ 4, 4, 1, 16] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 8 ] # - [ 2, 16 ] # - [ 4, 8 ] # - WorkGroup: # - [ 8, 32, 1 ] # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 16 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] - # sgemm NT - # ProblemType OperationType: GEMM DataType: s TransposeA: False TransposeB: True UseBeta: True Batched: True # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 1, 2] # - PrefetchGlobalRead: [True] # - ThreadTile: # # - [ 1, 32 ] # # - [ 2, 32 ] # # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # # - [ 4, 64 ] # - WorkGroup: # # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # # - [ 64, 4, 1 ] # # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 2, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 1, 4] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 4, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] - EdgeType: ["ShiftPtr"] - PrefetchLocalRead: [True] ForkParameters: - MatrixInstruction: - [ 4, 4, 1, 16] - PrefetchGlobalRead: [True] - ThreadTile: - [ 2, 8 ] # - [ 2, 16 ] # - [ 4, 8 ] - WorkGroup: - [ 8, 32, 1 ] # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] - WorkGroupMapping: [8] - GlobalSplitU: [1] - DepthU: [ 16 ] - VectorWidth: [1] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] - Range: [ [1031], [1031], [8], [1031] ] - # sgemm TT - # ProblemType OperationType: GEMM DataType: s TransposeA: True TransposeB: True UseBeta: True Batched: True # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [32, 32, 1, 2] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] # - [ 4, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] - # BenchmarkProblemSizeGroup - Standard InitialSolutionParameters: BenchmarkCommonParameters: - KernelLanguage: ["Assembly"] - EdgeType: ["ShiftPtr"] - PrefetchLocalRead: [True] ForkParameters: - MatrixInstruction: - [32, 32, 2, 1] - PrefetchGlobalRead: [True] - ThreadTile: - [ 1, 32 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 1, 64 ] # - [ 2, 64 ] - WorkGroup: - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] - WorkGroupMapping: [8] - GlobalSplitU: [1] - DepthU: [ 8 ] - VectorWidth: [1] BenchmarkForkParameters: JoinParameters: BenchmarkJoinParameters: BenchmarkFinalParameters: - ProblemSizes: - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 1, 4] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # # - [ 4, 16 ] # # - [ 8, 16 ] # # - [ 2, 32 ] # # - [ 4, 32 ] # # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # # - [ 32, 8, 1 ] # # - [ 64, 4, 1 ] # # - [ 128, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [16, 16, 4, 1] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 16 ] # - [ 4, 16 ] # - [ 8, 16 ] # - [ 2, 32 ] # - [ 4, 32 ] # - [ 2, 64 ] # - WorkGroup: # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - [ 64, 4, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 8 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ] # - # BenchmarkProblemSizeGroup - Standard # InitialSolutionParameters: # BenchmarkCommonParameters: # - KernelLanguage: ["Assembly"] # - EdgeType: ["ShiftPtr"] # - PrefetchLocalRead: [True] # ForkParameters: # - MatrixInstruction: # - [ 4, 4, 1, 16] # - PrefetchGlobalRead: [True] # - ThreadTile: # - [ 2, 8 ] # - [ 2, 16 ] # - [ 4, 8 ] # - WorkGroup: # - [ 8, 32, 1 ] # - [ 16, 16, 1 ] # - [ 32, 8, 1 ] # - WorkGroupMapping: [8] # - GlobalSplitU: [1] # - DepthU: [ 16 ] # - VectorWidth: [1] # BenchmarkForkParameters: # JoinParameters: # BenchmarkJoinParameters: # BenchmarkFinalParameters: # - ProblemSizes: # - Range: [ [1024], [1024], [1], [1024] ] # - Range: [ [1031], [1031], [8], [1031] ]