Hotfix: tune aquavanjaram942X BBS+HHS NN datatype (#1403)

* tune aquavanjaram942X BBS NN datatype

* Tune aquavanjaram942X HHS NN datatype

* retune aquavanjaram942X HHS NN datatype

* retune aquavanjaram942X HHS NN datatype with the setting of LoopDoWhile=false

Author: amgddm

library/src/blas3/Tensile/Logic/asm_full/aquavanjaram942X/aquavanjaram942_Cijk_Ailk_Bljk_BBS_BH.yaml

@@ -21419,6 +21419,293 @@
     _VectorStore: 1
     _WorkspaceSizePerElemC: 0
     _staggerStrideShift: 0
+  - 1LDSBuffer: 1
+    AggressivePerfMode: 1
+    AssertAIGreaterThanEqual: -1
+    AssertAILessThanEqual: -1
+    AssertAlphaValue: false
+    AssertBetaValue: false
+    AssertCEqualsD: false
+    AssertFree0ElementMultiple: 8
+    AssertFree1ElementMultiple: 8
+    AssertMinApproxSize: 3
+    AssertSizeEqual: {}
+    AssertSizeGreaterThan: {3: 128}
+    AssertSizeLessThan: {}
+    AssertSizeMultiple: {}
+    AssertStrideAEqual: {0: 1}
+    AssertStrideBEqual: {0: 1}
+    AssertStrideCEqual: {0: 1}
+    AssertStrideDEqual: {0: 1}
+    AssertSummationElementMultiple: 64
+    AssignedDerivedParameters: true
+    AssignedProblemIndependentDerivedParameters: true
+    AtomicAddC: false
+    BufferLoad: true
+    BufferStore: true
+    CheckDimOverflow: 0
+    CheckTensorDimAsserts: false
+    ClusterLocalRead: true
+    CodeObjectVersion: default
+    CustomKernelName: ''
+    DepthU: 32
+    DepthULdsDivisor: 1
+    DirectToLdsA: false
+    DirectToLdsB: false
+    DirectToVgprA: false
+    DirectToVgprB: false
+    DisableAtomicFail: 0
+    DisableKernelPieces: 0
+    DisableVgprOverlapping: false
+    EdgeType: ShiftPtr
+    EnableF32XdlMathOp: false
+    EnableMatrixInstruction: true
+    EnableMatrixInstructionStore: true
+    ExpandPointerSwap: true
+    ExtraLatencyForLR: 0
+    ExtraMiLatencyLeft: 0
+    ForceStoreSC1: false
+    Fp16AltImpl: false
+    Fp16AltImplRound: false
+    FractionalLoad: 0
+    GlobalLoadVectorWidthA: 8
+    GlobalLoadVectorWidthB: 8
+    GlobalRead2A: true
+    GlobalRead2B: true
+    GlobalReadCoalesceGroupA: true
+    GlobalReadCoalesceGroupB: true
+    GlobalReadCoalesceVectorA: true
+    GlobalReadCoalesceVectorB: true
+    GlobalReadPerMfma: 1
+    GlobalReadVectorWidth: 8
+    GlobalSplitU: 1
+    GlobalSplitUAlgorithm: SingleBuffer
+    GlobalSplitUAtomicAdd: false
+    GlobalSplitUSummationAssignmentRoundRobin: true
+    GlobalSplitUWorkGroupMappingRoundRobin: false
+    GlobalWriteVectorWidth: 2
+    GroupLoadStore: false
+    GuaranteeNoPartialA: true
+    GuaranteeNoPartialB: true
+    ISA: [9, 4, 2]
+    InnerUnroll: 1
+    InterleaveAlpha: 0
+    KernelLanguage: Assembly
+    LSCA: 256
+    LSCB: 32
+    LSPA: 2
+    LSPB: 64
+    LVCA: 32
+    LVCB: 4
+    LVPA: 1
+    LVPB: 8
+    LdcEqualsLdd: false
+    LdsBlockSizePerPadA: 0
+    LdsBlockSizePerPadB: 128
+    LdsInitCVgprs: false
+    LdsNumElements: 17408
+    LdsNumElementsAlignedA: 8192
+    LdsNumElementsAlignedB: 9216
+    LdsOffsetA: 0
+    LdsOffsetA_Blk: 32768
+    LdsOffsetB: 8192
+    LdsOffsetB_Blk: 40960
+    LdsPadA: 0
+    LdsPadB: 8
+    LocalDotLayout: 1
+    LocalRead2A: true
+    LocalRead2B: true
+    LocalReadVectorWidth: 8
+    LocalSplitU: 1
+    LocalWrite2A: true
+    LocalWrite2B: true
+    LocalWritePerMfma: -1
+    LocalWriteUseSgprA: false
+    LocalWriteUseSgprB: false
+    LoopDoWhile: false
+    LoopIters: 4
+    LoopTail: true
+    LoopUnroll: 32
+    MACInstruction: FMA
+    MFMA_BF16_1K: true
+    MIArchVgpr: false
+    MIBlock: [32, 32, 8, 1, 1, 1]
+    MIInputPerThread: 4
+    MIOutputVectorWidth: 4
+    MIRegPerOut: 1
+    MIWaveGroup: [4, 1]
+    MIWaveTile: [2, 8]
+    MIWaveTileA: 2
+    MIWaveTileB: 8
+    MacroTile0: 256
+    MacroTile1: 256
+    MacroTileA: 256
+    MacroTileB: 256
+    MacroTileShapeMax: 64
+    MacroTileShapeMin: 1
+    MagicDivAlg: 2
+    MatrixInstB: 1
+    MatrixInstBM: 1
+    MatrixInstBN: 1
+    MatrixInstK: 8
+    MatrixInstM: 32
+    MatrixInstN: 32
+    MatrixInstruction: [32, 32, 8, 1]
+    MaxOccupancy: 40
+    MaxVgprNumber: 256
+    MemoryModifierFormat: SC0
+    MinKForGSU: 256
+    MinVgprNumber: 0
+    NoLdsWriteCode: false
+    NoReject: false
+    NonTemporalA: 0
+    NonTemporalB: 0
+    NonTemporalC: 0
+    NonTemporalD: 0
+    NumElementsPerBatchStore: 16
+    NumElementsPerThread: 256
+    NumGlobalWriteVectorsPerThread: 128
+    NumLoadsA: 4
+    NumLoadsB: 4
+    NumLoadsCoalescedA: 1
+    NumLoadsCoalescedB: 1
+    NumLoadsPerpendicularA: 4
+    NumLoadsPerpendicularB: 4
+    NumThreads: 256
+    OptNoLoadLoop: 1
+    OptPreLoopVmcnt: 0
+    PackBatchDims: 0
+    PackFreeDims: 1
+    PackGranularity: 2
+    PackSummationDims: 0
+    PackedC0IdxChars: [I]
+    PackedC0IndicesX: [0]
+    PackedC1IdxChars: [J]
+    PackedC1IndicesX: [1]
+    PerformanceSyncLocation: -1
+    PerformanceWaitCount: -1
+    PerformanceWaitLocation: -1
+    PersistentKernel: 0
+    PersistentKernelAlongBatch: false
+    PrefetchAcrossPersistent: 0
+    PrefetchAcrossPersistentMode: 0
+    PrefetchGlobalRead: 2
+    PrefetchLocalRead: 5
+    ProblemType:
+      AllowNoFreeDims: false
+      AssignedDerivedParameters: true
+      Batched: true
+      ComplexConjugateA: false
+      ComplexConjugateB: false
+      ComputeDataType: 0
+      ConvolutionConfig: []
+      DataType: 7
+      DestDataType: 7
+      F32XdlMathOp: 0
+      Fp16AltImpl: false
+      Fp16AltImplRound: false
+      Fp32toFp8SWClip: true
+      Fp8NoPackUpConversion: false
+      HighPrecisionAccumulate: true
+      Index0: 0
+      Index01A: 0
+      Index01B: 1
+      Index1: 1
+      IndexAssignmentsA: [0, 3, 2]
+      IndexAssignmentsB: [3, 1, 2]
+      IndexAssignmentsLD: [4, 5, 6, 7]
+      IndexUnroll: 3
+      IndexUnrollA: 1
+      IndexUnrollB: 0
+      IndicesBatch: [2]
+      IndicesFree: [0, 1]
+      IndicesSummation: [3]
+      MirrorDimsA: []
+      MirrorDimsB: []
+      NumIndicesBatch: 1
+      NumIndicesC: 3
+      NumIndicesFree: 2
+      NumIndicesLD: 4
+      NumIndicesSummation: 1
+      OperationType: GEMM
+      SetConstStrideA: []
+      SetConstStrideB: []
+      SilentHighPrecisionAccumulate: false
+      StochasticRounding: false
+      StridedBatched: true
+      TLUA: true
+      TLUB: false
+      Tensor0: 0
+      Tensor1: 1
+      TileA: 0
+      TileAwareSelection: false
+      TileB: 1
+      TotalIndices: 4
+      TransposeA: false
+      TransposeB: false
+      UseBeta: true
+      UseInitialStridesAB: false
+      UseInitialStridesCD: false
+      ZeroPadA: []
+      ZeroPadB: []
+    ReplacementKernel: false
+    ScheduleGlobalRead: 1
+    ScheduleIterAlg: 3
+    ScheduleLocalWrite: 1
+    SolutionIndex: 75
+    SolutionNameMin: Cijk_Ailk_Bljk_BBS_BH_MT256x256x32_MI32x32x8x1_SN_WGM32
+    SourceSwap: true
+    StaggerU: 0
+    StaggerUMapping: 0
+    StaggerUStride: 0
+    StoreCInUnroll: false
+    StoreCInUnrollExact: false
+    StoreCInUnrollInterval: 1
+    StoreCInUnrollPostLoop: false
+    StorePriorityOpt: false
+    StoreRemapVectorWidth: 0
+    StoreSyncOpt: 4
+    StoreVectorWidth: 2
+    StreamK: 0
+    SubGroup0: 8
+    SubGroup1: 32
+    SubGroupA: 8
+    SubGroupB: 32
+    SuppressNoLoadLoop: false
+    ThreadSeparateGlobalReadA: 0
+    ThreadSeparateGlobalReadB: 0
+    ThreadTile: [2, 256]
+    ThreadTile0: 32
+    ThreadTile1: 8
+    ThreadTileA: 32
+    ThreadTileB: 8
+    TransposeLDS: 1
+    UnrollIncIsDepthU: 0
+    UnrollMajorLDSA: false
+    UnrollMajorLDSB: true
+    UnrollMemFence: false
+    Use64bShadowLimit: 1
+    UseInstOffsetForGRO: 0
+    UseSgprForGRO: -1
+    Valid: true
+    VectorAtomicWidth: 1
+    VectorStore: -1
+    VectorWidth: 2
+    VectorWidthA: 2
+    VectorWidthB: 2
+    VgprForLocalReadPacking: true
+    WaveSeparateGlobalReadA: 1
+    WaveSeparateGlobalReadB: 0
+    WavefrontSize: 64
+    WorkGroup: [128, 2, 1]
+    WorkGroupMapping: 32
+    WorkGroupMappingType: B
+    _DepthULds: 32
+    _GlobalAccumulation: null
+    _UseSgprForGRO: 1
+    _VectorStore: 1
+    _WorkspaceSizePerElemC: 0
+    _staggerStrideShift: 0
 - [2, 3, 0, 1]
 - - - [6144, 4864, 1, 4096]
     - [41, 74.92]
@@ -21580,6 +21867,8 @@
     - [73, 6.2]
   - - [9728, 64, 1, 1024]
     - [74, 10.2]
+  - - [98304, 98304, 1, 98304]
+    - [75, 83.29]
 - null
 - null
 - DeviceEfficiency