Matmul Feature Scenarios

In addition to the basic computing capabilities of Matmul described above, you also need to master the basic knowledge and operator implementation of Matmul. In addition, Matmul matrix programming provides processing capabilities and multiple functions applicable to different scenarios. The following table lists the specific scenarios and key APIs or parameters used. For details, see the corresponding sections.

**Table 1** Matmul features
Feature Description	Key API or Parameter
Multi-core aligned tiling	SetDim and EnableMultiCoreSplitK (multi-core K splitting)
Multi-core non-aligned tiling	SetTail, SetDim, and EnableMultiCoreSplitK (multi-core K splitting)
Asynchronous scenario processing	Iterate, GetTensorC, and IterateAll
Quantization and dequantization of matrix multiplication outputs	SetDequantType, SetQuantScalar, and SetQuantVector
Custom data movement	MatmulCallBackFunc, SetUserDefInfo, and SetSelfDefineData
Channel split of matrix multiplication outputs	isEnableChannelSplit in MatmulConfig
General matrix-vector multiplication	SetAType
4:2 sparse matrix multiplication	SetSparse and SetSparseIndex
Upper/Lower triangular matrix multiplication	MatmulPolicy template parameters
Matrix multiplication with TSCM inputs	DataCopy
N-direction alignment of matrix multiplication outputs	SetCType
Partial output for a single matrix multiplication	isPartialOutput in MatmulConfig
Independent running mechanism of AIC and AIV	enableMixDualMaster in MatmulConfig
MxMatmul scenario	SetScaleAType, SetScaleBType, SetMadType, SetTensorScaleA, and SetTensorScaleB

**Table 2** BatchMatmul features
Feature Description	Involved API
Basic functions of BatchMatmul	BatchMatmul in the NORMAL format: IterateBatch and SetBatchInfoForNormal
Basic functions of BatchMatmul	BatchMatmul in the BSNGD, SBNGD, and BNGS1S2 formats: IterateBatch, SetALayout, SetBLayout, SetCLayout, and SetBatchNum
BatchMatmul reusing bias matrix	GetMMConfig, IterateBatch, and SetBatchInfoForNormal

Parent topic: Matmul Kernel APIs