tvm_tirx.tirx.tile.views.Gemm

class tvm_tirx.tirx.tile.views.Gemm

Bases: TensorCall

General matrix multiplication: D = A * B * alpha + C * beta.

Parameters:

• D – Output matrix

• A – First input matrix

• B – Second input matrix

• C – Third input matrix (for bias)

• transpose_A – Whether to transpose A

• transpose_B – Whether to transpose B

• alpha – Scalar multiplier for A*B

• beta – Scalar multiplier for C

kind = 'gemm'

output

lhs

rhs

bias

transpose_A

transpose_B

alpha

beta

property srcs: list[Expr]

Get the source matrices.

property dsts: list[Expr]

Get the destination matrix.

get_private_tensors_trn(tensor_dict, sctx)

Parameters:

• op (TensorCall)

• tensor_dict (dict[Any, tuple[Bind, Stmt | None]])

• sctx (DispatchContext)

Return type:

dict[str, Any]