Linear Algebra Operators

GEMM

GemmOp

General matrix multiplication with support for transpose modes.

• Supported architectures: Ampere (SM 80/86), Hopper (SM 90)
• Supported dtypes: float16, bfloat16, float32

from tileops.ops import GemmOp

op = GemmOp(dtype=torch.float16)
C = op.forward(A, B)  # C = A @ B

Transpose modes: NN, NT, TN, TT

Also handles GEMV (matrix-vector) cases internally.