tilelang.tileop.gemm.gemm_wmma

GEMM implementation using AMD RDNA WMMA instructions (gfx11/gfx12).

Classes

GemmWMMA

GEMM using AMD RDNA WMMA instructions (16×16×16, warp-size=32).

Module Contents

class tilelang.tileop.gemm.gemm_wmma.GemmWMMA

Bases: tilelang.tileop.gemm.gemm_base.GemmBase

GEMM using AMD RDNA WMMA instructions (16×16×16, warp-size=32).

infer_layout(target, thread_nums)

Parameters:

• target (tvm.target.Target)

• thread_nums (int)

lower(layout_map, target, thread_bounds, thread_var)

Parameters:

• layout_map (dict)

• target (tvm.target.Target)

• thread_bounds (tvm.ir.Range)

• thread_var (tvm.tir.Var)

is_gemm_ss()

Return type:

bool

is_gemm_sr()

Return type:

bool

is_gemm_rs()

Return type:

bool

is_gemm_rr()

Return type:

bool