Method matmul

tests/python/relax/test_transform_fuse_ops.py:1669–1679

(x: T.Buffer((T.int64(16), T.int64(16)), "float32"), lv2: T.Buffer((T.int64(16), T.int64(16)), "float32"), T_matmul: T.Buffer((T.int64(16), T.int64(16)), "float32"))

        @T.prim_func(private=True, s_tir=True)
        def matmul(x: T.Buffer((T.int64(16), T.int64(16)), "float32"), lv2: T.Buffer((T.int64(16), T.int64(16)), "float32"), T_matmul: T.Buffer((T.int64(16), T.int64(16)), "float32")):
            T.func_attr({"tirx.noalias": True})
            # with T.sblock("root"):
            for ax0, ax1, k in T.grid(T.int64(16), T.int64(16), T.int64(16)):
                with T.sblock("T_matmul"):
                    v_ax0, v_ax1, v_k = T.axis.remap("SSR", [ax0, ax1, k])
                    T.reads(x[v_ax0, v_k], lv2[v_k, v_ax1])
                    T.writes(T_matmul[v_ax0, v_ax1])
                    with T.init():
                        T_matmul[v_ax0, v_ax1] = T.float32(0)
                    T_matmul[v_ax0, v_ax1] = T_matmul[v_ax0, v_ax1] + x[v_ax0, v_k] * lv2[v_k, v_ax1]

        @R.function
        def main(x: R.Tensor((16, 16), dtype="float32"), packed_params: R.Tuple(R.Tensor((16, 16), dtype="float16"), R.Tensor((16, 16), dtype="float16"))) -> R.Tensor((16, 16), dtype="float32"):