Lunderberg commented on code in PR #16596: URL: https://github.com/apache/tvm/pull/16596#discussion_r1499465097
########## src/relax/transform/reorder_permute_dims_after_concat.cc: ########## @@ -0,0 +1,173 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +/*! + * \file tvm/relax/transform/reorder_permute_dims_after_concat.cc + * \brief Reorder concat(permute_dims(A), permute_dims(B)) into permute_dims(concat(A,B)) + */ + +#include <tvm/relax/analysis.h> +#include <tvm/relax/dataflow_matcher.h> +#include <tvm/relax/expr.h> +#include <tvm/relax/expr_functor.h> +#include <tvm/relax/transform.h> + +#include <optional> +#include <unordered_set> +#include <vector> + +#include "../op/tensor/index.h" +#include "../op/tensor/linear_algebra.h" +#include "../op/tensor/manipulate.h" + +namespace tvm { +namespace relax { + +namespace { +std::tuple<DFPattern, TypedPackedFunc<Expr(Expr, Map<DFPattern, Expr>)>> CreatePatterns() { + // TODO(Lunderberg): Allow pattern-matching to handle a flexible + // number of arguments, each of which matches the same type of + // pattern. + size_t min_concat = 2; + size_t max_concat = 12; Review Comment: Entirely arbitrary at the moment. The choice of 12 as the max was to make it significantly larger than any actual case I've seen so far (3 for query/key/value), and because I'd been in the habit of using [this Rust utility](https://docs.rs/itertools/latest/itertools/traits/trait.HomogeneousTuple.html) in hobby projects. Because the long-term plan is to remove the restriction altogether, I'd actually lean away from having a global constant to document it. Having the global constant would allow other portions of the code to depend on it, which would make it harder to migrate to the long-term solution. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
