hzfan commented on a change in pull request #5092: [TIR][PASS] dtype rewrite for indexing variables URL: https://github.com/apache/incubator-tvm/pull/5092#discussion_r400252957
########## File path: src/tir/pass/narrow_datatype.cc ########## @@ -0,0 +1,373 @@ +/* + * 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 narrow_datatype.cc + * \brief narrow the datatype of indexing vars + */ + +#include <tvm/tir/ir_pass.h> +#include <tvm/tir/op.h> +#include "../../arith/ir_mutator_with_analyzer.h" +#include "../../arith/ir_visitor_with_analyzer.h" + +namespace tvm { +namespace tir { + +// This pass narrows indexing expressions (like StoreNode::Index) +// that trivially fit into i32 to i32. Considering that i32 indices +// may be more efficient on some backends (while i64 may be more +// efficient on others, like llvm), we may want this pass when i32 +// indices are more efficient. +// +// For Var v, we determine its dtype by examining all the PrimExpr +// that contains v, denoted by E = {e_0 = v, e_1, e_2, ..., e_k}. +// If all expressions in E fit into i32, then we think v can be narrowed +// to i32. +// +// To make an indexing expression i32, we must make sure that every +// component of that expression is of dtype i32. So besides Var, we +// rewrite the following inside an indexing expression +// - Var +// - IntImm +// - Cast +// +// Algorithm: +// - Use DataTypeVisitor to determine whether a Var can be narrowed or not. +// - Use DataTypeRewritter to rewrite the components of an indexing expression. + +using arith::Analyzer; +using arith::IRMutatorWithAnalyzer; +using arith::ConstIntBound; + +class DataTypeVisitor final : public StmtExprVisitor { + public: + explicit DataTypeVisitor(int target_bits) + : bits_(target_bits), target_bits_(target_bits) {} + + void VisitExpr(const PrimExpr& e) { + if (e.dtype().is_int()) { + int bits = max_bits_; + ConstIntBound bound = analyzer_.const_int_bound(e); + int64_t ubound = Downcast<IntImm, PrimExpr>(max_value(DataType::Int(target_bits_)))->value; + int64_t lbound = Downcast<IntImm, PrimExpr>(min_value(DataType::Int(target_bits_)))->value; + if (e.dtype().bits() <= target_bits_ || Review comment: Yes, I think this pass aligns with what you said. It only narrows and never promotes. This code means lower bits fits into higher bits. For example, Consider `e` is `i.i64 <= j.i64`, a bool expression with only 1 bit. So `e` fits into i32 and does not hinder narrowing i to i32. https://github.com/apache/incubator-tvm/pull/5092/files#diff-98ae729cf00e30cff311ed80b4a25df9R129 ensures dtype promotion does not occcur. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
