2 //Copyright (C) 2016 Google, Inc.
6 //Redistribution and use in source and binary forms, with or without
7 //modification, are permitted provided that the following conditions
10 // Redistributions of source code must retain the above copyright
11 // notice, this list of conditions and the following disclaimer.
13 // Redistributions in binary form must reproduce the above
14 // copyright notice, this list of conditions and the following
15 // disclaimer in the documentation and/or other materials provided
16 // with the distribution.
18 // Neither the name of 3Dlabs Inc. Ltd. nor the names of its
19 // contributors may be used to endorse or promote products derived
20 // from this software without specific prior written permission.
22 //THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 //"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 //LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 //FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 //COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 //INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 //BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 //LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
30 //CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 //LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
32 //ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 //POSSIBILITY OF SUCH DAMAGE.
36 // Author: Dejan Mircevski, Google
39 // The SPIR-V spec requires code blocks to appear in an order satisfying the
40 // dominator-tree direction (ie, dominator before the dominated). This is,
41 // actually, easy to achieve: any pre-order CFG traversal algorithm will do it.
42 // Because such algorithms visit a block only after traversing some path to it
43 // from the root, they necessarily visit the block's idom first.
45 // But not every graph-traversal algorithm outputs blocks in an order that
46 // appears logical to human readers. The problem is that unrelated branches may
47 // be interspersed with each other, and merge blocks may come before some of the
48 // branches being merged.
50 // A good, human-readable order of blocks may be achieved by performing
51 // depth-first search but delaying merge nodes until after all their branches
52 // have been visited. This is implemented below by the inReadableOrder()
58 #include <unordered_map>
64 // Traverses CFG in a readable order, invoking a pre-set callback on each block.
65 // Use by calling visit() on the root block.
66 class ReadableOrderTraverser {
68 explicit ReadableOrderTraverser(std::function<void(Block*)> callback)
69 : callback_(callback) {}
71 // Visits the block if it hasn't been visited already and isn't currently
72 // being delayed. Invokes callback(block), then descends into its successors.
73 // Delays merge-block processing until all the branches have been completed.
74 void visit(Block* block) {
76 if (visited_[block] || delayed_[block]) return;
78 visited_[block] = true;
79 Block* mergeBlock = nullptr;
80 auto mergeInst = block->getMergeInstruction();
82 Id mergeId = mergeInst->getIdOperand(0);
84 block->getParent().getParent().getInstruction(mergeId)->getBlock();
85 delayed_[mergeBlock] = true;
87 for (const auto succ : block->getSuccessors()) visit(succ);
89 delayed_[mergeBlock] = false;
95 std::function<void(Block*)> callback_;
96 // Whether a block has already been visited or is being delayed.
97 std::unordered_map<Block*, bool> visited_, delayed_;
101 void spv::inReadableOrder(Block* root, std::function<void(Block*)> callback) {
102 ReadableOrderTraverser(callback).visit(root);