Files
carbon-lang/toolchain/parse/tree_and_subtrees.cpp
T
Jon Ross-Perkins f67791cfee Separate subtree size information from parse nodes. (#4174)
Move subtree sizes over to TreeAndSubtrees, using the different
structure to represent the additional parse work that occurs, as well as
making it clear which functions require the extra information. My intent
is to make it hard to use this by accident.

The subtree size is still tracked during Parse::Tree construction. I
think a lot of that can be cleaned up, although we use it during
placeholder assignment so it may take some work. I wanted to see what
people thought about this before taking action on such a change.

I'm using a 1m line source file generated by #4124 for testing. Command
is `time bazel-bin/toolchain/install/prefix_root/bin/carbon compile
--phase=check --dump-mem-usage ~/tmp/data.carbon`

At head, what I'm seeing is:

```
...
parse_tree_.node_impls_:
  used_bytes:      61516116
  reserved_bytes:  61516116
...
Total:
  used_bytes:      447814230
  reserved_bytes:  551663894
...
1.43s user 0.14s system 99% cpu 1.565 total
```

With `Tree::Verify` disabled completely, it looks like:
```
parse_tree_.node_impls_:
  used_bytes:      41010744
  reserved_bytes:  41010744
...
Total:
  used_bytes:      427308858
  reserved_bytes:  531158522
...
1.20s user 0.13s system 99% cpu 1.332 total
```

Re-enabling just the basic verification (what is now `Tree::Verify`),
I'm seeing maybe 0.05s slower, but that's within noise for my system. I
do see variability in my timing results, and overall I think this is a
0.2s +/- 0.1s improvement versus the earlier (always testing `Extract`
code) implementation. That's opt; debug builds will be unaffected,
because the same checking occurs as before.

Note, the subtree size is a third of the node representation, which is
why I'm showing the decrease in memory usage here.
2024-07-31 19:39:45 +00:00

245 lines
8.4 KiB
C++

// Part of the Carbon Language project, under the Apache License v2.0 with LLVM
// Exceptions. See /LICENSE for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
#include "toolchain/parse/tree_and_subtrees.h"
namespace Carbon::Parse {
TreeAndSubtrees::TreeAndSubtrees(const Lex::TokenizedBuffer& tokens,
const Tree& tree)
: tokens_(&tokens), tree_(&tree) {
subtree_sizes_.reserve(tree_->size());
// A stack of nodes which haven't yet been used as children.
llvm::SmallVector<NodeId> size_stack;
for (auto n : tree.postorder()) {
// Nodes always include themselves.
int32_t size = 1;
auto kind = tree.node_kind(n);
if (kind.has_child_count()) {
// When the child count is set, remove the specific number from the stack.
CARBON_CHECK(static_cast<int32_t>(size_stack.size()) >=
kind.child_count())
<< "Need " << kind.child_count() << " children for " << kind
<< ", have " << size_stack.size() << " available";
for (auto i : llvm::seq(kind.child_count())) {
auto child = size_stack.pop_back_val();
CARBON_CHECK((size_t)child.index < subtree_sizes_.size());
size += subtree_sizes_[child.index];
if (kind.has_bracket() && i == kind.child_count() - 1) {
CARBON_CHECK(kind.bracket() == tree.node_kind(child))
<< "Node " << kind << " needs bracket " << kind.bracket()
<< ", found wrong bracket " << tree.node_kind(child);
}
}
} else {
while (true) {
CARBON_CHECK(!size_stack.empty())
<< "Node " << kind << " is missing bracket " << kind.bracket();
auto child = size_stack.pop_back_val();
size += subtree_sizes_[child.index];
if (kind.bracket() == tree.node_kind(child)) {
break;
}
}
}
size_stack.push_back(n);
subtree_sizes_.push_back(size);
}
CARBON_CHECK(static_cast<int>(subtree_sizes_.size()) == tree_->size());
// Remaining nodes should all be roots in the tree; make sure they line up.
CARBON_CHECK(size_stack.back().index ==
static_cast<int32_t>(tree_->size()) - 1)
<< size_stack.back() << " " << tree_->size() - 1;
int prev_index = -1;
for (const auto& n : size_stack) {
CARBON_CHECK(n.index - subtree_sizes_[n.index] == prev_index)
<< "NodeId " << n << " is a root " << tree_->node_kind(n)
<< " with subtree_size " << subtree_sizes_[n.index]
<< ", but previous root was at " << prev_index << ".";
prev_index = n.index;
}
}
auto TreeAndSubtrees::VerifyExtract(NodeId node_id, NodeKind kind,
ErrorBuilder* trace) const -> bool {
switch (kind) {
#define CARBON_PARSE_NODE_KIND(Name) \
case NodeKind::Name: \
return VerifyExtractAs<Name>(node_id, trace).has_value();
#include "toolchain/parse/node_kind.def"
}
}
auto TreeAndSubtrees::Verify() const -> ErrorOr<Success> {
// Validate that each node extracts successfully when not marked as having an
// error.
//
// Without this code, a 10 mloc test case of lex & parse takes 4.129 s ± 0.041
// s. With this additional verification, it takes 5.768 s ± 0.036 s.
for (NodeId n : tree_->postorder()) {
if (tree_->node_has_error(n)) {
continue;
}
auto node_kind = tree_->node_kind(n);
if (!VerifyExtract(n, node_kind, nullptr)) {
ErrorBuilder trace;
trace << llvm::formatv(
"NodeId #{0} couldn't be extracted as a {1}. Trace:\n", n, node_kind);
VerifyExtract(n, node_kind, &trace);
return trace;
}
}
// Validate the roots. Also ensures Tree::ExtractFile() doesn't error.
if (!TryExtractNodeFromChildren<File>(NodeId::Invalid, roots(), nullptr)) {
ErrorBuilder trace;
trace << "Roots of tree couldn't be extracted as a `File`. Trace:\n";
TryExtractNodeFromChildren<File>(NodeId::Invalid, roots(), &trace);
return trace;
}
return Success();
}
auto TreeAndSubtrees::postorder(NodeId n) const
-> llvm::iterator_range<Tree::PostorderIterator> {
// The postorder ends after this node, the root, and begins at the start of
// its subtree.
int start_index = n.index - subtree_sizes_[n.index] + 1;
return Tree::PostorderIterator::MakeRange(NodeId(start_index), n);
}
auto TreeAndSubtrees::children(NodeId n) const
-> llvm::iterator_range<SiblingIterator> {
CARBON_CHECK(n.is_valid());
int end_index = n.index - subtree_sizes_[n.index];
return llvm::iterator_range<SiblingIterator>(
SiblingIterator(*this, NodeId(n.index - 1)),
SiblingIterator(*this, NodeId(end_index)));
}
auto TreeAndSubtrees::roots() const -> llvm::iterator_range<SiblingIterator> {
return llvm::iterator_range<SiblingIterator>(
SiblingIterator(*this,
NodeId(static_cast<int>(subtree_sizes_.size()) - 1)),
SiblingIterator(*this, NodeId(-1)));
}
auto TreeAndSubtrees::PrintNode(llvm::raw_ostream& output, NodeId n, int depth,
bool preorder) const -> bool {
output.indent(2 * (depth + 2));
output << "{";
// If children are being added, include node_index in order to disambiguate
// nodes.
if (preorder) {
output << "node_index: " << n << ", ";
}
output << "kind: '" << tree_->node_kind(n) << "', text: '"
<< tokens_->GetTokenText(tree_->node_token(n)) << "'";
if (tree_->node_has_error(n)) {
output << ", has_error: yes";
}
if (subtree_sizes_[n.index] > 1) {
output << ", subtree_size: " << subtree_sizes_[n.index];
if (preorder) {
output << ", children: [\n";
return true;
}
}
output << "}";
return false;
}
auto TreeAndSubtrees::Print(llvm::raw_ostream& output) const -> void {
output << "- filename: " << tokens_->source().filename() << "\n"
<< " parse_tree: [\n";
// Walk the tree just to calculate depths for each node.
llvm::SmallVector<int> indents;
indents.resize(subtree_sizes_.size(), 0);
llvm::SmallVector<std::pair<NodeId, int>, 16> node_stack;
for (NodeId n : roots()) {
node_stack.push_back({n, 0});
}
while (!node_stack.empty()) {
NodeId n = NodeId::Invalid;
int depth;
std::tie(n, depth) = node_stack.pop_back_val();
for (NodeId sibling_n : children(n)) {
indents[sibling_n.index] = depth + 1;
node_stack.push_back({sibling_n, depth + 1});
}
}
for (NodeId n : tree_->postorder()) {
PrintNode(output, n, indents[n.index], /*preorder=*/false);
output << ",\n";
}
output << " ]\n";
}
auto TreeAndSubtrees::PrintPreorder(llvm::raw_ostream& output) const -> void {
output << "- filename: " << tokens_->source().filename() << "\n"
<< " parse_tree: [\n";
// The parse tree is stored in postorder. The preorder can be constructed
// by reversing the order of each level of siblings within an RPO. The
// sibling iterators are directly built around RPO and so can be used with a
// stack to produce preorder.
// The roots, like siblings, are in RPO (so reversed), but we add them in
// order here because we'll pop off the stack effectively reversing then.
llvm::SmallVector<std::pair<NodeId, int>, 16> node_stack;
for (NodeId n : roots()) {
node_stack.push_back({n, 0});
}
while (!node_stack.empty()) {
NodeId n = NodeId::Invalid;
int depth;
std::tie(n, depth) = node_stack.pop_back_val();
if (PrintNode(output, n, depth, /*preorder=*/true)) {
// Has children, so we descend. We append the children in order here as
// well because they will get reversed when popped off the stack.
for (NodeId sibling_n : children(n)) {
node_stack.push_back({sibling_n, depth + 1});
}
continue;
}
int next_depth = node_stack.empty() ? 0 : node_stack.back().second;
CARBON_CHECK(next_depth <= depth) << "Cannot have the next depth increase!";
for (int close_children_count : llvm::seq(0, depth - next_depth)) {
(void)close_children_count;
output << "]}";
}
// We always end with a comma and a new line as we'll move to the next
// node at whatever the current level ends up being.
output << " ,\n";
}
output << " ]\n";
}
auto TreeAndSubtrees::CollectMemUsage(MemUsage& mem_usage,
llvm::StringRef label) const -> void {
mem_usage.Add(MemUsage::ConcatLabel(label, "subtree_sizes_"), subtree_sizes_);
}
auto TreeAndSubtrees::SiblingIterator::Print(llvm::raw_ostream& output) const
-> void {
output << node_;
}
} // namespace Carbon::Parse