Regex: Refactor thread handling in ThreadedRegexVM

This commit is contained in:
Maxime Coste 2017-10-03 18:00:52 +08:00
parent 589cde67f0
commit 9fbafba4cb

View File

@ -100,13 +100,12 @@ struct ThreadedRegexVM
}; };
enum class StepResult { Consumed, Matched, Failed }; enum class StepResult { Consumed, Matched, Failed };
StepResult step(size_t thread_index) StepResult step(Thread& thread)
{ {
const auto prog_start = m_program.bytecode.data(); const auto prog_start = m_program.bytecode.data();
const auto prog_end = prog_start + m_program.bytecode.size(); const auto prog_end = prog_start + m_program.bytecode.size();
while (true) while (true)
{ {
auto& thread = m_threads[thread_index];
const Codepoint cp = m_pos == m_end ? 0 : *m_pos; const Codepoint cp = m_pos == m_end ? 0 : *m_pos;
const CompiledRegex::Op op = (CompiledRegex::Op)*thread.inst++; const CompiledRegex::Op op = (CompiledRegex::Op)*thread.inst++;
switch (op) switch (op)
@ -130,7 +129,7 @@ struct ThreadedRegexVM
auto child = prog_start + *reinterpret_cast<const CompiledRegex::Offset*>(thread.inst); auto child = prog_start + *reinterpret_cast<const CompiledRegex::Offset*>(thread.inst);
thread.inst = parent; thread.inst = parent;
++thread.saves->refcount; ++thread.saves->refcount;
m_threads.insert(m_threads.begin() + thread_index + 1, {child, thread.saves}); m_current_threads.push_back({child, thread.saves});
break; break;
} }
case CompiledRegex::Split_PrioritizeChild: case CompiledRegex::Split_PrioritizeChild:
@ -139,7 +138,7 @@ struct ThreadedRegexVM
auto child = prog_start + *reinterpret_cast<const CompiledRegex::Offset*>(thread.inst); auto child = prog_start + *reinterpret_cast<const CompiledRegex::Offset*>(thread.inst);
thread.inst = child; thread.inst = child;
++thread.saves->refcount; ++thread.saves->refcount;
m_threads.insert(m_threads.begin() + thread_index + 1, {parent, thread.saves}); m_current_threads.push_back({parent, thread.saves});
break; break;
} }
case CompiledRegex::Save: case CompiledRegex::Save:
@ -219,10 +218,12 @@ struct ThreadedRegexVM
bool exec(Iterator begin, Iterator end, RegexExecFlags flags) bool exec(Iterator begin, Iterator end, RegexExecFlags flags)
{ {
bool found_match = false; bool found_match = false;
m_threads.clear(); m_current_threads.clear();
m_next_threads.clear();
const auto start_offset = (flags & RegexExecFlags::Search) ? 0 : CompiledRegex::search_prefix_size; const auto start_offset = (flags & RegexExecFlags::Search) ? 0 : CompiledRegex::search_prefix_size;
m_saves.push_back(std::make_unique<Saves>(Saves{1, Vector<Iterator>(m_program.save_count, Iterator{})})); m_saves.push_back(std::make_unique<Saves>(Saves{1, Vector<Iterator>(m_program.save_count, Iterator{})}));
m_threads.push_back({m_program.bytecode.data() + start_offset, m_saves.back().get()}); m_current_threads.push_back({m_program.bytecode.data() + start_offset, m_saves.back().get()});
m_begin = begin; m_begin = begin;
m_end = end; m_end = end;
@ -238,60 +239,55 @@ struct ThreadedRegexVM
for (m_pos = Utf8It{m_begin, m_begin, m_end}; m_pos != m_end; ++m_pos) for (m_pos = Utf8It{m_begin, m_begin, m_end}; m_pos != m_end; ++m_pos)
{ {
for (int i = 0; i < m_threads.size(); ++i) while (not m_current_threads.empty())
{ {
const auto res = step(i); auto thread = m_current_threads.back();
if (res == StepResult::Matched) m_current_threads.pop_back();
switch (step(thread))
{ {
case StepResult::Matched:
if (not (flags & RegexExecFlags::Search) or // We are not at end, this is not a full match if (not (flags & RegexExecFlags::Search) or // We are not at end, this is not a full match
(flags & RegexExecFlags::NotInitialNull and m_pos == m_begin)) (flags & RegexExecFlags::NotInitialNull and m_pos == m_begin))
{ {
m_threads[i].inst = nullptr; release_saves(thread.saves);
release_saves(m_threads[i].saves);
continue; continue;
} }
m_captures = std::move(m_threads[i].saves->pos); m_captures = std::move(thread.saves->pos);
if (flags & RegexExecFlags::AnyMatch) if (flags & RegexExecFlags::AnyMatch)
return true; return true;
found_match = true; found_match = true;
m_threads.resize(i); // remove this and lower priority threads m_current_threads.clear(); // remove this and lower priority threads
} break;
else if (res == StepResult::Failed) case StepResult::Failed:
{ release_saves(thread.saves);
m_threads[i].inst = nullptr; break;
release_saves(m_threads[i].saves); case StepResult::Consumed:
} if (contains_that(m_next_threads, [&](auto& t) { return t.inst == thread.inst; }))
else release_saves(thread.saves);
{ else
auto it = m_threads.begin() + i; m_next_threads.push_back(thread);
if (std::find_if(m_threads.begin(), it, [inst = it->inst](auto& t) break;
{ return t.inst == inst; }) != it)
{
m_threads[i].inst = nullptr;
release_saves(m_threads[i].saves);
}
} }
} }
// Remove dead threads if (m_next_threads.empty())
m_threads.erase(std::remove_if(m_threads.begin(), m_threads.end(),
[](auto& t) { return t.inst == nullptr; }),
m_threads.end());
// we should never have more than one thread on the same instruction
kak_assert(m_threads.size() <= m_program.bytecode.size());
if (m_threads.empty())
return found_match; return found_match;
std::swap(m_current_threads, m_next_threads);
std::reverse(m_current_threads.begin(), m_current_threads.end());
} }
if (found_match) if (found_match)
return true; return true;
// Step remaining threads to see if they match without consuming anything else // Step remaining threads to see if they match without consuming anything else
for (int i = 0; i < m_threads.size(); ++i) while (not m_current_threads.empty())
{ {
if (step(i) == StepResult::Matched) auto thread = m_current_threads.back();
m_current_threads.pop_back();
if (step(thread) == StepResult::Matched)
{ {
m_captures = std::move(m_threads[i].saves->pos); m_captures = std::move(thread.saves->pos);
return true; return true;
} }
} }
@ -318,7 +314,8 @@ struct ThreadedRegexVM
} }
const CompiledRegex& m_program; const CompiledRegex& m_program;
Vector<Thread> m_threads; Vector<Thread> m_current_threads;
Vector<Thread> m_next_threads;
using Utf8It = utf8::iterator<Iterator>; using Utf8It = utf8::iterator<Iterator>;