mirrors/serenity
1
0
Fork 0
mirror of https://github.com/SerenityOS/serenity.git synced 2025-01-24 18:32:28 -05:00
Code Issues Projects Releases Packages Wiki Activity

LibRegex: Merge alternations based on blocks and not instructions

Browse source 
The instructions can have dependencies (e.g. Repeat), so only unify
equal blocks instead of consecutive instructions.
Fixes #11247.

Also adds the minimal test case(s) from that issue.
This commit is contained in:
Ali Mohammad Pur 2021-12-15 08:52:07 +03:30 • committed by Ali Mohammad Pur
parent 92233660b8
commit d2e51fafa9
3 changed files with 23 additions and 17 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
Tests/LibRegex/Regex.cpp
View file

@ -899,7 +899,9 @@ TEST_CASE(optimizer_atomic_groups)
// Alternative fuse // Alternative fuse
Tuple { "(abcfoo|abcbar|abcbaz).*x"sv, "abcbarx"sv, true }, Tuple { "(abcfoo|abcbar|abcbaz).*x"sv, "abcbarx"sv, true },
Tuple { "(a|a)"sv, "a"sv, true }, Tuple { "(a|a)"sv, "a"sv, true },
Tuple { "(a|)"sv, ""sv, true }, // Ensure that empty alternatives are not outright removed Tuple { "(a|)"sv, ""sv, true }, // Ensure that empty alternatives are not outright removed
Tuple { "a{2,3}|a{5,8}"sv, "abc"sv, false }, // Optimiser should not mess up the instruction stream by ignoring inter-insn dependencies, see #11247.
Tuple { "^(a{2,3}|a{5,8})$"sv, "aaaa"sv, false }, // Optimiser should not mess up the instruction stream by ignoring inter-insn dependencies, see #11247.
// ForkReplace shouldn't be applied where it would change the semantics // ForkReplace shouldn't be applied where it would change the semantics
Tuple { "(1+)\\1"sv, "11"sv, true }, Tuple { "(1+)\\1"sv, "11"sv, true },
Tuple { "(1+)1"sv, "11"sv, true }, Tuple { "(1+)1"sv, "11"sv, true },

5
Userland/Libraries/LibRegex/RegexMatcher.h
View file

@ -227,10 +227,11 @@ public:
return result.success; return result.success;
} }
using BasicBlockList = Vector<Detail::Block>;
static BasicBlockList split_basic_blocks(ByteCode const&);
private: private:
void run_optimization_passes(); void run_optimization_passes();
using BasicBlockList = Vector<Detail::Block>;
BasicBlockList split_basic_blocks();
void attempt_rewrite_loops_as_atomic_groups(BasicBlockList const&); void attempt_rewrite_loops_as_atomic_groups(BasicBlockList const&);
}; };

31
Userland/Libraries/LibRegex/RegexOptimizer.cpp
View file

@ -21,16 +21,15 @@ void Regex<Parser>::run_optimization_passes()
// Rewrite fork loops as atomic groups // Rewrite fork loops as atomic groups
// e.g. a*b -> (ATOMIC a*)b // e.g. a*b -> (ATOMIC a*)b
attempt_rewrite_loops_as_atomic_groups(split_basic_blocks()); attempt_rewrite_loops_as_atomic_groups(split_basic_blocks(parser_result.bytecode));
parser_result.bytecode.flatten(); parser_result.bytecode.flatten();
} }
template<typename Parser> template<typename Parser>
typename Regex<Parser>::BasicBlockList Regex<Parser>::split_basic_blocks() typename Regex<Parser>::BasicBlockList Regex<Parser>::split_basic_blocks(ByteCode const& bytecode)
{ {
BasicBlockList block_boundaries; BasicBlockList block_boundaries;
auto& bytecode = parser_result.bytecode;
size_t end_of_last_block = 0; size_t end_of_last_block = 0;
MatchState state; MatchState state;
@ -472,33 +471,37 @@ void Optimizer::append_alternation(ByteCode& target, ByteCode&& left, ByteCode&&
return; return;
} }
auto left_blocks = Regex<PosixBasicParser>::split_basic_blocks(left);
auto right_blocks = Regex<PosixBasicParser>::split_basic_blocks(right);
size_t left_skip = 0; size_t left_skip = 0;
MatchState state; MatchState state;
for (state.instruction_position = 0; state.instruction_position < left.size() && state.instruction_position < right.size();) { for (size_t block_index = 0; block_index < left_blocks.size() && block_index < right_blocks.size(); block_index++) {
auto left_size = left.get_opcode(state).size(); auto& left_block = left_blocks[block_index];
auto right_size = right.get_opcode(state).size(); auto& right_block = right_blocks[block_index];
if (left_size != right_size) auto left_end = block_index + 1 == left_blocks.size() ? left_block.end : left_blocks[block_index + 1].start;
auto right_end = block_index + 1 == right_blocks.size() ? right_block.end : right_blocks[block_index + 1].start;
if (left_end - left_block.start != right_end - right_block.start)
break; break;
if (left.spans().slice(state.instruction_position, left_size) == right.spans().slice(state.instruction_position, right_size)) if (left.spans().slice(left_block.start, left_end - left_block.start) != right.spans().slice(right_block.start, right_end - right_block.start))
left_skip = state.instruction_position + left_size;
else
break; break;
state.instruction_position += left_size; left_skip = left_end;
} }
dbgln_if(REGEX_DEBUG, "Skipping {}/{} bytecode entries from {}/{}", left_skip, 0, left.size(), right.size()); dbgln_if(REGEX_DEBUG, "Skipping {}/{} bytecode entries from {}/{}", left_skip, 0, left.size(), right.size());
if (left_skip) { if (left_skip > 0) {
target.extend(left.release_slice(0, left_skip)); target.extend(left.release_slice(left_blocks.first().start, left_skip));
right = right.release_slice(left_skip); right = right.release_slice(left_skip);
} }
auto left_size = left.size(); auto left_size = left.size();
target.empend(static_cast<ByteCodeValueType>(OpCodeId::ForkJump)); target.empend(static_cast<ByteCodeValueType>(OpCodeId::ForkJump));
target.empend(right.size() + (left_size ? 2 : 0)); // Jump to the _ALT label target.empend(right.size() + (left_size > 0 ? 2 : 0)); // Jump to the _ALT label
target.extend(move(right)); target.extend(move(right));