- Introduced a detailed comparison document highlighting the functional equivalence between ZScream (C#) and YAZE (C++) overworld loading logic. - Verified key areas such as tile loading, expansion detection, map decompression, and coordinate calculations, confirming consistent behavior across both implementations. - Documented differences and improvements in YAZE, including enhanced error handling and memory management. - Provided validation results from integration tests ensuring data integrity and compatibility with existing ROMs.
138 lines
4.5 KiB
Markdown
138 lines
4.5 KiB
Markdown
# DungeonEditor Parallel Optimization Implementation
|
|
|
|
## 🚀 **Parallelization Strategy Implemented**
|
|
|
|
### **Problem Identified**
|
|
- **DungeonEditor::LoadAllRooms**: **17,966ms (17.97 seconds)** - 99.9% of loading time
|
|
- Loading **296 rooms** sequentially, each involving complex operations
|
|
- Perfect candidate for parallelization due to independent room processing
|
|
|
|
### **Solution: Multi-Threaded Room Loading**
|
|
|
|
#### **Key Optimizations**
|
|
|
|
1. **Parallel Room Processing**
|
|
```cpp
|
|
// Load 296 rooms using up to 8 threads
|
|
const int max_concurrency = std::min(8, std::thread::hardware_concurrency());
|
|
const int rooms_per_thread = (296 + max_concurrency - 1) / max_concurrency;
|
|
```
|
|
|
|
2. **Thread-Safe Result Collection**
|
|
```cpp
|
|
std::mutex results_mutex;
|
|
std::vector<std::pair<int, zelda3::RoomSize>> room_size_results;
|
|
std::vector<std::pair<int, ImVec4>> room_palette_results;
|
|
```
|
|
|
|
3. **Optimized Thread Distribution**
|
|
- **8 threads maximum** (reasonable limit for room loading)
|
|
- **~37 rooms per thread** (296 ÷ 8 = 37 rooms per thread)
|
|
- **Hardware concurrency aware** (adapts to available CPU cores)
|
|
|
|
#### **Parallel Processing Flow**
|
|
|
|
```cpp
|
|
// Each thread processes a batch of rooms
|
|
for (int i = start_room; i < end_room; ++i) {
|
|
// 1. Load room data (expensive operation)
|
|
rooms[i] = zelda3::LoadRoomFromRom(rom_, i);
|
|
|
|
// 2. Calculate room size
|
|
auto room_size = zelda3::CalculateRoomSize(rom_, i);
|
|
|
|
// 3. Load room objects
|
|
rooms[i].LoadObjects();
|
|
|
|
// 4. Process palette (thread-safe collection)
|
|
// ... palette processing ...
|
|
}
|
|
```
|
|
|
|
#### **Thread Safety Features**
|
|
|
|
1. **Mutex Protection**: `std::mutex results_mutex` protects shared data structures
|
|
2. **Lock Guards**: `std::lock_guard<std::mutex>` ensures thread-safe result collection
|
|
3. **Independent Processing**: Each thread works on different room ranges
|
|
4. **Synchronized Results**: Results collected and sorted on main thread
|
|
|
|
### **Expected Performance Impact**
|
|
|
|
#### **Theoretical Speedup**
|
|
- **8x faster** with 8 threads (ideal case)
|
|
- **Realistic expectation**: **4-6x speedup** due to:
|
|
- Thread creation overhead
|
|
- Mutex contention
|
|
- Memory bandwidth limitations
|
|
- Cache coherency issues
|
|
|
|
#### **Expected Results**
|
|
- **Before**: 17,966ms (17.97 seconds)
|
|
- **After**: **2,000-4,500ms (2-4.5 seconds)**
|
|
- **Total Loading Time**: **2.5-5 seconds** (down from 18.6 seconds)
|
|
- **Overall Improvement**: **70-85% reduction** in loading time
|
|
|
|
### **Technical Implementation Details**
|
|
|
|
#### **Thread Management**
|
|
```cpp
|
|
std::vector<std::future<absl::Status>> futures;
|
|
|
|
for (int thread_id = 0; thread_id < max_concurrency; ++thread_id) {
|
|
auto task = [this, &rooms, thread_id, rooms_per_thread, ...]() -> absl::Status {
|
|
// Process room batch
|
|
return absl::OkStatus();
|
|
};
|
|
|
|
futures.emplace_back(std::async(std::launch::async, task));
|
|
}
|
|
|
|
// Wait for all threads to complete
|
|
for (auto& future : futures) {
|
|
RETURN_IF_ERROR(future.get());
|
|
}
|
|
```
|
|
|
|
#### **Result Processing**
|
|
```cpp
|
|
// Sort results by room ID for consistent ordering
|
|
std::sort(room_size_results.begin(), room_size_results.end(),
|
|
[](const auto& a, const auto& b) { return a.first < b.first; });
|
|
|
|
// Process collected results on main thread
|
|
for (const auto& [room_id, room_size] : room_size_results) {
|
|
room_size_pointers_.push_back(room_size.room_size_pointer);
|
|
// ... process results ...
|
|
}
|
|
```
|
|
|
|
### **Monitoring and Validation**
|
|
|
|
#### **Performance Timing Added**
|
|
- **DungeonRoomLoader::PostProcessResults**: Measures result processing time
|
|
- **Thread creation overhead**: Minimal compared to room loading time
|
|
- **Result collection time**: Expected to be <100ms
|
|
|
|
#### **Logging and Debugging**
|
|
```cpp
|
|
util::logf("Loading %d dungeon rooms using %d threads (%d rooms per thread)",
|
|
kTotalRooms, max_concurrency, rooms_per_thread);
|
|
```
|
|
|
|
### **Benefits of This Approach**
|
|
|
|
1. **Massive Performance Gain**: 70-85% reduction in loading time
|
|
2. **Scalable**: Automatically adapts to available CPU cores
|
|
3. **Thread-Safe**: Proper synchronization prevents data corruption
|
|
4. **Maintainable**: Clean separation of parallel processing and result collection
|
|
5. **Robust**: Error handling per thread with proper status propagation
|
|
|
|
### **Next Steps**
|
|
|
|
1. **Test Performance**: Run application and measure actual speedup
|
|
2. **Validate Results**: Ensure room data integrity is maintained
|
|
3. **Fine-tune**: Adjust thread count if needed based on results
|
|
4. **Monitor**: Watch for any threading issues or performance regressions
|
|
|
|
This parallel optimization should transform YAZE from a slow-loading application to a lightning-fast ROM editor!
|