Deduplicate GPU and CPU seed outputs

src/common.h

@@ -5,6 +5,7 @@
 #include <atomic>
 #include <thread>
 #include <array>
+#include <unordered_set>
 
 #ifndef OMISSION_LARGE_BIOMES
 #define OMISSION_LARGE_BIOMES 0
@@ -42,13 +43,28 @@ struct CpuOutput {
     int32_t score;
 };
 
+inline bool operator==(const CpuOutput &a, const CpuOutput &b) {
+    return a.seed == b.seed && a.x == b.x && a.z == b.z && a.score == b.score;
+}
+
+struct CpuOutputHash {
+    size_t operator()(const CpuOutput &output) const {
+        size_t h = std::hash<uint64_t>{}(output.seed);
+        h ^= std::hash<int32_t>{}(output.x) + 0x9e3779b97f4a7c15ULL + (h << 6) + (h >> 2);
+        h ^= std::hash<int32_t>{}(output.z) + 0x9e3779b97f4a7c15ULL + (h << 6) + (h >> 2);
+        h ^= std::hash<int32_t>{}(output.score) + 0x9e3779b97f4a7c15ULL + (h << 6) + (h >> 2);
+        return h;
+    }
+};
+
 struct GpuOutputs {
     std::queue<GpuOutput> queue;
     std::mutex mutex;
 };
 
 struct CpuOutputs {
     std::queue<CpuOutput> queue;
+    std::unordered_set<CpuOutput, CpuOutputHash> seen;
     std::mutex mutex;
 };
 

src/cpu.cpp

@@ -66,7 +66,9 @@ void CpuThread::run() {
 
         {
             std::lock_guard lock(outputs.mutex);
-            outputs.queue.push(output.value());
+            if (outputs.seen.insert(output.value()).second) {
+                outputs.queue.push(output.value());
+            }
         }
     }
 

src/gpu.cu

@@ -1844,6 +1844,15 @@ void GpuThread::run() {
       h_buffer.resize(len);
       TRY_CUDA(cudaMemcpy(h_buffer.data(), final_outputs.data, sizeof(*h_buffer.data()) * len, cudaMemcpyDeviceToHost));
 
+      std::sort(h_buffer.begin(), h_buffer.end(), [](const SeedPos &a, const SeedPos &b) {
+        if (a.seed_index != b.seed_index) return a.seed_index < b.seed_index;
+        if (a.x != b.x) return a.x < b.x;
+        return a.z < b.z;
+      });
+      h_buffer.erase(std::unique(h_buffer.begin(), h_buffer.end(), [](const SeedPos &a, const SeedPos &b) {
+        return a.seed_index == b.seed_index && a.x == b.x && a.z == b.z;
+      }), h_buffer.end());
+
       {
         std::lock_guard lock(outputs.mutex);
         for (const auto &result : h_buffer) {