adapt to efficient wait in cache

qdp_project/src/Benchmark.cpp

@@ -30,10 +30,10 @@ constexpr uint32_t ITERATION_COUNT = 5;
 
 #ifdef MODE_PREFETCH
 constexpr size_t CHUNK_SIZE_B = 64_MiB;
-constexpr uint32_t GROUP_COUNT = 8;
+constexpr uint32_t GROUP_COUNT = 32;
 constexpr uint32_t TC_SCANA = 1;
-constexpr uint32_t TC_SCANB = 8;
-constexpr uint32_t TC_AGGRJ = 8;
+constexpr uint32_t TC_SCANB = 1;
+constexpr uint32_t TC_AGGRJ = 1;
 constexpr bool PERFORM_CACHING = true;
 constexpr bool DATA_IN_HBM = false;
 constexpr char MODE_STRING[] = "prefetch";
@@ -115,15 +115,17 @@ void scan_b(size_t gid, size_t tid) {
     LAUNCH_.wait();
 
     if constexpr (PERFORM_CACHING) {
+        std::vector<std::unique_ptr<dsacache::CacheData>> data;
+
         for (size_t i = start; i < end; i++) {
             const size_t chunk_index = get_chunk_index(gid, 0);
             uint64_t* chunk_ptr = get_chunk<TC_AGGRJ>(DATA_B_, chunk_index, i);
 
-            const auto data = CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_AGGRJ);
-
-	        sched_yield();
+            data.emplace_back(std::move(CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_AGGRJ)));
+        }
 
-	        data->WaitOnCompletion();
+        for (auto& e : data) {
+            e->WaitOnCompletion();
         }
     }