push changes to benchmark

qdp_project/src/Benchmark.cpp

@@ -24,16 +24,16 @@
 ////////////////////////////////
 /// BENCHMARK SETUP
 
-constexpr size_t WL_SIZE_B = 2_GiB;
+constexpr size_t WL_SIZE_B = 4_GiB;
 constexpr uint32_t WARMUP_ITERATION_COUNT = 5;
 constexpr uint32_t ITERATION_COUNT = 5;
 
 #ifdef MODE_PREFETCH
-constexpr size_t CHUNK_SIZE_B = 128_MiB;
-constexpr uint32_t GROUP_COUNT = ;
+constexpr size_t CHUNK_SIZE_B = 64_MiB;
+constexpr uint32_t GROUP_COUNT = 8;
 constexpr uint32_t TC_SCANA = 1;
-constexpr uint32_t TC_SCANB = 1;
-constexpr uint32_t TC_AGGRJ = 1;
+constexpr uint32_t TC_SCANB = 8;
+constexpr uint32_t TC_AGGRJ = 8;
 constexpr bool PERFORM_CACHING = true;
 constexpr bool DATA_IN_HBM = false;
 constexpr char MODE_STRING[] = "prefetch";
@@ -71,6 +71,8 @@ constexpr size_t RUN_COUNT = CHUNK_COUNT / GROUP_COUNT;
 constexpr size_t MASK_ELEMENT_SIZE = 16;
 constexpr size_t MASK_STEP_SIZE = CHUNK_SIZE_ELEMENTS / MASK_ELEMENT_SIZE;
 
+static_assert(RUN_COUNT > 0);
+
 using filter = Filter<uint64_t, LT, load_mode::Stream, false>;
 using aggregation = Aggregation<uint64_t, Sum, load_mode::Stream>;
 
@@ -111,24 +113,24 @@ inline uint16_t* get_mask(uint16_t* base, const size_t chunk_index) {
 }
 
 void scan_b(size_t gid, size_t tid) {
-    using namespace std::chrono_literals;
+    constexpr size_t split = RUN_COUNT / TC_SCANB;
+    const size_t start = tid * split;
+    const size_t end = start + split;
 
     LAUNCH_.wait();
 
     if constexpr (PERFORM_CACHING) {
-        std::unique_ptr<dsacache::CacheData> data;
-
-        for (size_t i = 0; i < RUN_COUNT; i++) {
+        for (size_t i = start; i < end; i++) {
             const size_t chunk_index = get_chunk_index<TC_AGGRJ>(gid, tid, i);
             uint64_t* chunk_ptr = get_chunk(DATA_B_, chunk_index);
 
-            data = CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_AGGRJ);
-    	    std::this_thread::sleep_for(100ms);
+            const auto data = CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_AGGRJ);
+	    sched_yield();
+
 	    data->WaitOnCompletion();
         }
     }
 
-
     BARRIERS_[gid]->arrive_and_drop();
 }