remove overlap-execution barriers and run for the entire block

qdp_project/src/Benchmark.cpp

@@ -54,46 +54,44 @@ void scan_b(size_t gid, size_t tid) {
 
 void scan_a(size_t gid, size_t tid) {
     THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)].clear();
-    THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)].resize(RUN_COUNT);
+    THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)].resize(1);
 
     LAUNCH_.wait();
 
-    for (size_t i = 0; i < RUN_COUNT; i++) {
-        THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_BEGIN] = std::chrono::steady_clock::now();
+    THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_BEGIN] = std::chrono::steady_clock::now();
 
+    for (size_t i = 0; i < RUN_COUNT; i++) {
         const size_t chunk_index = get_chunk_index(gid,  i);
         uint64_t* chunk_ptr = get_chunk<TC_SCANA>(DATA_A_, chunk_index, tid);
         uint16_t* mask_ptr = get_mask<TC_SCANA>(MASK_A_, chunk_index, tid);
 
         filter::apply_same(mask_ptr, nullptr, chunk_ptr, CMP_A, SUBCHUNK_SIZE_B_SCANA);
-
-        THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_WAIT] = std::chrono::steady_clock::now();
-
-        BARRIERS_[gid]->arrive_and_wait();
-
-        THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_END] = std::chrono::steady_clock::now();
     }
 
+    THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_WAIT] = std::chrono::steady_clock::now();
+
     BARRIERS_[gid]->arrive_and_drop();
+
+    THREAD_TIMING_[SCANA_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_END] = std::chrono::steady_clock::now();
 }
 
 void aggr_j(size_t gid, size_t tid) {
     CACHE_HITS_[UniqueIndex(gid,tid)] = 0;
 
     THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)].clear();
-    THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)].resize(RUN_COUNT);
+    THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)].resize(1);
 
     __m512i aggregator = aggregation::OP::zero();
 
     LAUNCH_.wait();
 
-    for (size_t i = 0; i < RUN_COUNT; i++) {
-        THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_BEGIN] = std::chrono::steady_clock::now();
+    THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_BEGIN] = std::chrono::steady_clock::now();
 
-        BARRIERS_[gid]->arrive_and_wait();
+    BARRIERS_[gid]->arrive_and_drop();
 
-        THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_WAIT] = std::chrono::steady_clock::now();
+    THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_WAIT] = std::chrono::steady_clock::now();
 
+    for (size_t i = 0; i < RUN_COUNT; i++) {
         const size_t chunk_index = get_chunk_index(gid, i);
         uint64_t* chunk_ptr = get_chunk<TC_AGGRJ>(DATA_B_, chunk_index, tid);
         uint16_t* mask_ptr_a = get_mask<TC_AGGRJ>(MASK_A_, chunk_index, tid);
@@ -124,12 +122,11 @@ void aggr_j(size_t gid, size_t tid) {
         uint64_t tmp = _mm512_reduce_add_epi64(aggregator);
         aggregator = aggregation::apply_masked(aggregator, data_ptr, mask_ptr_a, SUBCHUNK_SIZE_B_AGGRJ);
 
-        THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][i][TIME_STAMP_END] = std::chrono::steady_clock::now();
     }
 
     aggregation::happly(&DATA_DST_[UniqueIndex(gid,tid)], aggregator);
 
-    BARRIERS_[gid]->arrive_and_drop();
+    THREAD_TIMING_[AGGRJ_TIMING_INDEX][UniqueIndex(gid,tid)][0][TIME_STAMP_END] = std::chrono::steady_clock::now();
 }
 
 int main() {
@@ -162,6 +159,7 @@ int main() {
         // which is configured for xeonmax with smart assignment
         uint64_t cache_flags = 0;
         cache_flags |= dsacache::FLAG_WAIT_WEAK;
+        cache_flags |= dsacache::FLAG_HANDLE_PF;
         CACHE_.SetFlags(cache_flags);
         CACHE_.Init(CachePlacementPolicy, CopyMethodPolicy);
     }