From 52132522a31cd6f68b380dbd88587edb8c9d991e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Constantin=20F=C3=BCrst?= <c@fuersten.info>
Date: Tue, 23 Jan 2024 22:09:58 +0100
Subject: [PATCH] fix scanb not working with less than aggrj threads

---
 qdp_project/src/Benchmark.cpp | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

diff --git a/qdp_project/src/Benchmark.cpp b/qdp_project/src/Benchmark.cpp
index 48f03ee..f68bbd5 100644
--- a/qdp_project/src/Benchmark.cpp
+++ b/qdp_project/src/Benchmark.cpp
@@ -28,10 +28,10 @@ constexpr uint64_t CMP_A = 50;
 constexpr uint32_t WARMUP_ITERATION_COUNT = 0;
 constexpr uint32_t ITERATION_COUNT = 2;
 constexpr uint32_t GROUP_COUNT = 2;
-constexpr uint32_t TC_SCANA = 1;
 constexpr uint32_t TC_SCANB = 1;
+constexpr uint32_t TC_SCANA = 1;
 constexpr uint32_t TC_AGGRJ = 1;
-constexpr bool PERFORM_CACHING = false;
+constexpr bool PERFORM_CACHING = true;
 constexpr bool DATA_IN_HBM = false;
 constexpr char MODE_STRING[] = "DramBase";
 
@@ -59,19 +59,19 @@ uint64_t* DATA_DST_;
 void scan_b(size_t gid, size_t tid) {
     LAUNCH_.wait();
 
-    uint32_t runs = CHUNK_COUNT / GROUP_COUNT + (CHUNK_COUNT % GROUP_COUNT > gid);
-
-    std::unique_ptr<dsacache::CacheData> data;
+    if constexpr (PERFORM_CACHING) {
+        uint32_t runs = CHUNK_COUNT / GROUP_COUNT + (CHUNK_COUNT % GROUP_COUNT > gid);
 
-    for(uint32_t i = 0; i < runs; ++i) {
-        // calculate pointers
-        size_t chunk_id = gid + GROUP_COUNT * i;
-        uint64_t*  chunk_ptr = get_sub_chunk_ptr(DATA_B_, chunk_id, CHUNK_SIZE_ELEMENTS, tid, TC_SCANB);
+        std::unique_ptr<dsacache::CacheData> data;
 
-        if constexpr (PERFORM_CACHING) {
-            data = CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_SCANB);
-            data->WaitOnCompletion();
+        for(uint32_t i = 0; i < runs; ++i) {
+            // calculate pointers
+            size_t chunk_id = gid + GROUP_COUNT * i;
+            uint64_t*  chunk_ptr = get_sub_chunk_ptr(DATA_B_, chunk_id, CHUNK_SIZE_ELEMENTS, tid, TC_AGGRJ);
+            data = CACHE_.Access(reinterpret_cast<uint8_t *>(chunk_ptr), CHUNK_SIZE_B / TC_AGGRJ);
         }
+
+        data->WaitOnCompletion();
     }
 
     BARRIERS_[gid]->arrive_and_drop();
@@ -204,7 +204,7 @@ int main() {
         for(std::thread& t : copy_pool) { t.join(); }
         for(std::thread& t : agg_pool) { t.join(); }
 
-        Aggregation<uint64_t, Sum, load_mode::Aligned>::apply(DATA_DST_, DATA_DST_, sizeof(uint64_t) * TC_AGGRJ * GROUP_COUNT);
+        Aggregation<uint64_t, Sum, load_mode::Aligned>::apply(DATA_nDST_, DATA_DST_, sizeof(uint64_t) * TC_AGGRJ * GROUP_COUNT);
 
         const auto time_end = std::chrono::steady_clock::now();