From ba6d5d00d97ba11ce4a61a2c98cf8b75131d8aa8 Mon Sep 17 00:00:00 2001
From: Alexey Milovidov <milovidov@yandex-team.ru>
Date: Thu, 30 Jul 2020 02:10:11 +0300
Subject: [PATCH] Apply one of the variants

---
 src/AggregateFunctions/IAggregateFunction.h | 57 ++++++++++++++-------
 1 file changed, 39 insertions(+), 18 deletions(-)

diff --git a/src/AggregateFunctions/IAggregateFunction.h b/src/AggregateFunctions/IAggregateFunction.h
index a0876d457b8..251cf521d1a 100644
--- a/src/AggregateFunctions/IAggregateFunction.h
+++ b/src/AggregateFunctions/IAggregateFunction.h
@@ -222,24 +222,6 @@ public:
             static_cast<const Derived *>(this)->add(places[i] + place_offset, columns, i, arena);
     }
 
-    void addBatchLookupTable8(
-        size_t batch_size,
-        AggregateDataPtr * places,
-        size_t place_offset,
-        std::function<void(AggregateDataPtr &)> init,
-        const UInt8 * key,
-        const IColumn ** columns,
-        Arena * arena) const override
-    {
-        for (size_t i = 0; i < batch_size; ++i)
-        {
-            AggregateDataPtr & place = places[key[i]];
-            if (unlikely(!place))
-                init(place);
-            static_cast<const Derived *>(this)->add(place + place_offset, columns, i, arena);
-        }
-    }
-
     void addBatchSinglePlace(size_t batch_size, AggregateDataPtr place, const IColumn ** columns, Arena * arena) const override
     {
         for (size_t i = 0; i < batch_size; ++i)
@@ -274,6 +256,45 @@ public:
             current_offset = next_offset;
         }
     }
+
+    void addBatchLookupTable8(
+        size_t batch_size,
+        AggregateDataPtr * map,
+        size_t place_offset,
+        std::function<void(AggregateDataPtr &)> init,
+        const UInt8 * key,
+        const IColumn ** columns,
+        Arena * arena) const override
+    {
+        static constexpr size_t UNROLL_COUNT = 8;
+
+        size_t i = 0;
+
+        size_t batch_size_unrolled = batch_size / UNROLL_COUNT * UNROLL_COUNT;
+        for (; i < batch_size_unrolled; i += UNROLL_COUNT)
+        {
+            AggregateDataPtr places[UNROLL_COUNT];
+            for (size_t j = 0; j < UNROLL_COUNT; ++j)
+            {
+                AggregateDataPtr & place = map[key[i + j]];
+                if (unlikely(!place))
+                    init(place);
+
+                places[j] = place;
+            }
+
+            for (size_t j = 0; j < UNROLL_COUNT; ++j)
+                static_cast<const Derived *>(this)->add(places[j] + place_offset, columns, i + j, arena);
+        }
+
+        for (; i < batch_size; ++i)
+        {
+            AggregateDataPtr & place = map[key[i]];
+            if (unlikely(!place))
+                init(place);
+            static_cast<const Derived *>(this)->add(place + place_offset, columns, i, arena);
+        }
+    }
 };