Change bit_ size to fix TPCDS performance

oap-project · Nov 25, 2022 · 9ecb4e6 · 9ecb4e6
1 parent ab283cc
commit 9ecb4e6
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/velox/functions/sparksql/aggregates/BloomFilterAggAggregate.cpp b/velox/functions/sparksql/aggregates/BloomFilterAggAggregate.cpp
@@ -209,7 +209,8 @@ class BloomFilterAggAggregate : public exec::Aggregate {
     // velox BloomFilter bit_ size is bits::nextPowerOfTwo(capacity) / 4, and
     // spark bit_ size is Math.ceil(numBits / 64.0) so there is equal bit_ size
     // using numBits_ / 16
-    capacity_ = numBits_ / 16;
+    // but with TPCDS test, for velox BloomFilter, this value should be 64
+    capacity_ = numBits_ / 64;
   }
 
   static void