Deinline the coldest path too.

2 files changed, 13 insertions, 3 deletions

diff --git a/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.cpp b/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.cpp
index 331f01d4a82..b13bf4a10e2 100644
--- a/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.cpp
+++ b/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.cpp
@@ -70,15 +70,23 @@ MultiBitVector<Update>::updateLastValueCold(uint32_t docId) noexcept
     }
     const uint32_t index(BitWord::wordNum(docId));
     if (docId >= _lastMaxDocIdLimitRequireFetch) {
-        uint32_t baseIndex = index & ~(NumWordsInBatch - 1);
-        _update(_accel, baseIndex*sizeof(Word), _bvs, _lastWords);
-        _lastMaxDocIdLimitRequireFetch = (baseIndex + NumWordsInBatch) * BitWord::WordLen;
+        fetchChunk(index);
     }
     _lastValue = _lastWords[index % NumWordsInBatch];
     _lastMaxDocIdLimit = (index + 1) * BitWord::WordLen;
     return false;
 }
 
+template<typename Update>
+void
+MultiBitVector<Update>::fetchChunk(uint32_t index) noexcept
+{
+    uint32_t baseIndex = index & ~(NumWordsInBatch - 1);
+    _update(_accel, baseIndex*sizeof(Word), _bvs, _lastWords);
+    _lastMaxDocIdLimitRequireFetch = (baseIndex + NumWordsInBatch) * BitWord::WordLen;
+}
+
+
 
 
 template<typename Update>
diff --git a/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.h b/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.h
index 66598e28e3e..0d9e2c4f25f 100644
--- a/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.h
+++ b/searchlib/src/vespa/searchlib/queryeval/multibitvectoriterator.h
@@ -44,6 +44,8 @@ private:
         return false;
     }
     VESPA_DLL_LOCAL bool updateLastValueCold(uint32_t docId) noexcept __attribute__((noinline));
+    VESPA_DLL_LOCAL void fetchChunk(uint32_t docId) noexcept __attribute__((noinline));
+
     using IAccelrated = vespalib::hwaccelrated::IAccelrated;
 
     Update              _update;