Merge pull request ClickHouse#67778 from ClickHouse/test_attr

Apply `preserve_most` attribute at some places in code

Author: rienath

base/base/defines.h

@@ -28,6 +28,12 @@
 #define NO_INLINE __attribute__((__noinline__))
 #define MAY_ALIAS __attribute__((__may_alias__))
 
+#if defined(__x86_64__) || defined(__aarch64__)
+#    define PRESERVE_MOST __attribute__((preserve_most))
+#else
+#    define PRESERVE_MOST
+#endif
+
 #if !defined(__x86_64__) && !defined(__aarch64__) && !defined(__PPC__) && !defined(__s390x__) && !(defined(__loongarch64)) && !(defined(__riscv) && (__riscv_xlen == 64))
 #    error "The only supported platforms are x86_64 and AArch64, PowerPC (work in progress), s390x (work in progress), loongarch64 (experimental) and RISC-V 64 (experimental)"
 #endif

src/Common/Arena.h

@@ -146,7 +146,7 @@ class Arena : private boost::noncopyable
     }
 
     /// Add next contiguous MemoryChunk of memory with size not less than specified.
-    void NO_INLINE addMemoryChunk(size_t min_size)
+    void PRESERVE_MOST NO_INLINE addMemoryChunk(size_t min_size)
     {
         size_t next_size = nextSize(min_size + pad_right);
         if (head.empty())

src/Common/HashTable/HashTable.h

@@ -502,7 +502,7 @@ class HashTable : private boost::noncopyable,
     }
 
     /// Increase the size of the buffer.
-    void resize(size_t for_num_elems = 0, size_t for_buf_size = 0)
+    void PRESERVE_MOST resize(size_t for_num_elems = 0, size_t for_buf_size = 0)
     {
 #ifdef DBMS_HASH_MAP_DEBUG_RESIZES
         Stopwatch watch;
@@ -962,10 +962,7 @@ class HashTable : private boost::noncopyable,
     }
 
 public:
-    void reserve(size_t num_elements)
-    {
-        resize(num_elements);
-    }
+    void PRESERVE_MOST reserve(size_t num_elements) { resize(num_elements); }
 
     /// Insert a value. In the case of any more complex values, it is better to use the `emplace` function.
     std::pair<LookupResult, bool> ALWAYS_INLINE insert(const value_type & x)

src/Common/PODArray.h

@@ -135,8 +135,8 @@ class PODArrayBase : private boost::noncopyable, private TAllocator    /// empty
         TAllocator::free(c_start - pad_left, allocated_bytes());
     }
 
-    template <typename ... TAllocatorParams>
-    void realloc(size_t bytes, TAllocatorParams &&... allocator_params)
+    template <typename... TAllocatorParams>
+    void PRESERVE_MOST realloc(size_t bytes, TAllocatorParams &&... allocator_params)
     {
         if (c_start == null)
         {
@@ -167,8 +167,8 @@ class PODArrayBase : private boost::noncopyable, private TAllocator    /// empty
         return (stack_threshold > 0) && (allocated_bytes() <= stack_threshold);
     }
 
-    template <typename ... TAllocatorParams>
-    void reserveForNextSize(TAllocatorParams &&... allocator_params)
+    template <typename... TAllocatorParams>
+    void PRESERVE_MOST reserveForNextSize(TAllocatorParams &&... allocator_params)
     {
         if (empty())
         {

src/Interpreters/AggregationCommon.h

@@ -174,7 +174,7 @@ static inline T ALWAYS_INLINE packFixed(
     static constexpr auto bitmap_size = std::tuple_size<KeysNullMap<T>>::value;
     static constexpr bool has_bitmap = bitmap_size > 0;
 
-    if (has_bitmap)
+    if constexpr (has_bitmap)
     {
         memcpy(bytes + offset, bitmap.data(), bitmap_size * sizeof(UInt8));
         offset += bitmap_size;

src/Interpreters/Aggregator.cpp

@@ -2078,7 +2078,7 @@ Aggregator::ConvertToBlockResVariant Aggregator::convertToBlockImplFinal(
     data.forEachValue(
         [&](const auto & key, auto & mapped)
         {
-            if (!out_cols.has_value())
+            if (unlikely(!out_cols.has_value()))
                 init_out_cols();
 
             const auto & key_sizes_ref = shuffled_key_sizes ? *shuffled_key_sizes : key_sizes;