enhance: optimize term expr performance (#45490)

issue: https://github.com/milvus-io/milvus/issues/45641
pr: https://github.com/milvus-io/milvus/pull/45491

---------

Signed-off-by: sunby <sunbingyi1992@gmail.com>

internal/core/src/common/init_c.cpp

@@ -139,7 +139,6 @@ InitDefaultDeleteDumpBatchSize(int32_t val) {
         val);
 }
 
-
 void
 InitTrace(CTraceConfig* config) {
     auto traceConfig = milvus::tracer::TraceConfig{config->exporter,

internal/core/src/exec/expression/Element.h

@@ -198,7 +198,7 @@ class FlatVectorElement : public MultiElement {
     In(const ValueType& value) const override {
         if (std::holds_alternative<T>(value)) {
             for (const auto& v : values_) {
-                if (v == value)
+                if (v == std::get<T>(value))
                     return true;
             }
         }

internal/core/src/exec/expression/Expr.h

@@ -908,7 +908,7 @@ class SegmentExpr : public Expr {
 
     template <typename T, typename FUNC, typename... ValTypes>
     VectorPtr
-    ProcessIndexChunks(FUNC func, ValTypes... values) {
+    ProcessIndexChunks(FUNC func, const ValTypes&... values) {
         typedef std::
             conditional_t<std::is_same_v<T, std::string_view>, std::string, T>
                 IndexInnerType;

internal/core/src/exec/expression/GISFunctionFilterExpr.h

@@ -66,7 +66,6 @@ class PhyGISFunctionFilterExpr : public SegmentExpr {
         }
     }
 
-
  private:
     VectorPtr
     EvalForIndexSegment();

internal/core/src/exec/expression/TermExpr.cpp

@@ -810,6 +810,7 @@ PhyTermFilterExpr::ExecVisitorImplForIndex() {
         return nullptr;
     }
 
+    if (!arg_inited_) {
         std::vector<IndexInnerType> vals;
         for (auto& val : expr_->vals_) {
             if constexpr (std::is_same_v<T, double>) {
@@ -822,17 +823,23 @@ PhyTermFilterExpr::ExecVisitorImplForIndex() {
 
             // Generic overflow handling for all types
             bool overflowed = false;
-        auto converted_val = GetValueFromProtoWithOverflow<T>(val, overflowed);
+            auto converted_val =
+                GetValueFromProtoWithOverflow<T>(val, overflowed);
             if (!overflowed) {
                 vals.emplace_back(converted_val);
             }
         }
+        arg_set_ = std::make_shared<FlatVectorElement<IndexInnerType>>(vals);
+        arg_inited_ = true;
+    }
     auto execute_sub_batch = [](Index* index_ptr,
                                 const std::vector<IndexInnerType>& vals) {
         TermIndexFunc<T> func;
         return func(index_ptr, vals.size(), vals.data());
     };
-    auto res = ProcessIndexChunks<T>(execute_sub_batch, vals);
+    auto args =
+        std::dynamic_pointer_cast<FlatVectorElement<IndexInnerType>>(arg_set_);
+    auto res = ProcessIndexChunks<T>(execute_sub_batch, args->values_);
     AssertInfo(res->size() == real_batch_size,
                "internal error: expr processed rows {} not equal "
                "expect batch size {}",

internal/core/src/segcore/DeletedRecord.h

@@ -214,7 +214,8 @@ class DeletedRecord {
         SortedDeleteList::Accessor accessor(deleted_lists_);
         int total_size = accessor.size();
 
-        while (total_size - dumped_entry_count_.load() > DELETE_DUMP_BATCH_SIZE) {
+        while (total_size - dumped_entry_count_.load() >
+               DELETE_DUMP_BATCH_SIZE) {
             int32_t bitsize = 0;
             if constexpr (is_sealed) {
                 bitsize = sealed_row_count_;
@@ -232,11 +233,14 @@ class DeletedRecord {
                                              snapshots_.back().second.size());
             }
 
-            while (total_size - dumped_entry_count_.load() > DELETE_DUMP_BATCH_SIZE &&
+            while (total_size - dumped_entry_count_.load() >
+                       DELETE_DUMP_BATCH_SIZE &&
                    it != accessor.end()) {
                 Timestamp dump_ts = 0;
 
-                for (auto size = 0; size < DELETE_DUMP_BATCH_SIZE && it != accessor.end(); ++it, ++size) {
+                for (auto size = 0;
+                     size < DELETE_DUMP_BATCH_SIZE && it != accessor.end();
+                     ++it, ++size) {
                     bitmap.set(it->second);
                     dump_ts = it->first;
                 }

internal/core/src/storage/FileWriter.h

@@ -294,8 +294,7 @@ class FileWriter {
     // for global configuration
     static WriteMode
         mode_;  // The write mode, which can be 'buffered' (default) or 'direct'.
-    static size_t
-        buffer_size_;
+    static size_t buffer_size_;
 
     // for rate limiter
     io::Priority priority_;

internal/core/unittest/test_utils/DataGen.h

@@ -343,8 +343,7 @@ GenerateRandomSparseFloatVector(size_t rows,
     return tensor;
 }
 
-inline std::string
-generateRandomPoint() {
+inline std::string generateRandomPoint() {
     return "POINT(" +
            std::to_string(static_cast<double>(rand()) / RAND_MAX * 360.0 -
                           180.0) +