MDEV-23162 Improve Protocol performance for numeric data

An alternative implementation (replacing the one based on repertoire).

This implementation makes Field send itself to Protocol_text using
data type specific Protocol methods rather than field->val_str()
followed by protocol_text->store_str().

As now Field sends itself in the same way to all protocol types
(e.g. Protocol_binary, Protocol_text, Protocol_local),
the method Field::send_binary() was renamed just to Field::send().

Note, this change introduces symmetry between Field and Item,
because Items also send themself using a single method Item::send(),
which is used for *all* protocol types.

Performance improvement is achieved by the fact that Protocol_text
implements these data type specific methods using store_numeric_string_aux()
rather than store_string_aux(). The conversion now happens only when
character_set_results is not ASCII compatible character sets
(e.g. UCS2, UTF16, UTF32).

In the old code (before any MDEV-23162 work, e.g. as of 10.5.4),
Protocol_text::store(Field*) used val_str() for all data types.
So the execution went through the character set conversion routines
even for numeric and temporal data types.

Benchmarking summary (see  details in MDEV-23478):

The new approach stably demonstrates additional improvement comparing
to the previous implementation (the smaller time - the better):

Original   - the commit before MDEV-23162
             be98036f

        1m9.336s
        1m9.290s
        1m9.300s

MDEV-23162 - the repertoire optimization

        1m6.101s
        1m5.988s
        1m6.264s

MDEV-23478 - this commit

        1m2.150s
        1m2.079s
        1m2.099s

sql/field.cc

@@ -1888,7 +1888,7 @@ void Field::copy_from_tmp(int row_offset)
 }
 
 
-bool Field::send_binary(Protocol *protocol)
+bool Field::send(Protocol *protocol)
 {
   char buff[MAX_FIELD_WIDTH];
   String tmp(buff,sizeof(buff),charset());
@@ -1897,6 +1897,18 @@ bool Field::send_binary(Protocol *protocol)
 }
 
 
+bool Field_num::send_numeric_zerofill_str(Protocol_text *protocol,
+                                          protocol_send_type_t send_type)
+{
+  DBUG_ASSERT(marked_for_read());
+  StringBuffer<MAX_FIELD_WIDTH> tmp(&my_charset_latin1);
+  val_str(&tmp);
+  return protocol->store_numeric_zerofill_str(tmp.ptr(),
+                                              tmp.length(),
+                                              send_type);
+}
+
+
 /**
    Check to see if field size is compatible with destination.
 
@@ -3855,11 +3867,16 @@ String *Field_tiny::val_str(String *val_buffer,
   return val_str_from_long(val_buffer, 5, -10, nr);
 }
 
-bool Field_tiny::send_binary(Protocol *protocol)
+bool Field_tiny::send(Protocol *protocol)
 {
-  return protocol->store_tiny((longlong) (int8) ptr[0]);
+  DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_TINY);
+  return protocol->store_tiny(Field_tiny::val_int());
 }
 
+
 int Field_tiny::cmp(const uchar *a_ptr, const uchar *b_ptr) const
 {
   signed char a,b;
@@ -4015,8 +4032,12 @@ String *Field_short::val_str(String *val_buffer,
 }
 
 
-bool Field_short::send_binary(Protocol *protocol)
+bool Field_short::send(Protocol *protocol)
 {
+  DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_SHORT);
   return protocol->store_short(Field_short::val_int());
 }
 
@@ -4198,9 +4219,12 @@ String *Field_int::val_str_from_long(String *val_buffer,
 }
 
 
-bool Field_medium::send_binary(Protocol *protocol)
+bool Field_medium::send(Protocol *protocol)
 {
   DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_LONG);
   return protocol->store_long(Field_medium::val_int());
 }
 
@@ -4369,12 +4393,16 @@ String *Field_long::val_str(String *val_buffer,
 }
 
 
-bool Field_long::send_binary(Protocol *protocol)
+bool Field_long::send(Protocol *protocol)
 {
   DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_LONG);
   return protocol->store_long(Field_long::val_int());
 }
 
+
 int Field_long::cmp(const uchar *a_ptr, const uchar *b_ptr) const
 {
   int32 a,b;
@@ -4506,9 +4534,12 @@ String *Field_longlong::val_str(String *val_buffer,
 }
 
 
-bool Field_longlong::send_binary(Protocol *protocol)
+bool Field_longlong::send(Protocol *protocol)
 {
   DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_LONGLONG);
   return protocol->store_longlong(Field_longlong::val_int(), unsigned_flag);
 }
 
@@ -4686,9 +4717,12 @@ void Field_float::sort_string(uchar *to,uint length __attribute__((unused)))
 }
 
 
-bool Field_float::send_binary(Protocol *protocol)
+bool Field_float::send(Protocol *protocol)
 {
   DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_FLOAT);
   return protocol->store_float((float) Field_float::val_real(), dec);
 }
 
@@ -4980,8 +5014,12 @@ String *Field_double::val_str(String *val_buffer,
   return val_buffer;
 }
 
-bool Field_double::send_binary(Protocol *protocol)
+bool Field_double::send(Protocol *protocol)
 {
+  DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if (unlikely(zerofill) && (txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_DOUBLE);
   return protocol->store_double(Field_double::val_real(), dec);
 }
 
@@ -5378,7 +5416,7 @@ bool Field_timestamp::get_date(MYSQL_TIME *ltime, date_mode_t fuzzydate)
 }
 
 
-bool Field_timestamp0::send_binary(Protocol *protocol)
+bool Field_timestamp0::send(Protocol *protocol)
 {
   MYSQL_TIME ltime;
   Field_timestamp0::get_date(&ltime, date_mode_t(0));
@@ -5538,7 +5576,7 @@ int Field_timestamp_with_dec::set_time()
   return 0;
 }
 
-bool Field_timestamp_with_dec::send_binary(Protocol *protocol)
+bool Field_timestamp_with_dec::send(Protocol *protocol)
 {
   MYSQL_TIME ltime;
   Field_timestamp::get_date(&ltime, date_mode_t(0));
@@ -6047,7 +6085,7 @@ bool Field_time0::get_date(MYSQL_TIME *ltime, date_mode_t fuzzydate)
 }
 
 
-bool Field_time::send_binary(Protocol *protocol)
+bool Field_time::send(Protocol *protocol)
 {
   MYSQL_TIME ltime;
   get_date(&ltime, Time::Options(TIME_TIME_ONLY, get_thd()));
@@ -6369,9 +6407,12 @@ int Field_year::store_time_dec(const MYSQL_TIME *ltime, uint dec_arg)
   return 0;
 }
 
-bool Field_year::send_binary(Protocol *protocol)
+bool Field_year::send(Protocol *protocol)
 {
   DBUG_ASSERT(marked_for_read());
+  Protocol_text *txt;
+  if ((txt= dynamic_cast<Protocol_text*>(protocol)))
+    return send_numeric_zerofill_str(txt, PROTOCOL_SEND_SHORT);
   ulonglong tmp= Field_year::val_int();
   return protocol->store_short(tmp);
 }
@@ -6506,7 +6547,7 @@ void Field_date::store_TIME(const MYSQL_TIME *ltime)
   int4store(ptr,tmp);
 }
 
-bool Field_date::send_binary(Protocol *protocol)
+bool Field_date::send(Protocol *protocol)
 {
   longlong tmp= Field_date::val_int();
   MYSQL_TIME tm;
@@ -6600,7 +6641,7 @@ void Field_newdate::store_TIME(const MYSQL_TIME *ltime)
 }
 
 
-bool Field_newdate::send_binary(Protocol *protocol)
+bool Field_newdate::send(Protocol *protocol)
 {
   MYSQL_TIME tm;
   Field_newdate::get_date(&tm, date_mode_t(0));
@@ -6772,7 +6813,7 @@ Field_datetime::conversion_depends_on_sql_mode(THD *thd, Item *expr) const
 }
 
 
-bool Field_datetime0::send_binary(Protocol *protocol)
+bool Field_datetime0::send(Protocol *protocol)
 {
   MYSQL_TIME tm;
   Field_datetime0::get_date(&tm, date_mode_t(0));
@@ -6900,7 +6941,7 @@ void Field_datetime_hires::store_TIME(const MYSQL_TIME *ltime)
   store_bigendian(packed, ptr, Field_datetime_hires::pack_length());
 }
 
-bool Field_datetime_with_dec::send_binary(Protocol *protocol)
+bool Field_datetime_with_dec::send(Protocol *protocol)
 {
   MYSQL_TIME ltime;
   get_date(&ltime, date_mode_t(0));
@@ -7109,6 +7150,23 @@ void Field_longstr::make_send_field(Send_field *field)
   }
 }
 
+
+/*
+  An optimized version that uses less stack than Field::send().
+*/
+bool Field_longstr::send(Protocol *protocol)
+{
+  String tmp;
+  val_str(&tmp, &tmp);
+  /*
+    Ensure this function is only used with classes that do not allocate
+    memory in val_str()
+  */
+  DBUG_ASSERT(tmp.alloced_length() == 0);
+  return protocol->store(tmp.ptr(), tmp.length(), tmp.charset());
+}
+
+
 	/* Copy a string and fill with space */
 
 int Field_string::store(const char *from, size_t length,CHARSET_INFO *cs)
@@ -7698,6 +7756,17 @@ my_decimal *Field_varstring::val_decimal(my_decimal *decimal_value)
 }
 
 
+/*
+  An optimized version that uses less stack and less temporary
+  variable initialization than Field_longstr::send()
+*/
+bool Field_varstring::send(Protocol *protocol)
+{
+  return protocol->store((const char *) get_data(), get_length(),
+                         field_charset());
+}
+
+
 #ifdef HAVE_valgrind
 void Field_varstring::mark_unused_memory_as_defined()
 {

sql/field.h

@@ -37,6 +37,7 @@
 class Send_field;
 class Copy_field;
 class Protocol;
+class Protocol_text;
 class Create_field;
 class Relay_log_info;
 class Field;
@@ -1576,7 +1577,7 @@ class Field: public Value_source
     ptr= old_ptr;
     return str;
   }
-  virtual bool send_binary(Protocol *protocol);
+  virtual bool send(Protocol *protocol);
 
   virtual uchar *pack(uchar *to, const uchar *from, uint max_length);
   /**
@@ -2005,6 +2006,9 @@ class Field_num :public Field {
     return (flags & UNSIGNED_FLAG) ? Binlog_type_info::SIGN_UNSIGNED :
                                      Binlog_type_info::SIGN_SIGNED;
   }
+  bool send_numeric_zerofill_str(Protocol_text *protocol,
+                                 protocol_send_type_t send_type);
+
 public:
   const uint8 dec;
   bool zerofill,unsigned_flag;	// Purify cannot handle bit fields
@@ -2194,6 +2198,7 @@ class Field_longstr :public Field_str
   int store_decimal(const my_decimal *d) override;
   uint32 max_data_length() const override;
   void make_send_field(Send_field *) override;
+  bool send(Protocol *protocol) override;
 
   bool is_varchar_and_in_write_set() const override
   {
@@ -2520,7 +2525,7 @@ class Field_tiny :public Field_int
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 1; }
@@ -2583,7 +2588,7 @@ class Field_short final :public Field_int
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 2; }
@@ -2630,7 +2635,7 @@ class Field_medium final :public Field_int
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 3; }
@@ -2681,7 +2686,7 @@ class Field_long final :public Field_int
   int reset() override { ptr[0]=ptr[1]=ptr[2]=ptr[3]=0; return 0; }
   double val_real() override;
   longlong val_int() override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   String *val_str(String *, String *) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
@@ -2743,7 +2748,7 @@ class Field_longlong :public Field_int
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 8; }
@@ -2843,7 +2848,7 @@ class Field_float final :public Field_real {
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff, uint length) override;
   uint32 pack_length() const override { return sizeof(float); }
@@ -2907,7 +2912,7 @@ class Field_double :public Field_real {
   longlong val_int() override final { return val_int_from_real(false); }
   ulonglong val_uint() override final { return (ulonglong) val_int_from_real(true); }
   String *val_str(String *, String *) override final;
-  bool send_binary(Protocol *protocol) override final;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override final;
   void sort_string(uchar *buff, uint length) override final;
   uint32 pack_length() const override final { return sizeof(double); }
@@ -3210,7 +3215,7 @@ class Field_timestamp0 :public Field_timestamp
   {
     return (double) Field_timestamp0::val_int();
   }
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 4; }
@@ -3265,7 +3270,7 @@ class Field_timestamp_with_dec :public Field_timestamp {
     DBUG_ASSERT(length == pack_length());
     memcpy(to, ptr, length);
   }
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   double val_real() override;
   my_decimal* val_decimal(my_decimal*) override;
   int set_time() override;
@@ -3403,7 +3408,7 @@ class Field_year final :public Field_tiny {
   longlong val_int() override;
   String *val_str(String *, String *) override;
   bool get_date(MYSQL_TIME *ltime, date_mode_t fuzzydate) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   Information_schema_numeric_attributes
     information_schema_numeric_attributes() const override
   {
@@ -3462,7 +3467,7 @@ class Field_date final :public Field_date_common
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 4; }
@@ -3501,7 +3506,7 @@ class Field_newdate final :public Field_date_common
   double val_real() override;
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 3; }
@@ -3561,7 +3566,7 @@ class Field_time :public Field_temporal {
   int store(longlong nr, bool unsigned_val) override;
   int  store_decimal(const my_decimal *) override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   void set_curdays(THD *thd);
   Field *new_key_field(MEM_ROOT *root, TABLE *new_table,
                        uchar *new_ptr, uint32 length,
@@ -3775,7 +3780,7 @@ class Field_datetime0 final :public Field_datetime
   }
   longlong val_int() override;
   String *val_str(String *, String *) override;
-  bool send_binary(Protocol *protocol) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *,const uchar *) const override;
   void sort_string(uchar *buff,uint length) override;
   uint32 pack_length() const override { return 8; }
@@ -3815,7 +3820,7 @@ class Field_datetime_with_dec :public Field_datetime {
   uint decimals() const override final { return dec; }
   enum ha_base_keytype key_type() const override final { return HA_KEYTYPE_BINARY; }
   void make_send_field(Send_field *field) override final;
-  bool send_binary(Protocol *protocol) override final;
+  bool send(Protocol *protocol) override final;
   uchar *pack(uchar *to, const uchar *from, uint max_length) override final
   { return Field::pack(to, from, max_length); }
   const uchar *unpack(uchar* to, const uchar *from, const uchar *from_end,
@@ -4153,6 +4158,7 @@ class Field_varstring :public Field_longstr {
   longlong val_int() override;
   String *val_str(String *, String *) override;
   my_decimal *val_decimal(my_decimal *) override;
+  bool send(Protocol *protocol) override;
   int cmp(const uchar *a,const uchar *b) const override;
   int cmp_prefix(const uchar *a, const uchar *b, size_t prefix_len) const
     override;
@@ -4217,6 +4223,15 @@ class Field_varstring_compressed final :public Field_varstring {
   double val_real() override;
   longlong val_int() override;
   uint size_of() const override { return sizeof *this; }
+  /*
+    We use the default Field::send() implementation,
+    because the derived optimized version (from Field_longstr)
+    is not suitable for compressed fields.
+  */
+  bool send(Protocol *protocol) override
+  {
+    return Field::send(protocol);
+  }
   enum_field_types binlog_type() const override
   { return MYSQL_TYPE_VARCHAR_COMPRESSED; }
   void sql_type(String &str) const override
@@ -4614,6 +4629,15 @@ class Field_blob_compressed final :public Field_blob {
   String *val_str(String *, String *) override;
   double val_real() override;
   longlong val_int() override;
+  /*
+    We use the default Field::send() implementation,
+    because the derived optimized version (from Field_longstr)
+    is not suitable for compressed fields.
+  */
+  bool send(Protocol *protocol) override
+  {
+    return Field::send(protocol);
+  }
   uint size_of() const override { return sizeof *this; }
   enum_field_types binlog_type() const override
   { return MYSQL_TYPE_BLOB_COMPRESSED; }

sql/protocol.cc

@@ -1232,6 +1232,22 @@ bool Protocol_text::store_str(const char *from, size_t length,
 }
 
 
+bool Protocol_text::store_numeric_zerofill_str(const char *from,
+                                               size_t length,
+                                               protocol_send_type_t send_type)
+{
+#ifndef DBUG_OFF
+  DBUG_PRINT("info",
+       ("Protocol_text::store_numeric_zerofill_str field %u : %.*b",
+        field_pos, (int) length, (length == 0 ? "" : from)));
+  DBUG_ASSERT(field_handlers == 0 || field_pos < field_count);
+  DBUG_ASSERT(valid_handler(field_pos, send_type));
+  field_pos++;
+#endif
+  return store_numeric_string_aux(from, length);
+}
+
+
 bool Protocol_text::store_tiny(longlong from)
 {
 #ifndef DBUG_OFF
@@ -1321,12 +1337,6 @@ bool Protocol_text::store(Field *field)
 {
   if (field->is_null())
     return store_null();
-#ifndef DBUG_OFF
-  field_pos++;
-#endif
-  char buff[MAX_FIELD_WIDTH];
-  String str(buff,sizeof(buff), &my_charset_bin);
-  CHARSET_INFO *tocs= this->thd->variables.character_set_results;
 #ifdef DBUG_ASSERT_EXISTS
   TABLE *table= field->table;
   my_bitmap_map *old_map= 0;
@@ -1334,13 +1344,14 @@ bool Protocol_text::store(Field *field)
     old_map= dbug_tmp_use_all_columns(table, table->read_set);
 #endif
 
-  field->val_str(&str);
+  bool rc= field->send(this);
+
 #ifdef DBUG_ASSERT_EXISTS
   if (old_map)
     dbug_tmp_restore_column_map(table->read_set, old_map);
 #endif
 
-  return store_string_aux(str.ptr(), str.length(), str.charset(), tocs);
+  return rc;
 }
 
 
@@ -1553,12 +1564,12 @@ bool Protocol_binary::store_double(double from, uint32 decimals)
 bool Protocol_binary::store(Field *field)
 {
   /*
-    We should not increment field_pos here as send_binary() will call another
+    We should not increment field_pos here as send() will call another
     protocol function to do this for us
   */
   if (field->is_null())
     return store_null();
-  return field->send_binary(this);
+  return field->send(this);
 }
 
 

sql/protocol.h

@@ -219,6 +219,10 @@ class Protocol_text final :public Protocol
   bool store(Field *field) override;
 
   bool send_out_parameters(List<Item_param> *sp_params) override;
+
+  bool store_numeric_zerofill_str(const char *from, size_t length,
+                                  protocol_send_type_t send_type);
+
 #ifdef EMBEDDED_LIBRARY
   void remove_last_row() override;
 #endif

sql/sql_prepare.cc

@@ -5390,7 +5390,7 @@ bool Protocol_local::store(Field *field)
 {
   if (field->is_null())
     return store_null();
-  return field->send_binary(this);
+  return field->send(this);
 }