8256274: C2: Optimize copying of the shared type dictionary

Reviewed-by: neliasso, kvn

Author: cl4es

src/hotspot/share/libadt/dict.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -47,7 +47,7 @@ static const short xsum[MAXID] = {3,8,17,34,67,132,261,264,269,278,295,328,393,5
 class bucket : public ResourceObj {
 public:
   uint _cnt, _max;              // Size of bucket
-  void **_keyvals;              // Array of keys and values
+  void** _keyvals;              // Array of keys and values
 };
 
 //------------------------------Dict-----------------------------------------
@@ -64,75 +64,70 @@ Dict::Dict(CmpKey initcmp, Hash inithash) : _arena(Thread::current()->resource_a
 
   _size = 16;                   // Size is a power of 2
   _cnt = 0;                     // Dictionary is empty
-  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size);
-  memset((void*)_bin,0,sizeof(bucket)*_size);
+  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket) * _size);
+  memset((void*)_bin, 0, sizeof(bucket) * _size);
 }
 
-Dict::Dict(CmpKey initcmp, Hash inithash, Arena *arena, int size)
+Dict::Dict(CmpKey initcmp, Hash inithash, Arena* arena, int size)
 : _arena(arena), _hash(inithash), _cmp(initcmp) {
   // Size is a power of 2
   _size = MAX2(16, round_up_power_of_2(size));
 
   _cnt = 0;                     // Dictionary is empty
-  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size);
-  memset((void*)_bin,0,sizeof(bucket)*_size);
+  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket) * _size);
+  memset((void*)_bin, 0, sizeof(bucket) * _size);
 }
 
-//------------------------------~Dict------------------------------------------
-// Delete an existing dictionary.
-Dict::~Dict() {
-  /*
-  tty->print("~Dict %d/%d: ",_cnt,_size);
-  for( uint i=0; i < _size; i++) // For complete new table do
-    tty->print("%d ",_bin[i]._cnt);
-  tty->print("\n");*/
-  /*for( uint i=0; i<_size; i++ ) {
-    FREE_FAST( _bin[i]._keyvals );
-    } */
+// Deep copy into arena of choice
+Dict::Dict(const Dict &d, Arena* arena)
+: _arena(arena), _size(d._size), _cnt(d._cnt), _hash(d._hash), _cmp(d._cmp) {
+  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket) * _size);
+  memcpy((void*)_bin, (void*)d._bin, sizeof(bucket) * _size);
+  for (uint i = 0; i < _size; i++) {
+    if (!_bin[i]._keyvals) {
+      continue;
+    }
+    _bin[i]._keyvals = (void**)_arena->Amalloc_4(sizeof(void*) * _bin[i]._max * 2);
+    memcpy(_bin[i]._keyvals, d._bin[i]._keyvals, _bin[i]._cnt * 2 * sizeof(void*));
+  }
 }
 
-//------------------------------Clear----------------------------------------
-// Zap to empty; ready for re-use
-void Dict::Clear() {
-  _cnt = 0;                     // Empty contents
-  for( uint i=0; i<_size; i++ )
-    _bin[i]._cnt = 0;           // Empty buckets, but leave allocated
-  // Leave _size & _bin alone, under the assumption that dictionary will
-  // grow to this size again.
-}
+//------------------------------~Dict------------------------------------------
+// Delete an existing dictionary.
+Dict::~Dict() { }
 
 //------------------------------doubhash---------------------------------------
 // Double hash table size.  If can't do so, just suffer.  If can, then run
 // thru old hash table, moving things to new table.  Note that since hash
 // table doubled, exactly 1 new bit is exposed in the mask - so everything
 // in the old table ends up on 1 of two lists in the new table; a hi and a
 // lo list depending on the value of the bit.
-void Dict::doubhash(void) {
+void Dict::doubhash() {
   uint oldsize = _size;
   _size <<= 1;                  // Double in size
   _bin = (bucket*)_arena->Arealloc(_bin, sizeof(bucket) * oldsize, sizeof(bucket) * _size);
   memset((void*)(&_bin[oldsize]), 0, oldsize * sizeof(bucket));
   // Rehash things to spread into new table
   for (uint i = 0; i < oldsize; i++) { // For complete OLD table do
-    bucket *b = &_bin[i];              // Handy shortcut for _bin[i]
+    bucket* b = &_bin[i];              // Handy shortcut for _bin[i]
     if (!b->_keyvals) continue;        // Skip empties fast
 
-    bucket *nb = &_bin[i+oldsize];     // New bucket shortcut
+    bucket* nb = &_bin[i+oldsize];     // New bucket shortcut
     uint j = b->_max;                  // Trim new bucket to nearest power of 2
     while (j > b->_cnt) { j >>= 1; }   // above old bucket _cnt
     if (!j) { j = 1; }                 // Handle zero-sized buckets
     nb->_max = j << 1;
     // Allocate worst case space for key-value pairs
-    nb->_keyvals = (void**)_arena->Amalloc_4(sizeof(void *) * nb->_max * 2);
+    nb->_keyvals = (void**)_arena->Amalloc_4(sizeof(void* ) * nb->_max * 2);
     uint nbcnt = 0;
 
-    for (j = 0; j < b->_cnt; ) {           // Rehash all keys in this bucket
-      void *key = b->_keyvals[j + j];
+    for (j = 0; j < b->_cnt;) {        // Rehash all keys in this bucket
+      void* key = b->_keyvals[j + j];
       if ((_hash(key) & (_size-1)) != i) { // Moving to hi bucket?
         nb->_keyvals[nbcnt + nbcnt] = key;
         nb->_keyvals[nbcnt + nbcnt + 1] = b->_keyvals[j + j + 1];
         nb->_cnt = nbcnt = nbcnt + 1;
-        b->_cnt--;                         // Remove key/value from lo bucket
+        b->_cnt--;                     // Remove key/value from lo bucket
         b->_keyvals[j + j] = b->_keyvals[b->_cnt + b->_cnt];
         b->_keyvals[j + j + 1] = b->_keyvals[b->_cnt + b->_cnt + 1];
         // Don't increment j, hash compacted element also.
@@ -143,135 +138,86 @@ void Dict::doubhash(void) {
   } // End of for all buckets
 }
 
-//------------------------------Dict-----------------------------------------
-// Deep copy a dictionary.
-Dict::Dict( const Dict &d ) : ResourceObj(d), _arena(d._arena), _size(d._size), _cnt(d._cnt), _hash(d._hash), _cmp(d._cmp) {
-  _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size);
-  memcpy( (void*)_bin, (void*)d._bin, sizeof(bucket)*_size );
-  for( uint i=0; i<_size; i++ ) {
-    if( !_bin[i]._keyvals ) continue;
-    _bin[i]._keyvals=(void**)_arena->Amalloc_4( sizeof(void *)*_bin[i]._max*2);
-    memcpy( _bin[i]._keyvals, d._bin[i]._keyvals,_bin[i]._cnt*2*sizeof(void*));
-  }
-}
-
-//------------------------------Dict-----------------------------------------
-// Deep copy a dictionary.
-Dict &Dict::operator =( const Dict &d ) {
-  if( _size < d._size ) {       // If must have more buckets
-    _arena = d._arena;
-    _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*_size, sizeof(bucket)*d._size );
-    memset( (void*)(&_bin[_size]), 0, (d._size-_size)*sizeof(bucket) );
-    _size = d._size;
-  }
-  uint i;
-  for( i=0; i<_size; i++ ) // All buckets are empty
-    _bin[i]._cnt = 0;           // But leave bucket allocations alone
-  _cnt = d._cnt;
-  *(Hash*)(&_hash) = d._hash;
-  *(CmpKey*)(&_cmp) = d._cmp;
-  for( i=0; i<_size; i++ ) {
-    bucket *b = &d._bin[i];     // Shortcut to source bucket
-    for( uint j=0; j<b->_cnt; j++ )
-      Insert( b->_keyvals[j+j], b->_keyvals[j+j+1] );
-  }
-  return *this;
-}
-
 //------------------------------Insert----------------------------------------
 // Insert or replace a key/value pair in the given dictionary.  If the
 // dictionary is too full, it's size is doubled.  The prior value being
 // replaced is returned (NULL if this is a 1st insertion of that key).  If
 // an old value is found, it's swapped with the prior key-value pair on the
 // list.  This moves a commonly searched-for value towards the list head.
-void *Dict::Insert(void *key, void *val, bool replace) {
-  uint hash = _hash( key );     // Get hash key
-  uint i = hash & (_size-1);    // Get hash key, corrected for size
-  bucket *b = &_bin[i];         // Handy shortcut
-  for( uint j=0; j<b->_cnt; j++ ) {
-    if( !_cmp(key,b->_keyvals[j+j]) ) {
+void*Dict::Insert(void* key, void* val, bool replace) {
+  uint hash = _hash(key);       // Get hash key
+  uint i = hash & (_size - 1);  // Get hash key, corrected for size
+  bucket* b = &_bin[i];
+  for (uint j = 0; j < b->_cnt; j++) {
+    if (!_cmp(key, b->_keyvals[j + j])) {
       if (!replace) {
-        return b->_keyvals[j+j+1];
+        return b->_keyvals[j + j + 1];
       } else {
-        void *prior = b->_keyvals[j+j+1];
-        b->_keyvals[j+j  ] = key;       // Insert current key-value
-        b->_keyvals[j+j+1] = val;
-        return prior;           // Return prior
+        void* prior = b->_keyvals[j + j + 1];
+        b->_keyvals[j + j    ] = key;
+        b->_keyvals[j + j + 1] = val;
+        return prior;
       }
     }
   }
-  if( ++_cnt > _size ) {        // Hash table is full
+  if (++_cnt > _size) {         // Hash table is full
     doubhash();                 // Grow whole table if too full
-    i = hash & (_size-1);       // Rehash
-    b = &_bin[i];               // Handy shortcut
+    i = hash & (_size - 1);     // Rehash
+    b = &_bin[i];
   }
-  if( b->_cnt == b->_max ) {    // Must grow bucket?
-    if( !b->_keyvals ) {
+  if (b->_cnt == b->_max) {     // Must grow bucket?
+    if (!b->_keyvals) {
       b->_max = 2;              // Initial bucket size
       b->_keyvals = (void**)_arena->Amalloc_4(sizeof(void*) * b->_max * 2);
     } else {
       b->_keyvals = (void**)_arena->Arealloc(b->_keyvals, sizeof(void*) * b->_max * 2, sizeof(void*) * b->_max * 4);
       b->_max <<= 1;            // Double bucket
     }
   }
-  b->_keyvals[b->_cnt+b->_cnt  ] = key;
-  b->_keyvals[b->_cnt+b->_cnt+1] = val;
+  b->_keyvals[b->_cnt + b->_cnt    ] = key;
+  b->_keyvals[b->_cnt + b->_cnt + 1] = val;
   b->_cnt++;
   return NULL;                  // Nothing found prior
 }
 
 //------------------------------Delete---------------------------------------
 // Find & remove a value from dictionary. Return old value.
-void *Dict::Delete(void *key) {
-  uint i = _hash( key ) & (_size-1);    // Get hash key, corrected for size
-  bucket *b = &_bin[i];         // Handy shortcut
-  for( uint j=0; j<b->_cnt; j++ )
-    if( !_cmp(key,b->_keyvals[j+j]) ) {
-      void *prior = b->_keyvals[j+j+1];
+void* Dict::Delete(void* key) {
+  uint i = _hash(key) & (_size - 1); // Get hash key, corrected for size
+  bucket* b = &_bin[i];         // Handy shortcut
+  for (uint j = 0; j < b->_cnt; j++) {
+    if (!_cmp(key, b->_keyvals[j + j])) {
+      void* prior = b->_keyvals[j + j + 1];
       b->_cnt--;                // Remove key/value from lo bucket
-      b->_keyvals[j+j  ] = b->_keyvals[b->_cnt+b->_cnt  ];
-      b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1];
+      b->_keyvals[j+j  ] = b->_keyvals[b->_cnt + b->_cnt    ];
+      b->_keyvals[j+j+1] = b->_keyvals[b->_cnt + b->_cnt + 1];
       _cnt--;                   // One less thing in table
       return prior;
     }
+  }
   return NULL;
 }
 
 //------------------------------FindDict-------------------------------------
 // Find a key-value pair in the given dictionary.  If not found, return NULL.
 // If found, move key-value pair towards head of list.
-void *Dict::operator [](const void *key) const {
-  uint i = _hash( key ) & (_size-1);    // Get hash key, corrected for size
-  bucket *b = &_bin[i];         // Handy shortcut
-  for( uint j=0; j<b->_cnt; j++ )
-    if( !_cmp(key,b->_keyvals[j+j]) )
-      return b->_keyvals[j+j+1];
-  return NULL;
-}
-
-//------------------------------CmpDict--------------------------------------
-// CmpDict compares two dictionaries; they must have the same keys (their
-// keys must match using CmpKey) and they must have the same values (pointer
-// comparison).  If so 1 is returned, if not 0 is returned.
-int32_t Dict::operator ==(const Dict &d2) const {
-  if( _cnt != d2._cnt ) return 0;
-  if( _hash != d2._hash ) return 0;
-  if( _cmp != d2._cmp ) return 0;
-  for( uint i=0; i < _size; i++) {      // For complete hash table do
-    bucket *b = &_bin[i];       // Handy shortcut
-    if( b->_cnt != d2._bin[i]._cnt ) return 0;
-    if( memcmp(b->_keyvals, d2._bin[i]._keyvals, b->_cnt*2*sizeof(void*) ) )
-      return 0;                 // Key-value pairs must match
+void* Dict::operator [](const void* key) const {
+  uint i = _hash(key) & (_size - 1); // Get hash key, corrected for size
+  bucket* b = &_bin[i];         // Handy shortcut
+  for (uint j = 0; j < b->_cnt; j++) {
+    if (!_cmp(key, b->_keyvals[j + j])) {
+      return b->_keyvals[j + j + 1];
+    }
   }
-  return 1;                     // All match, is OK
+  return NULL;
 }
 
 //------------------------------print------------------------------------------
 // Handier print routine
 void Dict::print() {
   DictI i(this); // Moved definition in iterator here because of g++.
   tty->print("Dict@" INTPTR_FORMAT "[%d] = {", p2i(this), _cnt);
-  for( ; i.test(); ++i ) {
+  for (; i.test(); ++i) {
     tty->print("(" INTPTR_FORMAT "," INTPTR_FORMAT "),", p2i(i._key), p2i(i._value));
   }
   tty->print_cr("}");
@@ -288,12 +234,12 @@ void Dict::print() {
 // be in the range 0-127 (I double & add 1 to force oddness).  Keys are
 // limited to MAXID characters in length.  Experimental evidence on 150K of
 // C text shows excellent spreading of values for any size hash table.
-int hashstr(const void *t) {
+int hashstr(const void* t) {
   char c, k = 0;
   int32_t sum = 0;
-  const char *s = (const char *)t;
+  const char* s = (const char*)t;
 
-  while( ((c = *s++) != '\0') && (k < MAXID-1) ) { // Get characters till null or MAXID-1
+  while (((c = *s++) != '\0') && (k < MAXID-1)) { // Get characters till null or MAXID-1
     c = (c << 1) + 1;             // Characters are always odd!
     sum += c + (c << shft[k++]);  // Universal hash function
   }
@@ -303,33 +249,37 @@ int hashstr(const void *t) {
 //------------------------------hashptr--------------------------------------
 // Slimey cheap hash function; no guaranteed performance.  Better than the
 // default for pointers, especially on MS-DOS machines.
-int hashptr(const void *key) {
+int hashptr(const void* key) {
   return ((intptr_t)key >> 2);
 }
 
 // Slimey cheap hash function; no guaranteed performance.
-int hashkey(const void *key) {
+int hashkey(const void* key) {
   return (intptr_t)key;
 }
 
 //------------------------------Key Comparator Functions---------------------
-int32_t cmpstr(const void *k1, const void *k2) {
-  return strcmp((const char *)k1,(const char *)k2);
+int32_t cmpstr(const void* k1, const void* k2) {
+  return strcmp((const char*)k1, (const char*)k2);
 }
 
 // Cheap key comparator.
-int32_t cmpkey(const void *key1, const void *key2) {
-  if (key1 == key2) return 0;
+int32_t cmpkey(const void* key1, const void* key2) {
+  if (key1 == key2) {
+    return 0;
+  }
   intptr_t delta = (intptr_t)key1 - (intptr_t)key2;
-  if (delta > 0) return 1;
+  if (delta > 0) {
+    return 1;
+  }
   return -1;
 }
 
 //=============================================================================
 //------------------------------reset------------------------------------------
 // Create an iterator and initialize the first variables.
-void DictI::reset( const Dict *dict ) {
-  _d = dict;                    // The dictionary
+void DictI::reset(const Dict* dict) {
+  _d = dict;
   _i = (uint)-1;                // Before the first bin
   _j = 0;                       // Nothing left in the current bin
   ++(*this);                    // Step to first real value
@@ -339,15 +289,17 @@ void DictI::reset( const Dict *dict ) {
 // Find the next key-value pair in the dictionary, or return a NULL key and
 // value.
 void DictI::operator ++(void) {
-  if( _j-- ) {                  // Still working in current bin?
-    _key   = _d->_bin[_i]._keyvals[_j+_j];
-    _value = _d->_bin[_i]._keyvals[_j+_j+1];
+  if (_j--) {                   // Still working in current bin?
+    _key   = _d->_bin[_i]._keyvals[_j + _j];
+    _value = _d->_bin[_i]._keyvals[_j + _j + 1];
     return;
   }
 
-  while( ++_i < _d->_size ) {   // Else scan for non-zero bucket
+  while (++_i < _d->_size) {   // Else scan for non-zero bucket
     _j = _d->_bin[_i]._cnt;
-    if( !_j ) continue;
+    if (!_j) {
+      continue;
+    }
     _j--;
     _key   = _d->_bin[_i]._keyvals[_j+_j];
     _value = _d->_bin[_i]._keyvals[_j+_j+1];