map.go

     1  // Copyright 2014 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  package runtime
     6  
     7  // This file contains the implementation of Go's map type.
     8  //
     9  // A map is just a hash table. The data is arranged
    10  // into an array of buckets. Each bucket contains up to
    11  // 8 key/elem pairs. The low-order bits of the hash are
    12  // used to select a bucket. Each bucket contains a few
    13  // high-order bits of each hash to distinguish the entries
    14  // within a single bucket.
    15  //
    16  // If more than 8 keys hash to a bucket, we chain on
    17  // extra buckets.
    18  //
    19  // When the hashtable grows, we allocate a new array
    20  // of buckets twice as big. Buckets are incrementally
    21  // copied from the old bucket array to the new bucket array.
    22  //
    23  // Map iterators walk through the array of buckets and
    24  // return the keys in walk order (bucket #, then overflow
    25  // chain order, then bucket index).  To maintain iteration
    26  // semantics, we never move keys within their bucket (if
    27  // we did, keys might be returned 0 or 2 times).  When
    28  // growing the table, iterators remain iterating through the
    29  // old table and must check the new table if the bucket
    30  // they are iterating through has been moved ("evacuated")
    31  // to the new table.
    32  
    33  // Picking loadFactor: too large and we have lots of overflow
    34  // buckets, too small and we waste a lot of space. I wrote
    35  // a simple program to check some stats for different loads:
    36  // (64-bit, 8 byte keys and elems)
    37  //  loadFactor    %overflow  bytes/entry     hitprobe    missprobe
    38  //        4.00         2.13        20.77         3.00         4.00
    39  //        4.50         4.05        17.30         3.25         4.50
    40  //        5.00         6.85        14.77         3.50         5.00
    41  //        5.50        10.55        12.94         3.75         5.50
    42  //        6.00        15.27        11.67         4.00         6.00
    43  //        6.50        20.90        10.79         4.25         6.50
    44  //        7.00        27.14        10.15         4.50         7.00
    45  //        7.50        34.03         9.73         4.75         7.50
    46  //        8.00        41.10         9.40         5.00         8.00
    47  //
    48  // %overflow   = percentage of buckets which have an overflow bucket
    49  // bytes/entry = overhead bytes used per key/elem pair
    50  // hitprobe    = # of entries to check when looking up a present key
    51  // missprobe   = # of entries to check when looking up an absent key
    52  //
    53  // Keep in mind this data is for maximally loaded tables, i.e. just
    54  // before the table grows. Typical tables will be somewhat less loaded.
    55  
    56  import (
    57  	"internal/abi"
    58  	"internal/goarch"
    59  	"internal/runtime/atomic"
    60  	"runtime/internal/math"
    61  	"unsafe"
    62  )
    63  
    64  const (
    65  	// Maximum number of key/elem pairs a bucket can hold.
    66  	bucketCntBits = abi.MapBucketCountBits
    67  
    68  	// Maximum average load of a bucket that triggers growth is bucketCnt*13/16 (about 80% full)
    69  	// Because of minimum alignment rules, bucketCnt is known to be at least 8.
    70  	// Represent as loadFactorNum/loadFactorDen, to allow integer math.
    71  	loadFactorDen = 2
    72  	loadFactorNum = loadFactorDen * abi.MapBucketCount * 13 / 16
    73  
    74  	// data offset should be the size of the bmap struct, but needs to be
    75  	// aligned correctly. For amd64p32 this means 64-bit alignment
    76  	// even though pointers are 32 bit.
    77  	dataOffset = unsafe.Offsetof(struct {
    78  		b bmap
    79  		v int64
    80  	}{}.v)
    81  
    82  	// Possible tophash values. We reserve a few possibilities for special marks.
    83  	// Each bucket (including its overflow buckets, if any) will have either all or none of its
    84  	// entries in the evacuated* states (except during the evacuate() method, which only happens
    85  	// during map writes and thus no one else can observe the map during that time).
    86  	emptyRest      = 0 // this cell is empty, and there are no more non-empty cells at higher indexes or overflows.
    87  	emptyOne       = 1 // this cell is empty
    88  	evacuatedX     = 2 // key/elem is valid.  Entry has been evacuated to first half of larger table.
    89  	evacuatedY     = 3 // same as above, but evacuated to second half of larger table.
    90  	evacuatedEmpty = 4 // cell is empty, bucket is evacuated.
    91  	minTopHash     = 5 // minimum tophash for a normal filled cell.
    92  
    93  	// flags
    94  	iterator     = 1 // there may be an iterator using buckets
    95  	oldIterator  = 2 // there may be an iterator using oldbuckets
    96  	hashWriting  = 4 // a goroutine is writing to the map
    97  	sameSizeGrow = 8 // the current map growth is to a new map of the same size
    98  
    99  	// sentinel bucket ID for iterator checks
   100  	noCheck = 1<<(8*goarch.PtrSize) - 1
   101  )
   102  
   103  // isEmpty reports whether the given tophash array entry represents an empty bucket entry.
   104  func isEmpty(x uint8) bool {
   105  	return x <= emptyOne
   106  }
   107  
   108  // A header for a Go map.
   109  type hmap struct {
   110  	// Note: the format of the hmap is also encoded in cmd/compile/internal/reflectdata/reflect.go.
   111  	// Make sure this stays in sync with the compiler's definition.
   112  	count     int // # live cells == size of map.  Must be first (used by len() builtin)
   113  	flags     uint8
   114  	B         uint8  // log_2 of # of buckets (can hold up to loadFactor * 2^B items)
   115  	noverflow uint16 // approximate number of overflow buckets; see incrnoverflow for details
   116  	hash0     uint32 // hash seed
   117  
   118  	buckets    unsafe.Pointer // array of 2^B Buckets. may be nil if count==0.
   119  	oldbuckets unsafe.Pointer // previous bucket array of half the size, non-nil only when growing
   120  	nevacuate  uintptr        // progress counter for evacuation (buckets less than this have been evacuated)
   121  
   122  	extra *mapextra // optional fields
   123  }
   124  
   125  // mapextra holds fields that are not present on all maps.
   126  type mapextra struct {
   127  	// If both key and elem do not contain pointers and are inline, then we mark bucket
   128  	// type as containing no pointers. This avoids scanning such maps.
   129  	// However, bmap.overflow is a pointer. In order to keep overflow buckets
   130  	// alive, we store pointers to all overflow buckets in hmap.extra.overflow and hmap.extra.oldoverflow.
   131  	// overflow and oldoverflow are only used if key and elem do not contain pointers.
   132  	// overflow contains overflow buckets for hmap.buckets.
   133  	// oldoverflow contains overflow buckets for hmap.oldbuckets.
   134  	// The indirection allows to store a pointer to the slice in hiter.
   135  	overflow    *[]*bmap
   136  	oldoverflow *[]*bmap
   137  
   138  	// nextOverflow holds a pointer to a free overflow bucket.
   139  	nextOverflow *bmap
   140  }
   141  
   142  // A bucket for a Go map.
   143  type bmap struct {
   144  	// tophash generally contains the top byte of the hash value
   145  	// for each key in this bucket. If tophash[0] < minTopHash,
   146  	// tophash[0] is a bucket evacuation state instead.
   147  	tophash [abi.MapBucketCount]uint8
   148  	// Followed by bucketCnt keys and then bucketCnt elems.
   149  	// NOTE: packing all the keys together and then all the elems together makes the
   150  	// code a bit more complicated than alternating key/elem/key/elem/... but it allows
   151  	// us to eliminate padding which would be needed for, e.g., map[int64]int8.
   152  	// Followed by an overflow pointer.
   153  }
   154  
   155  // A hash iteration structure.
   156  // If you modify hiter, also change cmd/compile/internal/reflectdata/reflect.go
   157  // and reflect/value.go to match the layout of this structure.
   158  type hiter struct {
   159  	key         unsafe.Pointer // Must be in first position.  Write nil to indicate iteration end (see cmd/compile/internal/walk/range.go).
   160  	elem        unsafe.Pointer // Must be in second position (see cmd/compile/internal/walk/range.go).
   161  	t           *maptype
   162  	h           *hmap
   163  	buckets     unsafe.Pointer // bucket ptr at hash_iter initialization time
   164  	bptr        *bmap          // current bucket
   165  	overflow    *[]*bmap       // keeps overflow buckets of hmap.buckets alive
   166  	oldoverflow *[]*bmap       // keeps overflow buckets of hmap.oldbuckets alive
   167  	startBucket uintptr        // bucket iteration started at
   168  	offset      uint8          // intra-bucket offset to start from during iteration (should be big enough to hold bucketCnt-1)
   169  	wrapped     bool           // already wrapped around from end of bucket array to beginning
   170  	B           uint8
   171  	i           uint8
   172  	bucket      uintptr
   173  	checkBucket uintptr
   174  }
   175  
   176  // bucketShift returns 1<<b, optimized for code generation.
   177  func bucketShift(b uint8) uintptr {
   178  	// Masking the shift amount allows overflow checks to be elided.
   179  	return uintptr(1) << (b & (goarch.PtrSize*8 - 1))
   180  }
   181  
   182  // bucketMask returns 1<<b - 1, optimized for code generation.
   183  func bucketMask(b uint8) uintptr {
   184  	return bucketShift(b) - 1
   185  }
   186  
   187  // tophash calculates the tophash value for hash.
   188  func tophash(hash uintptr) uint8 {
   189  	top := uint8(hash >> (goarch.PtrSize*8 - 8))
   190  	if top < minTopHash {
   191  		top += minTopHash
   192  	}
   193  	return top
   194  }
   195  
   196  func evacuated(b *bmap) bool {
   197  	h := b.tophash[0]
   198  	return h > emptyOne && h < minTopHash
   199  }
   200  
   201  func (b *bmap) overflow(t *maptype) *bmap {
   202  	return *(**bmap)(add(unsafe.Pointer(b), uintptr(t.BucketSize)-goarch.PtrSize))
   203  }
   204  
   205  func (b *bmap) setoverflow(t *maptype, ovf *bmap) {
   206  	*(**bmap)(add(unsafe.Pointer(b), uintptr(t.BucketSize)-goarch.PtrSize)) = ovf
   207  }
   208  
   209  func (b *bmap) keys() unsafe.Pointer {
   210  	return add(unsafe.Pointer(b), dataOffset)
   211  }
   212  
   213  // incrnoverflow increments h.noverflow.
   214  // noverflow counts the number of overflow buckets.
   215  // This is used to trigger same-size map growth.
   216  // See also tooManyOverflowBuckets.
   217  // To keep hmap small, noverflow is a uint16.
   218  // When there are few buckets, noverflow is an exact count.
   219  // When there are many buckets, noverflow is an approximate count.
   220  func (h *hmap) incrnoverflow() {
   221  	// We trigger same-size map growth if there are
   222  	// as many overflow buckets as buckets.
   223  	// We need to be able to count to 1<<h.B.
   224  	if h.B < 16 {
   225  		h.noverflow++
   226  		return
   227  	}
   228  	// Increment with probability 1/(1<<(h.B-15)).
   229  	// When we reach 1<<15 - 1, we will have approximately
   230  	// as many overflow buckets as buckets.
   231  	mask := uint32(1)<<(h.B-15) - 1
   232  	// Example: if h.B == 18, then mask == 7,
   233  	// and rand() & 7 == 0 with probability 1/8.
   234  	if uint32(rand())&mask == 0 {
   235  		h.noverflow++
   236  	}
   237  }
   238  
   239  func (h *hmap) newoverflow(t *maptype, b *bmap) *bmap {
   240  	var ovf *bmap
   241  	if h.extra != nil && h.extra.nextOverflow != nil {
   242  		// We have preallocated overflow buckets available.
   243  		// See makeBucketArray for more details.
   244  		ovf = h.extra.nextOverflow
   245  		if ovf.overflow(t) == nil {
   246  			// We're not at the end of the preallocated overflow buckets. Bump the pointer.
   247  			h.extra.nextOverflow = (*bmap)(add(unsafe.Pointer(ovf), uintptr(t.BucketSize)))
   248  		} else {
   249  			// This is the last preallocated overflow bucket.
   250  			// Reset the overflow pointer on this bucket,
   251  			// which was set to a non-nil sentinel value.
   252  			ovf.setoverflow(t, nil)
   253  			h.extra.nextOverflow = nil
   254  		}
   255  	} else {
   256  		ovf = (*bmap)(newobject(t.Bucket))
   257  	}
   258  	h.incrnoverflow()
   259  	if !t.Bucket.Pointers() {
   260  		h.createOverflow()
   261  		*h.extra.overflow = append(*h.extra.overflow, ovf)
   262  	}
   263  	b.setoverflow(t, ovf)
   264  	return ovf
   265  }
   266  
   267  func (h *hmap) createOverflow() {
   268  	if h.extra == nil {
   269  		h.extra = new(mapextra)
   270  	}
   271  	if h.extra.overflow == nil {
   272  		h.extra.overflow = new([]*bmap)
   273  	}
   274  }
   275  
   276  func makemap64(t *maptype, hint int64, h *hmap) *hmap {
   277  	if int64(int(hint)) != hint {
   278  		hint = 0
   279  	}
   280  	return makemap(t, int(hint), h)
   281  }
   282  
   283  // makemap_small implements Go map creation for make(map[k]v) and
   284  // make(map[k]v, hint) when hint is known to be at most bucketCnt
   285  // at compile time and the map needs to be allocated on the heap.
   286  //
   287  // makemap_small should be an internal detail,
   288  // but widely used packages access it using linkname.
   289  // Notable members of the hall of shame include:
   290  //   - github.com/bytedance/sonic
   291  //
   292  // Do not remove or change the type signature.
   293  // See go.dev/issue/67401.
   294  //
   295  //go:linkname makemap_small
   296  func makemap_small() *hmap {
   297  	h := new(hmap)
   298  	h.hash0 = uint32(rand())
   299  	return h
   300  }
   301  
   302  // makemap implements Go map creation for make(map[k]v, hint).
   303  // If the compiler has determined that the map or the first bucket
   304  // can be created on the stack, h and/or bucket may be non-nil.
   305  // If h != nil, the map can be created directly in h.
   306  // If h.buckets != nil, bucket pointed to can be used as the first bucket.
   307  //
   308  // makemap should be an internal detail,
   309  // but widely used packages access it using linkname.
   310  // Notable members of the hall of shame include:
   311  //   - github.com/cloudwego/frugal
   312  //   - github.com/ugorji/go/codec
   313  //
   314  // Do not remove or change the type signature.
   315  // See go.dev/issue/67401.
   316  //
   317  //go:linkname makemap
   318  func makemap(t *maptype, hint int, h *hmap) *hmap {
   319  	mem, overflow := math.MulUintptr(uintptr(hint), t.Bucket.Size_)
   320  	if overflow || mem > maxAlloc {
   321  		hint = 0
   322  	}
   323  
   324  	// initialize Hmap
   325  	if h == nil {
   326  		h = new(hmap)
   327  	}
   328  	h.hash0 = uint32(rand())
   329  
   330  	// Find the size parameter B which will hold the requested # of elements.
   331  	// For hint < 0 overLoadFactor returns false since hint < bucketCnt.
   332  	B := uint8(0)
   333  	for overLoadFactor(hint, B) {
   334  		B++
   335  	}
   336  	h.B = B
   337  
   338  	// allocate initial hash table
   339  	// if B == 0, the buckets field is allocated lazily later (in mapassign)
   340  	// If hint is large zeroing this memory could take a while.
   341  	if h.B != 0 {
   342  		var nextOverflow *bmap
   343  		h.buckets, nextOverflow = makeBucketArray(t, h.B, nil)
   344  		if nextOverflow != nil {
   345  			h.extra = new(mapextra)
   346  			h.extra.nextOverflow = nextOverflow
   347  		}
   348  	}
   349  
   350  	return h
   351  }
   352  
   353  // makeBucketArray initializes a backing array for map buckets.
   354  // 1<<b is the minimum number of buckets to allocate.
   355  // dirtyalloc should either be nil or a bucket array previously
   356  // allocated by makeBucketArray with the same t and b parameters.
   357  // If dirtyalloc is nil a new backing array will be alloced and
   358  // otherwise dirtyalloc will be cleared and reused as backing array.
   359  func makeBucketArray(t *maptype, b uint8, dirtyalloc unsafe.Pointer) (buckets unsafe.Pointer, nextOverflow *bmap) {
   360  	base := bucketShift(b)
   361  	nbuckets := base
   362  	// For small b, overflow buckets are unlikely.
   363  	// Avoid the overhead of the calculation.
   364  	if b >= 4 {
   365  		// Add on the estimated number of overflow buckets
   366  		// required to insert the median number of elements
   367  		// used with this value of b.
   368  		nbuckets += bucketShift(b - 4)
   369  		sz := t.Bucket.Size_ * nbuckets
   370  		up := roundupsize(sz, !t.Bucket.Pointers())
   371  		if up != sz {
   372  			nbuckets = up / t.Bucket.Size_
   373  		}
   374  	}
   375  
   376  	if dirtyalloc == nil {
   377  		buckets = newarray(t.Bucket, int(nbuckets))
   378  	} else {
   379  		// dirtyalloc was previously generated by
   380  		// the above newarray(t.Bucket, int(nbuckets))
   381  		// but may not be empty.
   382  		buckets = dirtyalloc
   383  		size := t.Bucket.Size_ * nbuckets
   384  		if t.Bucket.Pointers() {
   385  			memclrHasPointers(buckets, size)
   386  		} else {
   387  			memclrNoHeapPointers(buckets, size)
   388  		}
   389  	}
   390  
   391  	if base != nbuckets {
   392  		// We preallocated some overflow buckets.
   393  		// To keep the overhead of tracking these overflow buckets to a minimum,
   394  		// we use the convention that if a preallocated overflow bucket's overflow
   395  		// pointer is nil, then there are more available by bumping the pointer.
   396  		// We need a safe non-nil pointer for the last overflow bucket; just use buckets.
   397  		nextOverflow = (*bmap)(add(buckets, base*uintptr(t.BucketSize)))
   398  		last := (*bmap)(add(buckets, (nbuckets-1)*uintptr(t.BucketSize)))
   399  		last.setoverflow(t, (*bmap)(buckets))
   400  	}
   401  	return buckets, nextOverflow
   402  }
   403  
   404  // mapaccess1 returns a pointer to h[key].  Never returns nil, instead
   405  // it will return a reference to the zero object for the elem type if
   406  // the key is not in the map.
   407  // NOTE: The returned pointer may keep the whole map live, so don't
   408  // hold onto it for very long.
   409  func mapaccess1(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer {
   410  	if raceenabled && h != nil {
   411  		callerpc := getcallerpc()
   412  		pc := abi.FuncPCABIInternal(mapaccess1)
   413  		racereadpc(unsafe.Pointer(h), callerpc, pc)
   414  		raceReadObjectPC(t.Key, key, callerpc, pc)
   415  	}
   416  	if msanenabled && h != nil {
   417  		msanread(key, t.Key.Size_)
   418  	}
   419  	if asanenabled && h != nil {
   420  		asanread(key, t.Key.Size_)
   421  	}
   422  	if h == nil || h.count == 0 {
   423  		if err := mapKeyError(t, key); err != nil {
   424  			panic(err) // see issue 23734
   425  		}
   426  		return unsafe.Pointer(&zeroVal[0])
   427  	}
   428  	if h.flags&hashWriting != 0 {
   429  		fatal("concurrent map read and map write")
   430  	}
   431  	hash := t.Hasher(key, uintptr(h.hash0))
   432  	m := bucketMask(h.B)
   433  	b := (*bmap)(add(h.buckets, (hash&m)*uintptr(t.BucketSize)))
   434  	if c := h.oldbuckets; c != nil {
   435  		if !h.sameSizeGrow() {
   436  			// There used to be half as many buckets; mask down one more power of two.
   437  			m >>= 1
   438  		}
   439  		oldb := (*bmap)(add(c, (hash&m)*uintptr(t.BucketSize)))
   440  		if !evacuated(oldb) {
   441  			b = oldb
   442  		}
   443  	}
   444  	top := tophash(hash)
   445  bucketloop:
   446  	for ; b != nil; b = b.overflow(t) {
   447  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
   448  			if b.tophash[i] != top {
   449  				if b.tophash[i] == emptyRest {
   450  					break bucketloop
   451  				}
   452  				continue
   453  			}
   454  			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   455  			if t.IndirectKey() {
   456  				k = *((*unsafe.Pointer)(k))
   457  			}
   458  			if t.Key.Equal(key, k) {
   459  				e := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   460  				if t.IndirectElem() {
   461  					e = *((*unsafe.Pointer)(e))
   462  				}
   463  				return e
   464  			}
   465  		}
   466  	}
   467  	return unsafe.Pointer(&zeroVal[0])
   468  }
   469  
   470  // mapaccess2 should be an internal detail,
   471  // but widely used packages access it using linkname.
   472  // Notable members of the hall of shame include:
   473  //   - github.com/ugorji/go/codec
   474  //
   475  // Do not remove or change the type signature.
   476  // See go.dev/issue/67401.
   477  //
   478  //go:linkname mapaccess2
   479  func mapaccess2(t *maptype, h *hmap, key unsafe.Pointer) (unsafe.Pointer, bool) {
   480  	if raceenabled && h != nil {
   481  		callerpc := getcallerpc()
   482  		pc := abi.FuncPCABIInternal(mapaccess2)
   483  		racereadpc(unsafe.Pointer(h), callerpc, pc)
   484  		raceReadObjectPC(t.Key, key, callerpc, pc)
   485  	}
   486  	if msanenabled && h != nil {
   487  		msanread(key, t.Key.Size_)
   488  	}
   489  	if asanenabled && h != nil {
   490  		asanread(key, t.Key.Size_)
   491  	}
   492  	if h == nil || h.count == 0 {
   493  		if err := mapKeyError(t, key); err != nil {
   494  			panic(err) // see issue 23734
   495  		}
   496  		return unsafe.Pointer(&zeroVal[0]), false
   497  	}
   498  	if h.flags&hashWriting != 0 {
   499  		fatal("concurrent map read and map write")
   500  	}
   501  	hash := t.Hasher(key, uintptr(h.hash0))
   502  	m := bucketMask(h.B)
   503  	b := (*bmap)(add(h.buckets, (hash&m)*uintptr(t.BucketSize)))
   504  	if c := h.oldbuckets; c != nil {
   505  		if !h.sameSizeGrow() {
   506  			// There used to be half as many buckets; mask down one more power of two.
   507  			m >>= 1
   508  		}
   509  		oldb := (*bmap)(add(c, (hash&m)*uintptr(t.BucketSize)))
   510  		if !evacuated(oldb) {
   511  			b = oldb
   512  		}
   513  	}
   514  	top := tophash(hash)
   515  bucketloop:
   516  	for ; b != nil; b = b.overflow(t) {
   517  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
   518  			if b.tophash[i] != top {
   519  				if b.tophash[i] == emptyRest {
   520  					break bucketloop
   521  				}
   522  				continue
   523  			}
   524  			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   525  			if t.IndirectKey() {
   526  				k = *((*unsafe.Pointer)(k))
   527  			}
   528  			if t.Key.Equal(key, k) {
   529  				e := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   530  				if t.IndirectElem() {
   531  					e = *((*unsafe.Pointer)(e))
   532  				}
   533  				return e, true
   534  			}
   535  		}
   536  	}
   537  	return unsafe.Pointer(&zeroVal[0]), false
   538  }
   539  
   540  // returns both key and elem. Used by map iterator.
   541  func mapaccessK(t *maptype, h *hmap, key unsafe.Pointer) (unsafe.Pointer, unsafe.Pointer) {
   542  	if h == nil || h.count == 0 {
   543  		return nil, nil
   544  	}
   545  	hash := t.Hasher(key, uintptr(h.hash0))
   546  	m := bucketMask(h.B)
   547  	b := (*bmap)(add(h.buckets, (hash&m)*uintptr(t.BucketSize)))
   548  	if c := h.oldbuckets; c != nil {
   549  		if !h.sameSizeGrow() {
   550  			// There used to be half as many buckets; mask down one more power of two.
   551  			m >>= 1
   552  		}
   553  		oldb := (*bmap)(add(c, (hash&m)*uintptr(t.BucketSize)))
   554  		if !evacuated(oldb) {
   555  			b = oldb
   556  		}
   557  	}
   558  	top := tophash(hash)
   559  bucketloop:
   560  	for ; b != nil; b = b.overflow(t) {
   561  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
   562  			if b.tophash[i] != top {
   563  				if b.tophash[i] == emptyRest {
   564  					break bucketloop
   565  				}
   566  				continue
   567  			}
   568  			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   569  			if t.IndirectKey() {
   570  				k = *((*unsafe.Pointer)(k))
   571  			}
   572  			if t.Key.Equal(key, k) {
   573  				e := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   574  				if t.IndirectElem() {
   575  					e = *((*unsafe.Pointer)(e))
   576  				}
   577  				return k, e
   578  			}
   579  		}
   580  	}
   581  	return nil, nil
   582  }
   583  
   584  func mapaccess1_fat(t *maptype, h *hmap, key, zero unsafe.Pointer) unsafe.Pointer {
   585  	e := mapaccess1(t, h, key)
   586  	if e == unsafe.Pointer(&zeroVal[0]) {
   587  		return zero
   588  	}
   589  	return e
   590  }
   591  
   592  func mapaccess2_fat(t *maptype, h *hmap, key, zero unsafe.Pointer) (unsafe.Pointer, bool) {
   593  	e := mapaccess1(t, h, key)
   594  	if e == unsafe.Pointer(&zeroVal[0]) {
   595  		return zero, false
   596  	}
   597  	return e, true
   598  }
   599  
   600  // Like mapaccess, but allocates a slot for the key if it is not present in the map.
   601  //
   602  // mapassign should be an internal detail,
   603  // but widely used packages access it using linkname.
   604  // Notable members of the hall of shame include:
   605  //   - github.com/bytedance/sonic
   606  //   - github.com/cloudwego/frugal
   607  //   - github.com/RomiChan/protobuf
   608  //   - github.com/segmentio/encoding
   609  //   - github.com/ugorji/go/codec
   610  //
   611  // Do not remove or change the type signature.
   612  // See go.dev/issue/67401.
   613  //
   614  //go:linkname mapassign
   615  func mapassign(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer {
   616  	if h == nil {
   617  		panic(plainError("assignment to entry in nil map"))
   618  	}
   619  	if raceenabled {
   620  		callerpc := getcallerpc()
   621  		pc := abi.FuncPCABIInternal(mapassign)
   622  		racewritepc(unsafe.Pointer(h), callerpc, pc)
   623  		raceReadObjectPC(t.Key, key, callerpc, pc)
   624  	}
   625  	if msanenabled {
   626  		msanread(key, t.Key.Size_)
   627  	}
   628  	if asanenabled {
   629  		asanread(key, t.Key.Size_)
   630  	}
   631  	if h.flags&hashWriting != 0 {
   632  		fatal("concurrent map writes")
   633  	}
   634  	hash := t.Hasher(key, uintptr(h.hash0))
   635  
   636  	// Set hashWriting after calling t.hasher, since t.hasher may panic,
   637  	// in which case we have not actually done a write.
   638  	h.flags ^= hashWriting
   639  
   640  	if h.buckets == nil {
   641  		h.buckets = newobject(t.Bucket) // newarray(t.Bucket, 1)
   642  	}
   643  
   644  again:
   645  	bucket := hash & bucketMask(h.B)
   646  	if h.growing() {
   647  		growWork(t, h, bucket)
   648  	}
   649  	b := (*bmap)(add(h.buckets, bucket*uintptr(t.BucketSize)))
   650  	top := tophash(hash)
   651  
   652  	var inserti *uint8
   653  	var insertk unsafe.Pointer
   654  	var elem unsafe.Pointer
   655  bucketloop:
   656  	for {
   657  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
   658  			if b.tophash[i] != top {
   659  				if isEmpty(b.tophash[i]) && inserti == nil {
   660  					inserti = &b.tophash[i]
   661  					insertk = add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   662  					elem = add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   663  				}
   664  				if b.tophash[i] == emptyRest {
   665  					break bucketloop
   666  				}
   667  				continue
   668  			}
   669  			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   670  			if t.IndirectKey() {
   671  				k = *((*unsafe.Pointer)(k))
   672  			}
   673  			if !t.Key.Equal(key, k) {
   674  				continue
   675  			}
   676  			// already have a mapping for key. Update it.
   677  			if t.NeedKeyUpdate() {
   678  				typedmemmove(t.Key, k, key)
   679  			}
   680  			elem = add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   681  			goto done
   682  		}
   683  		ovf := b.overflow(t)
   684  		if ovf == nil {
   685  			break
   686  		}
   687  		b = ovf
   688  	}
   689  
   690  	// Did not find mapping for key. Allocate new cell & add entry.
   691  
   692  	// If we hit the max load factor or we have too many overflow buckets,
   693  	// and we're not already in the middle of growing, start growing.
   694  	if !h.growing() && (overLoadFactor(h.count+1, h.B) || tooManyOverflowBuckets(h.noverflow, h.B)) {
   695  		hashGrow(t, h)
   696  		goto again // Growing the table invalidates everything, so try again
   697  	}
   698  
   699  	if inserti == nil {
   700  		// The current bucket and all the overflow buckets connected to it are full, allocate a new one.
   701  		newb := h.newoverflow(t, b)
   702  		inserti = &newb.tophash[0]
   703  		insertk = add(unsafe.Pointer(newb), dataOffset)
   704  		elem = add(insertk, abi.MapBucketCount*uintptr(t.KeySize))
   705  	}
   706  
   707  	// store new key/elem at insert position
   708  	if t.IndirectKey() {
   709  		kmem := newobject(t.Key)
   710  		*(*unsafe.Pointer)(insertk) = kmem
   711  		insertk = kmem
   712  	}
   713  	if t.IndirectElem() {
   714  		vmem := newobject(t.Elem)
   715  		*(*unsafe.Pointer)(elem) = vmem
   716  	}
   717  	typedmemmove(t.Key, insertk, key)
   718  	*inserti = top
   719  	h.count++
   720  
   721  done:
   722  	if h.flags&hashWriting == 0 {
   723  		fatal("concurrent map writes")
   724  	}
   725  	h.flags &^= hashWriting
   726  	if t.IndirectElem() {
   727  		elem = *((*unsafe.Pointer)(elem))
   728  	}
   729  	return elem
   730  }
   731  
   732  // mapdelete should be an internal detail,
   733  // but widely used packages access it using linkname.
   734  // Notable members of the hall of shame include:
   735  //   - github.com/ugorji/go/codec
   736  //
   737  // Do not remove or change the type signature.
   738  // See go.dev/issue/67401.
   739  //
   740  //go:linkname mapdelete
   741  func mapdelete(t *maptype, h *hmap, key unsafe.Pointer) {
   742  	if raceenabled && h != nil {
   743  		callerpc := getcallerpc()
   744  		pc := abi.FuncPCABIInternal(mapdelete)
   745  		racewritepc(unsafe.Pointer(h), callerpc, pc)
   746  		raceReadObjectPC(t.Key, key, callerpc, pc)
   747  	}
   748  	if msanenabled && h != nil {
   749  		msanread(key, t.Key.Size_)
   750  	}
   751  	if asanenabled && h != nil {
   752  		asanread(key, t.Key.Size_)
   753  	}
   754  	if h == nil || h.count == 0 {
   755  		if err := mapKeyError(t, key); err != nil {
   756  			panic(err) // see issue 23734
   757  		}
   758  		return
   759  	}
   760  	if h.flags&hashWriting != 0 {
   761  		fatal("concurrent map writes")
   762  	}
   763  
   764  	hash := t.Hasher(key, uintptr(h.hash0))
   765  
   766  	// Set hashWriting after calling t.hasher, since t.hasher may panic,
   767  	// in which case we have not actually done a write (delete).
   768  	h.flags ^= hashWriting
   769  
   770  	bucket := hash & bucketMask(h.B)
   771  	if h.growing() {
   772  		growWork(t, h, bucket)
   773  	}
   774  	b := (*bmap)(add(h.buckets, bucket*uintptr(t.BucketSize)))
   775  	bOrig := b
   776  	top := tophash(hash)
   777  search:
   778  	for ; b != nil; b = b.overflow(t) {
   779  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
   780  			if b.tophash[i] != top {
   781  				if b.tophash[i] == emptyRest {
   782  					break search
   783  				}
   784  				continue
   785  			}
   786  			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.KeySize))
   787  			k2 := k
   788  			if t.IndirectKey() {
   789  				k2 = *((*unsafe.Pointer)(k2))
   790  			}
   791  			if !t.Key.Equal(key, k2) {
   792  				continue
   793  			}
   794  			// Only clear key if there are pointers in it.
   795  			if t.IndirectKey() {
   796  				*(*unsafe.Pointer)(k) = nil
   797  			} else if t.Key.Pointers() {
   798  				memclrHasPointers(k, t.Key.Size_)
   799  			}
   800  			e := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
   801  			if t.IndirectElem() {
   802  				*(*unsafe.Pointer)(e) = nil
   803  			} else if t.Elem.Pointers() {
   804  				memclrHasPointers(e, t.Elem.Size_)
   805  			} else {
   806  				memclrNoHeapPointers(e, t.Elem.Size_)
   807  			}
   808  			b.tophash[i] = emptyOne
   809  			// If the bucket now ends in a bunch of emptyOne states,
   810  			// change those to emptyRest states.
   811  			// It would be nice to make this a separate function, but
   812  			// for loops are not currently inlineable.
   813  			if i == abi.MapBucketCount-1 {
   814  				if b.overflow(t) != nil && b.overflow(t).tophash[0] != emptyRest {
   815  					goto notLast
   816  				}
   817  			} else {
   818  				if b.tophash[i+1] != emptyRest {
   819  					goto notLast
   820  				}
   821  			}
   822  			for {
   823  				b.tophash[i] = emptyRest
   824  				if i == 0 {
   825  					if b == bOrig {
   826  						break // beginning of initial bucket, we're done.
   827  					}
   828  					// Find previous bucket, continue at its last entry.
   829  					c := b
   830  					for b = bOrig; b.overflow(t) != c; b = b.overflow(t) {
   831  					}
   832  					i = abi.MapBucketCount - 1
   833  				} else {
   834  					i--
   835  				}
   836  				if b.tophash[i] != emptyOne {
   837  					break
   838  				}
   839  			}
   840  		notLast:
   841  			h.count--
   842  			// Reset the hash seed to make it more difficult for attackers to
   843  			// repeatedly trigger hash collisions. See issue 25237.
   844  			if h.count == 0 {
   845  				h.hash0 = uint32(rand())
   846  			}
   847  			break search
   848  		}
   849  	}
   850  
   851  	if h.flags&hashWriting == 0 {
   852  		fatal("concurrent map writes")
   853  	}
   854  	h.flags &^= hashWriting
   855  }
   856  
   857  // mapiterinit initializes the hiter struct used for ranging over maps.
   858  // The hiter struct pointed to by 'it' is allocated on the stack
   859  // by the compilers order pass or on the heap by reflect_mapiterinit.
   860  // Both need to have zeroed hiter since the struct contains pointers.
   861  //
   862  // mapiterinit should be an internal detail,
   863  // but widely used packages access it using linkname.
   864  // Notable members of the hall of shame include:
   865  //   - github.com/bytedance/sonic
   866  //   - github.com/cloudwego/frugal
   867  //   - github.com/goccy/go-json
   868  //   - github.com/RomiChan/protobuf
   869  //   - github.com/segmentio/encoding
   870  //   - github.com/ugorji/go/codec
   871  //   - github.com/wI2L/jettison
   872  //
   873  // Do not remove or change the type signature.
   874  // See go.dev/issue/67401.
   875  //
   876  //go:linkname mapiterinit
   877  func mapiterinit(t *maptype, h *hmap, it *hiter) {
   878  	if raceenabled && h != nil {
   879  		callerpc := getcallerpc()
   880  		racereadpc(unsafe.Pointer(h), callerpc, abi.FuncPCABIInternal(mapiterinit))
   881  	}
   882  
   883  	it.t = t
   884  	if h == nil || h.count == 0 {
   885  		return
   886  	}
   887  
   888  	if unsafe.Sizeof(hiter{})/goarch.PtrSize != 12 {
   889  		throw("hash_iter size incorrect") // see cmd/compile/internal/reflectdata/reflect.go
   890  	}
   891  	it.h = h
   892  
   893  	// grab snapshot of bucket state
   894  	it.B = h.B
   895  	it.buckets = h.buckets
   896  	if !t.Bucket.Pointers() {
   897  		// Allocate the current slice and remember pointers to both current and old.
   898  		// This preserves all relevant overflow buckets alive even if
   899  		// the table grows and/or overflow buckets are added to the table
   900  		// while we are iterating.
   901  		h.createOverflow()
   902  		it.overflow = h.extra.overflow
   903  		it.oldoverflow = h.extra.oldoverflow
   904  	}
   905  
   906  	// decide where to start
   907  	r := uintptr(rand())
   908  	it.startBucket = r & bucketMask(h.B)
   909  	it.offset = uint8(r >> h.B & (abi.MapBucketCount - 1))
   910  
   911  	// iterator state
   912  	it.bucket = it.startBucket
   913  
   914  	// Remember we have an iterator.
   915  	// Can run concurrently with another mapiterinit().
   916  	if old := h.flags; old&(iterator|oldIterator) != iterator|oldIterator {
   917  		atomic.Or8(&h.flags, iterator|oldIterator)
   918  	}
   919  
   920  	mapiternext(it)
   921  }
   922  
   923  // mapiternext should be an internal detail,
   924  // but widely used packages access it using linkname.
   925  // Notable members of the hall of shame include:
   926  //   - github.com/bytedance/sonic
   927  //   - github.com/cloudwego/frugal
   928  //   - github.com/RomiChan/protobuf
   929  //   - github.com/segmentio/encoding
   930  //   - github.com/ugorji/go/codec
   931  //   - gonum.org/v1/gonum
   932  //
   933  // Do not remove or change the type signature.
   934  // See go.dev/issue/67401.
   935  //
   936  //go:linkname mapiternext
   937  func mapiternext(it *hiter) {
   938  	h := it.h
   939  	if raceenabled {
   940  		callerpc := getcallerpc()
   941  		racereadpc(unsafe.Pointer(h), callerpc, abi.FuncPCABIInternal(mapiternext))
   942  	}
   943  	if h.flags&hashWriting != 0 {
   944  		fatal("concurrent map iteration and map write")
   945  	}
   946  	t := it.t
   947  	bucket := it.bucket
   948  	b := it.bptr
   949  	i := it.i
   950  	checkBucket := it.checkBucket
   951  
   952  next:
   953  	if b == nil {
   954  		if bucket == it.startBucket && it.wrapped {
   955  			// end of iteration
   956  			it.key = nil
   957  			it.elem = nil
   958  			return
   959  		}
   960  		if h.growing() && it.B == h.B {
   961  			// Iterator was started in the middle of a grow, and the grow isn't done yet.
   962  			// If the bucket we're looking at hasn't been filled in yet (i.e. the old
   963  			// bucket hasn't been evacuated) then we need to iterate through the old
   964  			// bucket and only return the ones that will be migrated to this bucket.
   965  			oldbucket := bucket & it.h.oldbucketmask()
   966  			b = (*bmap)(add(h.oldbuckets, oldbucket*uintptr(t.BucketSize)))
   967  			if !evacuated(b) {
   968  				checkBucket = bucket
   969  			} else {
   970  				b = (*bmap)(add(it.buckets, bucket*uintptr(t.BucketSize)))
   971  				checkBucket = noCheck
   972  			}
   973  		} else {
   974  			b = (*bmap)(add(it.buckets, bucket*uintptr(t.BucketSize)))
   975  			checkBucket = noCheck
   976  		}
   977  		bucket++
   978  		if bucket == bucketShift(it.B) {
   979  			bucket = 0
   980  			it.wrapped = true
   981  		}
   982  		i = 0
   983  	}
   984  	for ; i < abi.MapBucketCount; i++ {
   985  		offi := (i + it.offset) & (abi.MapBucketCount - 1)
   986  		if isEmpty(b.tophash[offi]) || b.tophash[offi] == evacuatedEmpty {
   987  			// TODO: emptyRest is hard to use here, as we start iterating
   988  			// in the middle of a bucket. It's feasible, just tricky.
   989  			continue
   990  		}
   991  		k := add(unsafe.Pointer(b), dataOffset+uintptr(offi)*uintptr(t.KeySize))
   992  		if t.IndirectKey() {
   993  			k = *((*unsafe.Pointer)(k))
   994  		}
   995  		e := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+uintptr(offi)*uintptr(t.ValueSize))
   996  		if checkBucket != noCheck && !h.sameSizeGrow() {
   997  			// Special case: iterator was started during a grow to a larger size
   998  			// and the grow is not done yet. We're working on a bucket whose
   999  			// oldbucket has not been evacuated yet. Or at least, it wasn't
  1000  			// evacuated when we started the bucket. So we're iterating
  1001  			// through the oldbucket, skipping any keys that will go
  1002  			// to the other new bucket (each oldbucket expands to two
  1003  			// buckets during a grow).
  1004  			if t.ReflexiveKey() || t.Key.Equal(k, k) {
  1005  				// If the item in the oldbucket is not destined for
  1006  				// the current new bucket in the iteration, skip it.
  1007  				hash := t.Hasher(k, uintptr(h.hash0))
  1008  				if hash&bucketMask(it.B) != checkBucket {
  1009  					continue
  1010  				}
  1011  			} else {
  1012  				// Hash isn't repeatable if k != k (NaNs).  We need a
  1013  				// repeatable and randomish choice of which direction
  1014  				// to send NaNs during evacuation. We'll use the low
  1015  				// bit of tophash to decide which way NaNs go.
  1016  				// NOTE: this case is why we need two evacuate tophash
  1017  				// values, evacuatedX and evacuatedY, that differ in
  1018  				// their low bit.
  1019  				if checkBucket>>(it.B-1) != uintptr(b.tophash[offi]&1) {
  1020  					continue
  1021  				}
  1022  			}
  1023  		}
  1024  		if (b.tophash[offi] != evacuatedX && b.tophash[offi] != evacuatedY) ||
  1025  			!(t.ReflexiveKey() || t.Key.Equal(k, k)) {
  1026  			// This is the golden data, we can return it.
  1027  			// OR
  1028  			// key!=key, so the entry can't be deleted or updated, so we can just return it.
  1029  			// That's lucky for us because when key!=key we can't look it up successfully.
  1030  			it.key = k
  1031  			if t.IndirectElem() {
  1032  				e = *((*unsafe.Pointer)(e))
  1033  			}
  1034  			it.elem = e
  1035  		} else {
  1036  			// The hash table has grown since the iterator was started.
  1037  			// The golden data for this key is now somewhere else.
  1038  			// Check the current hash table for the data.
  1039  			// This code handles the case where the key
  1040  			// has been deleted, updated, or deleted and reinserted.
  1041  			// NOTE: we need to regrab the key as it has potentially been
  1042  			// updated to an equal() but not identical key (e.g. +0.0 vs -0.0).
  1043  			rk, re := mapaccessK(t, h, k)
  1044  			if rk == nil {
  1045  				continue // key has been deleted
  1046  			}
  1047  			it.key = rk
  1048  			it.elem = re
  1049  		}
  1050  		it.bucket = bucket
  1051  		if it.bptr != b { // avoid unnecessary write barrier; see issue 14921
  1052  			it.bptr = b
  1053  		}
  1054  		it.i = i + 1
  1055  		it.checkBucket = checkBucket
  1056  		return
  1057  	}
  1058  	b = b.overflow(t)
  1059  	i = 0
  1060  	goto next
  1061  }
  1062  
  1063  // mapclear deletes all keys from a map.
  1064  // It is called by the compiler.
  1065  //
  1066  // mapclear should be an internal detail,
  1067  // but widely used packages access it using linkname.
  1068  // Notable members of the hall of shame include:
  1069  //   - github.com/cloudwego/frugal
  1070  //
  1071  // Do not remove or change the type signature.
  1072  // See go.dev/issue/67401.
  1073  //
  1074  //go:linkname mapclear
  1075  func mapclear(t *maptype, h *hmap) {
  1076  	if raceenabled && h != nil {
  1077  		callerpc := getcallerpc()
  1078  		pc := abi.FuncPCABIInternal(mapclear)
  1079  		racewritepc(unsafe.Pointer(h), callerpc, pc)
  1080  	}
  1081  
  1082  	if h == nil || h.count == 0 {
  1083  		return
  1084  	}
  1085  
  1086  	if h.flags&hashWriting != 0 {
  1087  		fatal("concurrent map writes")
  1088  	}
  1089  
  1090  	h.flags ^= hashWriting
  1091  
  1092  	// Mark buckets empty, so existing iterators can be terminated, see issue #59411.
  1093  	markBucketsEmpty := func(bucket unsafe.Pointer, mask uintptr) {
  1094  		for i := uintptr(0); i <= mask; i++ {
  1095  			b := (*bmap)(add(bucket, i*uintptr(t.BucketSize)))
  1096  			for ; b != nil; b = b.overflow(t) {
  1097  				for i := uintptr(0); i < abi.MapBucketCount; i++ {
  1098  					b.tophash[i] = emptyRest
  1099  				}
  1100  			}
  1101  		}
  1102  	}
  1103  	markBucketsEmpty(h.buckets, bucketMask(h.B))
  1104  	if oldBuckets := h.oldbuckets; oldBuckets != nil {
  1105  		markBucketsEmpty(oldBuckets, h.oldbucketmask())
  1106  	}
  1107  
  1108  	h.flags &^= sameSizeGrow
  1109  	h.oldbuckets = nil
  1110  	h.nevacuate = 0
  1111  	h.noverflow = 0
  1112  	h.count = 0
  1113  
  1114  	// Reset the hash seed to make it more difficult for attackers to
  1115  	// repeatedly trigger hash collisions. See issue 25237.
  1116  	h.hash0 = uint32(rand())
  1117  
  1118  	// Keep the mapextra allocation but clear any extra information.
  1119  	if h.extra != nil {
  1120  		*h.extra = mapextra{}
  1121  	}
  1122  
  1123  	// makeBucketArray clears the memory pointed to by h.buckets
  1124  	// and recovers any overflow buckets by generating them
  1125  	// as if h.buckets was newly alloced.
  1126  	_, nextOverflow := makeBucketArray(t, h.B, h.buckets)
  1127  	if nextOverflow != nil {
  1128  		// If overflow buckets are created then h.extra
  1129  		// will have been allocated during initial bucket creation.
  1130  		h.extra.nextOverflow = nextOverflow
  1131  	}
  1132  
  1133  	if h.flags&hashWriting == 0 {
  1134  		fatal("concurrent map writes")
  1135  	}
  1136  	h.flags &^= hashWriting
  1137  }
  1138  
  1139  func hashGrow(t *maptype, h *hmap) {
  1140  	// If we've hit the load factor, get bigger.
  1141  	// Otherwise, there are too many overflow buckets,
  1142  	// so keep the same number of buckets and "grow" laterally.
  1143  	bigger := uint8(1)
  1144  	if !overLoadFactor(h.count+1, h.B) {
  1145  		bigger = 0
  1146  		h.flags |= sameSizeGrow
  1147  	}
  1148  	oldbuckets := h.buckets
  1149  	newbuckets, nextOverflow := makeBucketArray(t, h.B+bigger, nil)
  1150  
  1151  	flags := h.flags &^ (iterator | oldIterator)
  1152  	if h.flags&iterator != 0 {
  1153  		flags |= oldIterator
  1154  	}
  1155  	// commit the grow (atomic wrt gc)
  1156  	h.B += bigger
  1157  	h.flags = flags
  1158  	h.oldbuckets = oldbuckets
  1159  	h.buckets = newbuckets
  1160  	h.nevacuate = 0
  1161  	h.noverflow = 0
  1162  
  1163  	if h.extra != nil && h.extra.overflow != nil {
  1164  		// Promote current overflow buckets to the old generation.
  1165  		if h.extra.oldoverflow != nil {
  1166  			throw("oldoverflow is not nil")
  1167  		}
  1168  		h.extra.oldoverflow = h.extra.overflow
  1169  		h.extra.overflow = nil
  1170  	}
  1171  	if nextOverflow != nil {
  1172  		if h.extra == nil {
  1173  			h.extra = new(mapextra)
  1174  		}
  1175  		h.extra.nextOverflow = nextOverflow
  1176  	}
  1177  
  1178  	// the actual copying of the hash table data is done incrementally
  1179  	// by growWork() and evacuate().
  1180  }
  1181  
  1182  // overLoadFactor reports whether count items placed in 1<<B buckets is over loadFactor.
  1183  func overLoadFactor(count int, B uint8) bool {
  1184  	return count > abi.MapBucketCount && uintptr(count) > loadFactorNum*(bucketShift(B)/loadFactorDen)
  1185  }
  1186  
  1187  // tooManyOverflowBuckets reports whether noverflow buckets is too many for a map with 1<<B buckets.
  1188  // Note that most of these overflow buckets must be in sparse use;
  1189  // if use was dense, then we'd have already triggered regular map growth.
  1190  func tooManyOverflowBuckets(noverflow uint16, B uint8) bool {
  1191  	// If the threshold is too low, we do extraneous work.
  1192  	// If the threshold is too high, maps that grow and shrink can hold on to lots of unused memory.
  1193  	// "too many" means (approximately) as many overflow buckets as regular buckets.
  1194  	// See incrnoverflow for more details.
  1195  	if B > 15 {
  1196  		B = 15
  1197  	}
  1198  	// The compiler doesn't see here that B < 16; mask B to generate shorter shift code.
  1199  	return noverflow >= uint16(1)<<(B&15)
  1200  }
  1201  
  1202  // growing reports whether h is growing. The growth may be to the same size or bigger.
  1203  func (h *hmap) growing() bool {
  1204  	return h.oldbuckets != nil
  1205  }
  1206  
  1207  // sameSizeGrow reports whether the current growth is to a map of the same size.
  1208  func (h *hmap) sameSizeGrow() bool {
  1209  	return h.flags&sameSizeGrow != 0
  1210  }
  1211  
  1212  //go:linkname sameSizeGrowForIssue69110Test
  1213  func sameSizeGrowForIssue69110Test(h *hmap) bool {
  1214  	return h.sameSizeGrow()
  1215  }
  1216  
  1217  // noldbuckets calculates the number of buckets prior to the current map growth.
  1218  func (h *hmap) noldbuckets() uintptr {
  1219  	oldB := h.B
  1220  	if !h.sameSizeGrow() {
  1221  		oldB--
  1222  	}
  1223  	return bucketShift(oldB)
  1224  }
  1225  
  1226  // oldbucketmask provides a mask that can be applied to calculate n % noldbuckets().
  1227  func (h *hmap) oldbucketmask() uintptr {
  1228  	return h.noldbuckets() - 1
  1229  }
  1230  
  1231  func growWork(t *maptype, h *hmap, bucket uintptr) {
  1232  	// make sure we evacuate the oldbucket corresponding
  1233  	// to the bucket we're about to use
  1234  	evacuate(t, h, bucket&h.oldbucketmask())
  1235  
  1236  	// evacuate one more oldbucket to make progress on growing
  1237  	if h.growing() {
  1238  		evacuate(t, h, h.nevacuate)
  1239  	}
  1240  }
  1241  
  1242  func bucketEvacuated(t *maptype, h *hmap, bucket uintptr) bool {
  1243  	b := (*bmap)(add(h.oldbuckets, bucket*uintptr(t.BucketSize)))
  1244  	return evacuated(b)
  1245  }
  1246  
  1247  // evacDst is an evacuation destination.
  1248  type evacDst struct {
  1249  	b *bmap          // current destination bucket
  1250  	i int            // key/elem index into b
  1251  	k unsafe.Pointer // pointer to current key storage
  1252  	e unsafe.Pointer // pointer to current elem storage
  1253  }
  1254  
  1255  func evacuate(t *maptype, h *hmap, oldbucket uintptr) {
  1256  	b := (*bmap)(add(h.oldbuckets, oldbucket*uintptr(t.BucketSize)))
  1257  	newbit := h.noldbuckets()
  1258  	if !evacuated(b) {
  1259  		// TODO: reuse overflow buckets instead of using new ones, if there
  1260  		// is no iterator using the old buckets.  (If !oldIterator.)
  1261  
  1262  		// xy contains the x and y (low and high) evacuation destinations.
  1263  		var xy [2]evacDst
  1264  		x := &xy[0]
  1265  		x.b = (*bmap)(add(h.buckets, oldbucket*uintptr(t.BucketSize)))
  1266  		x.k = add(unsafe.Pointer(x.b), dataOffset)
  1267  		x.e = add(x.k, abi.MapBucketCount*uintptr(t.KeySize))
  1268  
  1269  		if !h.sameSizeGrow() {
  1270  			// Only calculate y pointers if we're growing bigger.
  1271  			// Otherwise GC can see bad pointers.
  1272  			y := &xy[1]
  1273  			y.b = (*bmap)(add(h.buckets, (oldbucket+newbit)*uintptr(t.BucketSize)))
  1274  			y.k = add(unsafe.Pointer(y.b), dataOffset)
  1275  			y.e = add(y.k, abi.MapBucketCount*uintptr(t.KeySize))
  1276  		}
  1277  
  1278  		for ; b != nil; b = b.overflow(t) {
  1279  			k := add(unsafe.Pointer(b), dataOffset)
  1280  			e := add(k, abi.MapBucketCount*uintptr(t.KeySize))
  1281  			for i := 0; i < abi.MapBucketCount; i, k, e = i+1, add(k, uintptr(t.KeySize)), add(e, uintptr(t.ValueSize)) {
  1282  				top := b.tophash[i]
  1283  				if isEmpty(top) {
  1284  					b.tophash[i] = evacuatedEmpty
  1285  					continue
  1286  				}
  1287  				if top < minTopHash {
  1288  					throw("bad map state")
  1289  				}
  1290  				k2 := k
  1291  				if t.IndirectKey() {
  1292  					k2 = *((*unsafe.Pointer)(k2))
  1293  				}
  1294  				var useY uint8
  1295  				if !h.sameSizeGrow() {
  1296  					// Compute hash to make our evacuation decision (whether we need
  1297  					// to send this key/elem to bucket x or bucket y).
  1298  					hash := t.Hasher(k2, uintptr(h.hash0))
  1299  					if h.flags&iterator != 0 && !t.ReflexiveKey() && !t.Key.Equal(k2, k2) {
  1300  						// If key != key (NaNs), then the hash could be (and probably
  1301  						// will be) entirely different from the old hash. Moreover,
  1302  						// it isn't reproducible. Reproducibility is required in the
  1303  						// presence of iterators, as our evacuation decision must
  1304  						// match whatever decision the iterator made.
  1305  						// Fortunately, we have the freedom to send these keys either
  1306  						// way. Also, tophash is meaningless for these kinds of keys.
  1307  						// We let the low bit of tophash drive the evacuation decision.
  1308  						// We recompute a new random tophash for the next level so
  1309  						// these keys will get evenly distributed across all buckets
  1310  						// after multiple grows.
  1311  						useY = top & 1
  1312  						top = tophash(hash)
  1313  					} else {
  1314  						if hash&newbit != 0 {
  1315  							useY = 1
  1316  						}
  1317  					}
  1318  				}
  1319  
  1320  				if evacuatedX+1 != evacuatedY || evacuatedX^1 != evacuatedY {
  1321  					throw("bad evacuatedN")
  1322  				}
  1323  
  1324  				b.tophash[i] = evacuatedX + useY // evacuatedX + 1 == evacuatedY
  1325  				dst := &xy[useY]                 // evacuation destination
  1326  
  1327  				if dst.i == abi.MapBucketCount {
  1328  					dst.b = h.newoverflow(t, dst.b)
  1329  					dst.i = 0
  1330  					dst.k = add(unsafe.Pointer(dst.b), dataOffset)
  1331  					dst.e = add(dst.k, abi.MapBucketCount*uintptr(t.KeySize))
  1332  				}
  1333  				dst.b.tophash[dst.i&(abi.MapBucketCount-1)] = top // mask dst.i as an optimization, to avoid a bounds check
  1334  				if t.IndirectKey() {
  1335  					*(*unsafe.Pointer)(dst.k) = k2 // copy pointer
  1336  				} else {
  1337  					typedmemmove(t.Key, dst.k, k) // copy elem
  1338  				}
  1339  				if t.IndirectElem() {
  1340  					*(*unsafe.Pointer)(dst.e) = *(*unsafe.Pointer)(e)
  1341  				} else {
  1342  					typedmemmove(t.Elem, dst.e, e)
  1343  				}
  1344  				dst.i++
  1345  				// These updates might push these pointers past the end of the
  1346  				// key or elem arrays.  That's ok, as we have the overflow pointer
  1347  				// at the end of the bucket to protect against pointing past the
  1348  				// end of the bucket.
  1349  				dst.k = add(dst.k, uintptr(t.KeySize))
  1350  				dst.e = add(dst.e, uintptr(t.ValueSize))
  1351  			}
  1352  		}
  1353  		// Unlink the overflow buckets & clear key/elem to help GC.
  1354  		if h.flags&oldIterator == 0 && t.Bucket.Pointers() {
  1355  			b := add(h.oldbuckets, oldbucket*uintptr(t.BucketSize))
  1356  			// Preserve b.tophash because the evacuation
  1357  			// state is maintained there.
  1358  			ptr := add(b, dataOffset)
  1359  			n := uintptr(t.BucketSize) - dataOffset
  1360  			memclrHasPointers(ptr, n)
  1361  		}
  1362  	}
  1363  
  1364  	if oldbucket == h.nevacuate {
  1365  		advanceEvacuationMark(h, t, newbit)
  1366  	}
  1367  }
  1368  
  1369  func advanceEvacuationMark(h *hmap, t *maptype, newbit uintptr) {
  1370  	h.nevacuate++
  1371  	// Experiments suggest that 1024 is overkill by at least an order of magnitude.
  1372  	// Put it in there as a safeguard anyway, to ensure O(1) behavior.
  1373  	stop := h.nevacuate + 1024
  1374  	if stop > newbit {
  1375  		stop = newbit
  1376  	}
  1377  	for h.nevacuate != stop && bucketEvacuated(t, h, h.nevacuate) {
  1378  		h.nevacuate++
  1379  	}
  1380  	if h.nevacuate == newbit { // newbit == # of oldbuckets
  1381  		// Growing is all done. Free old main bucket array.
  1382  		h.oldbuckets = nil
  1383  		// Can discard old overflow buckets as well.
  1384  		// If they are still referenced by an iterator,
  1385  		// then the iterator holds a pointers to the slice.
  1386  		if h.extra != nil {
  1387  			h.extra.oldoverflow = nil
  1388  		}
  1389  		h.flags &^= sameSizeGrow
  1390  	}
  1391  }
  1392  
  1393  // Reflect stubs. Called from ../reflect/asm_*.s
  1394  
  1395  // reflect_makemap is for package reflect,
  1396  // but widely used packages access it using linkname.
  1397  // Notable members of the hall of shame include:
  1398  //   - gitee.com/quant1x/gox
  1399  //   - github.com/modern-go/reflect2
  1400  //   - github.com/goccy/go-json
  1401  //   - github.com/RomiChan/protobuf
  1402  //   - github.com/segmentio/encoding
  1403  //   - github.com/v2pro/plz
  1404  //
  1405  // Do not remove or change the type signature.
  1406  // See go.dev/issue/67401.
  1407  //
  1408  //go:linkname reflect_makemap reflect.makemap
  1409  func reflect_makemap(t *maptype, cap int) *hmap {
  1410  	// Check invariants and reflects math.
  1411  	if t.Key.Equal == nil {
  1412  		throw("runtime.reflect_makemap: unsupported map key type")
  1413  	}
  1414  	if t.Key.Size_ > abi.MapMaxKeyBytes && (!t.IndirectKey() || t.KeySize != uint8(goarch.PtrSize)) ||
  1415  		t.Key.Size_ <= abi.MapMaxKeyBytes && (t.IndirectKey() || t.KeySize != uint8(t.Key.Size_)) {
  1416  		throw("key size wrong")
  1417  	}
  1418  	if t.Elem.Size_ > abi.MapMaxElemBytes && (!t.IndirectElem() || t.ValueSize != uint8(goarch.PtrSize)) ||
  1419  		t.Elem.Size_ <= abi.MapMaxElemBytes && (t.IndirectElem() || t.ValueSize != uint8(t.Elem.Size_)) {
  1420  		throw("elem size wrong")
  1421  	}
  1422  	if t.Key.Align_ > abi.MapBucketCount {
  1423  		throw("key align too big")
  1424  	}
  1425  	if t.Elem.Align_ > abi.MapBucketCount {
  1426  		throw("elem align too big")
  1427  	}
  1428  	if t.Key.Size_%uintptr(t.Key.Align_) != 0 {
  1429  		throw("key size not a multiple of key align")
  1430  	}
  1431  	if t.Elem.Size_%uintptr(t.Elem.Align_) != 0 {
  1432  		throw("elem size not a multiple of elem align")
  1433  	}
  1434  	if abi.MapBucketCount < 8 {
  1435  		throw("bucketsize too small for proper alignment")
  1436  	}
  1437  	if dataOffset%uintptr(t.Key.Align_) != 0 {
  1438  		throw("need padding in bucket (key)")
  1439  	}
  1440  	if dataOffset%uintptr(t.Elem.Align_) != 0 {
  1441  		throw("need padding in bucket (elem)")
  1442  	}
  1443  
  1444  	return makemap(t, cap, nil)
  1445  }
  1446  
  1447  // reflect_mapaccess is for package reflect,
  1448  // but widely used packages access it using linkname.
  1449  // Notable members of the hall of shame include:
  1450  //   - gitee.com/quant1x/gox
  1451  //   - github.com/modern-go/reflect2
  1452  //   - github.com/v2pro/plz
  1453  //
  1454  // Do not remove or change the type signature.
  1455  // See go.dev/issue/67401.
  1456  //
  1457  //go:linkname reflect_mapaccess reflect.mapaccess
  1458  func reflect_mapaccess(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer {
  1459  	elem, ok := mapaccess2(t, h, key)
  1460  	if !ok {
  1461  		// reflect wants nil for a missing element
  1462  		elem = nil
  1463  	}
  1464  	return elem
  1465  }
  1466  
  1467  //go:linkname reflect_mapaccess_faststr reflect.mapaccess_faststr
  1468  func reflect_mapaccess_faststr(t *maptype, h *hmap, key string) unsafe.Pointer {
  1469  	elem, ok := mapaccess2_faststr(t, h, key)
  1470  	if !ok {
  1471  		// reflect wants nil for a missing element
  1472  		elem = nil
  1473  	}
  1474  	return elem
  1475  }
  1476  
  1477  // reflect_mapassign is for package reflect,
  1478  // but widely used packages access it using linkname.
  1479  // Notable members of the hall of shame include:
  1480  //   - gitee.com/quant1x/gox
  1481  //   - github.com/v2pro/plz
  1482  //
  1483  // Do not remove or change the type signature.
  1484  //
  1485  //go:linkname reflect_mapassign reflect.mapassign0
  1486  func reflect_mapassign(t *maptype, h *hmap, key unsafe.Pointer, elem unsafe.Pointer) {
  1487  	p := mapassign(t, h, key)
  1488  	typedmemmove(t.Elem, p, elem)
  1489  }
  1490  
  1491  //go:linkname reflect_mapassign_faststr reflect.mapassign_faststr0
  1492  func reflect_mapassign_faststr(t *maptype, h *hmap, key string, elem unsafe.Pointer) {
  1493  	p := mapassign_faststr(t, h, key)
  1494  	typedmemmove(t.Elem, p, elem)
  1495  }
  1496  
  1497  //go:linkname reflect_mapdelete reflect.mapdelete
  1498  func reflect_mapdelete(t *maptype, h *hmap, key unsafe.Pointer) {
  1499  	mapdelete(t, h, key)
  1500  }
  1501  
  1502  //go:linkname reflect_mapdelete_faststr reflect.mapdelete_faststr
  1503  func reflect_mapdelete_faststr(t *maptype, h *hmap, key string) {
  1504  	mapdelete_faststr(t, h, key)
  1505  }
  1506  
  1507  // reflect_mapiterinit is for package reflect,
  1508  // but widely used packages access it using linkname.
  1509  // Notable members of the hall of shame include:
  1510  //   - github.com/modern-go/reflect2
  1511  //   - gitee.com/quant1x/gox
  1512  //   - github.com/v2pro/plz
  1513  //   - github.com/wI2L/jettison
  1514  //
  1515  // Do not remove or change the type signature.
  1516  // See go.dev/issue/67401.
  1517  //
  1518  //go:linkname reflect_mapiterinit reflect.mapiterinit
  1519  func reflect_mapiterinit(t *maptype, h *hmap, it *hiter) {
  1520  	mapiterinit(t, h, it)
  1521  }
  1522  
  1523  // reflect_mapiternext is for package reflect,
  1524  // but widely used packages access it using linkname.
  1525  // Notable members of the hall of shame include:
  1526  //   - gitee.com/quant1x/gox
  1527  //   - github.com/modern-go/reflect2
  1528  //   - github.com/goccy/go-json
  1529  //   - github.com/v2pro/plz
  1530  //   - github.com/wI2L/jettison
  1531  //
  1532  // Do not remove or change the type signature.
  1533  // See go.dev/issue/67401.
  1534  //
  1535  //go:linkname reflect_mapiternext reflect.mapiternext
  1536  func reflect_mapiternext(it *hiter) {
  1537  	mapiternext(it)
  1538  }
  1539  
  1540  // reflect_mapiterkey is for package reflect,
  1541  // but widely used packages access it using linkname.
  1542  // Notable members of the hall of shame include:
  1543  //   - github.com/goccy/go-json
  1544  //   - gonum.org/v1/gonum
  1545  //
  1546  // Do not remove or change the type signature.
  1547  // See go.dev/issue/67401.
  1548  //
  1549  //go:linkname reflect_mapiterkey reflect.mapiterkey
  1550  func reflect_mapiterkey(it *hiter) unsafe.Pointer {
  1551  	return it.key
  1552  }
  1553  
  1554  // reflect_mapiterelem is for package reflect,
  1555  // but widely used packages access it using linkname.
  1556  // Notable members of the hall of shame include:
  1557  //   - github.com/goccy/go-json
  1558  //   - gonum.org/v1/gonum
  1559  //
  1560  // Do not remove or change the type signature.
  1561  // See go.dev/issue/67401.
  1562  //
  1563  //go:linkname reflect_mapiterelem reflect.mapiterelem
  1564  func reflect_mapiterelem(it *hiter) unsafe.Pointer {
  1565  	return it.elem
  1566  }
  1567  
  1568  // reflect_maplen is for package reflect,
  1569  // but widely used packages access it using linkname.
  1570  // Notable members of the hall of shame include:
  1571  //   - github.com/goccy/go-json
  1572  //   - github.com/wI2L/jettison
  1573  //
  1574  // Do not remove or change the type signature.
  1575  // See go.dev/issue/67401.
  1576  //
  1577  //go:linkname reflect_maplen reflect.maplen
  1578  func reflect_maplen(h *hmap) int {
  1579  	if h == nil {
  1580  		return 0
  1581  	}
  1582  	if raceenabled {
  1583  		callerpc := getcallerpc()
  1584  		racereadpc(unsafe.Pointer(h), callerpc, abi.FuncPCABIInternal(reflect_maplen))
  1585  	}
  1586  	return h.count
  1587  }
  1588  
  1589  //go:linkname reflect_mapclear reflect.mapclear
  1590  func reflect_mapclear(t *maptype, h *hmap) {
  1591  	mapclear(t, h)
  1592  }
  1593  
  1594  //go:linkname reflectlite_maplen internal/reflectlite.maplen
  1595  func reflectlite_maplen(h *hmap) int {
  1596  	if h == nil {
  1597  		return 0
  1598  	}
  1599  	if raceenabled {
  1600  		callerpc := getcallerpc()
  1601  		racereadpc(unsafe.Pointer(h), callerpc, abi.FuncPCABIInternal(reflect_maplen))
  1602  	}
  1603  	return h.count
  1604  }
  1605  
  1606  // mapinitnoop is a no-op function known the Go linker; if a given global
  1607  // map (of the right size) is determined to be dead, the linker will
  1608  // rewrite the relocation (from the package init func) from the outlined
  1609  // map init function to this symbol. Defined in assembly so as to avoid
  1610  // complications with instrumentation (coverage, etc).
  1611  func mapinitnoop()
  1612  
  1613  // mapclone for implementing maps.Clone
  1614  //
  1615  //go:linkname mapclone maps.clone
  1616  func mapclone(m any) any {
  1617  	e := efaceOf(&m)
  1618  	e.data = unsafe.Pointer(mapclone2((*maptype)(unsafe.Pointer(e._type)), (*hmap)(e.data)))
  1619  	return m
  1620  }
  1621  
  1622  // moveToBmap moves a bucket from src to dst. It returns the destination bucket or new destination bucket if it overflows
  1623  // and the pos that the next key/value will be written, if pos == bucketCnt means needs to written in overflow bucket.
  1624  func moveToBmap(t *maptype, h *hmap, dst *bmap, pos int, src *bmap) (*bmap, int) {
  1625  	for i := 0; i < abi.MapBucketCount; i++ {
  1626  		if isEmpty(src.tophash[i]) {
  1627  			continue
  1628  		}
  1629  
  1630  		for ; pos < abi.MapBucketCount; pos++ {
  1631  			if isEmpty(dst.tophash[pos]) {
  1632  				break
  1633  			}
  1634  		}
  1635  
  1636  		if pos == abi.MapBucketCount {
  1637  			dst = h.newoverflow(t, dst)
  1638  			pos = 0
  1639  		}
  1640  
  1641  		srcK := add(unsafe.Pointer(src), dataOffset+uintptr(i)*uintptr(t.KeySize))
  1642  		srcEle := add(unsafe.Pointer(src), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+uintptr(i)*uintptr(t.ValueSize))
  1643  		dstK := add(unsafe.Pointer(dst), dataOffset+uintptr(pos)*uintptr(t.KeySize))
  1644  		dstEle := add(unsafe.Pointer(dst), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+uintptr(pos)*uintptr(t.ValueSize))
  1645  
  1646  		dst.tophash[pos] = src.tophash[i]
  1647  		if t.IndirectKey() {
  1648  			srcK = *(*unsafe.Pointer)(srcK)
  1649  			if t.NeedKeyUpdate() {
  1650  				kStore := newobject(t.Key)
  1651  				typedmemmove(t.Key, kStore, srcK)
  1652  				srcK = kStore
  1653  			}
  1654  			// Note: if NeedKeyUpdate is false, then the memory
  1655  			// used to store the key is immutable, so we can share
  1656  			// it between the original map and its clone.
  1657  			*(*unsafe.Pointer)(dstK) = srcK
  1658  		} else {
  1659  			typedmemmove(t.Key, dstK, srcK)
  1660  		}
  1661  		if t.IndirectElem() {
  1662  			srcEle = *(*unsafe.Pointer)(srcEle)
  1663  			eStore := newobject(t.Elem)
  1664  			typedmemmove(t.Elem, eStore, srcEle)
  1665  			*(*unsafe.Pointer)(dstEle) = eStore
  1666  		} else {
  1667  			typedmemmove(t.Elem, dstEle, srcEle)
  1668  		}
  1669  		pos++
  1670  		h.count++
  1671  	}
  1672  	return dst, pos
  1673  }
  1674  
  1675  func mapclone2(t *maptype, src *hmap) *hmap {
  1676  	hint := src.count
  1677  	if overLoadFactor(hint, src.B) {
  1678  		// Note: in rare cases (e.g. during a same-sized grow) the map
  1679  		// can be overloaded. Make sure we don't allocate a destination
  1680  		// bucket array larger than the source bucket array.
  1681  		// This will cause the cloned map to be overloaded also,
  1682  		// but that's better than crashing. See issue 69110.
  1683  		hint = int(loadFactorNum * (bucketShift(src.B) / loadFactorDen))
  1684  	}
  1685  	dst := makemap(t, hint, nil)
  1686  	dst.hash0 = src.hash0
  1687  	dst.nevacuate = 0
  1688  	// flags do not need to be copied here, just like a new map has no flags.
  1689  
  1690  	if src.count == 0 {
  1691  		return dst
  1692  	}
  1693  
  1694  	if src.flags&hashWriting != 0 {
  1695  		fatal("concurrent map clone and map write")
  1696  	}
  1697  
  1698  	if src.B == 0 && !(t.IndirectKey() && t.NeedKeyUpdate()) && !t.IndirectElem() {
  1699  		// Quick copy for small maps.
  1700  		dst.buckets = newobject(t.Bucket)
  1701  		dst.count = src.count
  1702  		typedmemmove(t.Bucket, dst.buckets, src.buckets)
  1703  		return dst
  1704  	}
  1705  
  1706  	if dst.B == 0 {
  1707  		dst.buckets = newobject(t.Bucket)
  1708  	}
  1709  	dstArraySize := int(bucketShift(dst.B))
  1710  	srcArraySize := int(bucketShift(src.B))
  1711  	for i := 0; i < dstArraySize; i++ {
  1712  		dstBmap := (*bmap)(add(dst.buckets, uintptr(i*int(t.BucketSize))))
  1713  		pos := 0
  1714  		for j := 0; j < srcArraySize; j += dstArraySize {
  1715  			srcBmap := (*bmap)(add(src.buckets, uintptr((i+j)*int(t.BucketSize))))
  1716  			for srcBmap != nil {
  1717  				dstBmap, pos = moveToBmap(t, dst, dstBmap, pos, srcBmap)
  1718  				srcBmap = srcBmap.overflow(t)
  1719  			}
  1720  		}
  1721  	}
  1722  
  1723  	if src.oldbuckets == nil {
  1724  		return dst
  1725  	}
  1726  
  1727  	oldB := src.B
  1728  	srcOldbuckets := src.oldbuckets
  1729  	if !src.sameSizeGrow() {
  1730  		oldB--
  1731  	}
  1732  	oldSrcArraySize := int(bucketShift(oldB))
  1733  
  1734  	for i := 0; i < oldSrcArraySize; i++ {
  1735  		srcBmap := (*bmap)(add(srcOldbuckets, uintptr(i*int(t.BucketSize))))
  1736  		if evacuated(srcBmap) {
  1737  			continue
  1738  		}
  1739  
  1740  		if oldB >= dst.B { // main bucket bits in dst is less than oldB bits in src
  1741  			dstBmap := (*bmap)(add(dst.buckets, (uintptr(i)&bucketMask(dst.B))*uintptr(t.BucketSize)))
  1742  			for dstBmap.overflow(t) != nil {
  1743  				dstBmap = dstBmap.overflow(t)
  1744  			}
  1745  			pos := 0
  1746  			for srcBmap != nil {
  1747  				dstBmap, pos = moveToBmap(t, dst, dstBmap, pos, srcBmap)
  1748  				srcBmap = srcBmap.overflow(t)
  1749  			}
  1750  			continue
  1751  		}
  1752  
  1753  		// oldB < dst.B, so a single source bucket may go to multiple destination buckets.
  1754  		// Process entries one at a time.
  1755  		for srcBmap != nil {
  1756  			// move from oldBlucket to new bucket
  1757  			for i := uintptr(0); i < abi.MapBucketCount; i++ {
  1758  				if isEmpty(srcBmap.tophash[i]) {
  1759  					continue
  1760  				}
  1761  
  1762  				if src.flags&hashWriting != 0 {
  1763  					fatal("concurrent map clone and map write")
  1764  				}
  1765  
  1766  				srcK := add(unsafe.Pointer(srcBmap), dataOffset+i*uintptr(t.KeySize))
  1767  				if t.IndirectKey() {
  1768  					srcK = *((*unsafe.Pointer)(srcK))
  1769  				}
  1770  
  1771  				srcEle := add(unsafe.Pointer(srcBmap), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+i*uintptr(t.ValueSize))
  1772  				if t.IndirectElem() {
  1773  					srcEle = *((*unsafe.Pointer)(srcEle))
  1774  				}
  1775  				dstEle := mapassign(t, dst, srcK)
  1776  				typedmemmove(t.Elem, dstEle, srcEle)
  1777  			}
  1778  			srcBmap = srcBmap.overflow(t)
  1779  		}
  1780  	}
  1781  	return dst
  1782  }
  1783  
  1784  // keys for implementing maps.keys
  1785  //
  1786  //go:linkname keys maps.keys
  1787  func keys(m any, p unsafe.Pointer) {
  1788  	e := efaceOf(&m)
  1789  	t := (*maptype)(unsafe.Pointer(e._type))
  1790  	h := (*hmap)(e.data)
  1791  
  1792  	if h == nil || h.count == 0 {
  1793  		return
  1794  	}
  1795  	s := (*slice)(p)
  1796  	r := int(rand())
  1797  	offset := uint8(r >> h.B & (abi.MapBucketCount - 1))
  1798  	if h.B == 0 {
  1799  		copyKeys(t, h, (*bmap)(h.buckets), s, offset)
  1800  		return
  1801  	}
  1802  	arraySize := int(bucketShift(h.B))
  1803  	buckets := h.buckets
  1804  	for i := 0; i < arraySize; i++ {
  1805  		bucket := (i + r) & (arraySize - 1)
  1806  		b := (*bmap)(add(buckets, uintptr(bucket)*uintptr(t.BucketSize)))
  1807  		copyKeys(t, h, b, s, offset)
  1808  	}
  1809  
  1810  	if h.growing() {
  1811  		oldArraySize := int(h.noldbuckets())
  1812  		for i := 0; i < oldArraySize; i++ {
  1813  			bucket := (i + r) & (oldArraySize - 1)
  1814  			b := (*bmap)(add(h.oldbuckets, uintptr(bucket)*uintptr(t.BucketSize)))
  1815  			if evacuated(b) {
  1816  				continue
  1817  			}
  1818  			copyKeys(t, h, b, s, offset)
  1819  		}
  1820  	}
  1821  	return
  1822  }
  1823  
  1824  func copyKeys(t *maptype, h *hmap, b *bmap, s *slice, offset uint8) {
  1825  	for b != nil {
  1826  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
  1827  			offi := (i + uintptr(offset)) & (abi.MapBucketCount - 1)
  1828  			if isEmpty(b.tophash[offi]) {
  1829  				continue
  1830  			}
  1831  			if h.flags&hashWriting != 0 {
  1832  				fatal("concurrent map read and map write")
  1833  			}
  1834  			k := add(unsafe.Pointer(b), dataOffset+offi*uintptr(t.KeySize))
  1835  			if t.IndirectKey() {
  1836  				k = *((*unsafe.Pointer)(k))
  1837  			}
  1838  			if s.len >= s.cap {
  1839  				fatal("concurrent map read and map write")
  1840  			}
  1841  			typedmemmove(t.Key, add(s.array, uintptr(s.len)*uintptr(t.Key.Size())), k)
  1842  			s.len++
  1843  		}
  1844  		b = b.overflow(t)
  1845  	}
  1846  }
  1847  
  1848  // values for implementing maps.values
  1849  //
  1850  //go:linkname values maps.values
  1851  func values(m any, p unsafe.Pointer) {
  1852  	e := efaceOf(&m)
  1853  	t := (*maptype)(unsafe.Pointer(e._type))
  1854  	h := (*hmap)(e.data)
  1855  	if h == nil || h.count == 0 {
  1856  		return
  1857  	}
  1858  	s := (*slice)(p)
  1859  	r := int(rand())
  1860  	offset := uint8(r >> h.B & (abi.MapBucketCount - 1))
  1861  	if h.B == 0 {
  1862  		copyValues(t, h, (*bmap)(h.buckets), s, offset)
  1863  		return
  1864  	}
  1865  	arraySize := int(bucketShift(h.B))
  1866  	buckets := h.buckets
  1867  	for i := 0; i < arraySize; i++ {
  1868  		bucket := (i + r) & (arraySize - 1)
  1869  		b := (*bmap)(add(buckets, uintptr(bucket)*uintptr(t.BucketSize)))
  1870  		copyValues(t, h, b, s, offset)
  1871  	}
  1872  
  1873  	if h.growing() {
  1874  		oldArraySize := int(h.noldbuckets())
  1875  		for i := 0; i < oldArraySize; i++ {
  1876  			bucket := (i + r) & (oldArraySize - 1)
  1877  			b := (*bmap)(add(h.oldbuckets, uintptr(bucket)*uintptr(t.BucketSize)))
  1878  			if evacuated(b) {
  1879  				continue
  1880  			}
  1881  			copyValues(t, h, b, s, offset)
  1882  		}
  1883  	}
  1884  	return
  1885  }
  1886  
  1887  func copyValues(t *maptype, h *hmap, b *bmap, s *slice, offset uint8) {
  1888  	for b != nil {
  1889  		for i := uintptr(0); i < abi.MapBucketCount; i++ {
  1890  			offi := (i + uintptr(offset)) & (abi.MapBucketCount - 1)
  1891  			if isEmpty(b.tophash[offi]) {
  1892  				continue
  1893  			}
  1894  
  1895  			if h.flags&hashWriting != 0 {
  1896  				fatal("concurrent map read and map write")
  1897  			}
  1898  
  1899  			ele := add(unsafe.Pointer(b), dataOffset+abi.MapBucketCount*uintptr(t.KeySize)+offi*uintptr(t.ValueSize))
  1900  			if t.IndirectElem() {
  1901  				ele = *((*unsafe.Pointer)(ele))
  1902  			}
  1903  			if s.len >= s.cap {
  1904  				fatal("concurrent map read and map write")
  1905  			}
  1906  			typedmemmove(t.Elem, add(s.array, uintptr(s.len)*uintptr(t.Elem.Size())), ele)
  1907  			s.len++
  1908  		}
  1909  		b = b.overflow(t)
  1910  	}
  1911  }
  1912
View as plain text