runtime: implement growing hashmaps

Add support for growing hashmaps beyond their initial size.

src/runtime/hashmap.go

@@ -89,13 +89,20 @@ func hashmapLen(m *hashmap) int {
 // Set a specified key to a given value. Grow the map if necessary.
 //go:nobounds
 func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint32, keyEqual func(x, y unsafe.Pointer, n uintptr) bool) {
+	tophash := hashmapTopHash(hash)
+
+	if m.buckets == nil {
+		// No bucket was allocated yet, do so now.
+		m.buckets = unsafe.Pointer(hashmapInsertIntoNewBucket(m, key, value, tophash))
+		return
+	}
+
 	numBuckets := uintptr(1) << m.bucketBits
 	bucketNumber := (uintptr(hash) & (numBuckets - 1))
 	bucketSize := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*8
 	bucketAddr := uintptr(m.buckets) + bucketSize*bucketNumber
 	bucket := (*hashmapBucket)(unsafe.Pointer(bucketAddr))
-
+	var lastBucket *hashmapBucket
-	tophash := hashmapTopHash(hash)
 
 	// See whether the key already exists somewhere.
 	var emptySlotKey unsafe.Pointer
@@ -104,9 +111,9 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 	for bucket != nil {
 		for i := uintptr(0); i < 8; i++ {
 			slotKeyOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*uintptr(i)
-			slotKey := unsafe.Pointer(bucketAddr + slotKeyOffset)
+			slotKey := unsafe.Pointer(uintptr(unsafe.Pointer(bucket)) + slotKeyOffset)
 			slotValueOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*uintptr(i)
-			slotValue := unsafe.Pointer(bucketAddr + slotValueOffset)
+			slotValue := unsafe.Pointer(uintptr(unsafe.Pointer(bucket)) + slotValueOffset)
 			if bucket.tophash[i] == 0 && emptySlotKey == nil {
 				// Found an empty slot, store it for if we couldn't find an
 				// existing slot.
@@ -115,7 +122,7 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 				emptySlotTophash = &bucket.tophash[i]
 			}
 			if bucket.tophash[i] == tophash {
-				// Could be an existing value that's the same.
+				// Could be an existing key that's the same.
 				if keyEqual(key, slotKey, uintptr(m.keySize)) {
 					// found same key, replace it
 					memcpy(slotValue, value, uintptr(m.valueSize))
@@ -123,16 +130,37 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 				}
 			}
 		}
+		lastBucket = bucket
 		bucket = bucket.next
 	}
-	if emptySlotKey != nil {
+	if emptySlotKey == nil {
-		m.count++
+		// Add a new bucket to the bucket chain.
-		memcpy(emptySlotKey, key, uintptr(m.keySize))
+		// TODO: rebalance if necessary to avoid O(n) insert and lookup time.
-		memcpy(emptySlotValue, value, uintptr(m.valueSize))
+		lastBucket.next = (*hashmapBucket)(hashmapInsertIntoNewBucket(m, key, value, tophash))
-		*emptySlotTophash = tophash
 		return
 	}
-	panic("todo: hashmap: grow bucket")
+	m.count++
+	memcpy(emptySlotKey, key, uintptr(m.keySize))
+	memcpy(emptySlotValue, value, uintptr(m.valueSize))
+	*emptySlotTophash = tophash
+}
+
+// hashmapInsertIntoNewBucket creates a new bucket, inserts the given key and
+// value into the bucket, and returns a pointer to this bucket.
+func hashmapInsertIntoNewBucket(m *hashmap, key, value unsafe.Pointer, tophash uint8) *hashmapBucket {
+	bucketBufSize := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*8
+	bucketBuf := alloc(bucketBufSize)
+	// Insert into the first slot, which is empty as it has just been allocated.
+	slotKeyOffset := unsafe.Sizeof(hashmapBucket{})
+	slotKey := unsafe.Pointer(uintptr(bucketBuf) + slotKeyOffset)
+	slotValueOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8
+	slotValue := unsafe.Pointer(uintptr(bucketBuf) + slotValueOffset)
+	m.count++
+	memcpy(slotKey, key, uintptr(m.keySize))
+	memcpy(slotValue, value, uintptr(m.valueSize))
+	bucket := (*hashmapBucket)(bucketBuf)
+	bucket.tophash[0] = tophash
+	return bucket
 }
 
 // Get the value of a specified key, or zero the value if not found.

testdata/map.go

@@ -50,15 +50,15 @@ func main() {
 
 	// test preallocated map
 	squares := make(map[int]int, 200)
-	for i := 0; i < 100; i++ {
+	testBigMap(squares, 100)
-		squares[i] = i*i
-		for j := 0; j <= i; j++ {
-			if v := squares[j]; v != j*j {
-				println("unexpected value read back from squares map:", j, v)
-			}
-		}
-	}
 	println("tested preallocated map")
+
+	// test growing maps
+	squares = make(map[int]int, 0)
+	testBigMap(squares, 10)
+	squares = make(map[int]int, 20)
+	testBigMap(squares, 40)
+	println("tested growing of a map")
 }
 
 func readMap(m map[string]int, key string) {
@@ -73,3 +73,22 @@ func lookup(m map[string]int, key string) {
 	value, ok := m[key]
 	println("lookup with comma-ok:", key, value, ok)
 }
+
+func testBigMap(squares map[int]int, n int) {
+	for i := 0; i < n; i++ {
+		if len(squares) != i {
+			println("unexpected length:", len(squares), "at i =", i)
+		}
+		squares[i] = i*i
+		for j := 0; j <= i; j++ {
+			if v, ok := squares[j]; !ok || v != j*j {
+				if !ok {
+					println("key not found in squares map:", j)
+				} else {
+					println("unexpected value read back from squares map:", j, v)
+				}
+				return
+			}
+		}
+	}
+}

testdata/map.txt

@@ -55,3 +55,4 @@ true false 0
 4321
 5555
 tested preallocated map
+tested growing of a map