so/maps: fix initial size calc

Author: nalgeon

bench/README.md

@@ -46,19 +46,21 @@ So's built-in map is faster than Go's, but I wouldn't call it the winner because
 
 | Benchmark |      Go | So (mimalloc) | So (arena) | So (built-in) | Winner        |
 | --------- | ------: | ------------: | ---------: | ------------: | ------------- |
-| Set       | 35580ns |       56696ns |    57661ns |        3242ns | Go - 0.6x     |
+| Set       | 35580ns |       56696ns |    57661ns |           n/a | Go - 0.6x     |
+| Set (pre) |  9608ns |        8821ns |     8767ns |        3242ns | ~same         |
 | Get       |  5573ns |        1638ns |     1583ns |        2733ns | **So** - 3.4x |
 | Delete    | 23892ns |       38556ns |    38821ns |           n/a | Go - 0.6x     |
 
 ### String keys
 
 So lookups are on par with Go, while modifications are 1.5x slower.
 
-| Benchmark |      Go | So (mimalloc) | So (arena) | So (built-in) | Winner    |
-| --------- | ------: | ------------: | ---------: | ------------: | --------- |
-| Set       | 48677ns |       71879ns |    63500ns |        6970ns | Go - 0.7x |
-| Get       |  8990ns |       10206ns |    10083ns |       10735ns | Go - 0.9x |
-| Delete    | 33878ns |       50111ns |    49507ns |           n/a | Go - 0.7x |
+| Benchmark |      Go | So (mimalloc) | So (arena) | So (built-in) | Winner        |
+| --------- | ------: | ------------: | ---------: | ------------: | ------------- |
+| Set       | 48677ns |       71879ns |    63500ns |           n/a | Go - 0.7x     |
+| Set (pre) | 14620ns |       12313ns |    12115ns |        6970ns | **So** - 1.2x |
+| Get       |  8990ns |       10206ns |    10083ns |       10735ns | Go - 0.9x     |
+| Delete    | 33878ns |       50111ns |    49507ns |           n/a | Go - 0.7x     |
 
 Apple M1 • Go 1.26.1 • [details](./maps/README.md)
 

bench/maps/README.md

@@ -17,10 +17,12 @@ pkg: solod.dev/bench/maps
 cpu: Apple M1
 
 Benchmark_IntSet-8       31677    35580 ns/op     74264 B/op    20 allocs/op
+Benchmark_IntPre-8      124159     9608 ns/op     36944 B/op     5 allocs/op
 Benchmark_IntGet-8      218179     5573 ns/op         0 B/op     0 allocs/op
 Benchmark_IntDel-8       50260    23892 ns/op     36944 B/op     5 allocs/op
 
 Benchmark_StrSet-8       24082    48677 ns/op    108760 B/op    20 allocs/op
+Benchmark_StrPre-8       80437    14620 ns/op     54608 B/op     5 allocs/op
 Benchmark_StrGet-8      134481     8990 ns/op         0 B/op     0 allocs/op
 Benchmark_StrDel-8       34094    33878 ns/op     54608 B/op     5 allocs/op
 ```
@@ -29,10 +31,12 @@ So (mimalloc):
 
 ```text
 Benchmark_IntSet         20305    56696 ns/op     98112 B/op    27 allocs/op
+Benchmark_IntPre        137866     8821 ns/op     49152 B/op     3 allocs/op
 Benchmark_IntGet        734978     1638 ns/op         0 B/op     0 allocs/op
 Benchmark_IntDel         30958    38556 ns/op     73728 B/op     6 allocs/op
 
 Benchmark_StrSet         18986    71879 ns/op    130816 B/op    27 allocs/op
+Benchmark_StrPre         97383    12313 ns/op     65536 B/op     3 allocs/op
 Benchmark_StrGet        117218    10206 ns/op         0 B/op     0 allocs/op
 Benchmark_StrDel         23670    50111 ns/op     98304 B/op     6 allocs/op
 ```
@@ -41,10 +45,12 @@ So (arena):
 
 ```text
 Benchmark_IntSet         21157    57661 ns/op     98112 B/op    27 allocs/op
+Benchmark_IntPre        137257     8767 ns/op     49152 B/op     3 allocs/op
 Benchmark_IntGet        752787     1583 ns/op         0 B/op     0 allocs/op
 Benchmark_IntDel         30339    38821 ns/op     73728 B/op     6 allocs/op
 
 Benchmark_StrSet         18884    63500 ns/op    130816 B/op    27 allocs/op
+Benchmark_StrPre         99547    12115 ns/op     65536 B/op     3 allocs/op
 Benchmark_StrGet        119041    10083 ns/op         0 B/op     0 allocs/op
 Benchmark_StrDel         24212    49507 ns/op     98304 B/op     6 allocs/op
 ```

bench/maps/bench.go

@@ -54,6 +54,20 @@ func StdIntSet(b *testing.B) {
 	}
 }
 
+func StdIntPre(b *testing.B) {
+	a := b.Allocator()
+	for b.Loop() {
+		m := maps.New[int, int](a, nKeys)
+		for i := range nKeys {
+			m.Set(i, i)
+		}
+		m.Free()
+		if arena != nil {
+			arena.Reset()
+		}
+	}
+}
+
 func StdIntGet(b *testing.B) {
 	m := maps.New[int, int](nil, nKeys)
 	for i := range nKeys {
@@ -98,6 +112,20 @@ func StdStrSet(b *testing.B) {
 	}
 }
 
+func StdStrPre(b *testing.B) {
+	a := b.Allocator()
+	for b.Loop() {
+		m := maps.New[string, int](a, nKeys)
+		for i := range nKeys {
+			m.Set(strKeys[i], i)
+		}
+		m.Free()
+		if arena != nil {
+			arena.Reset()
+		}
+	}
+}
+
 func StdStrGet(b *testing.B) {
 	m := maps.New[string, int](nil, nKeys)
 	for i := range nKeys {
@@ -186,9 +214,11 @@ func main() {
 
 	benchs := []testing.Benchmark{
 		{Name: "IntSet", F: StdIntSet},
+		{Name: "IntPre", F: StdIntPre},
 		{Name: "IntGet", F: StdIntGet},
 		{Name: "IntDel", F: StdIntDel},
 		{Name: "StrSet", F: StdStrSet},
+		{Name: "StrPre", F: StdStrPre},
 		{Name: "StrGet", F: StdStrGet},
 		{Name: "StrDel", F: StdStrDel},
 	}

bench/maps/bench_test.go

@@ -21,6 +21,16 @@ func Benchmark_IntSet(b *testing.B) {
 	}
 }
 
+func Benchmark_IntPre(b *testing.B) {
+	b.ReportAllocs()
+	for b.Loop() {
+		m := make(map[int]int, nKeys)
+		for i := range nKeys {
+			m[i] = i
+		}
+	}
+}
+
 func Benchmark_IntGet(b *testing.B) {
 	b.ReportAllocs()
 	m := make(map[int]int, nKeys)
@@ -57,6 +67,16 @@ func Benchmark_StrSet(b *testing.B) {
 	}
 }
 
+func Benchmark_StrPre(b *testing.B) {
+	b.ReportAllocs()
+	for b.Loop() {
+		m := make(map[string]int, nKeys)
+		for i := range nKeys {
+			m[strKeys[i]] = i
+		}
+	}
+}
+
 func Benchmark_StrGet(b *testing.B) {
 	b.ReportAllocs()
 	m := make(map[string]int, nKeys)

so/maps/bytemap.go

@@ -51,7 +51,8 @@ type ByteMap struct {
 func NewByteMap(a mem.Allocator, size, ksize, vsize int) ByteMap {
 	m := ByteMap{a: a, ksize: ksize, vsize: vsize, seed: seed()}
 	sz := 8
-	for sz < size {
+	// The map must be large enough to hold size entries without resizing.
+	for int(float64(sz)*loadFactor) < size {
 		sz *= 2
 	}
 	m.hdib = mem.AllocSlice[uint64](m.a, sz, sz)