Haxmap Versions Save

Notes

• Improved hashing performance leading to greater efficiency overall
• Fixed cases when hash key is 0 valued, now tests like the case below will no longer fail

func TestHash0Collision(t *testing.T) {
	m := haxmap.New[string, int]()
	staticHasher := func(key string) uintptr {
		return 0
	}
	m.SetHasher(staticHasher)
	m.Set("1", 1)
	m.Set("2", 2)
	_, ok := m.Get("1")
	if !ok {
		t.Error("1 not found")
	}
	_, ok = m.Get("2")
	if !ok {
		t.Error("2 not found")
	}
}

Benchmarks

Benchmarks were performed against golang sync.Map and the latest cornelk-hashmap

All results were computed from benchstat of 20 runs (code available here)

1. Concurrent Reads Only

name                         time/op
HaxMapReadsOnly-8            9.90µs ±13%
GoSyncMapReadsOnly-8         22.2µs ±14%
CornelkMapReadsOnly-8        10.4µs ±16%

name                         alloc/op
HaxMapReadsOnly-8             0.00B
GoSyncMapReadsOnly-8          0.00B
CornelkMapReadsOnly-8         0.00B

name                         allocs/op
HaxMapReadsOnly-8              0.00
GoSyncMapReadsOnly-8           0.00
CornelkMapReadsOnly-8          0.00

2. Concurrent Reads with Writes

name                         time/op
HaxMapReadsWithWrites-8      11.2µs ±16%
GoSyncMapReadsWithWrites-8   25.6µs ±14%
CornelkMapReadsWithWrites-8  12.1µs ±12%

name                         alloc/op
HaxMapReadsWithWrites-8      1.35kB ± 8%
GoSyncMapReadsWithWrites-8   6.07kB ± 8%
CornelkMapReadsWithWrites-8  5.80kB ±10%

name                         allocs/op
HaxMapReadsWithWrites-8         168 ± 8%
GoSyncMapReadsWithWrites-8      562 ± 8%
CornelkMapReadsWithWrites-8     207 ±10%

From the above results it is evident that haxmap takes the least time, memory and allocations in all cases

• Huge performance gain obtained by switching bucket implementation to Harris lock-free list followed by additional optimizations from this paper
• Fixed issue #54 from cornelk-hashmap which caused data loss on concurrent writes during map resize operations
• Added support for 32-bit systems and more map key types

Benchmarks

Benchmarks were performed against golang sync.Map and the latest generics-enabled cornelk-hashmap

All results were computed from benchstat of 20 runs (code available here)

1. Concurrent Reads Only

name                         time/op
HaxMapReadsOnly-8            10.9µs ±11%
GoSyncMapReadsOnly-8         23.0µs ± 7%
CornelkMapReadsOnly-8        12.3µs ±11%

name                         alloc/op
HaxMapReadsOnly-8             0.00B
GoSyncMapReadsOnly-8          0.00B
CornelkMapReadsOnly-8         0.00B

name                         allocs/op
HaxMapReadsOnly-8              0.00
GoSyncMapReadsOnly-8           0.00
CornelkMapReadsOnly-8          0.00

2. Concurrent Reads with Writes

name                         time/op
HaxMapReadsWithWrites-8      12.6µs ±11%
GoSyncMapReadsWithWrites-8   25.6µs ±10%
CornelkMapReadsWithWrites-8  14.3µs ±13%

name                         alloc/op
HaxMapReadsWithWrites-8      1.46kB ± 3%
GoSyncMapReadsWithWrites-8   6.19kB ± 6%
CornelkMapReadsWithWrites-8  6.70kB ± 9%

name                         allocs/op
HaxMapReadsOnly-8              0.00
HaxMapReadsWithWrites-8         183 ± 3%
GoSyncMapReadsOnly-8           0.00
GoSyncMapReadsWithWrites-8      573 ± 7%
CornelkMapReadsOnly-8          0.00
CornelkMapReadsWithWrites-8     239 ± 9%

From the above results it is evident that haxmap takes the least time, memory and allocations in all cases