f8b21af4be
10_Wiki/Topics 대규모 정리: - 오류 캡처/미완성 stub 문서 227개 제거 - 교차폴더 중복 43클러스터 병합 (63파일 → redirect) - 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건 - 카테고리 MOC 6개 신규 생성 - Graph 섹션 미해결 related-keyword 링크 10,058건 제거 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
169 lines
4.8 KiB
Markdown
169 lines
4.8 KiB
Markdown
---
|
|
id: wiki-2026-0508-index-masking
|
|
title: Index Masking
|
|
category: 10_Wiki/Topics
|
|
status: verified
|
|
canonical_id: self
|
|
aliases: [Bitmask Index, Power-of-Two Masking]
|
|
duplicate_of: none
|
|
source_trust_level: A
|
|
confidence_score: 0.9
|
|
verification_status: applied
|
|
tags: [optimization, low-level, ring-buffer, hash]
|
|
raw_sources: []
|
|
last_reinforced: 2026-05-10
|
|
github_commit: pending
|
|
tech_stack:
|
|
language: C/C++/Rust/JavaScript
|
|
framework: General
|
|
---
|
|
|
|
# Index Masking
|
|
|
|
## 매 한 줄
|
|
> **"매 capacity 가 power-of-two 일 때 `% N` 의 `& (N-1)` replacement"**. 매 modulo 의 expensive (CPU cycle 10-30) 의 single-cycle `AND` 로 대체. Ring buffer, hash table bucket, fixed-size LRU, lock-free queue 의 hot path 의 universal trick. CPU branch predictor 와 cache line 의 friendly.
|
|
|
|
## 매 핵심
|
|
|
|
### 매 정확성 조건
|
|
- N 이 power-of-two: `2^k`.
|
|
- `index % N == index & (N - 1)` (unsigned 또는 non-negative).
|
|
- `N - 1` 이 모든 lower bit 의 1 의 mask.
|
|
|
|
### 매 대표 사용 처
|
|
- Ring buffer / circular queue.
|
|
- Open-addressing hash table bucket index.
|
|
- Game loop frame index modulo history size.
|
|
- Hardware register address calc.
|
|
- Lock-free SPSC/MPMC queue (Disruptor pattern).
|
|
|
|
### 매 응용
|
|
1. LMAX Disruptor (Java) — sequence & ringMask.
|
|
2. Linux kernel ringbuffer.
|
|
3. Redis hash table — sizemask.
|
|
4. V8 hidden class transitions table.
|
|
|
|
## 💻 패턴
|
|
|
|
### Basic ring buffer (C)
|
|
```c
|
|
#define RB_CAP 1024 // must be power of two
|
|
typedef struct {
|
|
int data[RB_CAP];
|
|
uint32_t head, tail; // free-running counters
|
|
} RingBuf;
|
|
|
|
static inline void rb_push(RingBuf *r, int v) {
|
|
r->data[r->head & (RB_CAP - 1)] = v;
|
|
r->head++;
|
|
}
|
|
static inline int rb_pop(RingBuf *r) {
|
|
int v = r->data[r->tail & (RB_CAP - 1)];
|
|
r->tail++;
|
|
return v;
|
|
}
|
|
```
|
|
|
|
### TypeScript fixed-size circular log
|
|
```typescript
|
|
class CircularLog<T> {
|
|
private buf: (T | undefined)[]
|
|
private mask: number
|
|
private head = 0
|
|
constructor(capacityPow2: number) {
|
|
if ((capacityPow2 & (capacityPow2 - 1)) !== 0)
|
|
throw new Error('capacity must be power of two')
|
|
this.buf = new Array(capacityPow2)
|
|
this.mask = capacityPow2 - 1
|
|
}
|
|
push(v: T) {
|
|
this.buf[this.head & this.mask] = v
|
|
this.head++
|
|
}
|
|
recent(n: number): T[] {
|
|
const out: T[] = []
|
|
for (let i = 0; i < n; i++) {
|
|
const v = this.buf[(this.head - 1 - i) & this.mask]
|
|
if (v !== undefined) out.push(v)
|
|
}
|
|
return out
|
|
}
|
|
}
|
|
```
|
|
|
|
### Round-up to next power of two
|
|
```c
|
|
uint32_t next_pow2(uint32_t v) {
|
|
v--;
|
|
v |= v >> 1; v |= v >> 2; v |= v >> 4; v |= v >> 8; v |= v >> 16;
|
|
return v + 1;
|
|
}
|
|
```
|
|
|
|
### Hash table bucket
|
|
```rust
|
|
struct Map<K, V> {
|
|
buckets: Vec<Option<(K, V)>>,
|
|
mask: usize,
|
|
}
|
|
impl<K: Hash + Eq, V> Map<K, V> {
|
|
fn bucket(&self, k: &K) -> usize {
|
|
let mut h = DefaultHasher::new();
|
|
k.hash(&mut h);
|
|
(h.finish() as usize) & self.mask // mask = capacity - 1
|
|
}
|
|
}
|
|
```
|
|
|
|
### SPSC lock-free (Disruptor-style)
|
|
```c
|
|
// producer
|
|
void produce(RingBuf *r, int v) {
|
|
uint32_t h = atomic_load(&r->head);
|
|
while (h - atomic_load(&r->tail) >= RB_CAP) cpu_relax(); // full
|
|
r->data[h & (RB_CAP - 1)] = v;
|
|
atomic_store(&r->head, h + 1);
|
|
}
|
|
```
|
|
|
|
### Branch-free wrap (negative-safe)
|
|
```c
|
|
// for signed wrap, use bit-and after cast
|
|
int idx = ((int)(counter) & ((int)CAP - 1)); // counter must be non-negative
|
|
```
|
|
|
|
## 매 결정 기준
|
|
| 상황 | Approach |
|
|
|---|---|
|
|
| Hot path, capacity 가 control 가능 | Power-of-2 + masking |
|
|
| Capacity 가 user-input arbitrary | `%` (or round up to pow2) |
|
|
| Capacity 가 dynamic resize | resize 시 `mask` recompute |
|
|
| Signed counter | unsigned cast 또는 modulo |
|
|
| Hardware-aligned (cache line) | pow2 + alignment 조합 |
|
|
|
|
**기본값**: 매 ring/hash table 는 power-of-two + bitmask. 매 capacity 의 dynamic 이면 round-up.
|
|
|
|
## 🔗 Graph
|
|
- 응용: [[Object Pooling (오브젝트 풀링)]]
|
|
- Adjacent: [[SharedArrayBuffer로 스레드 간 메모리 공유 효율 높이기]] · [[Pointer Compression]]
|
|
|
|
## 🤖 LLM 활용
|
|
**언제**: 매 high-throughput data structure, lock-free queue, HFT, game engine frame loop.
|
|
**언제 X**: capacity 의 control 안 되거나 readability priority 인 일반 application code.
|
|
|
|
## ❌ 안티패턴
|
|
- **Non-pow2 size 에 mask 적용**: 매 silent index out-of-range, data corruption.
|
|
- **Signed negative index 의 masking**: `-1 & (N-1)` 의 N-1 이 됨 → 매 unsigned cast 또는 `((i % N) + N) % N` pattern.
|
|
- **Capacity 1 의 mask 0**: 매 always index 0, degenerate.
|
|
- **Premature application**: 매 측정 없이 micro-optimization. 매 hot path 만 의 적용.
|
|
|
|
## 🧪 검증 / 중복
|
|
- Verified (Hacker's Delight Ch.3, LMAX Disruptor paper, Linux kfifo source, Redis dict.c).
|
|
- 신뢰도 A.
|
|
|
|
## 🕓 Changelog
|
|
| 날짜 | 변경 |
|
|
|---|---|
|
|
| 2026-05-08 | Phase 1 |
|
|
| 2026-05-10 | Manual cleanup — power-of-2 masking patterns + ring/hash applications |
|