Skip to content

docs: Update benchmark numbers in readme files #42

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
siyul-park merged 1 commit into from
May 25, 2026
Merged

docs: Update benchmark numbers in readme files #42

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,30 +34,30 @@ go get github.com/siyul-park/minivm

## Performance

Recursive `fib(35)` — linux/amd64, Intel Xeon @ 2.80 GHz, Go 1.26.2:
Recursive `fib(35)` — linux/amd64, Intel Xeon @ 2.10 GHz, Go 1.26.2:

| Runtime | ns/op | B/op | allocs/op | vs native Go | execution model |
|---|---|---|---|---|---|
| native Go | 51,947,220 | 0 | 0 | 1× | compiled |
| wazero | 84,807,148 | 16 | 2 | 1.6× | WASM JIT |
| **minivm** | **1,672,707,295** | **288** | **1** | **32×** | **threaded interpreter** |
| tengo | 2,665,298,176 | 312,800,180 | 39,088,180 | 51× | bytecode VM |
| gopher-lua | 4,081,167,978 | 971,008 | 3,793 | 79× | register VM |
| goja | 5,427,175,850 | 383,488 | 46,384 | 105× | bytecode VM |
| native Go | 56,441,552 | 0 | 0 | 1× | compiled |
| wazero | 84,601,941 | 16 | 2 | 1.5× | WASM → native JIT |
| **minivm** | **1,320,092,108** | **244** | **1** | **23×** | **threaded interpreter** |
| tengo | 2,276,648,719 | 312,797,200 | 39,088,175 | 40× | bytecode VM |
| gopher-lua | 3,002,897,021 | 971,008 | 3,793 | 53× | register VM |
| goja | 3,962,089,181 | 380,400 | 46,377 | 70× | bytecode VM |

Among interpreters without JIT, minivm is fastest in this benchmark: **1.6× tengo, 2.4× gopher-lua, 3.2× goja**. Allocation count stays near zero regardless of recursion depth; tengo accumulates 39M allocations at fib(35).
Among interpreters without JIT, minivm is fastest in this benchmark: **1.7× tengo, 2.3× gopher-lua, 3.0× goja**. Allocation count stays near zero regardless of recursion depth; tengo accumulates 39M allocations at fib(35).

wazero's lead is structural: it compiles WASM to native x86-64 at module load. minivm closes this gap on ARM64, where JIT promotes hot segments to native code.
wazero's advantage is structural: it compiles the WebAssembly module to native x86-64 at load time. minivm closes this gap on ARM64, where JIT promotes hot numeric segments to native code.

Single-instruction throughput (threaded interpreter):

| Workload | ns/op |
|---|---|
| i32/i64/f32/f64 arithmetic | ~20–22 |
| i32/i64/f32/f64 arithmetic | ~17–25 |
| branches (`br`, `br_if`) | ~20–24 |
| bytecode function call | ~26–29 |
| host function call | ~36 |
| array / struct operations | ~90–140 |
| array / struct operations | ~82–117 |

Full results: [`docs/benchmarks.md`](docs/benchmarks.md)

Expand Down
22 changes: 11 additions & 11 deletions README_kr.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,30 +34,30 @@ go get github.com/siyul-park/minivm

## 성능

재귀 `fib(35)` — linux/amd64, Intel Xeon @ 2.80 GHz, Go 1.26.2:
재귀 `fib(35)` — linux/amd64, Intel Xeon @ 2.10 GHz, Go 1.26.2:

| 런타임 | ns/op | B/op | allocs/op | vs native Go | 실행 모델 |
|---|---|---|---|---|---|
| native Go | 51,947,220 | 0 | 0 | 1× | 컴파일 |
| wazero | 84,807,148 | 16 | 2 | 1.6× | WASM JIT |
| **minivm** | **1,672,707,295** | **288** | **1** | **32×** | **스레디드 인터프리터** |
| tengo | 2,665,298,176 | 312,800,180 | 39,088,180 | 51× | 바이트코드 VM |
| gopher-lua | 4,081,167,978 | 971,008 | 3,793 | 79× | 레지스터 VM |
| goja | 5,427,175,850 | 383,488 | 46,384 | 105× | 바이트코드 VM |
| native Go | 56,441,552 | 0 | 0 | 1× | 컴파일 |
| wazero | 84,601,941 | 16 | 2 | 1.5× | WASM → 네이티브 JIT |
| **minivm** | **1,320,092,108** | **244** | **1** | **23×** | **스레디드 인터프리터** |
| tengo | 2,276,648,719 | 312,797,200 | 39,088,175 | 40× | 바이트코드 VM |
| gopher-lua | 3,002,897,021 | 971,008 | 3,793 | 53× | 레지스터 VM |
| goja | 3,962,089,181 | 380,400 | 46,377 | 70× | 바이트코드 VM |

JIT 없는 인터프리터 중 이 벤치마크에서는 minivm이 가장 빠릅니다: **tengo 대비 1.6×, gopher-lua 대비 2.4×, goja 대비 3.2×**. 재귀 깊이에 관계없이 할당 수가 거의 0에 가깝고, tengo는 fib(35)에서 3,900만 번 할당합니다.
JIT 없는 인터프리터 중 이 벤치마크에서는 minivm이 가장 빠릅니다: **tengo 대비 1.7×, gopher-lua 대비 2.3×, goja 대비 3.0×**. 재귀 깊이에 관계없이 할당 수가 거의 0에 가깝고, tengo는 fib(35)에서 3,900만 번 할당합니다.

wazero가 앞서는 이유는 모듈 로드 시점에 WebAssembly를 x86-64 네이티브 코드로 JIT 컴파일하기 때문입니다. ARM64에서는 minivm도 핫 세그먼트를 네이티브 코드로 승격하므로 이 격차가 좁혀집니다.
wazero가 앞서는 이유는 모듈 로드 시점에 WebAssembly를 x86-64 네이티브 코드로 JIT 컴파일하기 때문입니다. ARM64에서는 minivm도 핫 숫자 세그먼트를 네이티브 코드로 승격하므로 이 격차가 좁혀집니다.

단일 명령어 처리량 (스레디드 인터프리터):

| 워크로드 | ns/op |
|---|---|
| i32/i64/f32/f64 산술 | ~20–22 |
| i32/i64/f32/f64 산술 | ~17–25 |
| 분기 (`br`, `br_if`) | ~20–24 |
| 바이트코드 함수 호출 | ~26–29 |
| 호스트 함수 호출 | ~36 |
| 배열 / 구조체 연산 | ~90–140 |
| 배열 / 구조체 연산 | ~82–117 |

전체 측정 결과: [`docs/benchmarks.md`](docs/benchmarks.md)

Expand Down
Loading