superfastpy 26.7.3

Zero-configuration Python function accelerator. Just add @fast.

⚡ superfastpy

Zero-configuration Python function accelerator.
Just add @fast — superfastpy analyzes your code and applies the best backend automatically.
No compilation. No configuration. No learning curve.

---

🚀 Installation

pip install superfastpy

---

🔥 Quick Start

from superfastpy import fast


@fast
def sum_squares(arr):
    total = 0.0
    for i in range(len(arr)):
        total += arr[i] * arr[i]
    return total

That's it. superfastpy automatically detects the loop and compiles it to machine code via Numba JIT.

---

🧠 How it works

superfastpy analyzes your function's AST (Abstract Syntax Tree) at decoration time and selects the best backend:

Code Pattern	Backend	Description
Nested loops	numba_parallel	Auto-converts range → prange for true parallelism
Simple loops	numba	JIT-compiled to machine code
Array / math ops	numpy	Vectorized execution
Everything else	native	No overhead added

---

🎯 Decorators

@fast — Auto mode

@fast
def my_func(arr):
    ...

# Or with options
@fast(backend="numba", verbose=True)
def my_func(arr):
    ...

@fast.parallel — Force parallel

@fast.parallel
def dot_product(a, b):
    result = 0.0
    for i in range(len(a)):   # ← superfastpy auto-converts to prange
        result += a[i] * b[i]
    return result

superpy rewrites range → prange via AST transformation automatically. No manual changes needed.

@fast.cache — Accelerate + cache results

@fast.cache
def heavy_compute(arr):
    total = 0.0
    for i in range(len(arr)):
        total += arr[i] * arr[i]
    return total

heavy_compute(arr)  # computed
heavy_compute(arr)  # cache hit — instant return

---

🛠️ Utilities

from superfastpy import info, benchmark, cache_stats

info(my_func)  # prints backend info
benchmark(my_func, arr)  # measures average execution time
cache_stats(my_func)  # prints cache hit rate

---

📊 Benchmark Results

sum_squares    → numba          0.0008ms
matrix_sum     → numba_parallel 0.0100ms  (small array; parallel shines on large data)
parallel_dot   → numba_parallel 0.0088ms
heavy_compute  → numba+cache    0.0125ms  (cache hit rate: 75%)
add            → native         0.0002ms

---

📄 License

MIT License