Surge is a Swift library, similar to numpy, that uses the Accelerate framework to provide high-performance functions for matrix math, digital signal processing, and image manipulation.
Surge aims to bring Accelerate to the mainstream, making it as easy (and nearly as fast, in most cases) to perform computation over a set of numbers as for a single member.
Curious about the name Surge? Back in the mid 90's, Apple, IBM, and Motorola teamed up to create AltiVec (a.k.a the Velocity Engine), which provided a SIMD instruction set for the PowerPC architecture. When Apple made the switch to Intel CPUs, AltiVec was ported to the x86 architecture and rechristened Accelerate. The derivative of Accelerate (and second derivative of Velocity) is known as either jerk, jolt, surge, or lurch, hence the name of this library.
Use Swift Package Manager
Surge functions are named according to their corresponding "Math.h" functions, where applicable (omitting
fanddaffixes, since type information is communicated and enforced by the language's type system).
sumasummaxminmeanmeamgmeasqaddsubmuldivmodremaindersqrtsum of squared values
absceilcopysignfloorrecroundtrunc
convxcorr
expexp2loglog2log10logb
fft
sinhcoshtanhasinhacoshatanh
addmulinvtransposenegate
You can also initialize matrices with generator functions for the elements, thus enabling the user to create matrices from gaussian distributions for example.
pow
sincossincostanasinacosatanrad2degdeg2rad
import Surge
let n = [1.0, 2.0, 3.0, 4.0, 5.0]
let sum = Surge.sum(n) // 15.0import Surge
let a = [1.0, 3.0, 5.0, 7.0]
let b = [2.0, 4.0, 6.0, 8.0]
let product = Surge.mul(a, b) // [2.0, 12.0, 30.0, 56.0]Initial benchmarks on iOS devices and the iOS simulator indicate significant performance improvements over a conventional Swift implementation.
import Surge
let numbers: [Double] = ...
var sum: Double = 0.0
// Naïve Swift Implementation
sum = reduce(numbers, 0.0, +)
// Surge Implementation
sum = Surge.sum(numbers)(Time in milliseconds, Optimization Level -Ofast)
| n | Swift | Surge | Δ |
|---|---|---|---|
| 100 | 0.269081 | 0.004453 | ~60x |
| 100000 | 251.037254 | 0.028687 | ~9000x |
| 100000000 | 239474.689326 | 57.009841 | ~4000x |
Surge's performance characteristics have not yet been thoroughly evaluated, though initial benchmarks show incredible promise. Further investigation is definitely warranted.
Surge is available under the MIT license. See the LICENSE file for more info.