Using SIMD instructions inside a buffer iteration

UnsafeMutablePointer+EmptyCheck.swift

extension UnsafeMutablePointer<UInt8> {
  /// Returns the index of the first element in the buffer that is not `0`
	func firstIndexNotZero(size bufferSize: Int) -> Int? {
		let stride = MemoryLayout<SIMD16<UInt8>>.stride
		let simdSize = bufferSize / stride
		
		// Iterate buffer using 16-component SIMD vectors
		let simdIndex: Int? = withMemoryRebound(to: SIMD16<UInt8>.self, capacity: simdSize) { buffer in
			let zero = SIMD16<UInt8>(repeating: 0)
			return (0..<simdSize).first { buffer[$0] != zero }
		}
		// At least one byte is non-zero, find it within the chunk
		if let simdIndex {
			let index = simdIndex * stride
			return (index..<index + stride).first { self[$0] != 0 }!
		}
		
		// Check any remaining bytes
		let simdEnd = simdSize * stride
		return (simdEnd..<bufferSize).first { self[$0] != 0 }
	}
}

This function is called like this:

buffer.firstIndexNotZero(size: size)

And returns the same as

(0..<size).first(where: { buffer[$0] != 0 })

The implementation used SIMD instructions to speed up an otherwise really slow operation.

Note: If buffer was an Array not an UnsafeMutablePointer, a more natural way to write this would be buffer.firstIndex { $0 != 0 }. But this has them same runtime penalty of iterating every item separately...