Attachment: Download

Issue Description:

Implementing a QuadraticSequence from 2 standard Sequence as a generic protocol creates code that is 30x slower than a direct Sequence/Iterator implementation. Not sure if this is expected or if there are specifics in this code that explain the situation.

I was instructed to run it through SIL tools to inspect code. You'll find the original source code, SIL outout as attachement.

Note that the source code has both the generic protocol approach and the direct implementation which can be switched easily.

Original discussion (forum): https://forums.swift.org/t/performance-of-generic-protocol-extension-30x-slower-swift-4-2/18582

Here's the test code:

func testPerformanceXLoop() { 
    self.measure {
        let indices = XQuadraticIndexRange(rows: 1000, columns: 1000) var sum = 0 for i in indices { sum += i.0+i.1 } print("sum= \(sum)") } 
}