Benchmark {Float,Double,Float80}.description (#15234)

* Benchmark {Float,Double,Float80}.description

This just tests how fast we can format the standard
floating-point types.  It also includes a test that
uses the result, to ensure that future optimized
_creation_ of the string doesn't incidentally pessimize
_use_ of the results.

* Benchmark {Float,Double,Float80}.description

This just tests how fast we can format the standard
floating-point types.  It also includes a test that
uses the result, to ensure that future optimized
_creation_ of the string doesn't incidentally pessimize
_use_ of the results.

* Add benchmarks for values close to 1 + other review suggestions

Author: tbkka

benchmark/CMakeLists.txt

@@ -77,6 +77,7 @@ set(SWIFT_BENCH_MODULES
     single-source/Exclusivity
     single-source/ExistentialPerformance
     single-source/Fibonacci
+    single-source/FloatingPointPrinting
     single-source/Hanoi
     single-source/Hash
     single-source/HashQuadratic

benchmark/single-source/FloatingPointPrinting.swift

@@ -0,0 +1,204 @@
+//===--- FloatingPointPrinting.swift -----------------------------------===//
+//
+// This source file is part of the Swift.org open source project
+//
+// Copyright (c) 2018 Apple Inc. and the Swift project authors
+// Licensed under Apache License v2.0 with Runtime Library Exception
+//
+// See https://swift.org/LICENSE.txt for license information
+// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
+//
+//===----------------------------------------------------------------------===//
+
+// This test verifies the performance of generating a text description
+// from a binary floating-point value.
+
+import TestsUtils
+
+public let FloatingPointPrinting = [
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float_description_small",
+    runFunction: run_FloatingPointPrinting_Float_description_small,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Double_description_small",
+    runFunction: run_FloatingPointPrinting_Double_description_small,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float80_description_small",
+    runFunction: run_FloatingPointPrinting_Float80_description_small,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float_description_uniform",
+    runFunction: run_FloatingPointPrinting_Float_description_uniform,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Double_description_uniform",
+    runFunction: run_FloatingPointPrinting_Double_description_uniform,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float80_description_uniform",
+    runFunction: run_FloatingPointPrinting_Float80_description_uniform,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float_interpolated",
+    runFunction: run_FloatingPointPrinting_Float_interpolated,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Double_interpolated",
+    runFunction: run_FloatingPointPrinting_Double_interpolated,
+    tags: [.validation, .api, .runtime, .String]),
+
+  BenchmarkInfo(
+    name: "FloatingPointPrinting_Float80_interpolated",
+    runFunction: run_FloatingPointPrinting_Float80_interpolated,
+    tags: [.validation, .api, .runtime, .String])
+]
+
+// Generate descriptions for 100,000 values around 1.0.
+//
+// Note that some formatting algorithms behave very
+// differently for values around 1.0 than they do for
+// less-common extreme values.  Having a "small" test
+// and a "uniform" test exercises both cases.
+//
+// Dividing integers 1...100000 by 101 (a prime) yields floating-point
+// values from about 1e-2 to about 1e3, each with plenty of digits after
+// the decimal:
+
+@inline(never)
+public func run_FloatingPointPrinting_Float_description_small(_ N: Int) {
+  let count = 100_000
+  for _ in 0..<N {
+    for i in 1...count {
+      let f = Float(i) / 101.0
+      blackHole(f.description)
+    }
+  }
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Double_description_small(_ N: Int) {
+  let count = 100_000
+  for _ in 0..<N {
+    for i in 1...count {
+      let f = Double(i) / 101.0
+      blackHole(f.description)
+    }
+  }
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Float80_description_small(_ N: Int) {
+  let count = 100_000
+  for _ in 0..<N {
+    for i in 1...count {
+      let f = Float80(i) / 101.0
+      blackHole(f.description)
+    }
+  }
+}
+
+// Generate descriptions for 100,000 values spread evenly across
+// the full range of the type:
+
+@inline(never)
+public func run_FloatingPointPrinting_Float_description_uniform(_ N: Int) {
+  let count = 100_000
+  let step = UInt32.max / UInt32(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let raw = UInt32(i) * step
+      let f = Float(bitPattern: raw)
+      blackHole(f.description)
+    }
+  }
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Double_description_uniform(_ N: Int) {
+  let count = 100_000
+  let step = UInt64.max / UInt64(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let raw = UInt64(i) * step
+      let f = Double(bitPattern: raw)
+      blackHole(f.description)
+    }
+  }
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Float80_description_uniform(_ N: Int) {
+  let count = 100_000
+  let step = UInt64.max / UInt64(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let fraction = UInt64(i) * step
+      let exponent = UInt(i) % 32768
+      let f = Float80(sign: .plus, exponentBitPattern: exponent, significandBitPattern: fraction)
+      blackHole(f.description)
+    }
+  }
+}
+
+// The "interpolated" tests verify that any storage optimizations used while
+// producing the formatted numeric strings don't pessimize later use of the
+// result.
+
+@inline(never)
+public func run_FloatingPointPrinting_Float_interpolated(_ N: Int) {
+  let count = 100_000
+  let step = UInt32.max / UInt32(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let raw = UInt32(i) * step
+      let f = Float(bitPattern: raw)
+      blackHole("and the actual result was \(f)")
+    }
+  }
+  CheckResults(s.count > 1)
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Double_interpolated(_ N: Int) {
+  let count = 100_000
+  let step = UInt64.max / UInt64(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let raw = UInt64(i) * step
+      let f = Double(bitPattern: raw)
+      blackHole("and the actual result was \(f)")
+    }
+  }
+  CheckResults(s.count > 1)
+}
+
+@inline(never)
+public func run_FloatingPointPrinting_Float80_interpolated(_ N: Int) {
+  let count = 100_000
+  let step = UInt64.max / UInt64(count)
+  var s = ""
+  for _ in 0..<N {
+    for i in 0..<count {
+      let fraction = UInt64(i) * step
+      let exponent = UInt(i) % 32768
+      let f = Float80(sign: .plus, exponentBitPattern: exponent, significandBitPattern: fraction)
+      blackHole("and the actual result was \(f)")
+    }
+  }
+  CheckResults(s.count > 1)
+}
+

benchmark/utils/main.swift

@@ -64,6 +64,7 @@ import ErrorHandling
 import Exclusivity
 import ExistentialPerformance
 import Fibonacci
+import FloatingPointPrinting
 import Hanoi
 import Hash
 import HashQuadratic
@@ -215,6 +216,7 @@ registerBenchmark(ErrorHandling)
 registerBenchmark(Exclusivity)
 registerBenchmark(ExistentialPerformance)
 registerBenchmark(Fibonacci)
+registerBenchmark(FloatingPointPrinting)
 registerBenchmark(Hanoi)
 registerBenchmark(HashTest)
 registerBenchmark(HashQuadratic)