From 63233c251688fbeeec5786d034fe003c5ea2584c Mon Sep 17 00:00:00 2001
From: Anton Firszov <antonfir@gmail.com>
Date: Fri, 19 Oct 2018 13:38:03 +0200
Subject: [PATCH] benchmark conversion steps separately

---
 .../General/Vectorization/UInt32ToSingle.cs   | 66 +++++++++++++++++++
 .../Vectorization/WidenBytesToUInt32.cs       | 61 +++++++++++++++++
 2 files changed, 127 insertions(+)
 create mode 100644 tests/ImageSharp.Benchmarks/General/Vectorization/UInt32ToSingle.cs
 create mode 100644 tests/ImageSharp.Benchmarks/General/Vectorization/WidenBytesToUInt32.cs

diff --git a/tests/ImageSharp.Benchmarks/General/Vectorization/UInt32ToSingle.cs b/tests/ImageSharp.Benchmarks/General/Vectorization/UInt32ToSingle.cs
new file mode 100644
index 000000000..4a4b939b6
--- /dev/null
+++ b/tests/ImageSharp.Benchmarks/General/Vectorization/UInt32ToSingle.cs
@@ -0,0 +1,66 @@
+using System.Numerics;
+using System.Runtime.CompilerServices;
+
+using BenchmarkDotNet.Attributes;
+
+namespace SixLabors.ImageSharp.Benchmarks.General.Vectorization
+{
+    public class UInt32ToSingle
+    {
+        private float[] data;
+
+        private const int Count = 64;
+
+        [GlobalSetup]
+        public void Setup()
+        {
+            this.data = new float[Count];
+        }
+
+        [Benchmark(Baseline = true)]
+        public void MagicMethod()
+        {
+            ref Vector<float> b = ref Unsafe.As<float, Vector<float>>(ref this.data[0]);
+
+            int n = Count / Vector<float>.Count;
+
+            Vector<float> magick = new Vector<float>(32768.0f);
+            Vector<float> scale = new Vector<float>(255f) / new Vector<float>(256f);
+
+            for (int i = 0; i < n; i++)
+            {
+                // union { float f; uint32_t i; } u;
+                // u.f = 32768.0f + x * (255.0f / 256.0f);
+                // return (uint8_t)u.i;
+
+                ref Vector<float> d = ref Unsafe.Add(ref b, i);
+                Vector<float> x = d;
+                //x = Vector.Max(x, Vector<float>.Zero);
+                //x = Vector.Min(x, Vector<float>.One);
+
+                x = (x * scale) + magick;
+                d = x;
+            }
+        }
+
+        [Benchmark]
+        public void StandardSimd()
+        {
+            int n = Count / Vector<float>.Count;
+
+            ref Vector<float> b = ref Unsafe.As<float, Vector<float>>(ref this.data[0]);
+
+            var scale = new Vector<float>(1f / 255f);
+
+            for (int i = 0; i < n; i++)
+            {
+                ref Vector<float> df = ref Unsafe.Add(ref b, i);
+                Vector<uint> du = Unsafe.As<Vector<float>, Vector<uint>>(ref df);
+
+                Vector<float> v = Vector.ConvertToSingle(du);
+                v *= scale;
+                df = v;
+            }
+        }
+    }
+}
\ No newline at end of file
diff --git a/tests/ImageSharp.Benchmarks/General/Vectorization/WidenBytesToUInt32.cs b/tests/ImageSharp.Benchmarks/General/Vectorization/WidenBytesToUInt32.cs
new file mode 100644
index 000000000..f71f6ec1b
--- /dev/null
+++ b/tests/ImageSharp.Benchmarks/General/Vectorization/WidenBytesToUInt32.cs
@@ -0,0 +1,61 @@
+using System.Numerics;
+using System.Runtime.CompilerServices;
+
+using BenchmarkDotNet.Attributes;
+
+namespace SixLabors.ImageSharp.Benchmarks.General.Vectorization
+{
+    public class WidenBytesToUInt32
+    {
+        private byte[] source;
+
+        private uint[] dest;
+
+        private const int Count = 64;
+
+        [GlobalSetup]
+        public void Setup()
+        {
+            this.source = new byte[Count];
+            this.dest = new uint[Count];
+        }
+
+        [Benchmark(Baseline = true)]
+        public void Standard()
+        {
+            const int N = Count / 8;
+
+            ref SimdUtils.Octet.OfByte sBase = ref Unsafe.As<byte, SimdUtils.Octet.OfByte>(ref this.source[0]);
+            ref SimdUtils.Octet.OfUInt32 dBase = ref Unsafe.As<uint, SimdUtils.Octet.OfUInt32>(ref this.dest[0]);
+
+            for (int i = 0; i < N; i++)
+            {
+                Unsafe.Add(ref dBase, i).LoadFrom(ref Unsafe.Add(ref sBase, i));
+            }
+        }
+
+        [Benchmark]
+        public void Simd()
+        {
+            int n = Count / Vector<byte>.Count;
+
+            ref Vector<byte> sBase = ref Unsafe.As<byte, Vector<byte>>(ref this.source[0]);
+            ref Vector<uint> dBase = ref Unsafe.As<uint, Vector<uint>>(ref this.dest[0]);
+
+            for (int i = 0; i < n; i++)
+            {
+                Vector<byte> b = Unsafe.Add(ref sBase, i);
+
+                Vector.Widen(b, out Vector<ushort> s0, out Vector<ushort> s1);
+                Vector.Widen(s0, out Vector<uint> w0, out Vector<uint> w1);
+                Vector.Widen(s1, out Vector<uint> w2, out Vector<uint> w3);
+
+                ref Vector<uint> d = ref Unsafe.Add(ref dBase, i * 4);
+                d = w0;
+                Unsafe.Add(ref d, 1) = w1;
+                Unsafe.Add(ref d, 2) = w2;
+                Unsafe.Add(ref d, 3) = w3;
+            }
+        }
+    }
+}
\ No newline at end of file