transposebench.cs

using System;
using System.Diagnostics;
using System.Runtime.CompilerServices;

namespace CSTest
{
    class Program
    {
        static void Main(string[] args)
        {
            Console.WriteLine("Latency test:");
            for (int j = 0; j < 10; j++)
            {
                bool b = (j & 1) == 1;
                var sw = Stopwatch.StartNew();
                ulong x = 0;
                if (b)
                {
                    for (int i = 0; i < 1000000; i++)
                    {
                        x = Transpose(x);
                    }
                }
                else
                {
                    for (int i = 0; i < 1000000; i++)
                    {
                        x = Transpose2(x);
                    }
                }
                sw.Stop();
                sink = (uint)x;
                Console.WriteLine("{0}: {1}", b ? "plain" : "mul  ", sw.ElapsedTicks);
            }
            Console.WriteLine("Throughput test:");
            for (int j = 0; j < 10; j++)
            {
                bool b = (j & 1) == 1;
                var sw = Stopwatch.StartNew();
                ulong x = 0;
                if (b)
                {
                    for (int i = 0; i < 1000000; i++)
                    {
                        sink = (uint)Transpose(0);
                    }
                }
                else
                {
                    for (int i = 0; i < 1000000; i++)
                    {
                        sink = (uint)Transpose2(0);
                    }
                }
                sw.Stop();
                Console.WriteLine("{0}: {1}", b ? "plain" : "mul  ", sw.ElapsedTicks);
            }
        }

        volatile static uint sink;

        [MethodImpl(MethodImplOptions.NoInlining)]
        public static ulong Transpose2(ulong state)
        {
            ulong r0 = state & 0xF0000F0000F0000FL; //unchanged diagonals

            ulong t = (state & 0x0000F000FF000FF0UL) * ((1UL << 12) + (1UL << 24));
            ulong r1 = (state & 0x0F0000F0000F0000L) >> 12;
            r0 |= (state & 0x00F0000F00000000L) >> 24;
            r1 |= (state & 0x000F000000000000L) >> 36;
            r0 |= (state & 0x000000000000F000L) << 36;
            r1 |= t & 0x0FF000FF000F0000UL;

            return r0 | r1;
        }

        [MethodImpl(MethodImplOptions.NoInlining)]
        public static ulong Transpose(ulong state)
        {
            return state & 0xF0000F0000F0000FL //unchanged diagonals
                | (state & 0x0F0000F0000F0000L) >> 12
                | (state & 0x0000F0000F0000F0L) << 12
                | (state & 0x00F0000F00000000L) >> 24
                | (state & 0x00000000F0000F00L) << 24
                | (state & 0x000F000000000000L) >> 36
                | (state & 0x000000000000F000L) << 36;
        }
    }
}