Loads Per Cycle Test

LoadsPerCycleTest2.cs

 public class CpuLoadLimitTests
    {
        private readonly int[] _data = new int[1_000_000];

        [Benchmark(Baseline = true)]
        [MethodImpl(MethodImplOptions.AggressiveOptimization)]
        public unsafe int Sum_Simple()
        {
            var s0 = 0;
            var s1 = 0;
            var s2 = 0;
            var s3 = 0;

            fixed (int* data = &_data[0])
            {
                for (var i = 0; i < _data.Length; i += 4)
                {
                    unchecked
                    {
                        var p = &data[i];

                        s0 += p[0];
                        s1 += p[1];
                        s2 += p[2];
                        s3 += p[3];
                    }
                }

                return s0 + s1 + s2 + s3;
            }
        }

        [Benchmark]
        [MethodImpl(MethodImplOptions.AggressiveOptimization)]
        public unsafe int Sum_Tricky()
        {
            var s0 = 0;
            var s1 = 0;
            var s2 = 0;
            var s3 = 0;

            fixed (int* data = &_data[0])
            {
                for (var i = 0; i < _data.Length; i += 4)
                {
                    unchecked
                    {
                        var p = (long*)data;
                        var f = *p;
                        var s = *p + 1;

                        s0 += (int)(f >> 32);
                        s1 += (int)(f & 0xFFFF_FFFF);
                        s2 += (int)(s >> 32);
                        s3 += (int)(s & 0xFFFF_FFFF);

                        p += 2;
                    }
                }

                return s0 + s1 + s2 + s3;
            }
        }
    }