Pointer alignment

Pointer_alignment.cs

using System;
using System.Numerics;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;

namespace ConsoleApp1
{
    class Program
    {
        static unsafe void Main(string[] args)
        {
            double[] arr      = new double[1_000];  // is aligned (from GC / runtime)
            Span<double> span = arr.AsSpan(1);      // is not aligned by slicing, so next boundary will be at +3 = 4 again

            fixed (double* pArr = &MemoryMarshal.GetReference(span))
            {
                double* ptr = pArr;

                int elementsToAlign = PointerHelper.GetElementsToAlign<double>(ptr);
#if DEBUG
                Console.WriteLine(elementsToAlign);
#endif
                while (elementsToAlign > 0)
                {
                    elementsToAlign--;
                    ptr++;
                }

                elementsToAlign = PointerHelper.GetElementsToAlign<double>(ptr);
#if DEBUG
                Console.WriteLine(elementsToAlign);
#endif
            }
        }
    }
    //-------------------------------------------------------------------------
    internal static unsafe class PointerHelper
    {
        [MethodImpl(MethodImplOptions.AggressiveInlining)]
        public static int GetElementsToAlign<T>(void* ptr) where T : struct
        {
            //const int elementsPerByte = sizeof(double) / sizeof(byte);
            int elementsPerByte         = Unsafe.SizeOf<T>() / sizeof(byte);
            int sizeOfVector            = Unsafe.SizeOf<Vector<T>>();
            int vectorElements          = Vector<T>.Count;

            int address           = (int)ptr;
            int unalignedBytes    = address & (sizeOfVector - 1);
            int unalignedElements = unalignedBytes / elementsPerByte;

            // (vectorElements - unalignedElements) would be OK, but only in the case
            // unalignedElements > 0. For the 0-case the % has to be done.
            int elementsToAlign = (vectorElements - unalignedElements) & (vectorElements - 1);

            return elementsToAlign;

            /*
             * Bit hack for modulus: https://graphics.stanford.edu/~seander/bithacks.html#ModulusDivisionEasy
             * a % b = a & (b - 1)
             * is way faster :-)
             *
             * See also:
             * https://github.com/ahsonkhan/coreclr/blob/46b075fc1877e7087da53579c13c8c9069058b42/src/mscorlib/shared/System/SpanHelpers.Char.cs#L95-L97
             * https://github.com/ahsonkhan/coreclr/blob/46b075fc1877e7087da53579c13c8c9069058b42/src/mscorlib/shared/System/SpanHelpers.Char.cs#L131
             */
        }
    }
}

JIT will produce

; Assembly listing for method Program:Do(long):int
; Emitting BLENDED_CODE for X64 CPU with AVX
; optimized code
; rbp based frame
; fully interruptible
; Final local variable assignments
;
;  V00 arg0         [V00,T01] (  6, 12   )    long  ->  rdi        
;  V01 loc0         [V01,T00] (  7, 16   )     int  ->  rax        
;* V02 tmp0         [V02    ] (  0,  0   )     int  ->  zero-ref   
;* V03 tmp1         [V03    ] (  0,  0   )     int  ->  zero-ref   
;* V04 tmp2         [V04    ] (  0,  0   )     int  ->  zero-ref   
;  V05 tmp3         [V05,T04] (  2,  2   )     int  ->  rax        
;* V06 tmp4         [V06    ] (  0,  0   )     int  ->  zero-ref   
;* V07 tmp5         [V07    ] (  0,  0   )     int  ->  zero-ref   
;* V08 tmp6         [V08    ] (  0,  0   )     int  ->  zero-ref   
;  V09 tmp7         [V09,T05] (  2,  2   )     int  ->  rax        
;# V10 OutArgs      [V10    ] (  1,  1   )  lclBlk ( 0) [rsp+0x00]  
;  V11 rat0         [V11,T02] (  3,  6   )     int  ->  rax        
;  V12 rat1         [V12,T03] (  3,  6   )     int  ->  rax        
;
; Lcl frame size = 0

G_M56189_IG01:
       55                   push     rbp
       488BEC               mov      rbp, rsp

G_M56189_IG02:
       8BC7                 mov      eax, edi
       83E01F               and      eax, 31
       8BF0                 mov      esi, eax
       C1FE1F               sar      esi, 31
       83E607               and      esi, 7
       03F0                 add      esi, eax
       8BC6                 mov      eax, esi
       C1F803               sar      eax, 3
       F7D8                 neg      eax
       83C004               add      eax, 4
       83E003               and      eax, 3
       85C0                 test     eax, eax
       7E0A                 jle      SHORT G_M56189_IG04

G_M56189_IG03:
       FFC8                 dec      eax
       4883C708             add      rdi, 8
       85C0                 test     eax, eax
       7FF6                 jg       SHORT G_M56189_IG03

G_M56189_IG04:
       8BC7                 mov      eax, edi
       83E01F               and      eax, 31
       8BF8                 mov      edi, eax
       C1FF1F               sar      edi, 31
       83E707               and      edi, 7
       03F8                 add      edi, eax
       8BC7                 mov      eax, edi
       C1F803               sar      eax, 3
       F7D8                 neg      eax
       83C004               add      eax, 4
       83E003               and      eax, 3

G_M56189_IG05:
       5D                   pop      rbp
       C3                   ret      

; Total bytes of code 76, prolog size 4 for method Program:Do(long):int
; ============================================================