Optimize generated code for Fixed statement

[Anderman]

An extra statement is needed to use register vars

I have two functions

        private static unsafe int UnsafeSlow(byte[] src, int srcOffset, byte[] dst, int dstOffset)
        {
            var x = 0;
            fixed (byte* pSrc = &src[srcOffset])
            {
                x = *(pSrc + 0);
                x += *(pSrc + 1);
                return x;
            }
        }

With extra assignment

        private static unsafe int UnsafeFast(byte[] src, int srcOffset, byte[] dst, int dstOffset)
        {
            var x = 0;
            fixed (byte* pSrc1 = &src[srcOffset])
            {
                byte* pSrc = pSrc1;
                x = *(pSrc + 0);
                x += *(pSrc + 1);
                return x;
            }
        }

Code generated slow (line 9 and 11 do nothing)

            fixed (byte* pSrc = &src[srcOffset])
1  sub         rsp,28h  
2  xor         eax,eax  
3  mov         qword ptr [rsp+20h],rax  
4  cmp         edx,dword ptr [rcx+8]  
5  jae         000007FF622854D5  
6  movsxd      rax,edx  
7  lea         rax,[rcx+rax+10h]  
8  mov         qword ptr [rsp+20h],rax  
            {
                x = *(pSrc + 0);
9  mov         rax,qword ptr [rsp+20h]  
10  movzx       eax,byte ptr [rax]  
11  mov         rdx,qword ptr [rsp+20h]  
12  movzx       edx,byte ptr [rdx+1]  
13  add         eax,edx  
14  add         rsp,28h  
15  ret  

Fast version ( Line 9 does nothing)

1  sub         rsp,28h  
2  xor         eax,eax  
3  mov         qword ptr [rsp+20h],rax  
4  cmp         edx,dword ptr [rcx+8]  
5  jae         000007FF62285520  
6  movsxd      rax,edx  
7  lea         rax,[rcx+rax+10h]  
8  mov         qword ptr [rsp+20h],rax  
            {
                byte* pSrc = pSrc1;
9  mov         rax,qword ptr [rsp+20h]  
                x = *(pSrc + 0);
10  movzx       edx,byte ptr [rax]  
11  movzx       eax,byte ptr [rax+1]  
12  add         eax,edx  
13  add         rsp,28h  
14  ret  

category:cq
theme:pinning
skill-level:intermediate
cost:small

[deeprobin]

I can't reproduce this.
Sharplab

[teo-tsirpanis]

Closing, thanks for the heads-up @deeprobin. This issue seems to have been fixed since at least .NET Core 3.1.