Skip Null check and zero length checks of Pointers

[Anderman]

pointers to byte arrays are use as
fixed(byte* ptr arr)
the code generated will check if this is an null or zero length array

To skip these tests the statement can be written as
fixed(byte* ptr &arr[0])

see generated code

fixed(byte* ptr arr)

000007FF622857AB  mov         rax,rcx  
000007FF622857AE  test        rax,rax  
000007FF622857B1  je          000007FF622857BA  
000007FF622857B3  mov         ecx,dword ptr [rax+8]  
000007FF622857B6  test        ecx,ecx  
000007FF622857B8  jne         000007FF622857C3  
000007FF622857BA  xor         ecx,ecx  
000007FF622857BC  mov         qword ptr [rsp+20h],rcx  
000007FF622857C1  jmp         000007FF622857D1  
000007FF622857C3  cmp         ecx,0  
000007FF622857C6  jbe         000007FF622857FE  
000007FF622857C8  add         rax,10h  
000007FF622857CC  mov         qword ptr [rsp+20h],rax 

fixed(byte* ptr &arr[0])

000007FF6228582B  cmp         dword ptr [rcx+8],0  
000007FF6228582F  jbe         000007FF62285867  
000007FF62285831  add         rcx,10h  
000007FF62285835  mov         qword ptr [rsp+20h],rcx 

[benaadams]

Interesting :)

[Anderman]

This one is also nice to know Optimize generated code for Fixed statement. Looks like a clr or roslyn issue

[clrjunkie]

+1

[halter73]

@Anderman did #542 address all the places where we should use the fixed(byte* ptr &arr[0]) pattern?

[Anderman]

Yes, all that I could find. I skipped 1 or 2 because they didn't have much impact.

[Anderman]

@halter73 I think @benaadams did already change some in his PR

[halter73]

I also noticed @benaadams used the same pattern in at least one of his PRs.

Thanks! Just wanted to double check before closing.

[benaadams]

I also noticed @benaadams used the same pattern in at least one of his PRs.

@Anderman is making me up my game :)

[Anderman]

Ha, Thx there still is many things that can be improved

[halter73]

@Anderman No doubt there are many more things that can be improved, but we've decided we are going to lock down stop accepting new perf related PRs prior to our 1.0 release. This is meant to give us more time to focus on other things like stability and security.

[clrjunkie]

@halter73
👍