Optimize nullcheck for Interlocked.* operations

[EgorBo]

Closes #44087

int _field;

void Test() => Interlocked.Increment(ref _field);

; Assembly listing for method Program:Test():this
-      cmp      byte  ptr [rcx], cl
       add      rcx, 8
       lock     
       add      dword ptr [rcx], 1
       ret      
-; Total bytes of code 11
+; Total bytes of code 9

No need for an extra nullcheck when interlocked operation is an implicit nullcheck itself. Follows #44087 (comment) advice

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

---

int _field;

void Test() => Interlocked.Increment(ref _field);

; Assembly listing for method Program:Test():this
-      cmp      byte  ptr [rcx], cl
       add      rcx, 8
       lock     
       add      dword ptr [rcx], 1
       ret      
-; Total bytes of code 9
+; Total bytes of code 11

No need for extra nullcheck when interlocked operation is an implicit nullcheck itself. Follows #44087 (comment) advice

Author:	EgorBo
Assignees:	EgorBo
Labels:	area-CodeGen-coreclr
Milestone:	-

[AndyAyersMS]

How hard would it be to build an address mode here too?

[AndyAyersMS]

Also maybe take a look at this regression? Look like maybe we don't propagate non-null assertions from interlocked ops or something?

;; x64 libraries_tests.pmi
+45 (+7.85%) : 314641.dasm - System.Reflection.Internal.Tests.AbstractMemoryBlockTests+<>c__DisplayClass6_0:<DisposeThreadSafety>b__0():this

[EgorBo]

Will take a look this week