Reduce string interpolation overhead

[pentp]

For Nullable<T> where T:ISpanFormattable removes a lot of dead code for null values (ToString and AppendLiteral calls).
For reference types removes many redundant null checks and two interface casts (CORINFO_HELP_CHKCASTINTERFACE).

For non-null Nullable<T> the CORINFO_HELP_BOX_NULLABLE allocation still remains - this needs fixing in JIT.

[Copilot]

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

[EgorBo]

For non-null Nullable the CORINFO_HELP_BOX_NULLABLE allocation still remains - this needs fixing in JIT.

@pentp is there a minimal repro for it?

[pentp]

For non-null Nullable the CORINFO_HELP_BOX_NULLABLE allocation still remains - this needs fixing in JIT.

@pentp is there a minimal repro for it?

Minimal repro: https://godbolt.org/z/WcnoMG9GK

[EgorBo]

For non-null Nullable the CORINFO_HELP_BOX_NULLABLE allocation still remains - this needs fixing in JIT.

@pentp is there a minimal repro for it?

Minimal repro: https://godbolt.org/z/WcnoMG9GK

@AndyAyersMS looks like something EA could help with?

***** BB02 [0002]
STMT00004 ( 0x00D[E-] ... ??? ) <- INLRT @ 0x000[E-]
[000022] DAC-G------ * STORE_LCL_VAR ref V02 tmp1 [000019] --C-G------ \--* CALL nullcheck ref System.Guid:ToString(System.String,System.IFormatProvider):System.String:this
[000021] --C-G------ this +--* ADD byref [000015] --C-G------ | +--* CALL help ref CORINFO_HELP_BOX_NULLABLE
[000013] H---------- arg0 | | +--* CNS_INT(h) long 0x7ffab4be4648 class System.Nullable`1[System.Guid]
[000014] ----------- arg1 | | \--* LCL_ADDR byref V00 arg0 [+0]
[000020] ----------- | \--* CNS_INT long 8
[000017] ----------- arg1 +--* CNS_INT ref null
[000018] ----------- arg2 \--* CNS_INT ref null

[AndyAyersMS]

@AndyAyersMS looks like something EA could help with?

If we expand CORINFO_HELP_BOX_NULLABLE inline then perhaps... or we could try and handle it as an allocator and then inline expand it when it does not escape. That would require spilling the result of the helper call to a temp in the importer.

[stephentoub]

/ba-g failures are unrelated