JIT: Ignore source indir when checking interference for block store source address

[jakobbotsch]

For block stores we can contain both the source and destination address. Since the source is a value we will have an indirection on top of the address, such as

N015 ( 3, 12) [000014] I---------- t14 = CNS_INT(h) long 0xd1ffab1e static box ptr REG x1 $200
┌──▌ t14 long N017 ( 6, 14) [000001] #---G------ t1 = ▌ IND ref REG x1 $240
┌──▌ t1 ref N019 ( 8, 17) [000003] -c--G------ t3 = ▌ LEA(b+8) byref REG NA
┌──▌ t3 byref N021 ( 32, 39) [000004] nc-XG--N--- t4 = ▌ IND struct REG NA
N023 (???,???) [000025] Dc--------- t25 = LCL_FLD_ADDR byref V03 tmp1 [+0] NA REG NA
┌──▌ t25 byref ├──▌ t4 struct N025 ( 42, 46) [000017] sA--------- ▌ STORE_BLK struct<Xunit.StackFrameInfo, 16> (copy) (Unroll) REG NA

where [000004] is the source indirection and [000003] is the source address. The existing containment check was checking interference of [000003] with [000004], which is conservative given that the indirection itself is always contained.

This should fix some of the regressions seen in #78698.

[ghost]

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

Issue Details

---

For block stores we can contain both the source and destination address. Since the source is a value we will have an indirection on top of the address, such as

N017 ( 6, 14) [000001] #---G------ t1 = ▌ IND ref REG x1 $240
┌──▌ t1 ref
N019 ( 8, 17) [000003] -c--G------ t3 = ▌ LEA(b+8) byref REG NA
┌──▌ t3 byref
N021 ( 32, 39) [000004] nc-XG--N--- t4 = ▌ IND struct REG NA
N023 (???,???) [000025] Dc--------- t25 = LCL_FLD_ADDR byref V03 tmp1 [+0] NA REG NA
┌──▌ t25 byref
├──▌ t4 struct
N025 ( 42, 46) [000017] sA--------- ▌ STORE_BLK struct<Xunit.StackFrameInfo, 16> (copy) (Unroll) REG NA

where [000004] is the source indirection and [000003] is the source address. The existing containment check was checking interference of [000003] with [000004], which is conservative given that the indirection itself is always contained.

This should fix some of the regressions seen in #78698.

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

[jakobbotsch]

cc @dotnet/jit-contrib PTAL @SingleAccretion@EgorBo