jitstress failing widely with NRE

[jakobbotsch]

There are many NRE failures in the latest jitstress/libraries-jitstress runs.

[ghost]

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

Issue Details

---

There are many NRE failures in the latest jitstress/libraries-jitstress runs.

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

[jakobbotsch]

The first JIT commit that fails is 8ff1893. Oddly jitstress/libraries-jitstress both passed in #92710, but I'll take a closer look.

[jakobbotsch]

I made a last minute change in #92710 to remove an ordering side effect on a LCL_VAR where it wasn't supported, but that reveals that we're missing an ordering side effect in morph in some cases:

 fgMorphTree BB03, STMT00005 (before)
                [000018] DAC--------                         ▌  STORE_LCL_VAR ubyte  V02 tmp1         
                [000027] --C--------                         └──▌  EQ        int   
                [000036] ---XG------                            ├──▌  EQ        int   
                [000034] ---XG------                            │  ├──▌  AND       int   
                [000032] n--XG------                            │  │  ├──▌  IND       int   
                [000031] ---X-------                            │  │  │  └──▌  FIELD_ADDR byref  System.UriParser:_flags
                [000014] -----------                            │  │  │     └──▌  LCL_VAR   ref    V03 tmp2          (last use)
                [000033] -----------                            │  │  └──▌  CNS_INT   int    0x10000000
                [000035] -----------                            │  └──▌  CNS_INT   int    0
                [000026] -----------                            └──▌  CNS_INT   int    0
...

-     ( 15, 14) [000018] DA--GO-----                         ▌  STORE_LCL_VAR int    V02 tmp1         
-     ( 11, 11) [000036] ----GO-----                         └──▌  NE        int   
-     (  6,  9) [000034] ----GO-----                            ├──▌  AND       int   
-     (  4,  4) [000032] n---GO-----                            │  ├──▌  IND       int   
-     (  2,  2) [000049] -----O-N---                            │  │  └──▌  ADD       byref 
-     (  1,  1) [000047] -----O-----                            │  │     ├──▌  LCL_VAR   ref    V03 tmp2         
+     ( 15, 14) [000018] DA--G------                         ▌  STORE_LCL_VAR int    V02 tmp1         
+     ( 11, 11) [000036] ----G------                         └──▌  NE        int   
+     (  6,  9) [000034] ----G------                            ├──▌  AND       int   
+     (  4,  4) [000032] n---G------                            │  ├──▌  IND       int   
+     (  2,  2) [000049] -------N---                            │  │  └──▌  ADD       byref 
+     (  1,  1) [000047] -----------                            │  │     ├──▌  LCL_VAR   ref    V03 tmp2         
      (  1,  1) [000048] -----------                            │  │     └──▌  CNS_INT   long   16 Fseq[_flags]
      (  1,  4) [000033] -----------                            │  └──▌  CNS_INT   int    0x10000000
      (  1,  1) [000035] -----------                            └──▌  CNS_INT   int    0

We prove that the indirection [000032] is non-faulting, but that requires GTF_ORDER_SIDEEFF to be set on it. That was set on it previously only because it was propagated from [000047].