JIT: Forward sub does not kick in in some physical promotion cases

[jakobbotsch]

Example from @stephentoub:

public class Tests
{
    private ParsedStat _stat;

    public ulong GetTime()
    {
        ParsedStat stat = _stat;
        return stat.utime + stat.stime;
    }

    internal struct ParsedStat
    {
        internal int pid;
        internal string comm;
        internal char state;
        internal int ppid;
        internal int session;
        internal ulong utime;
        internal ulong stime;
        internal long nice;
        internal ulong starttime;
        internal ulong vsize;
        internal long rss;
        internal ulong rsslim;
    }
}

Codegen:

;  V00 this         [V00,T01] (  3,  3   )     ref  ->  rcx         this class-hnd single-def <Program>
;* V01 loc0         [V01    ] (  0,  0   )  struct (80) zero-ref    do-not-enreg[SF] <Program+ParsedStat>
;# V02 OutArgs      [V02    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;  V03 tmp1         [V03,T02] (  2,  2   )    long  ->  rax         single-def "V01.[008..016)"
;  V04 tmp2         [V04,T03] (  2,  2   )    long  ->  rcx         single-def "V01.[016..024)"
;  V05 tmp3         [V05,T00] (  3,  6   )   byref  ->  rcx         single-def "Spilling address for field-by-field copy"
;
; Lcl frame size = 0

G_M25807_IG01:  ;; offset=0x0000
						;; size=0 bbWeight=1 PerfScore 0.00
G_M25807_IG02:  ;; offset=0x0000
       add      rcx, 8
       mov      rax, qword ptr [rcx+0x08]
       mov      rcx, qword ptr [rcx+0x10]
       add      rax, rcx
						;; size=15 bbWeight=1 PerfScore 4.50
G_M25807_IG03:  ;; offset=0x000F
       ret      

Expected codegen:

mov rax, qword ptr [rcx+0x10]
add rax, qword ptr [rcx+0x18]
ret

The add rcx, 8 can be fixed via #88386. However, the containment issue happens because assignment decomposition produces commas:

***** BB01
STMT00000 ( 0x000[E-] ... 0x006 )
               [000026] -A-XG------                         ▌  COMMA     void  
               [000014] DA-X-------                         ├──▌  STORE_LCL_VAR byref  V05 tmp3         
               [000001] ---X-------                         │  └──▌  FIELD_ADDR byref  Program:_stat
               [000000] -----------                         │     └──▌  LCL_VAR   ref    V00 this          (last use)
               [000025] -A--G------                         └──▌  COMMA     void  
               [000020] DA--G------                            ├──▌  STORE_LCL_VAR long   V03 tmp1         
               [000019] n---G------                            │  └──▌  IND       long  
               [000018] -----------                            │     └──▌  ADD       byref 
               [000016] -----------                            │        ├──▌  LCL_VAR   byref  V05 tmp3         
               [000017] -----------                            │        └──▌  CNS_INT   long   8
               [000024] DA--G------                            └──▌  STORE_LCL_VAR long   V04 tmp2         
               [000023] n---G------                               └──▌  IND       long  
               [000022] -----------                                  └──▌  ADD       byref 
               [000015] -----------                                     ├──▌  LCL_VAR   byref  V05 tmp3         
               [000021] -----------                                     └──▌  CNS_INT   long   16

***** BB01
STMT00001 ( 0x007[E-] ... 0x014 )
               [000013] -----------                         ▌  RETURN    long  
               [000012] -----------                         └──▌  ADD       long  
               [000027] -----------                            ├──▌  LCL_VAR   long   V03 tmp1          (last use)
               [000028] -----------                            └──▌  LCL_VAR   long   V04 tmp2          (last use)

We should be able to produce top level statements instead to allow forward sub to kick in (or alternatively, teach forward sub about commas).

[ghost]

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

Issue Details

---

Example from @stephentoub:

public class Tests
{
    private ParsedStat _stat;

    public ulong GetTime()
    {
        ParsedStat stat = _stat;
        return stat.utime + stat.stime;
    }

    internal struct ParsedStat
    {
        internal int pid;
        internal string comm;
        internal char state;
        internal int ppid;
        internal int session;
        internal ulong utime;
        internal ulong stime;
        internal long nice;
        internal ulong starttime;
        internal ulong vsize;
        internal long rss;
        internal ulong rsslim;
    }
}

Codegen:

;  V00 this         [V00,T01] (  3,  3   )     ref  ->  rcx         this class-hnd single-def <Program>
;* V01 loc0         [V01    ] (  0,  0   )  struct (80) zero-ref    do-not-enreg[SF] <Program+ParsedStat>
;# V02 OutArgs      [V02    ] (  1,  1   )  struct ( 0) [rsp+0x00]  do-not-enreg[XS] addr-exposed "OutgoingArgSpace"
;  V03 tmp1         [V03,T02] (  2,  2   )    long  ->  rax         single-def "V01.[008..016)"
;  V04 tmp2         [V04,T03] (  2,  2   )    long  ->  rcx         single-def "V01.[016..024)"
;  V05 tmp3         [V05,T00] (  3,  6   )   byref  ->  rcx         single-def "Spilling address for field-by-field copy"
;
; Lcl frame size = 0

G_M25807_IG01:  ;; offset=0x0000
						;; size=0 bbWeight=1 PerfScore 0.00
G_M25807_IG02:  ;; offset=0x0000
       add      rcx, 8
       mov      rax, qword ptr [rcx+0x08]
       mov      rcx, qword ptr [rcx+0x10]
       add      rax, rcx
						;; size=15 bbWeight=1 PerfScore 4.50
G_M25807_IG03:  ;; offset=0x000F
       ret      

Expected codegen:

mov rax, qword ptr [rcx+0x10]
add rax, qword ptr [rcx+0x18]
ret

The add rcx, 8 can be fixed via #88386. However, the containment issue happens because assignment decomposition produces commas:

***** BB01
STMT00000 ( 0x000[E-] ... 0x006 )
               [000026] -A-XG------                         ▌  COMMA     void  
               [000014] DA-X-------                         ├──▌  STORE_LCL_VAR byref  V05 tmp3         
               [000001] ---X-------                         │  └──▌  FIELD_ADDR byref  Program:_stat
               [000000] -----------                         │     └──▌  LCL_VAR   ref    V00 this          (last use)
               [000025] -A--G------                         └──▌  COMMA     void  
               [000020] DA--G------                            ├──▌  STORE_LCL_VAR long   V03 tmp1         
               [000019] n---G------                            │  └──▌  IND       long  
               [000018] -----------                            │     └──▌  ADD       byref 
               [000016] -----------                            │        ├──▌  LCL_VAR   byref  V05 tmp3         
               [000017] -----------                            │        └──▌  CNS_INT   long   8
               [000024] DA--G------                            └──▌  STORE_LCL_VAR long   V04 tmp2         
               [000023] n---G------                               └──▌  IND       long  
               [000022] -----------                                  └──▌  ADD       byref 
               [000015] -----------                                     ├──▌  LCL_VAR   byref  V05 tmp3         
               [000021] -----------                                     └──▌  CNS_INT   long   16

***** BB01
STMT00001 ( 0x007[E-] ... 0x014 )
               [000013] -----------                         ▌  RETURN    long  
               [000012] -----------                         └──▌  ADD       long  
               [000027] -----------                            ├──▌  LCL_VAR   long   V03 tmp1          (last use)
               [000028] -----------                            └──▌  LCL_VAR   long   V04 tmp2          (last use)

We should be able to produce top level statements instead to allow forward sub to kick in (or alternatively, teach forward sub about commas).

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