Different results when using jit and interpreter on optimization level 2 and higher

[Itay2805]

I have found that the following code returns different results between interpreted and non-interpreted code, and specifically on jit with optimization level 2 and higher.

$ ./c2m test.mir -d -eg
  execution       -- 0 msec
exit code: 1
$ ./c2m test.mir -d -ei
  execution       -- 0 msec
exit code: 0
$ ./c2m test.mir -d -O0 -ei
  execution       -- 0 msec
exit code: 0

The offending code:

mod: module

malloc_proto: proto p, i64:size

import malloc

box_nullable_proto: proto p, p, blk0:8(arg0)

box_nullable:   func    p, p, blk0:8(arg0)
    local   i64:exception, i64:si0, i64:sv0, i64:ti0, i64:ti1
# 1 arg, 5 locals
    alloca  sv0, 8
L15:
    mov si0, sv0
    mov i64:(si0), i64:(arg0)
L16:
    mov ti0, si0
    mov si0, 0
    bf  L17, i8:(ti0)
    call    malloc_proto, malloc, si0, 24
    add ti1, si0, 16
    add ti0, ti0, 4
    mov i32:(ti1), i32:(ti0)
L17:
L18:
    mov ti0, si0
    ret 0, ti0
    endfunc

init_nullable_proto: proto p, p:this, i32:arg0

init_nullable:  func    p, p:this, i32:arg0
    local   i64:exception, i64:si0, i64:si1, i64:ti0, i64:ti1
# 2 args, 5 locals
L78:
    mov si0, this
L79:
    mov si1, 1
L80:
    mov ti0, si1
    mov ti1, si0
    mov i8:(ti1), ti0
L81:
    mov si0, this
L82:
    mov si1, arg0
L83:
    mov ti0, si1
    mov ti1, si0
    mov i32:4(ti1), ti0
L84:
    ret 0
    endfunc

main:   func    i32
    local   i64:exception, i64:si0, i64:ti0, i64:sv0
# 0 args, 4 locals
    alloca  sv0, 8
L5:
    mov si0, 123
L6:
    mov ti0, si0
    mov si0, sv0
    mov i64:(si0), 0
    call    init_nullable_proto, init_nullable, exception, si0, ti0
    bf  L7, exception
    ret exception
L7:
L8:
    mov ti0, si0
    call    box_nullable_proto, box_nullable, exception, si0, blk0:8(ti0)
    bf  L9, exception
    ret exception
L9:
L10:
    mov ti0, si0
    bf  L11, ti0
L12:
    mov si0, 0
L13:
    mov ti0, si0
    ret ti0
L11:
    mov si0, 1
L14:
    mov ti0, si0
    ret ti0
    endfunc

endmodule

[vnmakarov]

Thank you for reporting this and sorry for delay with other issues (I was on vacation).

Fixing this bug probably will take some time but I hope to fix this on next week.

I'll provide updates on this issue status.

[Itay2805]

No worries! Hope you had a nice vacation :)

[vnmakarov]

I've pushed a fix fd737c15a13bca5ad58d502eddaa7239188c30db

The problem was in sparse conditional constant propagation (in incorrect processing call insns as insns having multiple outputs).

[Itay2805]

Great work!