https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122215
Bug ID: 122215
Summary: Nondeterminism in ICF on riscv32
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: ipa
Assignee: unassigned at gcc dot gnu.org
Reporter: sjames at gcc dot gnu.org
CC: amonakov at gcc dot gnu.org
Target Milestone: ---
Created attachment 62524
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62524&action=edit
foo.c
This was reported in #gcc by cousteau in August and amonakov just reminded me
about it in the context of something else.
godbolt: https://gcc.godbolt.org/z/jE7791r1a (attached as foo.c).
The program has strange non-determinism from ICF where we sometimes have:
```
dual_reg_insn2:
li a4,0
li a5,0
some_custom_insn a0,a0,a4
ret
```
and other times (*) have:
```
dual_reg_insn2:
addi sp,sp,-16
sw s7,12(sp)
sw s8,8(sp)
li s7,0
li s8,0
mv a4,s7
mv a5,s8
lw s7,12(sp)
lw s8,8(sp)
some_custom_insn a0,a0,a4
addi sp,sp,16
jr ra
```
Playing around on godbolt, I can't repro it with trunk, but that might just be
luck (or it really is just latent now).
(*) Just rebuilding with an added comment or trivial change to *FLAGS like -gN
will do it.
