[Bug target/114399] New: [14] RISC-V rv32i miscompile

[Bug target/114399] New: [14] RISC-V rv32i miscompile

[patrick at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114399

            Bug ID: 114399
           Summary: [14] RISC-V rv32i miscompile
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: patrick at rivosinc dot com
  Target Milestone: ---

Testcase:
int a;
struct b {
  long long c;
};
struct d {
  short e;
  long long f;
};
union {
  struct b g;
  struct d h;
} i;
int main() {
  i.h.f++;
  a = i.g.c >> 1;
  __builtin_printf("%X\n", a);
}

Commands:
> gcc red.c -o red.out -m32 -fno-strict-aliasing -static
> ./red.out
80000000

> riscv64-unknown-linux-gnu-gcc -mabi=ilp32 -march=rv32i red.c -o red.out -fno-strict-aliasing -static
> /scratch/tc-testing/tc-mar-19/build-rv32-64-gcv/bin/qemu-riscv32 red.out
0

Discovered/tested with r14-9544-gc7a774edbf8 (not bisected)

Found via fuzzer.

I don't think this has any undefined behavior but the combo of union/struct has
me a bit wary.

[Bug target/114399] [14] RISC-V rv32i miscompile

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114399

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
32bit i686 long long has an alignment of 4 while most other targets have an
alignment of 8. 

Which you could see via sizeof.

```
int tt = sizeof(struct b);
int tt1 = sizeof(struct d);
```

i?86:
```
tt:
        .long   8
...
tt1:
        .long   12
```

RISCV32:
```
tt:
        .word   8
...
tt1:
        .word   16
```

Basically in the i?86 case, i.h.f++ will modify byte 4 of the union while on
most other targets, it will modify byte 8.

[Bug target/114399] [14] RISC-V rv32i miscompile

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114399

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
That is doing:
  long long f __attribute__((packed,aligned(4)));


Will produce the same result between the 2 targets.

Likewise:
  long long f __attribute__((packed,aligned(8)));


With  -Wpadded you can see the padding between e and f is added.

<source>:7:13: warning: padding struct to align 'd::f' [-Wpadded]


Just the padding is different between the 2 targets due to the alignment of
long long being 4 bytes (it is an old ABI where 4 bytes was the only alignment
at the time when long long was added, double have a similar thing too).

[Bug target/114399] [14] RISC-V rv32i miscompile

[patrick at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114399

Patrick O'Neill <patrick at rivosinc dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|INVALID                     |FIXED

--- Comment #3 from Patrick O'Neill <patrick at rivosinc dot com> ---
Ah thank you - that makes sense.
IIUC -malign-double is the option I'm looking for to set the padding to be the
same between x86/risc-v.

[Bug target/114399] [14] RISC-V rv32i miscompile

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114399

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|FIXED                       |INVALID

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
.