[Bug target/114476] New: [14] RISC-V rv64gcv vector: Runtime mismatch at -O3

[Bug target/114476] New: [14] RISC-V rv64gcv vector: Runtime mismatch at -O3

[patrick at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

            Bug ID: 114476
           Summary: [14] RISC-V rv64gcv vector: Runtime mismatch at -O3
           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:
signed char b = 9;
int c[23][23];
signed char d[23];
int main() {
  for (int h = 0; h < 23; h += 2) {
    c[h][h] = d[h];
    b *= 255;
  }
  __builtin_printf("%u\n", b);
}

Commands:
> /scratch/tc-testing/tc-mar-25/build-rv64gcv/bin/riscv64-unknown-linux-gnu-gcc -march=rv64gcv -O3 -fwrapv red.c -o red.out
> /scratch/tc-testing/tc-mar-25/build-rv64gcv/bin/qemu-riscv64 red.out
4294967287
> /scratch/tc-testing/tc-mar-25/build-rv64gcv/bin/riscv64-unknown-linux-gnu-gcc -march=rv64gcv -O2 -fwrapv red.c -o red.out
> /scratch/tc-testing/tc-mar-25/build-rv64gcv/bin/qemu-riscv64 red.out
9

Discovered/tested using r14-9654-g78b56a12dd0 (not bisected)

Found via fuzzer.

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-03-26
          Component|target                      |tree-optimization
     Ever confirmed|0                           |1
      Known to fail|                            |13.2.0
             Target|riscv*-*-*                  |riscv*-*-* aarch64
           Keywords|                            |needs-bisection
            Summary|[14] RISC-V rv64gcv vector: |[13/14 Regression] wrong
                   |Runtime mismatch at -O3     |code with -fwrapv -O3
                   |                            |-fno-vector-cost-mode (and
                   |                            |-march=armv9-a+sve2 on
                   |                            |aarch64 and -march=rv64gcv
                   |                            |on riscv)
             Status|UNCONFIRMED                 |NEW
   Target Milestone|---                         |14.0
      Known to work|                            |12.3.0

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Happens on aarch64 with `-march=armv9-a+sve2 -fno-vect-cost-model` even in GCC
13.2.0.

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note -fwrapv is important here ...

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Here is a testcase where we don't need -fwrapv:
```
signed int b = 9;
int c[23][23];
signed int d[23];
int main() {
  for (int h = 0; h < 23; h += 2) {
    c[h][h] = d[h];
    b= -b;
  }
  __builtin_printf("%u\n", b);
}

```

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> Here is a testcase where we don't need -fwrapv:
Ignore that, I was doing something wrong.

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|14.0                        |13.3
           Priority|P3                          |P2

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vector-cost-mode (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #5 from Robin Dapp <rdapp at gcc dot gnu.org> ---
So the result is -9 instead of 9 (or vice versa) and this happens (just) with
vectorization.  We only vectorize with -fwrapv.

From a first quick look, the following is what we have before vect:

(loop)
  <bb 3> [local count: 991171080]:
  ...
  # b_lsm.5_5 = PHI <_4(7), b_lsm.5_17(2)>
  ...
  _4 = -b_lsm.5_5;
(check)
    <bb 4> [local count: 82570744]:
  ...
  # b_lsm.5_22 = PHI <b_lsm.5_5(3)>
  ...
  if (b_lsm.5_22 != -9)

I.e. b gets negated with every iteration and we check the second to last
against -9.

With vectorization we have:
(init)
  <bb 2> [local count: 82570744]:
  b_lsm.5_17 = b;

(vectorized loop)
  <bb 3> [local count: 247712231]:
  ...
  # b_lsm.5_5 = PHI <_4(7), b_lsm.5_17(2)>
  ...
    _4 = -b_lsm.5_5;
  ...
  goto <bb 3>

(epilogue)
  <bb 10> [local count: 82570741]:
  ...
  # b_lsm.5_7 = PHI <_25(11), b_lsm.5_17(13)>
  ...
  _25 = -b_lsm.5_7;

(check)
  <bb 4> [local count: 82570744]:
  ...
  # b_lsm.5_22 = PHI <b_lsm.5_7(10)>
  if (b_lsm.5_22 != -9)

What looks odd here is that b_lsm.5_7's fallthrough argument is b_lsm.5_17 even
though we must have come through the vectorized loop (which negated b at least
once).  This makes us skip inversions.
Indeed, as b_lsm.5_22 is only dependent on the initial value of b it gets
optimized away and we compare b != -9.

Maybe I missed something but it looks like
  # b_lsm.5_7 = PHI <_25(11), b_lsm.5_17(13)>
should have b_lsm.5_5 or _4 as fallthrough argument.

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vect-cost-model (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I am going to mark this as a dup of bug 114485 since the issue is exactly the
same.

*** This bug has been marked as a duplicate of bug 114485 ***

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vect-cost-model (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[juzhe.zhong at rivai dot ai]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #7 from JuzheZhong <juzhe.zhong at rivai dot ai> ---
Hi, Robin.

Will you fix this bug ?

[Bug tree-optimization/114476] [13/14 Regression] wrong code with -fwrapv -O3 -fno-vect-cost-model (and -march=armv9-a+sve2 on aarch64 and -march=rv64gcv on riscv)

[rdapp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114476

--- Comment #8 from Robin Dapp <rdapp at gcc dot gnu.org> ---
I tried some things (for the related bug without -fwrapv) then got busy with
some other things.  I'm going to have another look later this week.