[Bug tree-optimization/122478] New: [16 Regression] Recent match.pd change breaks gimple checking

[law at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122478

            Bug ID: 122478
           Summary: [16 Regression] Recent match.pd change breaks gimple
                    checking
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: law at gcc dot gnu.org
  Target Milestone: ---

This change:

commit 5eafa8d16be87339acd11446283c571b2416a532 (HEAD)
Author: Artemiy Volkov <artemiy.vol...@arm.com>
Date:   Mon Oct 27 14:47:31 2025 +0000

    match.pd: simplify (view_convert (BIT_FIELD_REF))

    This patch adds a match.pd transformation to strip unnecessary
    view_converts of BIT_FIELD_REFs.  This is only valid when the type of the
    view_convert is a register type, and no loss of precision is involved when
    casting to an integer type; both of these conditions are checked for in
    the if clause of the new pattern.

    The change survives bootstrap + regtest on aarch64 and x86_64, and one
    additional test case has been added to gcc.dg/tree-ssa.

    Changes since v1:
            - Require is_gimple_reg_type (type) and remove conditions on @0
            - Reword the commit message to reflect code changes

    gcc/ChangeLog:

            * match.pd: Add pattern to simplify view_convert (BIT_FIELD_REF).

    gcc/testsuite/ChangeLog:

            * gcc.dg/tree-ssa/forwprop-42.c: New test.


Tests that now fail, but worked before (1 tests):

unix//-march=rv32gcv: gcc: gcc.target/riscv/rvv/autovec/pr114668.c -O3
-ftree-vectorize (test for excess errors)

In the log file we see:
Excess errors:
/home/jlaw/test/gcc/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr114668.c:18:5:
error: integral result type precision does not match field size of
'bit_field_ref'
_180 = BIT_FIELD_REF <mask__701.18_30, 1, 12>;
during GIMPLE pass: forwprop

You should be able to see this with a cross compiler, so you don't need a full
testing setup.  Compile the attached test with:

-O3 -march=rv32gcv -fno-vect-cost-model

char a;
int b;
short e[14];
char f[4][12544];
_Bool c[4][5];

__attribute__ ((noipa))
void foo (int a)
{
  if (a != 1)
    __builtin_abort ();
}

int main ()
{
  for (int i = 0; i < 4; ++i)
    for (int l = 0; l < 15; ++l)
      for (int m = 0; m < 15; ++m)
        f[i][l * m] = 3;
  for (int j = 0; j < 4; j += 1)
    for (int k = 3; k < 13; k += 3)
      for (_Bool l = 0; l < 1; l = 1)
        for (int m = 0; m < 4; m += 1)
          {
            a = 0;
            b -= e[k];
            c[j][m] = f[j][6];
          }
  for (long i = 2; i < 4; ++i)
    foo (c[3][3]);
}