[Bug tree-optimization/124272] New: unlink_stmt_vdef is sometimes not used with release_ssa_name/release_defs

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124272

            Bug ID: 124272
           Summary: unlink_stmt_vdef is sometimes not used with
                    release_ssa_name/release_defs
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Keywords: compile-time-hog, internal-improvement, memory-hog
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

I Noticed this while working on the ipa-fnsplit.cc patch.
We would leak some unused ssa names in this case.

gimple-ssa-sprintf.cc:
```
          /* Remove the call to the bounded function with a zero size
             (e.g., snprintf(0, 0, "%i", 123)) if there is no lhs.  */
          unlink_stmt_vdef (info.callstmt);
          gsi_remove (gsi, true);
          removed = true;
```

sanopt.cc:
```
                {
                  unlink_stmt_vdef (use_stmt);
                  gimple_stmt_iterator gsi2 = gsi_for_stmt (use_stmt);
                  gsi_remove (&gsi2, true);
                  remove = true;
                }
...
          unlink_stmt_vdef (stmt);
          gsi_remove (&gsi, true);
```

tree-eh.cc (cleanup_empty_eh_move_lp):
```
  unlink_stmt_vdef (gsi_stmt (gsi));
  gsi_remove (&gsi, true);

```

tree-parloops.cc (eliminate_local_variables_stmt):
```
      unlink_stmt_vdef (stmt);
      stmt = gimple_build_nop ();
      gsi_replace (gsi, stmt, false);
      dta.changed = true;
```

tree-ssa-dse.cc (delete_dead_or_redundant_call) (Just the first one, the second
is fine with release_defs):
```
  if (lhs)
    {
      tree ptr = gimple_call_arg (stmt, 0);
      gimple *new_stmt = gimple_build_assign (lhs, ptr);
      unlink_stmt_vdef (stmt);
      if (gsi_replace (gsi, new_stmt, true))
        bitmap_set_bit (need_eh_cleanup, bb->index);
    }
  else
    {
      /* Then we need to fix the operand of the consuming stmt.  */
      unlink_stmt_vdef (stmt);

      /* Remove the dead store.  */
      if (gsi_remove (gsi, true))
        bitmap_set_bit (need_eh_cleanup, bb->index);
      release_defs (stmt);
    }
```

tree-ssa-forwprop.cc (optimize_vector_load) (double check this one):
```

      gimple_stmt_iterator gsi2 = gsi_for_stmt (use_stmt);
      unlink_stmt_vdef (use_stmt);
      gsi_remove (&gsi2, true);
    }
```

tree-ssa-phiprop.cc:
```

          /* Unlinking the VDEF here is fine as we are sure that we process
             stmts in execution order due to aggregate copies having VDEFs
             and we emit loads on the edges in the very same order.
             We get multiple copies (or intermediate register loads) handled
             only by walking PHIs or immediate uses in a lucky order though,
             so we could signal the caller to re-start iterating over PHIs
             when we come here which would make it quadratic in the number
             of PHIs.  */
          unlink_stmt_vdef (use_stmt);
          gsi_remove (&gsi, true);
```

tree-ssa-strlen.cc (handle_builtin_memset) (like the tree-ssa-dse.cc one):
```
  unlink_stmt_vdef (memset_stmt);
  if (lhs)
    {
      gimple *assign = gimple_build_assign (lhs, ptr);
      gsi_replace (&m_gsi, assign, false);
    }
  else
    {
      gsi_remove (&m_gsi, true);
      release_defs (memset_stmt);
    }
```

tree-vect-slp.cc (vect_remove_slp_scalar_calls):
```
      unlink_stmt_vdef (stmt_info->stmt);
      gsi = gsi_for_stmt (stmt);
      vinfo->replace_stmt (&gsi, stmt_info, new_stmt);
      if (lhs)
        SSA_NAME_DEF_STMT (lhs) = new_stmt;
```


tree-vectorizer.cc (adjust_simduid_builtins):
```
            case IFN_GOMP_SIMD_ORDERED_START:
            case IFN_GOMP_SIMD_ORDERED_END:
              if (integer_onep (gimple_call_arg (stmt, 0)))
                {
                  enum built_in_function bcode
                    = (ifn == IFN_GOMP_SIMD_ORDERED_START
                       ? BUILT_IN_GOMP_ORDERED_START
                       : BUILT_IN_GOMP_ORDERED_END);
                  gimple *g
                    = gimple_build_call (builtin_decl_explicit (bcode), 0);
                  gimple_move_vops (g, stmt);
                  gsi_replace (&i, g, true);
                  continue;
                }
              gsi_remove (&i, true);
              unlink_stmt_vdef (stmt);
              continue;
```

ubsan.cc (ubsan_expand_bounds_ifn):
```
  /* Get rid of the UBSAN_BOUNDS call from the IR.  */
  unlink_stmt_vdef (stmt);
  gsi_remove (&gsi_orig, true);
```
(ubsan_expand_null_ifn):
```
      gsi_remove (gsip, true);
      /* Unlink the UBSAN_NULLs vops before replacing it.  */
      unlink_stmt_vdef (stmt);
      return true;
...
  gimple_stmt_iterator gsi2 = gsi_start_bb (then_bb);
  gimple_set_location (g, loc);
  gsi_insert_after (&gsi2, g, GSI_NEW_STMT);

  /* Unlink the UBSAN_NULLs vops before replacing it.  */
  unlink_stmt_vdef (stmt);

```
ubsan_expand_ptr_ifn



Check the ones in tree-ssa.cc (calls update_stmt afterwards which might release
it).

Check the one in omp-offload.cc .
Check the one in tree-ssa-sccvn.cc.