Re: [PATCH v1] tree-optimization/95821 - Convert strlen + strchr to memchr

[Noah Goldstein <goldstein.w.n@gmail.com>]

On Mon, Jun 20, 2022 at 10:29 AM Jakub Jelinek <jakub@redhat.com> wrote:
>
> On Mon, Jun 20, 2022 at 09:35:36AM -0700, Noah Goldstein via Gcc-patches wrote:
> > This patch allows for strchr(x, c) to the replace with memchr(x, c,
> > strlen(x) + 1) if strlen(x) has already been computed earlier in the
> > tree.
> >
> > Handles PR95821: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95821
> >
> > Since memchr doesn't need to re-find the null terminator it is faster
> > than strchr.
>
> Do you have a GCC Copyright assignment on file, or do you want to submit
> this under DCO ( https://gcc.gnu.org/dco.html )?  If the latter, there
> should be a Signed-off-by: line, both in the mail and later commit.
> >
> > bootstrapped and tested on x86_64-linux.
> >
> > gcc/
> >
>
> As it fixes a GCC bugzilla bug, the ChangeLog entry should start with
>         PR tree-optimization/95821
> line.
> >     * tree-ssa-strlen.cc: Emit memchr instead of strchr if strlen
> >      already computed.
>
> All the indented lines in ChangeLog should be indented by tab.
> You are modifying strlen_pass::handle_builtin_strchr function, so after
> tree-ssa-strlen.cc there should be that function name in parens:
>         * tree-ssa-strlen.cc (strlen_pass::handle_builtin_strchr): Emit
>         memchr ...

Fixed in v2.
>
> >
> > gcc/testsuite/
> >
> >     * c-c++-common/pr95821-1.c
> >     * c-c++-common/pr95821-2.c
> >     * c-c++-common/pr95821-3.c
> >     * c-c++-common/pr95821-4.c
> >     * c-c++-common/pr95821-5.c
> >     * c-c++-common/pr95821-6.c
>
> All the above lines should end with ": New test." after .c

Fixed in V2.
>
> > --- a/gcc/tree-ssa-strlen.cc
> > +++ b/gcc/tree-ssa-strlen.cc
>
> How does the patch relate to the one that H.J. attached in
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95821#c4 ?
>
> > @@ -2405,9 +2405,12 @@ strlen_pass::handle_builtin_strlen ()
> >      }
> >  }
> >
> > -/* Handle a strchr call.  If strlen of the first argument is known, replace
> > -   the strchr (x, 0) call with the endptr or x + strlen, otherwise remember
> > -   that lhs of the call is endptr and strlen of the argument is endptr - x.  */
> > +/* Handle a strchr call.  If strlen of the first argument is known,
> > +   replace the strchr (x, 0) call with the endptr or x + strlen,
> > +   otherwise remember that lhs of the call is endptr and strlen of the
> > +   argument is endptr - x.  If strlen of x is not know but has been
> > +   computed earlier in the tree then replace strchr(x, c) to
> > +   memchr(x, c, strlen + 1).  */
>
> Space before ( even in comments.

Fixed in V2.
>
>
>
> >  void
> >  strlen_pass::handle_builtin_strchr ()
> > @@ -2418,8 +2421,8 @@ strlen_pass::handle_builtin_strchr ()
> >    if (lhs == NULL_TREE)
> >      return;
> >
> > -  if (!integer_zerop (gimple_call_arg (stmt, 1)))
> > -    return;
> > +  tree chr = gimple_call_arg (stmt, 1);
> > +  bool is_strchr_zerop = integer_zerop (chr);
> >
> >    tree src = gimple_call_arg (stmt, 0);
> >
> > @@ -2452,32 +2455,56 @@ strlen_pass::handle_builtin_strchr ()
> >             fprintf (dump_file, "Optimizing: ");
> >             print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
> >           }
> > -       if (si != NULL && si->endptr != NULL_TREE)
> > +       if (!is_strchr_zerop)
> >           {
> > -           rhs = unshare_expr (si->endptr);
> > -           if (!useless_type_conversion_p (TREE_TYPE (lhs),
> > -                                           TREE_TYPE (rhs)))
> > -             rhs = fold_convert_loc (loc, TREE_TYPE (lhs), rhs);
> > +           /* If its not strchr(s, zerop) then try and convert to
> > +                    memchr if strlen has already been computed.  */
>
> Again, space before (.  The second line is weirdly formatted, should
> be indented below If.

Fixed in V2.
>
> > +           tree fn = builtin_decl_explicit (BUILT_IN_MEMCHR);
> > +           tree one = build_int_cst (TREE_TYPE (rhs), 1);
> > +           rhs = fold_build2_loc (loc, PLUS_EXPR, TREE_TYPE (rhs),
> > +                                  unshare_expr (rhs), one);
> > +           tree size = make_ssa_name (TREE_TYPE (rhs));
> > +           gassign *size_stmt = gimple_build_assign (size, rhs);
> > +           gsi_insert_before (&m_gsi, size_stmt, GSI_SAME_STMT);
> > +           rhs = size;
> > +           if (!update_gimple_call (&m_gsi, fn, 3, src, chr, rhs))
> > +             return;
>
> I think we should differentiate more.  If integer_nonzerop (chr)
> or perhaps better tree_expr_nonzero_p (chr), then it is better
> to optimize t = strlen (x); ... p = strchr (x, c); to
> t = strlen (x); ... p = memchr (x, c, t);
> the t + 1 is only needed if c might be zero.

Done in V2. Also added the optimizations if chr has zero-char bits.

Right now:

t=strlen (s);

strchr (s, 0) -> t;
strchr (s, 256) -> t;

strchr (s, 1234) -> memchr (s, 1234, t);
strchr (s, non_zero) -> memchr (s, non_zero, t);
strchr (s, unknown) -> memchr (s, unknown, t + 1);


>
> > +       /* Don't update strlen of lhs if search-char was non-zero.  */
>
> Wasn't known to be zero is the right thing.

Fixed in V2.
>
>         Jakub
>