[PATCH] Allow loop header copying when first iteration condition is known.

[Aldy Hernandez]

On Thu, Nov 11, 2021 at 8:30 AM Richard Biener
<richard.guenther@gmail.com> wrote:
>
> On Wed, Nov 10, 2021 at 9:42 PM Jeff Law <jeffreyalaw@gmail.com> wrote:
> >
> >
> >
> > On 11/10/2021 11:20 AM, Aldy Hernandez via Gcc-patches wrote:
> > > As discussed in the PR, the loop header copying pass avoids doing so
> > > when optimizing for size.  However, sometimes we can determine the
> > > loop entry conditional statically for the first iteration of the loop.
> > >
> > > This patch uses the path solver to determine the outgoing edge
> > > out of preheader->header->xx.  If so, it allows header copying.  Doing
> > > this in the loop optimizer saves us from doing gymnastics in the
> > > threader which doesn't have the context to determine if a loop
> > > transformation is profitable.
> > >
> > > I am only returning true in entry_loop_condition_is_static for
> > > a true conditional.  Technically a false conditional is also
> > > provably static, but allowing any boolean value causes a regression
> > > in gfortran.dg/vector_subscript_1.f90.
> > >
> > > I would have preferred not passing around the query object, but the
> > > layout of pass_ch and should_duplicate_loop_header_p make it a bit
> > > awkward to get it right without an outright refactor to the
> > > pass.
> > >
> > > Tested on x86-64 Linux.
> > >
> > > OK?
> > >
> > > gcc/ChangeLog:
> > >
> > >       PR tree-optimization/102906
> > >       * tree-ssa-loop-ch.c (entry_loop_condition_is_static): New.
> > >       (should_duplicate_loop_header_p): Call entry_loop_condition_is_static.
> > >       (class ch_base): Add m_ranger and m_query.
> > >       (ch_base::copy_headers): Pass m_query to
> > >       entry_loop_condition_is_static.
> > >       (pass_ch::execute): Allocate and deallocate m_ranger and
> > >       m_query.
> > >       (pass_ch_vect::execute): Same.
> > >
> > > gcc/testsuite/ChangeLog:
> > >
> > >       * gcc.dg/tree-ssa/pr102906.c: New test.
> > OK.  It also makes a nice little example of how to use a Ranger within
> > an existing pass.
>
> Note if you just test for the condition to be true it will only catch 50%
> of the desired cases since we have no idea whether the 'true' edge
> is the edge existing the loop or the edge remaining in the loop.
> For loop header copying we like to resolve statically to the edge
> remaining in the loop, so you want

Ahh, I figured there was some block shuffling needed.

I was cautious not to touch much because of the
gfortran.dg/vector_subscript_1.f90 regression, but now I see that the
test fails for all optimization levels except -Os.  With this fix we
properly fail for all levels.  I assume this is expected ;-).

>
> extract_true_false_edges_from_block (gimple_bb (last), &true_e, &false_e);
>
> /* If neither edge is the exit edge this is not a case we'd like to
>    special-case.  */
> if (!loop_exit_edge_p (l, true_e) && !loop_exit_edge_p (l, false_e))
>  return false;
>
> tree desired_static_value;
> if (loop_exit_edge_p (l, true_e))
>  desired_static_value = boolean_false_node;
> else
>   desired_static_value = boolean_true_node;
>
> and test for desired_static_value.

Thanks for the code!

OK pending tests?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Resolve-entry-loop-condition-for-the-edge-remaining-.patch
Type: text/x-patch
Size: 1835 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/gcc-patches/attachments/20211111/fdc98177/attachment-0001.bin>