Re: [PATCH] Abstract PHI and forwarder block checks in jump threader.

[Aldy Hernandez <aldyh@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 2899 bytes --]



On 9/7/21 2:59 PM, Richard Biener wrote:
> On September 7, 2021 12:02:27 PM GMT+02:00, Aldy Hernandez <aldyh@redhat.com> wrote:
>>
>>
>> On 9/6/21 9:19 AM, Richard Biener wrote:
>>> On Fri, Sep 3, 2021 at 3:59 PM Aldy Hernandez via Gcc-patches
>>> <gcc-patches@gcc.gnu.org> wrote:
>>>>
>>>> This patch abstracts out a couple common idioms in the forward
>>>> threader that I found useful while navigating the code base.
>>>>
>>>> Tested on x86-64 Linux.
>>>>
>>>> OK?
>>>>
>>>> gcc/ChangeLog:
>>>>
>>>>           * tree-ssa-threadedge.c (has_phis_p): New.
>>>>           (forwarder_block_p): New.
>>>>           (potentially_threadable_block): Call forwarder_block_p.
>>>>           (jump_threader::thread_around_empty_blocks): Call has_phis_p.
>>>>           (jump_threader::thread_through_normal_block): Call
>>>>           forwarder_block_p.
>>>> ---
>>>>    gcc/tree-ssa-threadedge.c | 25 +++++++++++++++++++------
>>>>    1 file changed, 19 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/gcc/tree-ssa-threadedge.c b/gcc/tree-ssa-threadedge.c
>>>> index e57f6d3e39c..3db54a199fd 100644
>>>> --- a/gcc/tree-ssa-threadedge.c
>>>> +++ b/gcc/tree-ssa-threadedge.c
>>>> @@ -95,6 +95,21 @@ jump_threader::thread_through_all_blocks (bool may_peel_loop_headers)
>>>>      return m_registry->thread_through_all_blocks (may_peel_loop_headers);
>>>>    }
>>>>
>>>> +static inline bool
>>>> +has_phis_p (basic_block bb)
>>>> +{
>>>> +  return !gsi_end_p (gsi_start_phis (bb));
>>>
>>> gimple_seq_empty_p (phi_nodes (bb)) shoud be cheaper.  Do virtual PHIs
>>> count as PHIs for you?
>>
>> I don't know.  The goal was to abstract some common idioms without
>> changing existing behavior, but if my abstractions confuse other
>> readers, perhaps I should revert my patch.
>>
>> FWIW, my initial motivation here was to merge the path profitability
>> code between the forward and backward threaders.  It seems the forward
>> threader is more permissive than the backward threader, even though the
>> latter can thread more paths than it's allowed (per profitable_path_p).
>>
>>>
>>>> +}
>>>> +
>>>> +/* Return TRUE for a forwarder block which is defined as having PHIs
>>>> +   but no instructions.  */
>>>> +
>>>> +static bool
>>>> +forwarder_block_p (basic_block bb)
>>>
>>> There exists a function with exactly the same signature in cfgrtl.h, likewise
>>> several similar implementations might exist elsewhere.
>>
>> Ughh, that's definitely not good.
>>
>>>
>>> Your definition is also quite odd, not matching what one would expect
>>> (the PHI requirement).  The tree-cfgcleanup.c variant has
>>> tree_forwarder_block_p which is explicit about this.
>>>
>>> Btw, gsi_start_nondebug_bb does not ignore labels.
>>
>> Would a name like empty_block_with_phis_p be more appropriate?
> 
> I think so.  That said, my main concern ist the clash with the same named function.

Agreed.

OK for trunk?

Aldy


[-- Attachment #2: p.patch --]
[-- Type: text/x-patch, Size: 1818 bytes --]

commit 77ac56456d5db150d6a71eaca918f19d2b478f82
Author: Aldy Hernandez <aldyh@redhat.com>
Date:   Tue Sep 7 15:20:23 2021 +0200

    Rename forwarder_block_p in treading code to empty_block_with_phis_p.
    
    gcc/ChangeLog:
    
            * tree-ssa-threadedge.c (forwarder_block_p): Rename to...
            (empty_block_with_phis_p): ...this.
            (potentially_threadable_block): Same.
            (jump_threader::thread_through_normal_block): Same.

diff --git a/gcc/tree-ssa-threadedge.c b/gcc/tree-ssa-threadedge.c
index 3db54a199fd..3c7cdc58b93 100644
--- a/gcc/tree-ssa-threadedge.c
+++ b/gcc/tree-ssa-threadedge.c
@@ -101,11 +101,10 @@ has_phis_p (basic_block bb)
   return !gsi_end_p (gsi_start_phis (bb));
 }
 
-/* Return TRUE for a forwarder block which is defined as having PHIs
-   but no instructions.  */
+/* Return TRUE for a block with PHIs but no statements.  */
 
 static bool
-forwarder_block_p (basic_block bb)
+empty_block_with_phis_p (basic_block bb)
 {
   return gsi_end_p (gsi_start_nondebug_bb (bb)) && has_phis_p (bb);
 }
@@ -123,7 +122,7 @@ potentially_threadable_block (basic_block bb)
      to the loop header.   We want to thread through them as we can
      sometimes thread to the loop exit, which is obviously profitable.
      The interesting case here is when the block has PHIs.  */
-  if (forwarder_block_p (bb))
+  if (empty_block_with_phis_p (bb))
     return true;
 
   /* If BB has a single successor or a single predecessor, then
@@ -1008,7 +1007,7 @@ jump_threader::thread_through_normal_block (vec<jump_thread_edge *> *path,
     {
       /* First case.  The statement simply doesn't have any instructions, but
 	 does have PHIs.  */
-      if (forwarder_block_p (e->dest))
+      if (empty_block_with_phis_p (e->dest))
 	return 0;
 
       /* Second case.  */