Re: [PATCH 1/2] correct BB frequencies after loop changed

[Jiufu Guo <guojiufu@linux.ibm.com>]

Jeff Law <law@redhat.com> writes:

> On 11/18/20 12:28 AM, Richard Biener wrote:
>> On Tue, 17 Nov 2020, Jeff Law wrote:
>>
>>> Minor questions for Jan and Richi embedded below...
>>>
>>> On 10/9/20 4:12 AM, guojiufu via Gcc-patches wrote:
>>>> When investigating the issue from https://gcc.gnu.org/pipermail/gcc-patches/2020-July/549786.html
>>>> I find the BB COUNTs of loop seems are not accurate in some case.
>>>> For example:
>>>>
>>>> In below figure:
>>>>
>>>>
>>>>                COUNT:268435456<bb 2>  pre-header
>>>>                         |
>>>>                         |  .--------------------.
>>>>                         |  |                    |
>>>>                         V  v                    |
>>>>                COUNT:805306369<bb 3>            |
>>>>                        / \                      |
>>>>                    33%/   \                     |
>>>>                      /     \                    |
>>>>                     v       v                   |
>>>> COUNT:268435456<bb 10>  COUNT:536870911<bb 15>  | 
>>>>     exit-edge                 |   latch         |
>>>>                               ._________________.
>>>>
>>>> Those COUNTs have below equations:
>>>> COUNT of exit-edge:268435456 = COUNT of pre-header:268435456
>>>> COUNT of exit-edge:268435456 = COUNT of header:805306369 * 33
>>>> COUNT of header:805306369 = COUNT of pre-header:268435456 + COUNT of latch:536870911
>>>>
>>>>
>>>> While after pcom:
>>>>
>>>>                COUNT:268435456<bb 2>  pre-header
>>>>                         |
>>>>                         |  .--------------------.
>>>>                         |  |                    |
>>>>                         V  v                    |
>>>>                COUNT:268435456<bb 3>            |
>>>>                        / \                      |
>>>>                    50%/   \                     |
>>>>                      /     \                    |
>>>>                     v       v                   |
>>>> COUNT:134217728<bb 10>  COUNT:134217728<bb 15>  | 
>>>>     exit-edge                 |   latch         |
>>>>                               ._________________.
>>>>
>>>> COUNT<bb 3> != COUNT<bb 2> + COUNT<bb 15>
>>>> COUNT<bb 10> != COUNT<bb2>
>>>>
>>>> In some cases, the probility of exit-edge is easy to estimate, then
>>>> those COUNTs of other BBs in loop can be re-caculated.
>>>>
>>>> Bootstrap and regtest pass on ppc64le. Is this ok for trunk?
>>>>
>>>> Jiufu
>>>>
>>>> gcc/ChangeLog:
>>>> 2020-10-09  Jiufu Guo   <guojiufu@linux.ibm.com>
>>>>
>>>> 	* cfgloopmanip.h (recompute_loop_frequencies): New function.
>>>> 	* cfgloopmanip.c (recompute_loop_frequencies): New implementation.
>>>> 	* tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Call
>>>> 	recompute_loop_frequencies.
>>>>
>>>> ---
>>>>  gcc/cfgloopmanip.c        | 53 +++++++++++++++++++++++++++++++++++++++
>>>>  gcc/cfgloopmanip.h        |  2 +-
>>>>  gcc/tree-ssa-loop-manip.c | 28 +++------------------
>>>>  3 files changed, 57 insertions(+), 26 deletions(-)
>>>>
>>>> diff --git a/gcc/cfgloopmanip.c b/gcc/cfgloopmanip.c
>>>> index 73134a20e33..b0ca82a67fd 100644
>>>> --- a/gcc/cfgloopmanip.c
>>>> +++ b/gcc/cfgloopmanip.c
>>>> @@ -31,6 +31,7 @@ along with GCC; see the file COPYING3.  If not see
>>>>  #include "gimplify-me.h"
>>>>  #include "tree-ssa-loop-manip.h"
>>>>  #include "dumpfile.h"
>>>> +#include "cfgrtl.h"
>>>>  
>>>>  static void copy_loops_to (class loop **, int,
>>>>  			   class loop *);
>>>> @@ -1773,3 +1774,55 @@ loop_version (class loop *loop,
>>>>  
>>>>    return nloop;
>>>>  }
>>>> +
>>>> +/* Recalculate the COUNTs of BBs in LOOP, if the probability of exit edge
>>>> +   is NEW_PROB.  */
>>>> +
>>>> +bool
>>>> +recompute_loop_frequencies (class loop *loop, profile_probability new_prob)
>>>> +{
>>>> +  edge exit = single_exit (loop);
>>>> +  if (!exit)
>>>> +    return false;
>>>> +
>>>> +  edge e;
>>>> +  edge_iterator ei;
>>>> +  edge non_exit;
>>>> +  basic_block * bbs;
>>>> +  profile_count exit_count = loop_preheader_edge (loop)->count ();
>>>> +  profile_probability exit_p = exit_count.probability_in (loop->header->count);
>>>> +  profile_count base_count = loop->header->count;
>>>> +  profile_count after_num = base_count.apply_probability (exit_p);
>>>> +  profile_count after_den = base_count.apply_probability (new_prob);
>>>> +
>>>> +  /* Update BB counts in loop body.
>>>> +     COUNT<exit> = COUNT<preheader>
>>>> +     COUNT<exit> = COUNT<header> * exit_edge_probility
>>>> +     The COUNT<new_header> = COUNT<old_header> * old_exit_p / new_prob.  */
>>>> +  bbs = get_loop_body (loop);
>>>> +  scale_bbs_frequencies_profile_count (bbs, loop->num_nodes, after_num,
>>>> +						     after_den);
>>>> +  free (bbs);
>>>> +
>>>> +  /* Update probability and count of the BB besides exit edge (maybe latch).  */
>>>> +  FOR_EACH_EDGE (e, ei, exit->src->succs)
>>>> +    if (e != exit)
>>>> +      break;
>>>> +  non_exit = e;
>>> Are we sure that exit->src has just two successors (will that case be
>>> canonicalized before we get here?).? If it has > 2 successors, then I'm
>>> pretty sure the frequencies get mucked up.? Richi could probably answer
>>> whether or not the block with the loop exit edge can have > 2 successors.
>> There's nothing preventing more than two edges in the SRC generally
>> (the exit could be an edge off a switch).
> That's precisely the case I was concerned about.
>
>>   But of course this function
>> is very likely called on loops that are countable which means niter
>> analysis has to handle it which in turn means all exits are controlled
>> by simple conditions on IVs.
> Thanks.  It sounds like it's unlikely we'll have > 2 out edges.
>>
>> I guess adding a gcc_assert (EDGE_COUNT (exit->src->succs) == 2) can't 
>> hurt (with a comment reflecting the above).
> Sounds good to me.   Just catching this case if it happens is good
> enough for me -- if it trips we can come back and adjust the code to
> distribute across the edges.
With this gcc_assert, run bootstrap and regression test, no failure
occur.
For this patch, in the original code, there is code:
-  new_nonexit = single_pred_edge (loop->latch);
-  prob = new_nonexit->probability;
-  new_nonexit->probability = new_exit->probability.invert ();
Which is also assume 2 successors.  So, it may relative safe.

Thanks,
Jiufu Guo.

>
> So if Jan could chime in on the downstream edge updates question then I
> think we'd be ready to move forward.
>
> jeff