public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: "Martin Liška" <mliska@suse.cz>
To: gcc-patches@gcc.gnu.org
Cc: Jan Hubicka <hubicka@ucw.cz>
Subject: Re: [PATCH] Improve profile handling in switch lowering.
Date: Mon, 28 Nov 2022 10:53:58 +0100	[thread overview]
Message-ID: <7eb66e9c-e412-73ea-136c-6788b69daab8@suse.cz> (raw)
In-Reply-To: <07de3a4b-0250-fbee-dbe6-db16728a0e27@suse.cz>

PING^4

On 4/7/22 16:00, Martin Liška wrote:
> PING^3
> 
> On 3/24/22 11:22, Martin Liška wrote:
>> PING^2
>>
>> On 3/4/22 14:47, Martin Liška wrote:
>>> PING^1
>>>
>>> On 1/26/22 12:11, Martin Liška wrote:
>>>> Hello.
>>>>
>>>> Right now, switch lowering does not update basic_block::count values
>>>> so that they are uninitiliazed. Moreover, I've updated probability scaling
>>>> when a more complex expansion happens. There are still some situations where
>>>> the profile is a bit imprecise, but the patch improves rapidly the current situation.
>>>>
>>>> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
>>>>
>>>> Ready to be installed?
>>>> Thanks,
>>>> Martin
>>>>
>>>>      PR tree-optimization/101301
>>>>      PR tree-optimization/103680
>>>>
>>>> gcc/ChangeLog:
>>>>
>>>>      * tree-switch-conversion.cc (bit_test_cluster::emit):
>>>>      Handle correctly remaining probability.
>>>>      (switch_decision_tree::try_switch_expansion): Fix BB's count
>>>>      where a cluster expansion happens.
>>>>      (switch_decision_tree::emit_cmp_and_jump_insns): Fill up also
>>>>      BB count.
>>>>      (switch_decision_tree::do_jump_if_equal): Likewise.
>>>>      (switch_decision_tree::emit_case_nodes): Handle special case
>>>>      for BT expansion which can also fallback to a default BB.
>>>>      * tree-switch-conversion.h (cluster::cluster): Add
>>>>      m_default_prob probability.
>>>> ---
>>>>   gcc/tree-switch-conversion.cc | 51 ++++++++++++++++++++++++-----------
>>>>   gcc/tree-switch-conversion.h  |  8 +++++-
>>>>   2 files changed, 43 insertions(+), 16 deletions(-)
>>>>
>>>> diff --git a/gcc/tree-switch-conversion.cc b/gcc/tree-switch-conversion.cc
>>>> index 670397c87e4..d6679e8dee3 100644
>>>> --- a/gcc/tree-switch-conversion.cc
>>>> +++ b/gcc/tree-switch-conversion.cc
>>>> @@ -1538,10 +1538,12 @@ bit_test_cluster::emit (tree index_expr, tree index_type,
>>>>         test[k].target_bb = n->m_case_bb;
>>>>         test[k].label = n->m_case_label_expr;
>>>>         test[k].bits = 0;
>>>> +      test[k].prob = profile_probability::never ();
>>>>         count++;
>>>>       }
>>>>
>>>>         test[k].bits += n->get_range (n->get_low (), n->get_high ());
>>>> +      test[k].prob += n->m_prob;
>>>>
>>>>         lo = tree_to_uhwi (int_const_binop (MINUS_EXPR, n->get_low (), minval));
>>>>         if (n->get_high () == NULL_TREE)
>>>> @@ -1629,6 +1631,11 @@ bit_test_cluster::emit (tree index_expr, tree index_type,
>>>>                     /*simple=*/true, NULL_TREE,
>>>>                     /*before=*/true, GSI_SAME_STMT);
>>>>
>>>> +  profile_probability subtree_prob = m_subtree_prob;
>>>> +  profile_probability default_prob = m_default_prob;
>>>> +  if (!default_prob.initialized_p ())
>>>> +    default_prob = m_subtree_prob.invert ();
>>>> +
>>>>     if (m_handles_entire_switch && entry_test_needed)
>>>>       {
>>>>         tree range = int_const_binop (MINUS_EXPR, maxval, minval);
>>>> @@ -1639,9 +1646,10 @@ bit_test_cluster::emit (tree index_expr, tree index_type,
>>>>                       /*simple=*/true, NULL_TREE,
>>>>                       /*before=*/true, GSI_SAME_STMT);
>>>>         tmp = fold_build2 (GT_EXPR, boolean_type_node, idx, range);
>>>> +      default_prob = default_prob.apply_scale (1, 2);
>>>>         basic_block new_bb
>>>>       = hoist_edge_and_branch_if_true (&gsi, tmp, default_bb,
>>>> -                     profile_probability::unlikely ());
>>>> +                     default_prob);
>>>>         gsi = gsi_last_bb (new_bb);
>>>>       }
>>>>
>>>> @@ -1662,14 +1670,12 @@ bit_test_cluster::emit (tree index_expr, tree index_type,
>>>>     else
>>>>       csui = tmp;
>>>>
>>>> -  profile_probability prob = profile_probability::always ();
>>>> -
>>>>     /* for each unique set of cases:
>>>>          if (const & csui) goto target  */
>>>>     for (k = 0; k < count; k++)
>>>>       {
>>>> -      prob = profile_probability::always ().apply_scale (test[k].bits,
>>>> -                             bt_range);
>>>> +      profile_probability prob = test[k].prob / (subtree_prob + default_prob);
>>>> +      subtree_prob -= test[k].prob;
>>>>         bt_range -= test[k].bits;
>>>>         tmp = wide_int_to_tree (word_type_node, test[k].mask);
>>>>         tmp = fold_build2 (BIT_AND_EXPR, word_type_node, csui, tmp);
>>>> @@ -1908,9 +1914,13 @@ switch_decision_tree::try_switch_expansion (vec<cluster *> &clusters)
>>>>         /* Emit cluster-specific switch handling.  */
>>>>         for (unsigned i = 0; i < clusters.length (); i++)
>>>>       if (clusters[i]->get_type () != SIMPLE_CASE)
>>>> -      clusters[i]->emit (index_expr, index_type,
>>>> -                 gimple_switch_default_label (m_switch),
>>>> -                 m_default_bb, gimple_location (m_switch));
>>>> +      {
>>>> +        edge e = single_pred_edge (clusters[i]->m_case_bb);
>>>> +        e->dest->count = e->src->count.apply_probability (e->probability);
>>>> +        clusters[i]->emit (index_expr, index_type,
>>>> +                   gimple_switch_default_label (m_switch),
>>>> +                   m_default_bb, gimple_location (m_switch));
>>>> +      }
>>>>       }
>>>>
>>>>     fix_phi_operands_for_edges ();
>>>> @@ -2162,6 +2172,7 @@ switch_decision_tree::emit_cmp_and_jump_insns (basic_block bb, tree op0,
>>>>     edge false_edge = split_block (bb, cond);
>>>>     false_edge->flags = EDGE_FALSE_VALUE;
>>>>     false_edge->probability = prob.invert ();
>>>> +  false_edge->dest->count = bb->count.apply_probability (prob.invert ());
>>>>
>>>>     edge true_edge = make_edge (bb, label_bb, EDGE_TRUE_VALUE);
>>>>     true_edge->probability = prob;
>>>> @@ -2192,6 +2203,7 @@ switch_decision_tree::do_jump_if_equal (basic_block bb, tree op0, tree op1,
>>>>     edge false_edge = split_block (bb, cond);
>>>>     false_edge->flags = EDGE_FALSE_VALUE;
>>>>     false_edge->probability = prob.invert ();
>>>> +  false_edge->dest->count = bb->count.apply_probability (prob.invert ());
>>>>
>>>>     edge true_edge = make_edge (bb, label_bb, EDGE_TRUE_VALUE);
>>>>     true_edge->probability = prob;
>>>> @@ -2227,7 +2239,7 @@ switch_decision_tree::emit_case_nodes (basic_block bb, tree index,
>>>>                    node->m_c->m_case_bb, p, loc);
>>>>         /* Since this case is taken at this point, reduce its weight from
>>>>        subtree_weight.  */
>>>> -      node->m_c->m_subtree_prob -= p;
>>>> +      node->m_c->m_subtree_prob -= node->m_c->m_prob;
>>>>
>>>>         if (node->m_left != NULL && node->m_right != NULL)
>>>>       {
>>>> @@ -2246,6 +2258,7 @@ switch_decision_tree::emit_case_nodes (basic_block bb, tree index,
>>>>              / (node->m_c->m_subtree_prob + default_prob));
>>>>             bb = do_jump_if_equal (bb, index, node->m_right->m_c->get_low (),
>>>>                        node->m_right->m_c->m_case_bb, p, loc);
>>>> +          node->m_c->m_subtree_prob -= node->m_right->m_c->m_prob;
>>>>
>>>>             p = (node->m_left->m_c->m_prob
>>>>              / (node->m_c->m_subtree_prob + default_prob));
>>>> @@ -2262,6 +2275,7 @@ switch_decision_tree::emit_case_nodes (basic_block bb, tree index,
>>>>             p = ((node->m_right->m_c->m_subtree_prob
>>>>               + default_prob.apply_scale (1, 2))
>>>>              / (node->m_c->m_subtree_prob + default_prob));
>>>> +          test_bb->count = bb->count.apply_probability (p);
>>>>             bb = emit_cmp_and_jump_insns (bb, index, node->m_c->get_high (),
>>>>                           GT_EXPR, test_bb, p, loc);
>>>>             default_prob = default_prob.apply_scale (1, 2);
>>>> @@ -2348,21 +2362,28 @@ switch_decision_tree::emit_case_nodes (basic_block bb, tree index,
>>>>        is the one to branch to.  */
>>>>         if (node->has_child () || node->m_c->get_type () != SIMPLE_CASE)
>>>>       {
>>>> +      bool is_bt = node->m_c->get_type () == BIT_TEST;
>>>> +      int parts = is_bt ? 3 : 2;
>>>> +
>>>>         /* Branch to a label where we will handle it later.  */
>>>>         basic_block test_bb = split_edge (single_succ_edge (bb));
>>>>         redirect_edge_succ (single_pred_edge (test_bb),
>>>>                     single_succ_edge (bb)->dest);
>>>>
>>>> -
>>>> -       profile_probability right_prob = profile_probability::never ();
>>>> -       if (node->m_right)
>>>> -         right_prob = node->m_right->m_c->m_subtree_prob;
>>>> -      p = ((right_prob + default_prob.apply_scale (1, 2))
>>>> +      profile_probability right_prob = profile_probability::never ();
>>>> +      if (node->m_right)
>>>> +        right_prob = node->m_right->m_c->m_subtree_prob;
>>>> +      p = ((right_prob + default_prob.apply_scale (1, parts))
>>>>              / (node->m_c->m_subtree_prob + default_prob));
>>>> +      test_bb->count = bb->count.apply_probability (p);
>>>>
>>>>         bb = emit_cmp_and_jump_insns (bb, index, node->m_c->get_high (),
>>>>                       GT_EXPR, test_bb, p, loc);
>>>> -      default_prob = default_prob.apply_scale (1, 2);
>>>> +
>>>> +      default_prob = default_prob.apply_scale (1, parts);
>>>> +      node->m_c->m_subtree_prob -= right_prob;
>>>> +      if (is_bt)
>>>> +        node->m_c->m_default_prob = default_prob;
>>>>
>>>>         /* Value belongs to this node or to the left-hand subtree.  */
>>>>         p = node->m_c->m_prob / (node->m_c->m_subtree_prob + default_prob);
>>>> diff --git a/gcc/tree-switch-conversion.h b/gcc/tree-switch-conversion.h
>>>> index e969c051a05..5f3f99353ba 100644
>>>> --- a/gcc/tree-switch-conversion.h
>>>> +++ b/gcc/tree-switch-conversion.h
>>>> @@ -102,6 +102,10 @@ public:
>>>>     /* Probability of reaching subtree rooted at this node.  */
>>>>     profile_probability m_subtree_prob;
>>>>
>>>> +  /* Probability of default case when reaching the node.
>>>> +     It is used by bit-test right now.  */
>>>> +  profile_probability m_default_prob;
>>>> +
>>>>   protected:
>>>>     /* Default constructor.  */
>>>>     cluster () {}
>>>> @@ -110,7 +114,8 @@ protected:
>>>>   cluster::cluster (tree case_label_expr, basic_block case_bb,
>>>>             profile_probability prob, profile_probability subtree_prob):
>>>>     m_case_label_expr (case_label_expr), m_case_bb (case_bb), m_prob (prob),
>>>> -  m_subtree_prob (subtree_prob)
>>>> +  m_subtree_prob (subtree_prob),
>>>> +  m_default_prob (profile_probability::uninitialized ())
>>>>   {
>>>>   }
>>>>
>>>> @@ -542,6 +547,7 @@ public:
>>>>     basic_block target_bb;
>>>>     tree label;
>>>>     int bits;
>>>> +  profile_probability prob;
>>>>
>>>>     /* Comparison function for qsort to order bit tests by decreasing
>>>>        probability of execution.  */
>>>
>>
> 


  reply	other threads:[~2022-11-28  9:54 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-01-26 11:11 Martin Liška
2022-03-04 13:47 ` Martin Liška
2022-03-24 10:22   ` Martin Liška
2022-04-07 14:00     ` Martin Liška
2022-11-28  9:53       ` Martin Liška [this message]
2022-11-28 17:38 ` Jeff Law

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=7eb66e9c-e412-73ea-136c-6788b69daab8@suse.cz \
    --to=mliska@suse.cz \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=hubicka@ucw.cz \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).