public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Jan Hubicka <hubicka@ucw.cz>
To: gcc-patches@gcc.gnu.org
Subject: Do not combine PRED_LOOP_GUARD and PRED_LOOP_GUARD_WITH_RECURSION
Date: Tue, 11 Aug 2020 12:03:48 +0200	[thread overview]
Message-ID: <20200811100348.GA2190@kam.mff.cuni.cz> (raw)

Hi,
this patch avoids both PRED_LOOP_GUARD and PRED_LOOP_GUARD_WITH_RECURSION to be
attached to one edge.  We have logic that prevents same predictor to apply to
one edge twice, but since we split LOOP_GUARD to two more specialized cases,
this no longer fires.

Double prediction happens in exchange benchmark and leads to unrealistically
low hitrates on some edges which in turn leads to bad IPA profile and misguides
ipa-cp.

Unforutnately it seems that the bad profile also leads to bit better
performance by disabling some of loop stuff, but that really ought to be done
in some meaningful way, not by an accident.

Bootstrapped/regtested x86_64-linux, comitted.

Honza

gcc/ChangeLog:

2020-08-11  Jan Hubicka  <hubicka@ucw.cz>

	* predict.c (not_loop_guard_equal_edge_p): New function.
	(maybe_predict_edge): New function.
	(predict_paths_for_bb): Use it.
	(predict_paths_leading_to_edge): Use it.

gcc/testsuite/ChangeLog:

2020-08-11  Jan Hubicka  <hubicka@ucw.cz>

	* gcc.dg/ipa/ipa-clone-2.c: Lower threshold from 500 to 400.

diff --git a/gcc/predict.c b/gcc/predict.c
index 2164a06e083..4c4bba54939 100644
--- a/gcc/predict.c
+++ b/gcc/predict.c
@@ -3122,6 +3122,35 @@ tree_guess_outgoing_edge_probabilities (basic_block bb)
   bb_predictions = NULL;
 }
 \f
+/* Filter function predicate that returns true for a edge predicate P
+   if its edge is equal to DATA.  */
+
+static bool
+not_loop_guard_equal_edge_p (edge_prediction *p, void *data)
+{
+  return p->ep_edge != (edge)data || p->ep_predictor != PRED_LOOP_GUARD;
+}
+
+/* Predict edge E with PRED unless it is already predicted by some predictor
+   considered equivalent.  */
+
+static void
+maybe_predict_edge (edge e, enum br_predictor pred, enum prediction taken)
+{
+  if (edge_predicted_by_p (e, pred, taken))
+    return;
+  if (pred == PRED_LOOP_GUARD
+      && edge_predicted_by_p (e, PRED_LOOP_GUARD_WITH_RECURSION, taken))
+    return;
+  /* Consider PRED_LOOP_GUARD_WITH_RECURSION superrior to LOOP_GUARD.  */
+  if (pred == PRED_LOOP_GUARD_WITH_RECURSION)
+    {
+      edge_prediction **preds = bb_predictions->get (e->src);
+      if (preds)
+	filter_predictions (preds, not_loop_guard_equal_edge_p, e);
+    }
+  predict_edge_def (e, pred, taken);
+}
 /* Predict edges to successors of CUR whose sources are not postdominated by
    BB by PRED and recurse to all postdominators.  */
 
@@ -3177,10 +3206,7 @@ predict_paths_for_bb (basic_block cur, basic_block bb,
 	 regions that are only reachable by abnormal edges.  We simply
 	 prevent visiting given BB twice.  */
       if (found)
-	{
-	  if (!edge_predicted_by_p (e, pred, taken))
-            predict_edge_def (e, pred, taken);
-	}
+	maybe_predict_edge (e, pred, taken);
       else if (bitmap_set_bit (visited, e->src->index))
 	predict_paths_for_bb (e->src, e->src, pred, taken, visited, in_loop);
     }
@@ -3223,7 +3249,7 @@ predict_paths_leading_to_edge (edge e, enum br_predictor pred,
   if (!has_nonloop_edge)
     predict_paths_for_bb (bb, bb, pred, taken, auto_bitmap (), in_loop);
   else
-    predict_edge_def (e, pred, taken);
+    maybe_predict_edge (e, pred, taken);
 }
 \f
 /* This is used to carry information about basic blocks.  It is
diff --git a/gcc/testsuite/gcc.dg/ipa/ipa-clone-2.c b/gcc/testsuite/gcc.dg/ipa/ipa-clone-2.c
index d513020ee8b..53ae25a1e24 100644
--- a/gcc/testsuite/gcc.dg/ipa/ipa-clone-2.c
+++ b/gcc/testsuite/gcc.dg/ipa/ipa-clone-2.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */
-/* { dg-options "-O3 -fdump-ipa-cp-details -fno-early-inlining --param ipa-cp-max-recursive-depth=8" } */
+/* { dg-options "-O3 -fdump-ipa-cp-details -fno-early-inlining --param ipa-cp-max-recursive-depth=8 --param=ipa-cp-eval-threshold=400" } */
 
 int fn();
 

             reply	other threads:[~2020-08-11 10:03 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-08-11 10:03 Jan Hubicka [this message]
2020-08-12 11:02 ` Tamar Christina
2020-08-13 14:32   ` Martin Liška

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=20200811100348.GA2190@kam.mff.cuni.cz \
    --to=hubicka@ucw.cz \
    --cc=gcc-patches@gcc.gnu.org \
    /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).