public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r13-4723] middle-end/108086 - avoid quadraticness in copy_edges_for_bb
@ 2022-12-15 17:43 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2022-12-15 17:43 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:d49b2a3a1dbbe3ac9ae134bc5742be7fdb417d32

commit r13-4723-gd49b2a3a1dbbe3ac9ae134bc5742be7fdb417d32
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Dec 15 13:42:16 2022 +0100

    middle-end/108086 - avoid quadraticness in copy_edges_for_bb
    
    For the testcase in PR108086 it's visible that we split blocks
    multiple times when inlining and that causes us to adjust the
    block tail stmt BBs multiple times, once for each split.  The
    fix is to walk backwards and split from the tail instead.
    
    For a reduced testcase this improves compile-time at -O by 4%.
    
            PR middle-end/108086
            * tree-inline.cc (copy_edges_for_bb): Walk stmts backwards for
            splitting the block to avoid quadratic behavior with setting
            stmts BB on multliple splits.

Diff:
---
 gcc/tree-inline.cc | 34 ++++++++++++++++++----------------
 1 file changed, 18 insertions(+), 16 deletions(-)

diff --git a/gcc/tree-inline.cc b/gcc/tree-inline.cc
index addfe7fcbcc..0360f1f1e57 100644
--- a/gcc/tree-inline.cc
+++ b/gcc/tree-inline.cc
@@ -2569,13 +2569,17 @@ copy_edges_for_bb (basic_block bb, profile_count num, profile_count den,
 	  && !old_edge->src->aux)
 	new_bb->count -= old_edge->count ().apply_scale (num, den);
 
-  for (si = gsi_start_bb (new_bb); !gsi_end_p (si);)
+  /* Walk stmts from end to start so that splitting will adjust the BB
+     pointer for each stmt at most once, even when we split the block
+     multiple times.  */
+  bool seen_nondebug = false;
+  for (si = gsi_last_bb (new_bb); !gsi_end_p (si);)
     {
       bool can_throw, nonlocal_goto;
       gimple *copy_stmt = gsi_stmt (si);
 
       /* Do this before the possible split_block.  */
-      gsi_next (&si);
+      gsi_prev (&si);
 
       /* If this tree could throw an exception, there are two
          cases where we need to add abnormal edge(s): the
@@ -2595,25 +2599,23 @@ copy_edges_for_bb (basic_block bb, profile_count num, profile_count den,
 
       if (can_throw || nonlocal_goto)
 	{
-	  if (!gsi_end_p (si))
-	    {
-	      while (!gsi_end_p (si) && is_gimple_debug (gsi_stmt (si)))
-		gsi_next (&si);
-	      if (gsi_end_p (si))
-		need_debug_cleanup = true;
-	    }
-	  if (!gsi_end_p (si))
-	    /* Note that bb's predecessor edges aren't necessarily
-	       right at this point; split_block doesn't care.  */
+	  /* If there's only debug insns after copy_stmt don't split
+	     the block but instead mark the block for cleanup.  */
+	  if (!seen_nondebug)
+	    need_debug_cleanup = true;
+	  else
 	    {
+	      /* Note that bb's predecessor edges aren't necessarily
+		 right at this point; split_block doesn't care.  */
 	      edge e = split_block (new_bb, copy_stmt);
-
-	      new_bb = e->dest;
-	      new_bb->aux = e->src->aux;
-	      si = gsi_start_bb (new_bb);
+	      e->dest->aux = new_bb->aux;
+	      seen_nondebug = false;
 	    }
 	}
 
+      if (!is_gimple_debug (copy_stmt))
+	seen_nondebug = true;
+
       bool update_probs = false;
 
       if (gimple_code (copy_stmt) == GIMPLE_EH_DISPATCH)

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2022-12-15 17:43 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-12-15 17:43 [gcc r13-4723] middle-end/108086 - avoid quadraticness in copy_edges_for_bb Richard Biener

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).