c++/1687: Extreme compile time regression from 2.95.2

[Zack Weinberg]

The following reply was made to PR c++/1687; it has been noted by GNATS.

From: Zack Weinberg <zack@codesourcery.com>
To: Wolfgang Bangerth <bangerth@ticam.utexas.edu>
Cc: gcc-gnats@gcc.gnu.org, kelley.cook@home.com, gcc-bugs@gcc.gnu.org,
	pbienst@gcc.gnu.org, Zack Weinberg <zack@codesourcery.com>
Subject: Re: c++/1687: Extreme compile time regression from 2.95.2
Date: Sun, 10 Nov 2002 14:14:13 -0800

 On Sun, Nov 10, 2002 at 03:51:12PM -0600, Wolfgang Bangerth wrote:
 > 
 > Zack,
 > you had a patch for this problem, as mentioned in the audit. Do you know 
 > whether it was applied? I cannot find it in the ChangeLogs.
 
 Yes, the patch was applied.  It's in cp/ChangeLog:
 
 2001-04-24  Zack Weinberg  <zackw@stanford.edu>
 
         * cp/optimize.c: Include hashtab.h.
         (struct inline_data): Add tree_pruner.
         (expand_call_inline, expand_calls_inline): Use it when calling
         walk_tree.
         (optimize_function): Initialize and free tree_pruner.
 
 This code is now in tree-inline.c, by the way
 
 > However, at -O3 it still takes forever, now with both C and C++, which 
 > seems a further regression (since previously this held only for C++). I 
 > don't trust this for various reasons, so maybe someone can confirm this 
 > with -O3?
 
 C being slow is a consequence of its using the tree-based inliner
 (previously only C++ did).  I don't know why -O3 would cause further
 trouble.  You could build cc1plus profiled (make clean in both
 libiberty and gcc, make all in libiberty with CFLAGS="-g -O2 -profile",
 make cc1plus in gcc with CFLAGS="-g -O2 -profile"), run it on the test
 case, and take a look at the gprof output.
 
 zw