[Bug libstdc++/7076] iostreams very slow compared with earlier gcc versions

[Bug libstdc++/7076] iostreams very slow compared with earlier gcc versions

[pme@gcc.gnu.org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


pme@gcc.gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |daniel.lemire@nrc.gc.ca


------- Additional Comments From pme@gcc.gnu.org  2003-06-01 20:10 -------
*** Bug 10672 has been marked as a duplicate of this bug. ***



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libstdc++/7076] iostreams very slow compared with earlier gcc versions

[pme@gcc.gnu.org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


pme@gcc.gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |wj178839@wp.pl


------- Additional Comments From pme@gcc.gnu.org  2003-06-01 20:11 -------
*** Bug 8761 has been marked as a duplicate of this bug. ***



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[giovannibajo@libero.it]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


giovannibajo@libero.it changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|0000-00-00 00:00:00         |2003-06-09 18:21:42
               date|                            |
            Summary|iostreams very slow compared|[3.3/3.4 regression]
                   |with earlier gcc versions   |iostreams very slow compared
                   |                            |with earlier gcc versions
   Target Milestone|---                         |3.3.1


------- Additional Comments From giovannibajo@libero.it  2003-06-09 18:21 -------
Reconfirmed with current mainline. Benjamin, I don't see this bug fixed in the 
mainline, not even adding ios::sync_with_stdio(false). It still takes more than 
7 seconds with mainline 20030604 on my computer, instead of less than a second 
like the GCC295 executable takes.

I rated this as a 3.3/3.4 regression.

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[bkoz at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


bkoz at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.3.1                       |3.3.2


------- Additional Comments From bkoz at gcc dot gnu dot org  2003-06-24 19:41 -------

This work is in progress, and will not be ported back to 3.3.x until
after 3.3.1.  I'll update this before 3.3.2

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[pinskia at physics dot uc dot edu]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


pinskia at physics dot uc dot edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2003-06-09 18:21:54         |2003-08-03 19:25:21
               date|                            |

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[mmitchel at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076


mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.3.2                       |3.3.3


------- Additional Comments From mmitchel at gcc dot gnu dot org  2003-10-16 09:44 -------
Postponed until GCC 3.3.3.

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[gdr at gcc dot gnu dot org]

------- Additional Comments From gdr at gcc dot gnu dot org  2003-12-21 18:04 -------
Benjamin, 
Do you have plan to fix this for 3.3.3?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2003-12-23 19:11 -------
Will be too invasive at this point for 3.3.3.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.3.3                       |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |minor


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[ian at airs dot com]

------- Additional Comments From ian at airs dot com  2004-01-13 04:16 -------
I just tried this test case on gcc 2.95.3 and current mainline.  I find that
current mainline is slower than 2.95.3, but it is less than twice as slow (0.42
seconds for 2.95.3, 0.76 seconds for mainline).  At this point I don't think
this counts as a significant gcc 3.4 regression.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[bkoz at gcc dot gnu dot org]

------- Additional Comments From bkoz at gcc dot gnu dot org  2004-01-13 16:10 -------

I agree with you Ian. 

We were keeping this open for a long time as a status marker for the io
perfomance work. However, now libstdc++ has a performance testing testsuite, and
we use that to measure performance. (There is a test case for this issue.)

-benjamin


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[paolo at gcc dot gnu dot org]

------- Additional Comments From paolo at gcc dot gnu dot org  2004-01-13 19:46 -------
Fixed for 3.4.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[gdr at integrable-solutions dot net]

------- Additional Comments From gdr at integrable-solutions dot net  2004-01-13 20:02 -------
Subject: Re:  [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

"paolo at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org> writes:

| Fixed for 3.4.

I guess there is no hope to have it fixed for 3.3.x.  Right?

-- Gaby


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[paolo at gcc dot gnu dot org]

------- Additional Comments From paolo at gcc dot gnu dot org  2004-01-13 20:17 -------
Hi Gaby.

3_3 is already quite good at integer types puts. It would be really nice
to have for 3_3 too the analogous improvement for gets (we are talking about
a factor or 6-7x ;)

Unfortunately, I'm afraid that the below changes the ABI, but I would be
*very* happy to be corrected (there is big interest inside SUSE too)

What do you think?

Paolo.

/////////////

2003-12-08  Paolo Carlini  <pcarlini@suse.de>

	* config/locale/generic/c_locale.cc (__convert_to_v(long&),
	__convert_to_v(unsigned long&), __convert_to_v(long long&),
	__convert_to_v(unsigned long long&): Remove.
	(__convert_to_v(float&), __convert_to_v(double&),
	__convert_to_v(long double&)): Remove last unused parameter.
	* config/locale/gnu/c_locale.cc: Likewise.
	* include/bits/locale_facets.h: Likewise.
	(num_get::_M_extract_int): Update declaration.
	(class __num_base): Tweak _S_ie and _S_iE according to the
	_S_atoms_in changes.
	* include/bits/locale_facets.tcc (num_get::_M_extract_int):
	Provide directly the final _ValueT, not a string: accumulate
	the result checking for overflow at each digit.
	(num_get::do_get(bool&), num_get::do_get(long&),
	num_get::do_get(unsigned short&), num_get::do_get(unsigned int&),
	num_get::do_get(unsigned long&), num_get::do_get(long long&),
	num_get::do_get(unsigned long long&), num_get::do_get(void*&)):
	Use the new _M_extract_int, simplify.
	* src/locale-inst.cc (num_get::_M_extract_int(long&),
	num_get::_M_extract_int(unsigned short&),
	num_get::_M_extract_int(unsigned int&),
	num_get::_M_extract_int(unsigned long&),
	num_get::_M_extract_int(long long&),
	num_get::_M_extract_int(unsigned long long&): Instantiate.
	* src/locale_facets.cc (__num_base::_S_atoms_in): Reorder
        the chars in the natural order: abcdefABCDEF.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gdr at integrable-solutions
                   |                            |dot net


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[gdr at integrable-solutions dot net]

------- Additional Comments From gdr at integrable-solutions dot net  2004-01-13 20:27 -------
Subject: Re:  [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

"paolo at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org> writes:

| Hi Gaby.
| 
| 3_3 is already quite good at integer types puts. It would be really nice
| to have for 3_3 too the analogous improvement for gets (we are talking about
| a factor or 6-7x ;)
| 
| Unfortunately, I'm afraid that the below changes the ABI, but I would be

which portions do you think may imply ABI change?
The reason I'm asking is for someone double-checking, because there is
no point is having you spend time on something that will not be
acceptable for 3_3.

-- Gaby


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[paolo at gcc dot gnu dot org]

------- Additional Comments From paolo at gcc dot gnu dot org  2004-01-13 20:37 -------
Well, two punctualizations:
1- Actually, 3_3 is way behind mainline for puts :( This is because the
   locale cache was disabled due to leaks that couldn't be fixed in time.
   But, its much faster than 3.0 and 3.1.
2- About the recent code for gets: I have removed many functions and instantiated
   some more templates. Sorry for my naivete about such matters. First blush
   the patch looks quite invasive, but I'm not sure if actually breaks the ABI.
   If there are chances that it doesn't I would be *very* happy to work on it.
   we can expect a good improvement even without the caching, probably 2-3x.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[paolo at gcc dot gnu dot org]

------- Additional Comments From paolo at gcc dot gnu dot org  2004-01-13 20:49 -------
... also, num_get::_M_extract_int has got a different signature :-(

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[gdr at integrable-solutions dot net]

------- Additional Comments From gdr at integrable-solutions dot net  2004-01-13 21:28 -------
Subject: Re:  [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

"paolo at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org> writes:

| ... also, num_get::_M_extract_int has got a different signature :-(

OK, this settle the issue.  We won't have a speedup for 3.3.x -- unless
someone wants to provoke an ABI debates.

Thanks, Paolo. 

-- Gaby 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[bkoz at redhat dot com]

------- Additional Comments From bkoz at redhat dot com  2004-01-13 23:17 -------
Subject: Re:  [3.3/3.4 regression] iostreams very slow
 compared with earlier gcc versions


>OK, this settle the issue.  We won't have a speedup for 3.3.x -- unless
>someone wants to provoke an ABI debates.

Just to clarify: changing the ABI in 3.[23] is out of the question. 

We're trying to put as much as possible back into 3.3, but it's not
always possible to do this.

I think the state of this bug report is that for buffered io, 3.4 is
faster than 2.x. For unbuffered, 3.4 is slower than 2.x but less than
earlier 3.x releases.

-benjamin


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076

[Bug libstdc++/7076] [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

[joe dot buck at synopsys dot com]

------- Additional Comments From joe dot buck at synopsys dot com  2004-01-13 23:21 -------
Subject: Re:  [3.3/3.4 regression] iostreams very slow compared with earlier gcc versions

On Tue, Jan 13, 2004 at 11:17:50PM -0000, bkoz at redhat dot com wrote:
> I think the state of this bug report is that for buffered io, 3.4 is
> faster than 2.x. For unbuffered, 3.4 is slower than 2.x but less than
> earlier 3.x releases.

Great.  Those who require speed will use buffered I/O, using explicit
flush calls if needed.  As far as I'm concerned, if a new design is faster
in buffered mode and slower in unbuffered, it is superior from a
performance point of view.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=7076