--- Comment #25 from paolo dot carlini at oracle dot com 2010-09-10 16:01
---
Good. Please give me a couple of days to come to your code. Note, since you
don't have a Copyright Assignment on file, it will be difficult to fully
acknowledge your work in the ChangeLog. Thus, I would sugges
--- Comment #24 from tstarling at wikimedia dot org 2010-09-10 15:25
---
Created an attachment (id=21766)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21766&action=view)
dynamic_cast<> hack
The attached patch uses a dynamic_cast<> hack to avoid the need to break the
ABI. I added
--- Comment #23 from tstarling at wikimedia dot org 2010-09-10 00:17
---
(In reply to comment #21)
> Anyway, not sure which STL getline we are talking about here, because e.g.
> src/istream.cc getline seems to access the stdio buffer directly:
> streamsize __size = std
--- Comment #22 from paolo dot carlini at oracle dot com 2010-09-09 16:08
---
Jakub, when, by default, cin & co boil down to stdio_sync_filebuf, the
underlying basic_streambuf is unbuffered, everything is unbuffered in the C++
library.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id
--- Comment #21 from jakub at gcc dot gnu dot org 2010-09-09 15:07 ---
#if __GNUC__ >= 3
# define _IO_BE(expr, res) __builtin_expect ((expr), res)
#else
# define _IO_BE(expr, res) (expr)
#endif
#define _IO_getc_unlocked(_fp) \
(_IO_BE ((_fp)->_IO_read_ptr >= (_fp)->_IO_read_end,
--- Comment #20 from paolo dot carlini at oracle dot com 2010-09-09 14:53
---
Good about POSIX, we would add a configure time test with some hope to enable
the mechanism outside Linux too. Anyway, I'm sure your kind of loop would
improve the performance a lot - if only we could have it
--- Comment #19 from tstarling at wikimedia dot org 2010-09-09 14:28
---
(In reply to comment #16)
> The *_unlocked versions are faster a lot actually, at least for the one
> character ops, because no locking is performed and the calls are inlined.
> But the question is whether libstdc+
--- Comment #18 from tstarling at wikimedia dot org 2010-09-09 14:12
---
Created an attachment (id=21752)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=21752&action=view)
gprof output
I haven't managed to get libstdc++ to compile with -pg, but compiling the test
program with -sta
--- Comment #17 from paolo dot carlini at oracle dot com 2010-09-09 10:42
---
At some point I tried quickly replacing some getc, did notice an improvement of
course, but of the order of magnitude I mentioned. Worth further investigating
sure (and simple, just replace in stdio_sync_ and
--- Comment #16 from jakub at gcc dot gnu dot org 2010-09-09 10:33 ---
The *_unlocked versions are faster a lot actually, at least for the one
character ops, because no locking is performed and the calls are inlined.
But the question is whether libstdc++ can use them, unless there is som
--- Comment #15 from paolo dot carlini at oracle dot com 2010-09-09 09:25
---
If you write a patch it would be of course looked at. But *please* try first
something that doesn't break the ABI, because we have *no* idea when you
changes would be applied otherwise. About the *_unlocked fu
--- Comment #14 from tstarling at wikimedia dot org 2010-09-09 02:31
---
(In reply to comment #11)
> So? We are not changing glibc here. The C++ library does *not* use buffering
> in
> the synced mode, and it does otherwise, for fstreams in particular. Where do
> you think the performa
--- Comment #13 from jakub at gcc dot gnu dot org 2010-09-08 14:48 ---
And, getdelim insist on allocating resp. reallocating the buffer. That is of
course usually desirable when used from C, but I'm afraid it isn't in this
case, where you have user provided buffer with fixed size. You
--- Comment #12 from paolo dot carlini at oracle dot com 2010-09-08 10:20
---
By the way, getdelim is not standard, thus would work only on linux, even more
special casing. More importantly, fgets *stores* newline and '\0', at variance
with getline, I don't think it can be used as-is as
--- Comment #11 from paolo dot carlini at oracle dot com 2010-09-08 09:59
---
(In reply to comment #8)
> But a 9-10x difference doesn't sound reasonable to me. The synced mode is not
> unbuffered, before or after my suggested change, it uses the internal buffer
> in
> glibc.
So? We ar
--- Comment #10 from paolo dot carlini at oracle dot com 2010-09-08 09:56
---
(In reply to comment #8)
> Maybe you should tell that to Paolo Carlini, who closed bug 15002 as "resolved
> fixed" in 2004,
And it *is* fixed. Did you actually open the testcases? Just plain fstreams,
thus no
16 matches
Mail list logo
