--- Comment #36 from jacob at math dot jussieu dot fr 2008-02-27 16:58
---
That's great; from the assembly code I take it that you are referring tothe
last benchmark.cpp; I was referring to the first one. Again, my 4.3 is one
month old so maybe things have further improved
--- Comment #34 from jacob at math dot jussieu dot fr 2008-02-27 16:36
---
I retried today with
g++-4.3 (Ubuntu 4.3-20080126-1ubuntu1) 4.3.0 20080126 (experimental) [trunk
revision 131874]
and my benchmark ran in 25s compared to 20s with g++-4.2.3.
So there is certainly a lot of
--- Comment #5 from jacob at math dot jussieu dot fr 2008-02-27 16:22
---
Oh sure, I hadn't thought about that, but it makes perfect sense to not
"flatten" callees whose body source code is not available.
I tried g++-4.3 and confirm that it is much faster and doesn
--- Comment #3 from jacob at math dot jussieu dot fr 2008-02-27 15:49
---
Thank you for the quick reply. I'm glad to hear that it's fixed and much
improved in 4.3.
I tried what you said about __attribute__((flatten)) (I copied and pasted your
code) and that didn&#x
--- Comment #1 from jacob at math dot jussieu dot fr 2008-02-27 14:48
---
Created an attachment (id=15238)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15238&action=view)
test case referred to in first post
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35393
normal
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jacob at math dot jussieu dot fr
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35393
--- Comment #27 from jacob at math dot jussieu dot fr 2007-11-09 12:55
---
OK, I'm using ubuntu and there's no 4.3 package, and last time I tried
compiling a snapshot it failed. I hope someone else will be able to run the
benchmark and post the result.
--
http://g
--- Comment #25 from jacob at math dot jussieu dot fr 2007-11-09 12:49
---
Huge thanks to Richard Guenther for the fix!
I don't have a 4.3 installed here but I trust you if you say it's fixed :)
Also thanks to everybody who helped, especially to Paolo Bonzini for the
minima
--- Comment #21 from jacob at math dot jussieu dot fr 2007-11-07 20:58
---
Hi, I'm the guy behind Eigen, from which the initial testcase is taken.
Would it help the compiler if I removed most of the "const" keywords in my
code?
I am asking because of the followin
--- Comment #7 from jacob at math dot jussieu dot fr 2007-09-30 16:41
---
OK, the person who reported the speed regression with g++-4.3 (Michael Olbrich)
will open a bug report. I don't actually have g++-4.3 compiled so I'm not the
right person to do so.
Thanks,
Benoit
--- Comment #5 from jacob at math dot jussieu dot fr 2007-09-30 14:58
---
OK, found it, it was a very stupid error by me. Declared the matrix's array
with the wrong size, so some write eventually wrote outside of the matrix's
array, thus overwriting a reference.
Closing the
--- Comment #4 from jacob at math dot jussieu dot fr 2007-09-30 13:47
---
I had tried to make a shorter testcase but didn't find any that caused a crash.
But yes, the shorter testcase that you found indeed causes a crash here. With
both 4.1 and 4.2.
So indeed, that's prob
--- Comment #2 from jacob at math dot jussieu dot fr 2007-09-30 09:16
---
Here are some thoughts about why it is so fast with g++-4.2, perhaps related to
why it segfaults.
My library is an Expression Templates library. So when you do m1+m2 with
matrices m1 and m2, instead of computing
--- Comment #1 from jacob at math dot jussieu dot fr 2007-09-30 08:57
---
Created an attachment (id=14271)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=14271&action=view)
The archive allowing to reproduce the bug.
Of course, I forgot the attachment :)
--
http://gcc.
t this
didn't happen with g++-4.1 suggests to me that this might be a bug in g++-4.2.
Of course I might be wrong, I'm not an expert.
Cheers,
Benoit
--
Summary: segfault in program compiled by g++ 4.2, corrupted
reference
Product: gcc
Vers
--- Comment #18 from jacob at math dot jussieu dot fr 2006-12-14 16:47
---
(In reply to comment #15)
> If the loop bounds are compile-time constants you can use template
> metaprogramming to unroll them.
>
I shouldn't elaborate on this as this is not the subject of t
--- Comment #17 from jacob at math dot jussieu dot fr 2006-12-14 16:34
---
(In reply to comment #15)
> If the loop bounds are compile-time constants you can use template
> metaprogramming to unroll them.
>
ah right, I initially misread "unroll" as "unnest&quo
--- Comment #16 from jacob at math dot jussieu dot fr 2006-12-14 16:28
---
(In reply to comment #15)
> If the loop bounds are compile-time constants you can use template
> metaprogramming to unroll them.
>
That is true, I will think about that.
I was also mentionnin
--- Comment #14 from jacob at math dot jussieu dot fr 2006-12-14 16:24
---
I forgot to say that size() is an inline function returning a constant that is
known at compile-time (a template parameter). Otherwise, of course, I wouldn't
expect these loops to get unrolled...
--
--- Comment #13 from jacob at math dot jussieu dot fr 2006-12-14 16:22
---
(In reply to comment #12)
> Yes. However, all this is only for my reduced testcase without the use of
> the C++ class. For the original testcase, the issues Andrew P. identified
> are still true.
OK
--- Comment #11 from jacob at math dot jussieu dot fr 2006-12-14 15:56
---
Very interesting, thanks... so does it mean that gcc did loop unrolling after
all? (sorry, I'm a newbie when it comes to compiler/assembler stuff).
And the speed difference was only caused by the com
--- Comment #6 from jacob at math dot jussieu dot fr 2006-12-13 20:24
---
(In reply to comment #4)
> most likely not in 4.2. I think something could be done in 4.3 timeframe.
ah, ok. Thanks for the information.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30201
--- Comment #5 from jacob at math dot jussieu dot fr 2006-12-13 20:22
---
Nope... with -O3 -ffast-math I get 1.9 seconds in average (this is a laptop
with CPU frequency scaling, so it's difficult to get precise numbers). Adding
-funroll-loops in addition to -ffast-math doesn'
--- Comment #2 from jacob at math dot jussieu dot fr 2006-12-13 18:54
---
Not understanding much in compiler stuff, I tried what you said, namely replace
the loop with
for( int i = 0; i < 3; i++ )
for( int j = 0; j < 3; j++ )
{
bool a = (
esn't unroll nested loops
Product: gcc
Version: 4.1.1
Status: UNCONFIRMED
Severity: major
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jacob at math dot jussieu dot fr
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30201
25 matches
Mail list logo
