compiler settings in AdaGIDE

compiler settings in AdaGIDE

[Ada novice]

Hi,
    I'm using the AdaGIDE editor (version 7.45.2) together with the
GNAT AdaCore libre compiler (release 2010) on a Win XP machine. I
would like my codes to run as fast as possible and here is the content
of a typical gnat.ago file that I use. Please let me know what
improvements I can make in order for an Ada program to run in the
minimum amount of time. I understand that setting the release mode
makes a program bypass some checks and make it run faster.


-gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

WINDOWS_TARGET

-- Begin compiler switches
RELEASE <- active (current) build mode
TRUE <- are gcc/gnatmake/main/target-dir of Release := these from
Debug mode ?
debug o2 false
debug inlining false
debug unroll_loops false
debug suppress_all_checks false
debug debug_info true
debug integer_overflow_check true
debug stack_check false
debug trace_back false
debug all_validity_checks false
debug strip_all_symbols false
debug profiling false
debug verbose_details false
debug all_warnings true
debug build_in_place false
release o2 true
release inlining true
release unroll_loops true
release suppress_all_checks true
release debug_info false
release integer_overflow_check false
release stack_check false
release trace_back false
release all_validity_checks false
release strip_all_symbols true
release profiling false
release verbose_details true
release all_warnings true
release build_in_place false
-- End compiler switches
-- Begin Release mode commands


-- End Release mode commands
-- Begin run options
prompt_arguments false
ad_trace_back false
-- End run options


Thanks.
YC

Re: compiler settings in AdaGIDE

[Gautier write-only]

In the "Compiler Options" box, you can add combinations of the
following:
  -fpeel-loops -ftracer -funswitch-loops -fweb -frename-registers
and look what happens. Also -O3 instead of -O2 is worth a look.
For floating-point maths, you can try
  -mfpmath=sse -msse2 (or -msse3)
If you look on the web for those switches, there are often ideas of
other ones...
In your sources, you can check (one at a time) the effects of the
Inline pragma.

HTH
G.

Re: compiler settings in AdaGIDE

[Ada novice]

Hi,
   Thanks for your kind help. As I mentioned earlier, I use

-gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

and so the -O3 is already there though the .ago file doesn't state it.
But perhaps in AdaGIDE, the option -O3 is equivalent to:

release o2 true
release inlining true

as written in the .ago file in my earlier message.

I have tried to put all of your options at once so as to have:

-gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato -fpeel-loops -ftracer
-
funswitch-loops -fweb -frename-registers -mfpmath=sse -msse3

but I didn't see improvements in the execution time. For the sample
program that I'm testing, the options that I used previously run the
code in 1538 s and with the new options (combination of yours and
mine), it took 1546 s.

The problem is that there's not much information available on the
compiler options on the web. But I shall investigate each of the
option that you supplied and also it can be a good idea to not put all
these options at the same time. For now I read that -mfpmath=sse -
msse3 should be very suitable to increase the execution speed.

Thanks again.

YC

Re: compiler settings in AdaGIDE

[Georg Bauhaus]

On 23.07.10 16:17, Ada novice wrote:
> Hi,
>    Thanks for your kind help. As I mentioned earlier, I use
> 
> -gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

-O3 may already trigger inlining an loop unrolling.
The GNAT Guides, including GCC docs, explain this.
They have been installed with GNAT on your computer. The versions
installed with your compiler should match the software installed,
perhaps more than any other version of these same docs you find
on the net. Last time I checked the sections on optimization
in the GNAT docs recommend -O2 -funroll-loops IIRC.

One thing you could do---after showing that your program
is correct---is remove numeric overflow checking
(-gnato).  Or turn off all checks (-gnatp).  The program is no
standard Ada program then, but since your program is correct... :-)

Another thing is playing with object sizes.
Lots of risk and disappointment ahead, though!
Sometimes a definition of a type can be altered such that
its representation uses only 32 bits on some 64 bit computer,
which may or may not be good for your program.


> and so the -O3 is already there though the .ago file doesn't state it.
> I have tried to put all of your options at once so as to have:
> 
> -gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato -fpeel-loops -ftracer -
> funswitch-loops -fweb -frename-registers -mfpmath=sse -msse3

Does your program use floating point types?



Georg

Re: compiler settings in AdaGIDE

[Robert A Duff]

Georg Bauhaus <rm.dash-bauhaus@futureapps.de> writes:

> On 23.07.10 16:17, Ada novice wrote:
>> Hi,
>>    Thanks for your kind help. As I mentioned earlier, I use
>> 
>> -gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

If you want speed, then you don't want -gnatVa.

> One thing you could do---after showing that your program
> is correct---is remove numeric overflow checking
> (-gnato).

To clarify: -gnato turns overflow checking ON -- it's off
by default.

>...Or turn off all checks (-gnatp).  The program is no
> standard Ada program then, but since your program is correct... :-)

It's still a standard Ada program.

- Bob

Re: compiler settings in AdaGIDE

[Georg Bauhaus]

On 23.07.10 17:16, Robert A Duff wrote:

>> ...Or turn off all checks (-gnatp).  The program is no
>> standard Ada program then, but since your program is correct... :-)
> 
> It's still a standard Ada program.

Without -gnato?

Re: compiler settings in AdaGIDE

[sjw]

On Jul 23, 4:22 pm, Georg Bauhaus <rm.dash-bauh...@futureapps.de>
wrote:
> On 23.07.10 17:16, Robert A Duff wrote:
>
> >> ...Or turn off all checks (-gnatp).  The program is no
> >> standard Ada program then, but since your program is correct... :-)
>
> > It's still a standard Ada program.
>
> Without -gnato?

-gnato is one of the things you have to do to make GNAT a conforming
Ada compiler.

Re: compiler settings in AdaGIDE

[Robert A Duff]

sjw <simon.j.wright@mac.com> writes:

> On Jul 23, 4:22�pm, Georg Bauhaus <rm.dash-bauh...@futureapps.de>
> wrote:
>> On 23.07.10 17:16, Robert A Duff wrote:
>>
>> >> ...Or turn off all checks (-gnatp). �The program is no
>> >> standard Ada program then, but since your program is correct... :-)
>>
>> > It's still a standard Ada program.
>>
>> Without -gnato?

Sure.  How could compiler switches affect some property
("standardness") of the text of an Ada program?

> -gnato is one of the things you have to do to make GNAT a conforming
> Ada compiler.

Well, not really.  If you don't say -gnato, then GNAT is implicitly
assuming "pragma Suppress(Overflow_Check);".  That's a standard
feature of Ada, and the compiler writer gets to decide how
source text is represented.  If you don't want that text as
part of your program, say "-gnato".

Similarly, if you say "-gnatn", GNAT is implicitly including
"pragma Suppress(All_Checks);" as part of your program text.

I realize this argument is a bit of a cheat.

And I think overflow checks should be turned on by default.
But I don't think it's a conformance issue, formally speaking
-- GNAT conforms with or without -gnatn or -gnato.

Without -gnatE (dynamic elaboration checks), on the other hand, GNAT is
nonconforming, because it rejects some legal programs, and you can't
play the above "source representation" game.  I think it's good
that it's nonconforming in this case, because the default behavior
(static elaboration checks) is better than the standard way,
and it doesn't harm portability.

- Bob

Re: compiler settings in AdaGIDE

[Ada novice]

Thanks for all the reading! It seems that I better keep -gnato (as I
don't want to remove any numeric overlfow checking).

Georg: Yes, I'm using floating point types (Long_float).

I'll take away the -gnatVa as Robert suggested.


Quote from Georg: "Last time I checked the sections on optimization
in the GNAT docs recommend -O2 -funroll-loops IIRC"

What's IIRC?

In AdaGIDE, the optimization is mentioned in Options > settings as
level O2. But writing -O2 -funroll-loops implies O3. Am I right? So
then writing -O3 -funroll-loops contains redundant information.

YC

Re: compiler settings in AdaGIDE

[Simon Wright]

Ada novice <posts@gmx.us> writes:

> What's IIRC?

If I Remember Correctly.

Re: compiler settings in AdaGIDE

[Dmitry A. Kazakov]

On Fri, 23 Jul 2010 14:03:19 -0400, Robert A Duff wrote:

> sjw <simon.j.wright@mac.com> writes:
> 
>> -gnato is one of the things you have to do to make GNAT a conforming
>> Ada compiler.
> 
> Well, not really.  If you don't say -gnato, then GNAT is implicitly
> assuming "pragma Suppress(Overflow_Check);".

Objection. Ada compiler should not assume anything not explicitly
specified. Example: what about assuming

   raise Program_Error;

at each even source line?

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: compiler settings in AdaGIDE

[Robert A Duff]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

>> Well, not really.  If you don't say -gnato, then GNAT is implicitly
>> assuming "pragma Suppress(Overflow_Check);".
>
> Objection. Ada compiler should not assume anything not explicitly
> specified.

I agree 100%.  Overflow checks should be on by default,
and you should have to do something explicit to turn them off.

But we were talking about standard conformance, which is not
quite the same as what compilers "should" do.

>... Example: what about assuming
>
>    raise Program_Error;
>
> at each even source line?

Well, so long as there's an option to turn that malfeature off,
it's standard conforming.  But it's not a good idea, and the users
of that compiler (if there are any) will complain loudly. ;-)

- Bob

P.S. I work for AdaCore, so I should note that my opinion
that overflow checks should be turned on by default is
my own personal opinion.  AdaCore disagrees.  The reason
is that they are fairly expensive.

Re: compiler settings in AdaGIDE

[Peter C. Chapin]

On 2010-07-23 17:01, Robert A Duff wrote:

> P.S. I work for AdaCore, so I should note that my opinion
> that overflow checks should be turned on by default is
> my own personal opinion.  AdaCore disagrees.  The reason
> is that they are fairly expensive.

Isn't that hardware dependent (the expense, I mean)? My understanding
was that one some architectures overflow checking could be done with
minimal... even zero... impact on a correct program.

Peter

Re: compiler settings in AdaGIDE

[BrianG]

Robert A Duff wrote:

> P.S. I work for AdaCore, so I should note that my opinion
> that overflow checks should be turned on by default is
> my own personal opinion.  AdaCore disagrees.  The reason
> is that they are fairly expensive.

I assume you mean that the checks are expensive?  :-):-):-)
(Sorry, I couldn't resist!)

--I can learn to resist
-- Anything but temptation
--I can learn to co-exist
-- With anything but pain
--I can learn to compromise
-- Anything but my desires
--I can learn to get along
-- With all the things I can't explain

Re: compiler settings in AdaGIDE

[Simon Wright]

Robert A Duff <bobduff@shell01.TheWorld.com> writes:

> Similarly, if you say "-gnatn", GNAT is implicitly including
> "pragma Suppress(All_Checks);" as part of your program text.

-gnatp ?

Re: compiler settings in AdaGIDE

[Robert A Duff]

Simon Wright <simon@pushface.org> writes:

> Robert A Duff <bobduff@shell01.TheWorld.com> writes:
>
>> Similarly, if you say "-gnatn", GNAT is implicitly including
>> "pragma Suppress(All_Checks);" as part of your program text.
>
> -gnatp ?

Yeah, thanks for the correction.

- Bob

Re: compiler settings in AdaGIDE

[Ada novice]

On Jul 23, 10:52 pm, Robert A Duff <bobd...@shell01.TheWorld.com>
wrote:
> Simon Wright <si...@pushface.org> writes:
> > Robert A Duff <bobd...@shell01.TheWorld.com> writes:
>
> >> Similarly, if you say "-gnatn", GNAT is implicitly including
> >> "pragma Suppress(All_Checks);" as part of your program text.
>

So removing -gnatn is a good thing if one wants to enable checks?

YC

Re: compiler settings in AdaGIDE

[Robert A Duff]

Ada novice <posts@gmx.us> writes:

> On Jul 23, 10:52�pm, Robert A Duff <bobd...@shell01.TheWorld.com>
> wrote:
>> Simon Wright <si...@pushface.org> writes:
>> > Robert A Duff <bobd...@shell01.TheWorld.com> writes:
>>
>> >> Similarly, if you say "-gnatn", GNAT is implicitly including
>> >> "pragma Suppress(All_Checks);" as part of your program text.
>
> So removing -gnatn is a good thing if one wants to enable checks?

No.  I meant -gnatp, not -gnatn.  Sorry!

- Bob

Re: compiler settings in AdaGIDE

[Ada novice]

On Jul 24, 2:03 pm, Robert A Duff <bobd...@shell01.TheWorld.com>
wrote:

>
> No.  I meant -gnatp, not -gnatn.  Sorry!
>
> - Bob


Thanks for the clarifications. So I have the options

-O3 -gnatn -funroll-loops -gnatf -gnato -fpeel-loops -ftracer -
funswitch-loops -fweb -frename-registers -mfpmath=sse -msse3

to play with.

Re: compiler settings in AdaGIDE

[jonathan]

On Jul 23, 10:52 am, Ada novice <po...@gmx.us> wrote:
> Hi,
>     I'm using the AdaGIDE editor (version 7.45.2) together with the
> GNAT AdaCore libre compiler (release 2010) on a Win XP machine. I
> would like my codes to run as fast as possible and here is the content
> of a typical gnat.ago file that I use. Please let me know what
> improvements I can make in order for an Ada program to run in the
> minimum amount of time.
>
> Thanks.
> YC


I wrote a quick benchmark using our favorite
complex number eigen-routine, so now we can test
some these things rather quickly.

The benchmark does 100 calls to

  Complex_Eigenvalues.Eigen (P , W , V , FAIL);

using 121 x 121 complex matrices.

I compiled it with both the 2009 and 2010 GNAT GPL
compilers and timed it on a fast Intel PC
(Xeon X5460 @ 3.16GHz) with:

   time ./bench

Let's look at the best results first, and compare the
'09 and '10 GNAT GPL compilers:

   2010 GNAT GPL:
   gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll-
loops
   3.61 seconds

   2009 GNAT GPL:
   gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll-
loops
   4.57 seconds

This came as a nice surprise to me!
We can learn some more about the compiler switches by
toggling them. I'll stick to the 2010 GNAT GPL from now on:

   change -O2 to -O3:
   (gnatmake bench.adb -gnatnp -O3 -march=native -ffast-math -funroll-
loops)
   running time changes from 3.61 to 3.63 seconds

   remove -funroll-loops:
   (gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math)
   running time changes from 3.61 to 3.66 seconds

   remove -ffast-math:
   (gnatmake bench.adb -gnatnp -O2 -march=native)
   running time changes from 3.61 to 4.35 seconds

The -ffast-math had an amazing affect. I've never seen that
before ... maybe an interaction with complex number arithmetic.

Now let's check -gnato:

   add -gnato:
   (gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll-
loops -gnato)
   running time changes from 3.61 to 3.64 seconds

(Good news again.)

I suspect that the -mfpmath=sse is the default on Intel. If you
are on an Intel processor, then you have the option of running
on the 387 stack by using -mfpmath=387. The compiler even tries
to use both sse and 387 if you use -mfpmath=387,sse, or so the
man pages say. (-mfpmath=387,sse isn't very good yet .. usually
slower.)

Finally, I always use the more portable:

   type Real is digits 15;

It will give you standard double precision without
much loss in speed as far as I can tell.
To get single precision:

   type Real is digits 6;

Its not much faster than the 15 digit version in the
present benchmark. My experience is that it is almost
always a bad idea to do much arithmetic (especially
something like eigenvalue calculations), in single
precision ... maybe at best useful for minimizing data
storage size if data is known to be inaccurate.)

I wasn't planning to do any more benchmarking than this,
but the 2009 vs 2010 results were so surprising and so
welcome that I took the time to do one more test.
You can find the code in the following public directory:

http://web.am.qub.ac.uk/users/j.parker/bench_depository/

You might also want to look at some other math
routines I keep at:

http://web.am.qub.ac.uk/users/j.parker/miscellany/

On this set of floating point routines, by the way,
here's the best I could come up with for maximum speed:

gnatmake xxx.adb -gnatnp -O2 -march=native -ffast-math -funroll-loops

I've never found anything much better .. I always start with
this set of switches (-gnatnp -O2 -march=native -ffast-math
-funroll-loops) and turn other switches on and off to see
if they help.

In the benchmark directory you can find some programs I've
either collected or written over the years. I'll compare
Ada and Fortran 90 versions of a jacobi iterative eigen-
decomposition for symmetric real matrices. The Ada version
has departed slightly from the fortran version over the years,
but the two are still very close ... do almost the same amount
of computation in the same way. Mostly I just want to
compare the 2009 and the 2010 GNAT GPL though - here is
a test on a 500 x 500 Moler's matrix:

   2010 GNAT GPL:
   gnatmake jacobi_eigen_bench_1 -gnatnp -O2 -march=native -ffast-math
-funroll-loops
   2.036 seconds

   2009 GNAT GPL:
   gnatmake jacobi_eigen_bench_1 -gnatnp -O2 -march=native -ffast-math
-funroll-loops
   2.676 seconds

Again, I found this so surprising that I copied over the
test routine  jacobi_eigen_tst_1.adb  from another directory,
just to make sure that the jacobi packages were actually
working. You may want to verify this. You'll need packages
matrix_sampler_3.ads and matrix_sampler_3.adb. For testing
I start with:

  gnatmake jacobi_eigen_tst_1.adb  -gnato -gnatVa -gnata

To make the Fortran comparison I used (1) jacobi_eigen.f90,
(2) a recent version (11.0) of the the Intel Fortran compiler,
and (3) a not very recent version of gfortran (based on
gcc 4.3.2):

   ifort -fast jacobi_eigen.f90

   gfortran jacobi_eigen.f90 -O3 -march=native -funroll-loops -ffast-
math

Results on a 500 x 500 matrix:

   2.038 seconds  (2010 GNAT GPL)
   2.012 seconds  (gfortran (based on gcc 4.3.2))
   1.824 seconds  (ifort -fast, version 11.0)

Actually, INTEL's ifort is usually 20% or more faster than
than the gcc's. It can do miraculous things with nested loops
but in this case I suspect that it may be doing some algebra on
the Jacobi inner loops and then factoring out some constants
(you get gcc speed if you suppress that kind of thing with
the ifort  -fp-model strict   switch).

J.

Re: compiler settings in AdaGIDE

[Ada novice]

Many thanks for all these precious information. The settings

-gnatnp -O2 -march=native -ffast-math -funroll-loops

was only 1 s faster (1537 s) as compared to using

-gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

which took 1538 s. I have two questions:

1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
all checks?

2. In the specification -march="machine architecture here", you use
native for the machine architecture. Does this apply to Intel
processors as well?


And thank you very much for making accessible the programs on your
website. This is indeed very generous on your part. I find some very
useful codes there.


YC

Re: compiler settings in AdaGIDE

[Jeffrey R. Carter]

On 07/25/2010 06:29 AM, Ada novice wrote:
>
> 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
> all checks?

No, but it is a little faster. You asked about maximizing speed, not about 
safety. Of course, none of this is going to make a significant difference, as 
you found out yourself. Significant speed changes come about from improved 
algorithms. The correct approach is to implement correctly, clearly, and safely, 
then measure against your timing requirements (which are unspecified), and only 
make changes if the result fails to meet the timing requirements.

-- 
Jeff Carter
"I soiled my armor, I was so scared."
Monty Python & the Holy Grail
71

Re: compiler settings in AdaGIDE

[Ada novice]

On Jul 25, 9:30 pm, "Jeffrey R. Carter"

<spam.jrcarter....@spam.acm.org> wrote:
> On 07/25/2010 06:29 AM, Ada novice wrote:
> > 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
> > all checks?

> No, but it is a little faster. You asked about maximizing speed, not about
> safety. Of course, none of this is going to make a significant difference, as
> you found out yourself. Significant speed changes come about from improved
> algorithms. The correct approach is to implement correctly, clearly, and safely,
> then measure against your timing requirements (which are unspecified), and only
> make changes if the result fails to meet the timing requirements.

Thanks. I understand. I would prefer to keep the -gnatn just to be
safe.

YC

Re: compiler settings in AdaGIDE

[sjw]

On Jul 26, 3:13 pm, Ada novice <po...@gmx.us> wrote:
> On Jul 25, 9:30 pm, "Jeffrey R. Carter"
>
> <spam.jrcarter....@spam.acm.org> wrote:
> > On 07/25/2010 06:29 AM, Ada novice wrote:
> > > 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
> > > all checks?
> > No, but it is a little faster. You asked about maximizing speed, not about
> > safety. Of course, none of this is going to make a significant difference, as
> > you found out yourself. Significant speed changes come about from improved
> > algorithms. The correct approach is to implement correctly, clearly, and safely,
> > then measure against your timing requirements (which are unspecified), and only
> > make changes if the result fails to meet the timing requirements.
>
> Thanks. I understand. I would prefer to keep the -gnatn just to be
> safe.

-gnatp to suppress checks. -gnatn is about inlining.

Re: compiler settings in AdaGIDE

[Dmitry A. Kazakov]

On Mon, 26 Jul 2010 08:48:58 -0700 (PDT), sjw wrote:

> On Jul 26, 3:13�pm, Ada novice <po...@gmx.us> wrote:
>> On Jul 25, 9:30 pm, "Jeffrey R. Carter"
>>
>> <spam.jrcarter....@spam.acm.org> wrote:
>>> On 07/25/2010 06:29 AM, Ada novice wrote:
>>> > 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
>>> > all checks?
>>> No, but it is a little faster. You asked about maximizing speed, not about
>>> safety. Of course, none of this is going to make a significant difference, as
>>> you found out yourself. Significant speed changes come about from improved
>>> algorithms. The correct approach is to implement correctly, clearly, and safely,
>>> then measure against your timing requirements (which are unspecified), and only
>>> make changes if the result fails to meet the timing requirements.
>>
>> Thanks. I understand. I would prefer to keep the -gnatn just to be
>> safe.
> 
> -gnatp to suppress checks. -gnatn is about inlining.

The gentleman's set is:

debug

-gnatVa  all validity checks on
-gnatf  show all errors messages
-gnato  integer overflow check
-g  add debug info
-fstack-check  check stack overflow

release

-gnatVa
-gnatf
-gnato
-O2  medium level optimization
-gnatn  inline as defined by the user

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: compiler settings in AdaGIDE

[Ada novice]

On Jul 26, 5:48 pm, sjw <simon.j.wri...@mac.com> wrote:

>
> -gnatp to suppress checks. -gnatn is about inlining.

Yes, sorry for the mix up.

YC

Re: compiler settings in AdaGIDE

[jonathan]

On Jul 25, 2:29 pm, Ada novice <po...@gmx.us> wrote:
> Many thanks for all these precious information. The settings
>
> -gnatnp -O2 -march=native -ffast-math -funroll-loops
>
> was only 1 s faster (1537 s) as compared to using
>
> -gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato

Interesting result. I repeated the 100 calls to
COMPLEX_EIGENVALUES.Eigen ( P, W, V, FAIL) test on a
121 x 121 sized matrix.
I still get the same big decrease in running time from the
-ffast-math switch. Generic_Complex_Eigenvalues
seems sensitive to -ffast-math, but I've never seen any other
program care much about it.

> 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
> all checks?

If I remove the -gnatp  then running time goes from 3.61 to 6.1 sec.
If I remove the -gnatnp then running time goes from 3.61 to 6.7 sec.
(I suspect that you know that -gnatnp is really -gnatn -gnatp).

On a high percentage of ordinary problems (usually data
analysis) I would not care at all about the increased
running time, and I would not remove the checks. There
is another class of program I write and these run for days
or weeks. These I never stop optimizing and testing, and
all checks are removed!

> 2. In the specification -march="machine architecture here", you use
> native for the machine architecture. Does this apply to Intel
> processors as well?

I just checked the gcc man pages and found nothing at all about
-march=native.  It helps a little on all the machines I've
used and occasionally in the past it has helped enormously.  Just
now I removed -march=native from the Generic_Complex_Eigenvalues
test and running time went from 3.61 sec to 3.58 sec. When
I remove it from the Jacobi benchmark, the program slows down
slightly. So the switch seems to do something on my intel machine,
but all I can say now is that the difference is very small and
unpredictable.

J.

Re: compiler settings in AdaGIDE

[Ada novice]

On Jul 26, 12:57 am, jonathan <johns...@googlemail.com> wrote:
> On Jul 25, 2:29 pm, Ada novice <po...@gmx.us> wrote:
>
> > Many thanks for all these precious information. The settings
>
> > -gnatnp -O2 -march=native -ffast-math -funroll-loops
>
> > was only 1 s faster (1537 s) as compared to using
>
> > -gnatVa -O3 -gnatn -funroll-loops -gnatf -gnato
>
> Interesting result. I repeated the 100 calls to
> COMPLEX_EIGENVALUES.Eigen ( P, W, V, FAIL) test on a
> 121 x 121 sized matrix.
> I still get the same big decrease in running time from the
> -ffast-math switch.

Maybe what I have in my code is not affected by the -ffast-math
switch.


> On a high percentage of ordinary problems (usually data
> analysis) I would not care at all about the increased
> running time, and I would not remove the checks. There
> is another class of program I write and these run for days
> or weeks. These I never stop optimizing and testing, and
> all checks are removed!

I would prefer to keep the -gnatn.

> > 2. In the specification -march="machine architecture here", you use
> > native for the machine architecture. Does this apply to Intel
> > processors as well?
>
> I just checked the gcc man pages and found nothing at all about
> -march=native.  It helps a little on all the machines I've
> used and occasionally in the past it has helped enormously.  Just
> now I removed -march=native from the Generic_Complex_Eigenvalues
> test and running time went from 3.61 sec to 3.58 sec. When
> I remove it from the Jacobi benchmark, the program slows down
> slightly. So the switch seems to do something on my intel machine,
> but all I can say now is that the difference is very small and
> unpredictable.


Thanks for this information.

YC

Re: compiler settings in AdaGIDE

[jonathan]

On Jul 25, 2:29 pm, Ada novice <po...@gmx.us> wrote:

> 1. I see that you use -gnatnp. Is this safe to do so i.e. to suppress
> all checks?


I could not resist testing the -gnatp switch on my other benchmark,
the Jacobi eigendecomposition.  So I removed -gnatp and added -gnato,
and running time went from 2.04 sec to 2.35 sec. Not bad at all!

J.

Making measurements (Was: compiler settings in AdaGIDE)

[Jacob Sparre Andersen]

Jonathan <johnscpg@googlemail.com> wrote:

>    2010 GNAT GPL:
>    gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll-
> loops
>    3.61 seconds

I hope this is a average or median value from several measurements.

Which of the two?  Did you estimate the uncertaincy of your
measurements?

Jacob
--
Jacob Sparre Andersen Research & Innovation
Vesterbrogade 148K, 1. th.
1620 K�benhavn V
Danmark

Phone:    +45 21 49 08 04
E-mail:   jacob@jacob-sparre.dk
Web-sted: http://www.jacob-sparre.dk/

Re: Making measurements (Was: compiler settings in AdaGIDE)

[jonathan]

On Jul 26, 9:02 am, Jacob Sparre Andersen <ja...@jacob-sparre.dk>
wrote:
> Jonathan <johns...@googlemail.com> wrote:
> >    2010 GNAT GPL:
> >    gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll-
> > loops
> >    3.61 seconds
>
> I hope this is a average or median value from several measurements.
>
> Which of the two?  Did you estimate the uncertaincy of your
> measurements?
>
> Jacob
> --
> Jacob Sparre Andersen Research & Innovation
> Vesterbrogade 148K, 1. th.
> 1620 København V
> Danmark
>
> Phone:    +45 21 49 08 04
> E-mail:   ja...@jacob-sparre.dk
> Web-sted:http://www.jacob-sparre.dk/

I did a few runs with the linux: time ./bench
on an empty machine and got numbers like this:

  0m3.612s  0m3.604s  0m3.620s  0m3.612s  0m3.620s  0m3.604s

Some programs fluctuate hugely in running time .. never
figured out why.

J.

Re: Making measurements

[Jacob Sparre Andersen]

Jonathan <johnscpg@googlemail.com> wrote:

> I did a few runs with the linux: time ./bench
> on an empty machine and got numbers like this:
>
>   0m3.612s  0m3.604s  0m3.620s  0m3.612s  0m3.620s  0m3.604s

Looks like we have a nice symmetric case where the average and the
median are the same. ;-)

> Some programs fluctuate hugely in running time .. never figured out
> why.

It may be something as simple as which core the program happens to be
running on.  Or which operating system processes are running in the
background.

In general it is a good idea to report an average of at least three
timing measurements and the standard deviation of the measurements.

Jacob
--
Jacob Sparre Andersen Research & Innovation
Vesterbrogade 148K, 1. th.
1620 K�benhavn V
Danmark

Phone:    +45 21 49 08 04
E-mail:   jacob@jacob-sparre.dk
Web-sted: http://www.jacob-sparre.dk/

Re: compiler settings in AdaGIDE

[Colin Paul Gloster]

On Sat, 24 Jul 2010, Jonathan sent:

|----------------------------------------------------------------------|
|"[..]                                                                 |
|                                                                      |
|   time ./bench                                                       |
|                                                                      |
|Let's look at the best results first, and compare the                 |
|'09 and '10 GNAT GPL compilers:                                       |
|                                                                      |
|   2010 GNAT GPL:                                                     |
|   gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll- |
|loops                                                                 |
|   3.61 seconds                                                       |
|                                                                      |
|   2009 GNAT GPL:                                                     |
|   gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math -funroll- |
|loops                                                                 |
|   4.57 seconds                                                       |
|                                                                      |
|This came as a nice surprise to me!                                   |
|We can learn some more about the compiler switches by                 |
|toggling them. I'll stick to the 2010 GNAT GPL from now on:           |
|                                                                      |
|   change -O2 to -O3:                                                 |
|   (gnatmake bench.adb -gnatnp -O3 -march=native -ffast-math -funroll-|
|loops)                                                                |
|   running time changes from 3.61 to 3.63 seconds                     |
|                                                                      |
|   remove -funroll-loops:                                             |
|   (gnatmake bench.adb -gnatnp -O2 -march=native -ffast-math)         |
|   running time changes from 3.61 to 3.66 seconds                     |
|                                                                      |
|   remove -ffast-math:                                                |
|   (gnatmake bench.adb -gnatnp -O2 -march=native)                     |
|   running time changes from 3.61 to 4.35 seconds                     |
|                                                                      |
|The -ffast-math had an amazing affect. I've never seen that           |
|before ... maybe an interaction with complex number arithmetic.       |
|                                                                      |
|[..]"                                                                 |
|----------------------------------------------------------------------|

The GNU Compiler Collection's  -ffast-math  switch used to provide a
significant increase in speed for various frontends excluding GNAT. In
my experience at least, this effect had no special relationship with
complex numbers.

Re: compiler settings in AdaGIDE

[Ada novice]

In this document:

http://www.benchit.org/downloads/documents/comparing_gcc42_gcc43.pdf

some guidelines are given for increased performance. Tow points worthy
to note:

1. The -ftree-vectorize option is now on by default under -O3.
2. Tuning for Intel Core 2 processors is available via -mtune=core2
and -march=core2.

Have anyone tried these options?

YC