X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f
X-Recipient: djgpp AT delorie DOT com
Message-ID: <4B2E7C60.2000600@iki.fi>
Date: Sun, 20 Dec 2009 21:34:56 +0200
From: Andris Pavenis <andris DOT pavenis AT iki DOT fi>
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; fi; rv:1.9.1.5) Gecko/20091204 Thunderbird/3.0
MIME-Version: 1.0
To: djgpp AT delorie DOT com
Subject: Re: ANNOUNCE: DJGPP port of GMP-4.3.1
References: <200912171904 DOT nBHJ42qV008905 AT delorie DOT com> <6d1f3dc2-9a0c-4a2b-9b23-3b746c8d8700 AT v30g2000yqm DOT googlegroups DOT com>
In-Reply-To: <6d1f3dc2-9a0c-4a2b-9b23-3b746c8d8700@v30g2000yqm.googlegroups.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Reply-To: djgpp AT delorie DOT com

20.12.2009 3:59, Rugxulo kirjoitti:
> Hi,
>
> On Dec 17, 12:56 pm, Andris Pavenis<andris DOT pave DOT  DOT  DOT  AT iki DOT fi>  wrote:
>    
>> This is announcement of DJGPP port of GMP-4.3.1
>>
>> GNU MP is a portable library written in C for arbitrary precision
>> arithmetic on integers, rational numbers, and floating-point numbers.
>>
>> See http://gmplib.org/ for more details.
>>      
> Dumb question, but why don't they make "--enable-fat" the default for
> x86? I just built it, and it's a little larger, but it should
> (ideally) be much faster, esp. for time-intensive things that use it
> (e.g. GCC) since it checks at runtime what cpu-specific code to run.
>
> (your compile for /beta/):
> -rw-a--     3.0 fat   517550 b- defX 09-Dec-13 18:23 lib/libgmp.a
> -rw-a--     3.0 fat    22976 b- defX 09-Dec-13 18:23 lib/libgmpxx.a
> -rw-a--     3.0 fat   247714 b- defX 09-Dec-13 18:23 lib/libmp.a
>
> (my compile using /beta/):
> 12/19/2009  06:50 PM           651,238 libgmp.a
> 12/19/2009  06:50 PM            23,300 libgmpxx.a
> 12/19/2009  06:51 PM           377,556 libmp.a
>
> http://rapidshare.com/files/323273044/gmp431b-beta-fat.zip.html
> MD5: C961D95121DCCC8A76A243A0F609798C
>
>    
Well,  did some tests (DJGPP v204pre only) using gmp-chudnovski.c 
(http://gmplib.org/pi-with-gmp.html).
In both cases the task was to compute pi with 100000000 digits.

1) the configure option --enable-fat is added:

D:\DJDEV\gmp\test>gmp-chudnovsky.exe 100000000
#terms=7051366, depth=24
sieve   time =  1.483
..................................................

bs      time = 660.055
    gcd  time =  0.000
div     time = 55.000
sqrt    time = 29.615
mul     time = 21.264
total   time = 767.582
    P size=145605885 digits (1.456059)
    Q size=145605879 digits (1.456059)

2) the library from earlier build (no --enable-fat)

D:\DJDEV\gmp\test>gmp-chudnovsky.exe 100000000
#terms=7051366, depth=24
sieve   time =  1.483
..................................................

bs      time = 925.659
    gcd  time =  0.000
div     time = 74.011
sqrt    time = 43.407
mul     time = 28.571
total   time = 1073.296
    P size=145605885 digits (1.456059)
    Q size=145605879 digits (1.456059)

As one can see, there is some improvement when --enable-fat has been 
specified.

I do not believe that compilation speed of GCC is noticeably dependant 
on optimization level of GMP.
As far as I understand GMP and MPFR is being using for propagating 
constants (if one specifies
some constant expression, GCC tries to calculate value in compile time 
instead of genarating code
for evaluation this expression). I guess that the number of such 
constant expressions is usually too small
to cause any noticable increase of compilation time dues to 
non-optimality of GMP.

Andris