gmp-4.3.0 is out

[KriZp]

I hit the same problem, and changed the configure script of gmp-ecm in line 24196 to read

  sprintf (buffer, "%d.%d.%d", __GNU_MP_VERSION, __GNU_MP_VERSION_MINOR, __GNU_MP_VERSION_PATCHLEVEL);

instead of

  sprintf (buffer, "%d.%d", __GNU_MP_VERSION, __GNU_MP_VERSION_MINOR);

After that it compiled.

[ET_]

Originally Posted by M0CZY

I have been trying to compile GMP-ECM using the new GMP-4.3.0 under Linux, but I have hit a snag.

The GMP installs correctly, and passes 'make check', but in the ECM stage, when I get to './configure --with-gmp=/usr/local', it outputs some stuff then finishes with

checking if gmp.h version and libgmp version are the same... (4.3/4.3.0) no
configure: error: 'gmp.h' and 'libgmp' have different versions, you have to reinstall GMP properly.
user@user-desktop:~/ecm-6.2.2$

I received the same difference in versions when I compiled under Windows using MinGW/Msys, but the program ignored it and continued working.

Could this be a bug in the new version of GMP?

Did you try "make install" after "make" and "make check"?

Luigi

[xilman]

Originally Posted by M0CZY

I have been trying to compile GMP-ECM using the new GMP-4.3.0 under Linux, but I have hit a snag.

Could this be a bug in the new version of GMP?

It's not a bug, it's a mis-feature.

The same change to the GMP version structure has hit the maintainers of a number of other packages similarly. Mail was sent to the GMP developers' list earlier today.

Paul

[M0CZY]

Did you try "make install" after "make" and "make check"?

Yes.
After 'make' and 'make check', under Ubuntu Linux I did 'sudo make install'
and 'make check' again, and it passed all the tests again.
I understand that it installs by default into /usr/local

[akruppa]

Ugh. The new format for the version string (always three parts from now on, including patchlevel even if it's zero) breaks the test. This needs a fix kinda soonish.

Alex

[Jeff Gilchrist]

Originally Posted by CRGreathouse

I was more interested in GMP vs MPIR, but that's good to know. 'Word on the street' was that the last versions of GMP and MPIR were about the same speed; will that still hold, or will GMP pull ahead?

They are about the same speed, GMP 4.3.0 is slightly faster now on AMD and Core2 than MPIR 1.1. Here are some benchmarks that use some common functions in a loop:

mpz_mul(t, r, r);
mpz_sub_ui(t, t, 2);
mpz_mod(r, t, p);

Intel Xeon E5405 @ 2.0GHz (Family 6, Model 23, Stepping 6)
Linux 64bit, gcc version 4.1.2
83338 bits (25088 digits)
Library		Time1		Time2
----------	---------	---------
GMP 4.3.0	2m34.857s	2m34.835s
MPIR 1.1.0	2m58.627s	2m58.706s
MPIR 1.0.0	3m00.482s	3m00.461s
GMP 4.2.4	5m38.156s	5m38.173s

Intel Xeon E5440 @ 2.83GHz (Family 6, Model 23, Stepping 10)
Linux 64bit, gcc version 4.1.2
83338 bits (25088 digits)
Library		Time1		Time2
----------	---------	---------
GMP 4.3.0	1m49.819s	1m49.436s
MPIR 1.1.0	2m05.877s	2m05.959s
MPIR 1.0.0	2m07.519s	2m07.499s
GMP 4.2.4	3m58.960s	3m58.976s

AMD Opteron 248 @ 2.2GHz (Family 15, Model 37, Stepping 1)
Linux 64bit, gcc version 3.4.6
83338 bits (25088 digits)
Library		Time1		Time2
----------	---------	---------
GMP 4.3.0	1m35.091s	1m34.815s
MPIR 1.1.0	1m36.480s	1m35.923s
MPIR 1.0.0	1m37.501s	1m36.857s
GMP 4.2.4	2m56.352s	2m55.702s

Jeff.

[Jeff Gilchrist]

Here are some GMP-ECM benchmarks comparing the two on Core2. Since --enable-asm-redc is used, GMP is mostly taken out of the equation anyways.

Intel Xeon E5405 @ 2.0GHz (Family 6, Model 23, Stepping 6)
Linux 64bit, gcc version 4.1.2, --enable-asm-redc

ECM
Factoring: 1877138824359859508015524119652506869600959721781289179190693027302028679377371001561
B1=20000000
Sigma: 980060817

MPIR 1.1.0
Step 1 took 61816ms | Step 1 took 61889ms
Step 2 took 27306ms | Step 2 took 27222ms

GMP 4.3.0
Step 1 took 58398ms | Step 1 took 58419ms
Step 2 took 26361ms | Step 2 took 26473ms


P-1
Factoring: 1877138824359859508015524119652506869600959721781289179190693027302028679377371001561
B1=20000000
x0: 524328229

MPIR 1.1.0
Step 1 took 5823ms | Step 1 took 5831ms
Step 2 took 6895ms | Step 2 took 6918ms

GMP 4.3.0
Step 1 took 6012ms | Step 1 took 5999ms
Step 2 took 6904ms | Step 2 took 6886ms


P+1
Factoring: 1877138824359859508015524119652506869600959721781289179190693027302028679377371001561
B1=20000000
x0: 524328229

MPIR 1.1.0
Step 1 took 12413ms | Step 1 took 12354ms
Step 2 took 10254ms | Step 2 took 10236ms

GMP 4.3.0
Step 1 took 11844ms | Step 1 took 12221ms
Step 2 took 10294ms | Step 2 took 10259ms

[akruppa]

An rc for 6.2.3 is at
http://www.loria.fr/~kruppaal/ecm-6.2.3d.tar.gz

From NEWS:

Changes between ecm-6.2.2 and ecm-6.2.3:
* Fixed incompatibility with GMP 4.3.0 when testing version in configure
* SSE2 asm code for Visual C added in stage 2 NTT code
* Small improvement to x86_64 mulredc asm code, slight speedup on Core 2
* Fixed incorrect carry propagation in subquadratic REDC code which
  could lead to incorrect arithmetic in rare cases
* Fixed memory leak with -v parameter when factor was found in ECM stage 1
* Fixed bug which caused only one ECM curve to be run in spite of -c
  parameter if input line did not end in newline
* Assembler mulredc code enabled by default on suitable architectures

The config.guess/config.sub scripts of autotools were updated, too... I hope this didn't break too much. Tests on different platforms, especially MinGW, MSVC, and Mac would be welcome.

Alex

[Jeff Gilchrist]

Originally Posted by akruppa

The config.guess/config.sub scripts of autotools were updated, too... I hope this didn't break too much. Tests on different platforms, especially MinGW, MSVC, and Mac would be welcome.

I have only had a chance to try it on cygwin on a Q9550 Core2 but for some reason configure thinks it is a pentium3 so I don't get any SSE2 code.

Jeff.

[akruppa]

What does uname -m say? If that returns, e.g., i686 as the cpu type, then configure uses Pentium3 code by default. You could override that with configure --build=i786.

Alex

[10metreh]

Originally Posted by akruppa

* Fixed bug which caused only one ECM curve to be run in spite of -c parameter if input line did not end in newline