How I install fastECPP onto my Ubuntu Machines

[VBCurtis]

Originally Posted by Stargate38

Anyone know a solution? I'm doing this in WSL, under Windows 10. I plan on using fastECPP instead of Primo from now on.

Give up on windows and move your technical work to linux permanently. :)

[EdH]

Here are some timing comparisons for a single machine with 4 cores and 8 threads running ecpp and Primo:

Candidate size: 300 decimal digits (mm:ss):
ecpp single core: 00:04
ecpp-mpi 4 cores: 00:24
primo 8 cores: 00:03.65
Candidate size: 1000 decimal digits (mm:ss):
ecpp single core: 01:34
ecpp-mpi 4 cores: 01:42
primo 8 threads: 00:47.57
Candidate size: 2000 decimal digits (mm:ss):
ecpp single core: 21:29
ecpp-mpi 4 cores: 11:42
primo 8 threads: 13:13
Candidate size: 3000 decimal digits (mm:ss):
ecpp single core: 129:08
ecpp-mpi 4 cores: 45:44
primo 8 threads: 75:20

Conclusion: Primo clearly runs faster than fastECPP until well over 1000 decimal digits. By 2000 decimal digits, fastECPP with MPI becomes faster.

[EdH]

Originally Posted by Stargate38

I was able to compile mpfrcx successfully, but when I try to compile cm, I get the following on the "make check" step:

../config/test-driver: line 112: 10398 Segmentation fault      (core dumped) "$@" >> "$log_file" 2>&1
FAIL: tecpp
============================================================================
Testsuite summary for cm 0.4.1dev
============================================================================
# TOTAL: 3
# PASS:  2
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0
============================================================================
See tests/test-suite.log
Please report to andreas.enge@inria.fr
============================================================================
make[3]: *** [Makefile:682: test-suite.log] Error 1
make[3]: Leaving directory '/root/Math/cm-0.4.1dev-41c4bce/cm-0.4.1dev/tests'
make[2]: *** [Makefile:790: check-TESTS] Error 2
make[2]: Leaving directory '/root/Math/cm-0.4.1dev-41c4bce/cm-0.4.1dev/tests'
make[1]: *** [Makefile:877: check-am] Error 2
make[1]: Leaving directory '/root/Math/cm-0.4.1dev-41c4bce/cm-0.4.1dev/tests'
make: *** [Makefile:407: check-recursive] Error 1

I've attached the log file in question. Anyone know a solution? I'm doing this in WSL, under Windows 10. I plan on using fastECPP instead of Primo from now on.

I just experienced the exact same error when trying to install the cm dev version on a Debian system that I had already been running the other version on. I needed to use make uninstall from the root user and make clean as the non-root user. I then removed the directory and started again from the tar file. All went well with that try.

This had nothing to do with WSL. This was a pure Debian system.

[EdH]

For anyone interested, I have added a BASH script to the first post, that can process factordb PRPs automatically using fastECPP.

ETA: I added a random value to be added to the candidate download. This should help avoid collisions between multiple workers in the same range of the db. The "srand" value can be considered the range for the PRPs, beginning at the start value.

[frank1285]

I have the same issue as discussed earlier:

mv -f .deps/ecpp.Tpo .deps/ecpp.Plo
/bin/bash ../libtool --tag=CC --mode=link gcc -std=c99 -g -O2 -version-info 0:0:0 -o libcm.la -rpath /usr/local/lib get_version.lo timer.lo file.lo nt.lo qdev.lo modular.lo modpol.lo mpzx.lo classgroup.lo modclass.lo curve.lo param.lo class.lo jmodp.lo pari.lo ecpp.lo -lpari -lmpfrcx -lmpc -lmpfr -lgmp -lz -lm
libtool: link: gcc -shared -fPIC -DPIC .libs/get_version.o .libs/timer.o .libs/file.o .libs/nt.o .libs/qdev.o .libs/modular.o .libs/modpol.o .libs/mpzx.o .libs/classgroup.o .libs/modclass.o .libs/curve.o .libs/param.o .libs/class.o .libs/jmodp.o .libs/pari.o .libs/ecpp.o -lpari -lmpfrcx -lmpc -lmpfr -lgmp -lz -lm -g -O2 -Wl,-soname -Wl,libcm.so.0 -o .libs/libcm.so.0.0.0
/usr/bin/ld: pari_mainstack: TLS definition in /usr/lib/gcc/x86_64-linux-gnu/7/../../../x86_64-linux-gnu/libpari.so section .tbss mismatches non-TLS reference in .libs/pari.o
/usr/lib/gcc/x86_64-linux-gnu/7/../../../x86_64-linux-gnu/libpari.so: error adding symbols: Bad value
collect2: error: ld returned 1 exit status
Makefile:458: recipe for target 'libcm.la' failed
make[2]: *** [libcm.la] Error 1
make[2]: Leaving directory '/home/frank/cm-0.4.1dev/lib'
Makefile:407: recipe for target 'all-recursive' failed
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory '/home/frank/cm-0.4.1dev'
Makefile:339: recipe for target 'all' failed
make: *** [all] Error 2

However I can't resolve it with make uninstall, make clean, removal of directory and start from tar file again as the suggestion below named. I installed Pari 2.15.1 which seemed to help an earlier post with similar issues, see 1 and 2. I tried it both on a Ubuntu 18.04 and a Ubuntu 22.10 machine. Same result, same error.

Originally Posted by EdH

I just experienced the exact same error when trying to install the cm dev version on a Debian system that I had already been running the other version on. I needed to use make uninstall from the root user and make clean as the non-root user. I then removed the directory and started again from the tar file. All went well with that try.

This had nothing to do with WSL. This was a pure Debian system.

Does anyone have an idea on how to solve this? Much appreciated already!

[paulunderwood]

Originally Posted by frank1285

I installed Pari 2.15.1

Does anyone have an idea on how to solve this? Much appreciated already!

Did you run ./Configure and sudo make install for Pari? Maybe a silly question. If so, does this happen cleanly?
Have you exported LD_LiBRARY_PATH to include /usr/local/lib so that the system knows where to find the new Pari for reference to compilation?

[frank1285]

Originally Posted by paulunderwood

Did you run ./Configure and sudo make install for Pari? Maybe a silly question. If so, does this happen cleanly?
Have you exported LD_LiBRARY_PATH to include /usr/local/lib so that the system knows where to find the new Pari for reference to compilation?

Thanks for your answer Paul! For Pari I ran ./Configure and sudo make install without any (visible) problems. Everything goes well up to the point of Make of CM-0.4.1dev.

I also added /usr/local/lib to LD_LIBRARY_PATH indeed, so that should be fine as well.

I plan to run it on google cloud platform, maybe that could have something to do with it?

[paulunderwood]

Originally Posted by frank1285

Thanks for your answer Paul! For Pari I ran ./Configure and sudo make install without any (visible) problems. Everything goes well up to the point of Make of CM-0.4.1dev.

I also added /usr/local/lib to LD_LIBRARY_PATH indeed, so that should be fine as well.

I plan to run it on google cloud platform, maybe that could have something to do with it?

Did you export LD_LIBRARY_PATH before compilation of CM?

[frank1285]

Originally Posted by paulunderwood

Did you export LD_LIBRARY_PATH before compilation of CM?

Yes, I did!

[paulunderwood]

Originally Posted by frank1285

Yes, I did!

Weird. The CM compilation error clearly says Pari and TLS is at fault. I have only ever seen this when trying with an old Pari. I am stumped!

I found this https://askubuntu.com/questions/3786...6-64-linux-gnu

You could add -L/usr/local/lib to CM's Makefile at CFLAGS ??

[paulunderwood]

Originally Posted by paulunderwood

Weird. The CM compilation error clearly says Pari and TLS is at fault. I have only ever seen this when trying with an old Pari. I am stumped!

I found this https://askubuntu.com/questions/3786...6-64-linux-gnu

You could add -L/usr/local/lib to CM's Makefile at CFLAGS ??

Another thing you could try is sudo apt-get remove pari-gp libpari-dev I think those are correct. If neccessary, reinstall Pari that you downloaded and try again to compile CM.