Compiling gmp-ecm

[charybdis]

Originally Posted by ewmayer

Sorry - too much orthogonal advice flying around. Just grabbed the 7.0.4 release, still don't see the above files:

[ewmayer@localhost ~]$ tar xjf ecm-7.0.4.tar.bz2 
...

ecm-7.0.4.tar.bz2 is not the "release tarball" that WraithX is referring to. There are four links on that page labelled "source code". Below them is one labelled "release tarball". That is the one that contains configure and install-sh. It compiles successfully for me.

[ewmayer]

Originally Posted by charybdis

ecm-7.0.4.tar.bz2 is not the "release tarball" that WraithX is referring to. There are four links on that page labelled "source code". Below them is one labelled "release tarball". That is the one that contains configure and install-sh. It compiles successfully for me.

I refer to all such compressed-tarchives of e.g. my Mlucas code as "release tarballs" - but finally got the one named so under the 4 you mention, building now. Thx!

Update: Build/test/install completed successfully, but a timing-comparison for the same input and bounds shows a 10x performance hit vs 1 core of my old Haswell - Ugh. I expected "only" a 4-5x hit based on similar comparative timings for the GMP gcd on GIMPS-wavefront-sized inputs.

[kotenok2000]

I grabbed 7.0.4 release from https://members.loria.fr/PZimmermann...ecm/devel.html

[PBMcL]

I have managed to answer my own question from another thread.
Current sources are gmp-6.3.0, ecm-7.0.4, and p95v308b15.source. My systems are skylake/20.04 and Raptorlake/22.04.

0. Build GMP, if not already available. The following assumes it's installed in the /usr/local folder.
1. Build gwnum per the instructions in the p95v308b15.source/gwnum/readme.txt file. My 64-bit systems use the "make -f make64" Terminal command.
2. Switch to the ecm-7.0.4 folder and run ./configure to include gwnum. On my systems I use

./configure --with-gmp=/usr/local --with-gwnum=../p95v308b15.source/gwnum

3. (The tricky part it took me a while to figure out.) Open the (just created) makefile in the ecm-7.0.4 folder in a text editor. Search for "libgmp.a" and add the following to that line: "[gmp path]libgmp.so". On my systems the line

GMPLIB = /usr/local/lib/libgmp.a

changes to

GMPLIB = /usr/local/lib/libgmp.a /usr/local/lib/libgmp.so

Now search for "-lpthread". Where it appears on the LIBS line, add "-ldl" right after it. On my system the line becomes

LIBS = -lrt -lm -lm -lm -lm -lm ../p95v308b15.source/gwnum/gwnum.a -lpthread -ldl

Save and close the makefile. On the Terminal where you executed ./configure... for ecm-7.0.4, run 'make'.
ecm-7.04 with gwnum should now compile and link without linker errors. (We hope!)

For some reason known only to the linker gods, the libgmp.so file is only needed to resolve ONE reference, to mpz_import.

Running "make check" reveals that there's still work to do. The group order option "-go n" is broken.

[PBMcL]

Open /ecm-7.0.4/Fgw.c in a text editor.
Replace all of the lines from #346 to #368 (inclusive)

346>  mpz_t gw_x, gw_z, gw_A;
...
368>   mpz_init2 (gw_z, (gw_n+1)*gw_log_2(gw_b)+64);

with the following:

  mpz_t gw_x, gw_z, gw_A, tmp;
  int youpi;

  /* P->y must never be zero if/when calling ecm_mul */
  if( P->y->_mp_size == 0 )
    mpres_set_ui (P->y, 1, modulus);

  if (mpz_cmp_ui (go, 1) > 0)
    {
      mpres_t b;
      mpres_init (b, modulus);
      mpres_add_ui (b, P->A, 2, modulus);
      mpres_div_2exp (b, b, 2, modulus); /* b == (A+2)/4 */
      ecm_mul (P->x, P->y, go, modulus, b);
      mpres_clear (b, modulus);
    }
  
  outputf (OUTPUT_VERBOSE, 
           "Using gwnum_ecmStage1(%.0f, %d, %d, %d, %.0f, %ld)\n",
           gw_k, gw_b, gw_n, gw_c, B1, gw_B1done);

  /* reconstruct k*b^n+c to get true size */
  mpz_init_set_ui (tmp, gw_b);
  mpz_pow_ui (tmp, tmp, gw_n);
  mpz_mul_ui (tmp, tmp, (unsigned long)gw_k);
  if (gw_c >= 0)
    mpz_add_ui (tmp, tmp, gw_c);
  else
    mpz_sub_ui (tmp, tmp, (gw_c * -1));

  /* Allocate enough memory for any residue (mod k*b^n+c) for x, z */
  mpz_init2 (gw_x, 8*sizeof(mp_size_t)*(tmp->_mp_size));
  mpz_init2 (gw_z, 8*sizeof(mp_size_t)*(tmp->_mp_size));