Msieve v1.48 feedback

[chris2be8]

msieve normally says "elapsed time hh:mm:ss" when it finishes. But I didn't get it after it ended with "too few cycles, matrix probably cannot build" or "matrix must have more columns than rows" or "matrix needs more columns than rows; try adding 2-3Invalid argumentore relations" messages. Admittedly the runs were not all the latest version of msieve.

Chris K

[jasonp]

This is a symptom of a longstanding problem with the library: on an error, it quits immediately. It would be much better to stop running the current job, set a failure status and return normally but that requires all sorts of contortions to make the control flow uniform.

[jasonp]

I've committed changes that now make the library report the SVN rev set at build time. This should hopefully avoid bumping up the version number to indicate that a given build is experimental.

I have also, for the first time in years, actually updated the readme files. Give them a look and let me know if I missed anything (the NFS postprocessing chapter is getting written now). From now on, the readme files will get bundled with all the binary downloads.

[debrouxl]

I've just noticed that for the "-D" option to work, the following patch needs to be applied:

Index: demo.c
===================================================================
--- demo.c      (révision 579)
+++ demo.c      (copie de travail)
@@ -331,7 +331,7 @@
        buf[0] = 0;
        while (i < argc) {
                if (argv[i][0] == (char)('-')) {
-                       switch(tolower(argv[i][1])) {
+                       switch(argv[i][1]) {
                        case 'h':
                        case '?':
                                print_usage(argv[0]);

Without that change, "-D" is interpreted as "-d" (deadline)

Linear algebra on the 3.4M matrix produced by filtering at target density 70, despite the -D 90 parameter, on the significantly oversieved task (SNFS 211, 66.3M raw relations, 56.1M unique) will complete in ~22h with 3 threads on the Athlon II X4 640. ~1h of wall clock time has already been spent on LA, so it doesn't make much sense to abort it, re-run filtering and run LA on the smaller matrix...
After getting the factors tomorrow, I may run filtering at target density 90.

And I think that in the help text, after the addition of -D, the double newline has become misplaced:

@@ -111,9 +111,9 @@
 #ifdef HAVE_CUDA
                 "   -g <num>  use GPU <num>, 0 <= num < (# graphics cards)>\n"
 #endif
-                "   -t <num>  use at most <num> threads\n\n"
+                "   -t <num>  use at most <num> threads\n"
                 "   -D <num>  make filtering target <num> nonzeros per\n"
-                "             matrix column (default value if 0)\n"
+                "             matrix column (default value if 0)\n\n"
                 " elliptic curve options:\n"
                 "   -e        perform 'deep' ECM, seek factors > 15 digits\n\n"
                 " quadratic sieve options:\n"

[jasonp]

Doh! Will fix...