CUDALucas (a.k.a. MaclucasFFTW/CUDA 2.3/CUFFTW)

[chalsall]

Originally Posted by Dubslow

The signal() code is identical to previous versions of CUDALucas except for the string printed.

You may want to do some research on the sigaction() function (although there is no direct Windows equivalent). signal() is a portability nightmare.

[Dubslow]

Originally Posted by chalsall

You may want to do some research on the sigaction() function (although there is no direct Windows equivalent). signal() is a portability nightmare.

:shrug: It's what msft has been using, and no one's complained so far. (I'm guessing it's been in there a long time, since at least 1.2 in all likelihood.)

[chalsall]

Originally Posted by Dubslow

:shrug: It's what msft has been using, and no one's complained so far. (I'm guessing it's been in there a long time, since at least 1.2 in all likelihood.)

:shrug: Just because it's being used doesn't necessarily mean it's correct in every environment. Once a month bug anyone?

Quoting the man page for signal(2):

The only portable use of signal() is to set a signal’s disposition to SIG_DFL or SIG_IGN. The semantics when using signal() to establish a signal handler vary across systems (and POSIX.1 explicitly permits this variation); do not use it for this purpose.

[Redarm]

Originally Posted by Dubslow

I didn't change a thing math-wise. Perhaps you chose a better FFT length? Edit: Could be new drivers, the 680 is still rather new. Or, if previously you were using arch>=2.0 but are now using arch=1.3, you might see some performance gains.

PS: Are you in Windows, and if so, does it crash for you like above?

i used the compiled versions for windows by flashjh, without the worktodo-file

edit: with worktodo.txt it will crash

[Dubslow]

Originally Posted by Redarm

i used the compiled versions for windows by flashjh, without the worktodo-file

edit: with worktodo.txt it will crash

Okay, well if you read the intermediate posts, that bug has been fixed and now we're just waiting for flash to recompile.

[LaurV]

Found my issue. In v2.02 you have:

enum ASSIGNMENT_ERRORS clear_assignment(char *filename, int exponent)
{
  <snip>
  
#ifdef linux  
  f_in = fopen(filename, "r");
  if (NULL == f_in)
    return CANT_OPEN_WORKFILE;
  
  f_out = fopen("__worktodo__.tmp", "w");
  if (NULL == f_out)
  {
    fclose(f_in);
    return CANT_OPEN_TEMPFILE;
  }
#else
  errno_t err;
  err = fopen_s(&f_in, filename, "r");
  if (err)
    return CANT_OPEN_WORKFILE;
  
  err=0; err = fopen_s(&f_out, "__worktodo.tmp", "w");
  if (err)
  {
    fclose(f_in);
    return CANT_OPEN_TEMPFILE;
  }
#endif

This seems to be fixed in v2.03, I saw you made a new "_fopen()" and moved the OS-specific inside. Waiting for binaries of v2.03 (don't want to complicate the things even more by attempting a compilation by myself, I let flash do it, he has more experience :P). Meantime 2.02 produced first correct residues, two DCs (all switches in command line, no ini used).

[Dubslow]

Originally Posted by LaurV

Found my issue. In v2.02 you have:

enum ASSIGNMENT_ERRORS clear_assignment(char *filename, int exponent)
{
  <snip>
  
#ifdef linux  
  f_in = fopen(filename, "r");
  if (NULL == f_in)
    return CANT_OPEN_WORKFILE;
  
  f_out = fopen("__worktodo__.tmp", "w");
  if (NULL == f_out)
  {
    fclose(f_in);
    return CANT_OPEN_TEMPFILE;
  }
#else
  errno_t err;
  err = fopen_s(&f_in, filename, "r");
  if (err)
    return CANT_OPEN_WORKFILE;
  
  err=0; err = fopen_s(&f_out, "__worktodo.tmp", "w");
  if (err)
  {
    fclose(f_in);
    return CANT_OPEN_TEMPFILE;
  }
#endif

This seems to be fixed in v2.03, I saw you made a new "_fopen()" and moved the OS-specific inside. Waiting for binaries of v2.03 (don't want to complicate the things even more by attempting a compilation by myself, I let flash do it, he has more experience :P). Meantime 2.02 produced first correct residues, two DCs (all switches in command line, no ini used).

Ah. In that case, it was probably just a transcription error on flash's part when he inserted the fopen_s. When I redefined all that crap under _fopen, I deleted all the Windows #ifdefs and used the Linux ones, and apparently I deleted the error as well.

By the way throughout this mess I think I've had 3 for 3 with my hacks.

By the other way, was I correct about you wanting print_time_from_seconds to always pad the ETA with zeros to a constant length?

PS Try hitting the ^C yourself.
___________________________________________________________________

PPS

Originally Posted by LaurV

No idea what that means till I will see it in action
The "err=" and "ms/iter" is what I was thinking of. The ETA looks good as it is in v2.02. If you pad it with zeroes or better spaces up to 3:2:2 digits (in case is shorter) it would be perfect. But this is just nitpicking...

Originally Posted by LaurV

edit2: next cosmetic thing would be to use a fixed format (6, 5, even 4 decimals would be enough) for error display, and 4 decimals for ETA.

The ms/iter was already at a steady field width; like I said in my original response, err is also now at a fixed field width. ETA would look like
15 hrs: "15:42:24"
5 hrs: "05:42:24"
0 hrs 42 mns: "00:42:24"
7 mins "00:07:24"
24 seconds "00:00:24"
4 seconds "00:00:04"

I personally think 3:2:2 is a bit excessive, since the vast majority of users do DCs.
___________________________________________________________________

PPPS

Originally Posted by LaurV

Remember I am still running v2.02 :P
Most probably it will say "caught. Bla Bla" without "^C". I don't want to interrupt the work right now only for that.

edit: could not resist... I was right. Maybe you have done some changes in v2.03, but they didn't reach me yet...

Hmm... it should be "^C caught. ..." because when you hit ^C yourself, it appears on the terminal, then " caught..." makes it look like one print. (And no, I haven't changed that part since I initially made 2.00a.)
___________________________________________________________________

PPPPS

Originally Posted by LaurV

ETA is already 2:2:2, so if you think 3:2:2 is too much, then there is nothing to do here. For me it looks ok either way.

Well if it gets below an hour, the hour field isn't printed at all and we get a 2:2, not a 2:2:2. That's what I was wondering about.

//From apsen
void
print_time_from_seconds (int sec)
{
  if (sec > 3600)
    {
      printf ("%d", sec / 3600);
      sec %= 3600;
      printf (":%02d", sec / 60);
    }
  else
    printf ("%d", sec / 60);
  sec %= 60;
  printf (":%02d", sec);
}

Reattaching 2.03 from previous page.

[LaurV]

Originally Posted by Dubslow

By the other way, was I correct about you wanting print_time_from_seconds to always pad the ETA with zeros to a constant length?

No idea what that means till I will see it in action
The "err=" and "ms/iter" is what I was thinking of. The ETA looks good as it is in v2.02. If you pad it with zeroes or better spaces up to 3:2:2 digits (in case is shorter) it would be perfect. But this is just nitpicking...

Thanks for the PM, my post count is inflated too, but who cares :D, I saw your reply, I wrongly wrote ETA there. I was thinking to ms/iter with 4 decimals, fixed. ETA is already 2:2:2, so if you think 3:2:2 is too much, then there is nothing to do here. For me it looks ok either way.

[LaurV]

Originally Posted by Dubslow

PS Try hitting the ^C yourself.

Remember I am still running v2.02 :P
Most probably it will say "caught. Bla Bla" without "^C". I don't want to interrupt the work right now only for that.

edit: could not resist... I was right. Maybe you have done some changes in v2.03, but they didn't reach me yet...

[flashjh]

Attached CUDALucas 2.03 x64 binaries - Tested


- I was able to run with or without .ini file
- Worktodo.txt works fine.
- Command line still takes precedence
- Test=XXXXXXXX in worktodo.txt works fine
- Test=N/A,XXXXXXXX,XX,X works fine
- Test=AID,XXXXXXXX,XX,X works fine
- I tested DoubleCheck with all these also, they work fine

@Dubslow - everything compiled straight out, but the makefile.win needed a small change OUT = NAME -> OUT = $(NAME) on line 6 and I still need /Tp for now

This is CUDA 4.0 | sm20 & sm_21 (see next posts for CUDA 3.2 & source)

Edit: With .ini and worktodo.txt in directory, I can run just CUDALucas.exe and it works fine.

BTW - Thanks for all the hard work everyone!

Edit2:

I just edited my worktodo.txt with M86243 and restarted with CUDALucas.exe:

 
M( 86243 )P, n = 4608, CUDALucas v2.03
Continuing work from a partial result of M26105XXX fft length = 1572864 iteration = 13559297

It successfully found the Prime, cleared the exponent from worktodo.txt and continued with my current exponent - awesome!

[flashjh]

CUDALucas 2.03 x64 CUDA 3.2 | sm_13