llrCUDA

[msft]

Original llrpi380devsrc.zip Bug on 64bitLinux.
32bitLinux:

llrpi380devsrc/linuxllr$ ./llrpi -d  -q"1000065*2^220897-1"
Starting Lucas Lehmer Riesel prime test of 1000065*2^220897-1
Using zero-padded rational base DWT, FFT length = 32768
V1 = 15 ; Computing U0...done.
...
1000065*2^220897-1 is prime!  Time : 2812.600 sec.

64bitLinux:

llrpi380devsrc/linuxllr$ gdb llrpi
GNU gdb 6.8-debian
Copyright (C) 2008 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu"...
(gdb)  r -d -q"1000065*2^220897-1"
Starting program: /home/msft/llr.cuda/llrpi380devsrc/linuxllr/llrpi -d -q"100006
5*2^220897-1"
Starting Lucas Lehmer Riesel prime test of 1000065*2^220897-1
Using zero-padded rational base DWT, FFT length = 32768
V1 = 15 ; Computing U0...
Program received signal SIGSEGV, Segmentation fault.
0x0000000000444762 in modred (x=0x7f0fec466010) at gwpnumi.c:613
613             *pscr++ = carry;                                        // Zero
12 double words in scratch area
(gdb) where
#0  0x0000000000444762 in modred (x=0x7f0fec466010) at gwpnumi.c:613
#1  0x0000000000445206 in lucas_square (x=0x7f0fec466010, N=32768,
    error_log=1, noadd=0, nomul=0) at gwpnumi.c:794
#2  0x0000000000448148 in gwpsquare (s=0x7f0fec466010) at gwpnumi.c:1582
#3  0x0000000000420da0 in isLLRP (format=18, sgk=0x7fffaab33c80 "1000065",
    b_else=1, n=220897, binput=2, ninput=220897, shift=0, res=0x7fffaab24ef0)
    at Llr.c:6935
#4  0x000000000042b205 in process_num (format=18,
    sgk=0x7fffaab33c80 "1000065", base=2, n=220897, incr=-1, shift=0,
    res=0x7fffaab24ef0) at Llr.c:9540
#5  0x000000000042e926 in primeContinue () at Llr.c:10445
#6  0x0000000000431054 in linuxContinue (
    error_message=0x4595b8 "Another llrpi is already running!\n")
    at lprime.c:638
#7  0x0000000000430ccb in main (argc=3, argv=0x7fffaab39728) at lprime.c:465
(gdb)

llrpi380devslinux.zip is correct result on 64bitLinux.

[msft]

Only Support k*2^n-1(small k).

$ time ./llrCUDA -d -q3*2^2312734-1
Starting Lucas Lehmer Riesel prime test of 3*2^2312734-1
Using real irrational base DWT, FFT length = 262144
V1 = 9 ; Computing U0...done.

3*2^2312734-1 is prime! Time : 3290.047 sec.

real 54m53.325s
user 32m47.340s
sys 19m33.530s

[mdettweiler]

Originally Posted by msft

Only Support k*2^n-1(small k).

$ time ./llrCUDA -d -q3*2^2312734-1
Starting Lucas Lehmer Riesel prime test of 3*2^2312734-1
Using real irrational base DWT, FFT length = 262144
V1 = 9 ; Computing U0...done.

3*2^2312734-1 is prime! Time : 3290.047 sec.

real 54m53.325s
user 32m47.340s
sys 19m33.530s

Just curious, how small is "small k" in this context? <1000? <100000?

[frmky]

So at this point, version 0.21 does Proth tests and 0.29 does LLR tests?

Timing on the GTX 480:

[cluster@node01 llrcuda.0.29]$ time ./llrCUDA -d -q3*2^2312734-1
Starting Lucas Lehmer Riesel prime test of 3*2^2312734-1
Using real irrational base DWT, FFT length = 262144
V1 = 9 ; Computing U0...done.

3*2^2312734-1 is prime! Time : 1932.870 sec.

real 32m16.940s
user 22m54.242s
sys 9m22.607s

The reported time per iteration was 0.833 ms.

[msft]

Originally Posted by msft

Original llrpi380devsrc.zip Bug on 64bitLinux.
64bitLinux:

llrpi380devsrc/linuxllr$ gdb llrpi
(gdb)  r -d -q"1000065*2^220897-1"
Starting program: /home/msft/llr.cuda/llrpi380devsrc/linuxllr/llrpi -d -q"100006
5*2^220897-1"
Starting Lucas Lehmer Riesel prime test of 1000065*2^220897-1
Using zero-padded rational base DWT, FFT length = 32768
V1 = 15 ; Computing U0...
Program received signal SIGSEGV, Segmentation fault.
0x0000000000444762 in modred (x=0x7f0fec466010) at gwpnumi.c:613
613             *pscr++ = carry;                                        // Zero
12 double words in scratch area

Fix it.

gwpnumi.cu:             //msft scral = (double *) ((temp + 7) & 0xFFFFFFF8);    // id. double word aligned
gwpnumi.cu:             scral = (double *) ((temp + 7) & 0xFFFFFFFFFFFFFFF8);   // id. double word aligned

[msft]

Hi ,mdettweiler

Originally Posted by mdettweiler

Just curious, how small is "small k" in this context? <1000? <100000?

Fix it.
Support k*2^n-1.

1000065*2^220897-1 is prime! Time : 3059.684 sec.
3*2^164987-1 is prime! Time : 42.702 sec.

[msft]

Hi ,frmky

Originally Posted by frmky

So at this point, version 0.21 does Proth tests and 0.29 does LLR tests?

Currently.

[msft]

Add Proth test.

[nuggetprime]

Just courious,will there be a version for GPUs with only compute capability 1.2(no double precision) too?

[msft]

Hi ,nuggetprime

Originally Posted by nuggetprime

Just courious,will there be a version for GPUs with only compute capability 1.2(no double precision) too?

Need DP.

[mdettweiler]

Originally Posted by msft

Hi ,mdettweiler

Fix it.
Support k*2^n-1.

1000065*2^220897-1 is prime! Time : 3059.684 sec.
3*2^164987-1 is prime! Time : 42.702 sec.

Cool! That easily covers the k-ranges tested by many of the larger projects (NPLB, RPS, PrimeGrid, even TPS to some degree).

Meanwhile: Gary's GPU is still not functioning yet, but we're working on it. (Yes, I've been saying that for a while. ) We finally decided to give up on Ubuntu 10.04--I think I must have done something to royally mess it up early on with regard to the GPU drivers. We're going to go back to 8.04 which we know worked earlier with the GPU.