20221013, 21:54  #34  
"Ben"
Feb 2007
E95_{16} Posts 
Quote:
Code:
time ./gnfslasieve4I15e v f 100000000 c 1000 r snfs_100000k_I15.txt gnfslasieve4I15e (with asm64,avx512 mmxtd,avx512 lasetup,avx512 lasched,avx512 sieve1,avx512 ecm): L1_BITS=15 Warning: lowering FB_bound to 99999999. FBsize 14496980+0 (deg 6), 5761454+0 (deg 1) total yield: 3292, q=100001029 (0.09862 sec/rel) ETA 0h00m) 54 Special q, 422 reduction iterations reports: 132797745>29057200>27213568>21932573>8129900>4480475 Number of relations with k rational and l algebraic primes for (k,l)=: Total yield: 3292 0/0 mpqs failures, 44981/2888 vain mpqs milliseconds total: Sieve 138570 Sched 0 medsched 34580 TD 77570 (Init 3540, MPQS 25900) SieveChange 20, lasieve_setup 73940 TD side 0: init/small/medium/large/search: 1630 10550 970 7180 9370 sieve: init/small/medium/large/search: 1480 17190 930 45940 4380 TD side 1: init/small/medium/large/search: 520 6840 1030 5140 4570 sieve: init/small/medium/large/search: 1130 20930 990 44170 1430 321.303u 4.323s 5:26.59 99.7% 0+0k 2840+752io 1pf+0w Last fiddled with by bsquared on 20221013 at 21:55 

20221014, 04:23  #35 
"Curtis"
Feb 2005
Riverside, CA
3^{3}×11×19 Posts 
I recognise that you're not offering to become the official ggnfs dev, but your tinyecm speed enhancements make ggnfs massively more interesting than it was a few months ago for cuttingedge work.
The cutting edge would benefit from the 16e siever working properly with J 16 flag, which would make it effectively 16.5e. This flag works on 15e J 15, and sometimes works as 16e J 16 but sometimes crashes. There's a small chance those crashes can be fixed, and even a chance your new code happens to remedy the code path that caused the intermittent crashing. If J 16 can be used, we in principle could factor SNFS350 with ggnfs, or GNFS235ish. Of course, we can just use CADO for extralarge sieve regions... but a new ggnfs revision holds out hope to be BOINCified to extend the life of the big nfs@home queue. 
20221014, 10:21  #36  
Just call me Henry
"David"
Sep 2007
Liverpool (GMT/BST)
2^{4}·13·29 Posts 
Quote:
Ben is rewriting much of the asm code as it is. How much more difficult would it be to create a version that would support I>16 compared with the work he is currently doing? Would we expect this to use an insane amount of memory(2x + another 2x for 17e vs 16e is my guess) or run loads slower(1.5x to 2x is my guess)? Multithreading is the other improvement that would be seriously useful especially if it doesn't increase memory useage much over a single thread(and reduces it over multiple parallel runs). Both of these ideas would probably be quite a bit of work but it appears that the ggnfs siever still has some milage vs the CADO siever. Last fiddled with by henryzz on 20221014 at 10:22 

20221014, 13:16  #37  
"Ben"
Feb 2007
3,733 Posts 
Quote:


20221014, 13:35  #38 
"Bo Chen"
Oct 2005
Wuhan,China
3×61 Posts 
The new binary seems a bit faster again, but relations sieved still less than official ggnfs.
533 version Code:
chenbo@chenbo:~/D/chenbo/my/math/cpp/yafu/yafu533/yafumaster/factor/lasieve4_64$ ./gnfslasieve4I16e v f 316000000 c 1000 o R1340L_16e_a_316000000_316001000.out a R1340L_poly.txt R gnfslasieve4I16e (with asm64,avx512 mmxtd,avx512 lasetup,avx512 lasched,avx512 sieve1): L1_BITS=15 Resuming with f 316000000 c 1000 Warning: lowering FB_bound to 315999999. FBsize 17068601+0 (deg 8), 26355865+0 (deg 1) total yield: 1016, q=316001009 (0.74686 sec/rel) ETA 0h00m) 48 Special q, 369 reduction iterations reports: 196252933>18535990>16754616>14918491>7199150>2604600 Number of relations with k rational and l algebraic primes for (k,l)=: Total yield: 1016 0/0 mpqs failures, 615/5273 vain mpqs milliseconds total: Sieve 153464 Sched 322285 medsched 342 TD 150804 (Init 2311, MPQS 47217) SieveChange 34, lasieve_setup 131883 TD side 0: init/small/medium/large/search: 2054 25496 442 18705 8146 sieve: init/small/medium/large/search: 3440 40229 355 22448 7287 TD side 1: init/small/medium/large/search: 2634 18516 440 17998 6464 sieve: init/small/medium/large/search: 1880 52551 356 23756 1162 chenbo@chenbo:~/D/chenbo/my/math/cpp/yafu/yafu533/yafumaster/factor/lasieve4_64$ Code:
chenbo@chenbo:~/D/chenbo/my/math/ggnfs/Repunit/R1340L/R1340L_3$ ./cmd_R1340L_31.sh ./gnfslasieve4I16e_bsquared2 v f 316000000 c 1000 o R1340L_16e_a_316000000_316001000.out a R1340L_poly.txt R gnfslasieve4I16e (with asm64,avx512 mmxtd,avx512 lasetup,avx512 lasched,avx512 sieve1): L1_BITS=15 Resuming with f 316000000 c 1000 Warning: lowering FB_bound to 315999999. FBsize 17068601+0 (deg 8), 26355865+0 (deg 1) total yield: 1046, q=316001009 (0.70359 sec/rel) ETA 0h00m) 48 Special q, 369 reduction iterations reports: 196252933>18535990>16754616>14918491>7199150>2604600 Number of relations with k rational and l algebraic primes for (k,l)=: Total yield: 1046 0/0 mpqs failures, 620/5273 vain mpqs milliseconds total: Sieve 383352 Sched 0 medsched 98730 TD 121986 (Init 2296, MPQS 17974) SieveChange 35, lasieve_setup 131862 TD side 0: init/small/medium/large/search: 2081 25563 445 18850 8096 sieve: init/small/medium/large/search: 3398 40614 366 136886 7334 TD side 1: init/small/medium/large/search: 2637 18572 436 18108 6545 sieve: init/small/medium/large/search: 1959 52980 368 138268 1179 chenbo@chenbo:~/D/chenbo/my/math/ggnfs/Repunit/R1340L/R1340L_3$ official 441 version (compiled by myself using Edh's method) Code:
chenbo@chenbo:~/D/chenbo/my/math/ggnfs/Repunit/R1340L/R1340L_3$ ./cmd_R1340L_31.sh ./gnfslasieve4I16e_boc_L15 v f 316000000 c 1000 o R1340L_16e_a_316000000_316001000.out a R1340L_poly.txt R gnfslasieve4I16e (with asm64): L1_BITS=15, SVN $Revision: 430 $ Resuming with f 316000000 c 1000 Warning: lowering FB_bound to 315999999. FBsize 17068601+0 (deg 8), 26355865+0 (deg 1) total yield: 1242, q=316001009 (0.92359 sec/rel) ETA 0h00m) 48 Special q, 369 reduction iterations reports: 196265020>18544215>16754559>14918439>7199123>2604605 Number of relations with k rational and l algebraic primes for (k,l)=: Total yield: 1242 3/0 mpqs failures, 59841/21024 vain mpqs milliseconds total: Sieve 533510 Sched 0 medsched 91584 TD 166418 (Init 2157, MPQS 55174) SieveChange 355597 TD side 0: init/small/medium/large/search: 5852 24449 412 17979 7413 sieve: init/small/medium/large/search: 3872 44722 347 207775 7318 TD side 1: init/small/medium/large/search: 11239 17804 415 17312 5834 sieve: init/small/medium/large/search: 2917 56320 347 208746 1147 chenbo@chenbo:~/D/chenbo/my/math/ggnfs/Repunit/R1340L/R1340L_3$ 
20221014, 13:52  #39  
"Ben"
Feb 2007
E95_{16} Posts 
Quote:
So. The real answer is that I don't know to what degree this will impact an overall factorization quite yet. It will depend on the parameters chosen. If someone can suggest an easier candidate number that uses 3LP, I can try to run some tests with a variety of parameter choices. 

20221014, 15:32  #40  
"Ben"
Feb 2007
3,733 Posts 
Quote:
Code:
time ./gnfslasieve4I16e v f 316000000 c 1000 a R1340L_poly.txt o R1340L_16e_a_316000000_316001000.out.2 gnfslasieve4I16e (with asm64,avx512 mmxtd,avx512 lasetup,avx512 lasched,avx512 sieve1,avx512 ecm): L1_BITS=15 Warning: lowering FB_bound to 315999999. FBsize 26351441+0 (deg 8), 26355865+0 (deg 1) total yield: 1142, q=316001009 (0.83914 sec/rel) ETA 0h00m) 48 Special q, 369 reduction iterations reports: 239715573>22542070>20471524>18368663>7200755>2605199 Number of relations with k rational and l algebraic primes for (k,l)=: Total yield: 1142 0/0 mpqs failures, 697/5271 vain mpqs milliseconds total: Sieve 210380 Sched 414030 medsched 1040 TD 156050 (Init 3290, MPQS 25890) SieveChange 50, lasieve_setup 176760 TD side 0: init/small/medium/large/search: 2660 31640 1260 22950 13090 sieve: init/small/medium/large/search: 4160 50670 960 35690 11850 TD side 1: init/small/medium/large/search: 3150 22390 1310 21520 5860 sieve: init/small/medium/large/search: 3900 66380 1120 33230 2420 945.404u 15.583s 16:01.43 99.9% 0+0k 2840+288io 1pf+0w Also, FYI, when using tinyecm there is no longer a 96bit limitation on mfbr/a. With AVX512_ECM the limit is 104 and otherwise it is 128. Maybe this is helpful for 3LP jobs with lpbr/a > 32. 

20221014, 15:54  #41 
"Curtis"
Feb 2005
Riverside, CA
3^{3}·11·19 Posts 

20221014, 16:17  #42  
"Ben"
Feb 2007
3,733 Posts 
Quote:
Code:
time ./gnfslasieve4I17e v f 316000000 c 500 a R1340L_poly.txt o R1340L_17e_a_316000000_316000500.out.1 gnfslasieve4I17e (with asm64,avx512 mmxtd,avx512 lasetup,avx512 lasched,avx512 sieve1,avx512 ecm): L1_BITS=16 Warning: lowering FB_bound to 315999999. FBsize 26351441+0 (deg 8), 26355865+0 (deg 1) Recurrence init: A=65536 exceeds 65535 2.490u 0.193s 0:02.92 91.7% 0+0k 2088+0io 1pf+0w 

20221015, 00:44  #43 
"Bo Chen"
Oct 2005
Wuhan,China
3·61 Posts 
1. Validity check.
For safety reason, I write a small program and have verified that for the small test relation set (R1340L, q=316M, c 1000), the avx512 ggnfs version's relations are all belong to the official ggnfs version's. 2. Other run result. Another two persons run the 533 binary under Linux, all failed. Linux on AMDcomputer: AVX512error Linux on Intelcomputer: illegal command. I am curious if without using avx512 instruction set, whether ggnfs is still faster when using ecmtiny. And if there is a method to detect the CPU not has avx512, it would be better to use native code automatically. 3. About the local build. When I build gnfslasieve4I16e, it pop some errors, I change the Makefile under lasieve_64/asm, add a line "CC=icc", then build again, the gnfslasieve4I16e could build successfully, and also could run smoothly. When building, there is a warning says a library has no static version and it run using dynamic version. 
20221015, 01:16  #44 
Apr 2020
929 Posts 
This limitation was already removed in lasieve5. Have you been modifying lasieve4 or lasieve5?

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
yafu ignoring yafu.ini  chris2be8  YAFU  9  20220217 17:52 
YAFU + GGNFS Confirmation  nivek000  YAFU  1  20211210 22:35 
Running YAFU via Aliqueit doesn't find yafu.ini  EdH  YAFU  8  20180314 17:22 
GGNFS or something better?  ZetaFlux  Factoring  1  20070807 22:40 
ggnfs  ATH  Factoring  3  20060812 22:50 