Mfakto cl_barret15_73_gs_2 - Not A Factor?
 

Mfakto cl_barret15_73_gs_2 kernel on Mfakto

Working on the very last remaining 47M exponent.

processing: TF factor 2742459209866315019473 for M47079733 (2[SUP]71[/SUP]-2[SUP]72[/SUP])
Error code: 45, error text: 2742459209866315019473 is not a factor of M47079733

Retried on two different Fury Xs in different machines, both indicated this factor which primenet rejects.

Wolfram alpha says the remainder is 2742459209866315019471 (2 off from even)

Re-ran this on mfaktc on one GPU of a Titan Z:

no factor for M47079733 from 2^71 to 2^72 [mfaktc 0.21 barrett76_mul32_gs]

I saw this error only when I had a PSU going crazy. I assume that mfakto would behave similarly to mfaktc in this regard. The "normal" response of mfaktc, for me, if the card is too OC'd or the voltage is too low, is to crash. Not a Factor has been a very rare error for me.

EDIT: I am ignorant of CL Lucas, except for what I read here, so the causes might be different from CUDA.

I TF'ed it on a GTX 760, for giggles:

no factor for M47079733 from 2^71 to 2^72 [mfaktc 0.21 barrett76_mul32_gs]

I'll run it on my HD7950

[CODE](19:03) gp > (1<<47079733+1)/2742459209866315019473
%68 = 24138552060634017262794635021044514699753531753848
(19:03) gp > floor(%)===%
%69 = 1[/CODE]

edit:the full result is cut off. I also tested it earlier through PARI and the way TF is done on mersennes.

[CODE](13:27) gp > a=binary(47079733);f=2742459209866315019473;b=Mod(1,f);for(x=1,#a,b=b^2;if(a[x]==1,b*=2));b-1
%57 = Mod(2742459209866315019471, 2742459209866315019473)[/CODE]

We might be on to a bug in mfakto, my HD7950 also reports the factor:
AMD Catalyst 15.7
Windows 7 64bit

[code]
Runtime options
Inifile mfakto.ini
Verbosity 1
SieveOnGPU yes
MoreClasses yes
GPUSievePrimes 82486
GPUSieveProcessSize 24Ki bits
GPUSieveSize 96Mi bits
FlushInterval 8
WorkFile worktodo.txt
ResultsFile results.txt
Checkpoints enabled
CheckpointDelay 300s
Stages enabled
StopAfterFactor bitlevel
PrintMode compact
V5UserID VictordeHollander
ComputerID PCVICTOR
TimeStampInResults yes
VectorSize 2
GPUType GCN
SmallExp no
UseBinfile mfakto_Kernels.elf
Compiletime options
Select device - Get device info - Loading binary kernel file mfakto_Kernels.elf
Compiling kernels.

OpenCL device info
name Tahiti (Advanced Micro Devices, Inc.)
device (driver) version OpenCL 1.2 AMD-APP (1800.8) (1800.8 (VM))
maximum threads per block 256
maximum threads per grid 16777216
number of multiprocessors 28 (1792 compute elements)
clock rate 900MHz

Automatic parameters
threads per grid 256
optimizing kernels for GCN

Started a simple selftest ...
Selftest statistics
number of tests 108
successful tests 108

selftest PASSED![/code][code]M47079733 has a factor: 2742459209866315019473 [TF:71:72:mfakto 0.14-Win cl_barrett15_73_gs_2][/code]


((2^47079733-1) mod 2742459209866315019473
= 2742459209866315019471
That is indeed 2 off as people earlier mentioned. Could it be a precision problem with the barrett15_73 ?

[QUOTE=VictordeHolland;419465]We might be on to a bug in mfakto, my HD7950 also reports the factor:
AMD Catalyst 15.7
Windows 7 64bit

[code]
Runtime options
Inifile mfakto.ini
Verbosity 1
SieveOnGPU yes
MoreClasses yes
GPUSievePrimes 82486
GPUSieveProcessSize 24Ki bits
GPUSieveSize 96Mi bits
FlushInterval 8
WorkFile worktodo.txt
ResultsFile results.txt
Checkpoints enabled
CheckpointDelay 300s
Stages enabled
StopAfterFactor bitlevel
PrintMode compact
V5UserID VictordeHollander
ComputerID PCVICTOR
TimeStampInResults yes
VectorSize 2
GPUType GCN
SmallExp no
UseBinfile mfakto_Kernels.elf
Compiletime options
Select device - Get device info - Loading binary kernel file mfakto_Kernels.elf
Compiling kernels.

OpenCL device info
name Tahiti (Advanced Micro Devices, Inc.)
device (driver) version OpenCL 1.2 AMD-APP (1800.8) (1800.8 (VM))
maximum threads per block 256
maximum threads per grid 16777216
number of multiprocessors 28 (1792 compute elements)
clock rate 900MHz

Automatic parameters
threads per grid 256
optimizing kernels for GCN

Started a simple selftest ...
Selftest statistics
number of tests 108
successful tests 108

selftest PASSED![/code][code]M47079733 has a factor: 2742459209866315019473 [TF:71:72:mfakto 0.14-Win cl_barrett15_73_gs_2][/code][/QUOTE]

Oh dear, if that's true.

But doesn't Primenet test the factors reported when you check them in?

[QUOTE=Gordon;419467]Oh dear, if that's true.

But doesn't Primenet test the factors reported when you check them in?[/QUOTE]

Yes. It does. I was immediately informed that these factors were spurious when I checked them in. The form they reported in was not typical, either, and lines may have been repeated. I might even have the results.txt for them.....

As I said, in this case, my PSU had pretty much lost regulation. I did a lot of adjusting before :gah:, it sunk in that Something was Seriously Wrong. Swapping out the PSU fixed things.

Well sir, you may have run in a very rare bug... My HD7970 finds the fake factor too.

[CODE]mfakto 0.15pre2-Win (64bit build)

Runtime options
Inifile mfakto.ini
Verbosity 2
SieveOnGPU yes
MoreClasses yes
GPUSievePrimes 120000
GPUSieveProcessSize 16Ki bits
GPUSieveSize 64Mi bits
FlushInterval 0
WorkFile worktodo.txt
ResultsFile results.txt
Checkpoints enabled
CheckpointDelay 1800s
Stages enabled
StopAfterFactor class
PrintMode compact
V5UserID none
ComputerID none
ProgressHeader "Date Time | class Pct | time ETA | GHz-d/day Sieve Wait"
ProgressFormat "%d %T | %C %p%% | %t %e | %g %s %W%%"
TimeStampInResults yes
VectorSize 2
GPUType AUTO
SmallExp yes
UseBinfile mfakto_Kernels.elf
Compiletime options

Select device - Get device info:
Device 1/1: Tahiti (Advanced Micro Devices, Inc.),
device version: OpenCL 1.2 AMD-APP (1084.4), driver version: 1084.4 (VM)
Extensions: cl_khr_fp64 cl_amd_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_e
xtended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr
_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addre
ssable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_
amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_media_ops2 cl_amd_popcnt cl_khr_d3d10_shari
ng cl_amd_c1x_atomics
Global memory:2147483648, Global memory cache: 16384, local memory: 32768, workgroup size:
256, Work dimensions: 3[256, 256, 256, 0, 0] , Max clock speed:1050, compute units:32

OpenCL device info
name Tahiti (Advanced Micro Devices, Inc.)
device (driver) version OpenCL 1.2 AMD-APP (1084.4) (1084.4 (VM))
maximum threads per block 256
maximum threads per grid 16777216
number of multiprocessors 32 (2048 compute elements)
clock rate 1050MHz

Automatic parameters
threads per grid 256
optimizing kernels for GCN

Loading binary kernel file mfakto_Kernels.elf
Compiling kernels (build options: "-I. -DVECTOR_SIZE=2 -DGCN -O3 -DMORE_CLASSES -DCL_GPU_S
IEVE -DSMALL_EXP").
Started a simple selftest ...
Selftest statistics
number of tests 130
successful tests 130

selftest PASSED!

got assignment: exp=47079733 bit_min=71 bit_max=72 (10.16 GHz-days)
Starting trial factoring M47079733 from 2^71 to 2^72 (10.16GHz-days)
k_min = 25076429823060 - k_max = 50152859648435
Using GPU kernel "cl_barrett15_73_gs_2"
No checkpoint file "M47079733.ckp" found.
Date Time | class Pct | time ETA | GHz-d/day Sieve Wait
Dec 17 19:21 | 4452 96.7% | 2.215 1m11s | 412.76 120373 0.00%
M47079733 has a factor: 2742459209866315019473

found 1 factor for M47079733 from 2^71 to 2^72 (partially tested) [mfakto 0.15pre2-Win cl_barrett15_73_gs_2]
tf(): total time spent: 33m 54.032s (431.50 GHz-days / day)
[/CODE]OTOH, false positive is not big deal, they are verified and hyper-verified. A false negative would be much worse, because that would mean three times wasting the time and the resources: once to do effectively the TF which missed the factor, twice to do the LL, because if the factor was missed, you would need to LL it, and last, third time, to DC it.

So, Bdot, are you there? Are we losing factors?

[QUOTE=LaurV;419510]Are we losing factors?[/QUOTE]

We could look at the success rates for mfakto and see if it's lower than mfaktc.

Also, if the bug is found we should really spend time figuring out how likely it is that other factors were missed, and which they might be -- it might be worth spending time redoing certain classes of work.

I'm away right now, but someone could try with the mfakto.ini set to use a different GCN optimization / kernel.

I now believe I have had this happen one other time but at the time did not think much of it. I can try to find that entry in my old logs.

The positive side is it found a factor of 2^p+1, which I believe I read somewhere in the factoring math that the algorithm could find those factors but the were trivial to check for. Perhaps that's what science man was saying?

I have the kernel source open, might try stepping through it at that k value to look for rounding errors/loss of precision.