Source?
 

Where can I find the source for mfakto?

[QUOTE=bayanne;476518]Where can I find the source for mfakto?[/QUOTE]

[url]http://www.mersenneforum.org/mfakto/[/url]

[QUOTE=chalsall;476554][URL]http://www.mersenneforum.org/mfakto/[/URL][/QUOTE]
[url]https://github.com/Bdot42/mfakto[/url] also has the latest changes ...

It seems that running AMD APP SDK 2.5 or above is a prerequisite ... which is Windows or Linux

Well, mfakto only uses the OpenCL APIs (if I remember correctly). If Apple provides some kind of OpenCL API in one of their Devkits, then this should also satisfy the missing symbols.

Unable to progress with mfakto on AMD 7000 series
 

Hi, I'm not sure if this is the best place for this question but I seem to be stuck trying to get mfakto to run on my old Dell machine with an AMD Radeon HD 7000 series card (exact model unknown since it's... long forgotten, and the Catalyst Center says just that).

When I double-click or run mfakto-x64.exe or mfakto-x86.exe using default settings in mfakto.ini, it runs some self-tests and then stops at this:

(Verbosity set to 4 to try to debug)
[CODE]
Selftest statistics
number of tests 108
successful tests 108

selftest PASSED!

got assignment: exp=159457759 bit_min=71 bit_max=72 (3.00 GHz-days)
Starting trial factoring M159457759 from 2^71 to 2^72 (3.00GHz-days)
k_min = 7403789112720 - k_max = 14807578234150
Using GPU kernel "cl_barrett32_77_gs_4"
No checkpoint file "M159457759.ckp" found.[/CODE]

mfakto version is "mfakto-0.14-win"
OS is Windows 10 v1703
CPU is Intel i7-3770

Info from AMD Catalyst Control Center:
Catalyst version: 15.7.1
Driver Packaging version: 15.20.1062.1004
OpenGL version: 6.14.10.13399

Grabbed from mfakto-x64.exe output on the OpenGL part:
[CODE]OpenCL device info
name Caicos (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 2 (160 compute elements)
clock rate 775MHz

Automatic parameters
threads per grid 256
optimizing kernels for VLIW5[/CODE]

I have tried different TF assignments, all get stuck at the same place. I monitored the GPU activity using Process Explorer, which showed no activity when the TF begins after the self tests ended when running either the x86 or x64 version of the .exe.

Any advice?

I'm running Prime95 on the same PC and it's running well. mfakto is stuck at the clause shown above regardless of Prime95 activity (or lack thereof). -st and -st2 turned up no errors (GPU activity registered in proc explorer during st).

You could always try with the latest version ([url=https://download.mersenne.ca/mfakto/mfakto-0.15pre6]v0.15pre6[/url]) and see if the problem has already been fixed.

[QUOTE=James Heinrich;477169]You could always try with the latest version ([url=https://download.mersenne.ca/mfakto/mfakto-0.15pre6]v0.15pre6[/url]) and see if the problem has already been fixed.[/QUOTE]

Thanks for the suggestion, James!

I tried 0.15pre6, but the result is still the same.

However, I discovered that when I switch off GPU sieve (SieveOnGPU=0) for both 0.14 and 0.15pre6, mfakto will run.

It appears that sieving with GPU would stall the computation on my machine/card. Specifying GPUType and other parameters in mfakto.ini doesn't seem to help in getting past that stalling.

In the spirit of experimentation, knowing that 0.15pre6 supports Intel integrated GPUs (this machine has both Intel and AMD GPUs), I tried running 0.15pre6 using the Intel platform (-d 21) with SieveOnGPU=1, and the computation would run without issue (albeit I have to specify the GPUType=Intel in order to run on the Intel chip).

Could this be a bug somewhere? Or I just didn't set the right settings?

I've been getting a similar problem when running the beta version on my Radeon R7 M460, except mfakto will run for a while before getting stuck. My computer also has the additional problem of crashing with the [c]THREAD_STUCK_IN_DEVICE_DRIVER[/c] error.

[QUOTE=Stampeder;477227]
However, I discovered that when I switch off GPU sieve (SieveOnGPU=0) for both 0.14 and 0.15pre6, mfakto will run.

...

Could this be a bug somewhere? Or I just didn't set the right settings?[/QUOTE]

Hi Stampeder,

your Caicos GPU is still a VLIW5 architecture, which is pretty weak at GPU sieving anyway. With just 2 compute cores, the expected performance would also be rather low. But the real problem is most likely insufficient local memory in the GPU. Probably mfakto does not calculate the required memory correctly and tries to use more than what is available. The AMD driver would just hang up in such cases (instead of throwing an error).

You can try to reduce the GPUSieveSize , GPUSieveProcessSize and GPUSievePrimes values towards their minimum ... Maybe this allows mfakto to run.

But as mentioned, for this GPU I'd recommend CPU-Sieving anyway.

Cheers,
Bdot

[QUOTE=ixfd64;477240]I've been getting a similar problem when running the beta version on my Radeon R7 M460, except mfakto will run for a while before getting stuck. My computer also has the additional problem of crashing with the [c]THREAD_STUCK_IN_DEVICE_DRIVER[/c] error.[/QUOTE]

This could indicate that the single kernel invocations take longer than some timeout that the OS is willing to wait until it can use the GPU for showing the screen. Try lowering the GPUSieve* parameters as that will shorten the time the GPU is blocked.