Manual Testing LL result syntax (where to find documentation)

[preda]

I'm working on an OpenCL LL checker, and I'd need some information about the syntax of the result lines for manual testing.

I looked at the syntax produced by CUDALucas, which looks like this:
M( 68365469 )C, 0x36d76bdb156e76__, offset = 12047241, n = 4096K, CUDALucas v2.05.1, AID: 002E82D09E92AB807F86C3E7106CB418

I tried to copy it, replacing the "program name/version" (i.e. CUDALucas vx.x.x) with a different name, but the line was not accepted. I ended up submitting a "fake CUDALucas" result (which was accepted).

Is there a process for registering a new program name/version?

Thank you for pointing me to this information.

[ATH]

You better hide the last 2 characters in the residue for exponents that have not been successfully double checked.

[preda]

Until I find a solution (how to register a new program name), I'm probably going to submit as "CUDALucas v0.0.314" and with offset=0. This allows some degree of identification I assume, though it's just a temporary hack.

[Mark Rose]

I would strongly suggest using a unique program name. It's not too difficult to add the detection code for your program, but they need to know what its output looks like to add it.

When you have what your output looks like, let James know and he can fix the parsing.

[VictordeHolland]

Originally Posted by Mark Rose

I would strongly suggest using a unique program name.

+1 to this.

Don't get me wrong, I think it's great you made another OpenCL testing program. But it would be better if it gets a distinct program name/tag in the database to identify results submitted with it. If there is a bugs with Cllucas or your program they could be easier tracked. Does your program use a shift?

Out of curiosity, did you recheck the Mersenne Primes with your program? And if so, do you get the right residues (00000000....) ?

[preda]

The result line of my program is exactly the same as CUDALucas save for the program name (because I see no reason to produce a different result). The program right now doesn't use offset, thus offset=0 always in the result.

The program name is "gpuowl", followed by a two-segment version string e.g. "gpuowl v0.1".

M( 77000201 )C, 0x5561a29ad401481c, offset = 0, n = 4096K, gpuowl v0.1, AID: FAF7794A597507B167BE68936175A48E

For a prime, I expect the result should be:
M( 77000000 )P, 0x0000000000000000, offset = 0, n = 4096K, gpuowl v0.1, AID: xxx

The program did check correctly M37,156,667 (to 0 residue, 0 all), and did double-check correctly a few 77M candidates, so it's working at least to some degree.

[GP2]

I wonder if this OpenCL program might eventually be adapted to run on an FPGA such as the Xilinx UltraScale+.

Amazon is in the process of rolling out FPGA instances on AWS cloud, although this is still in preview.

[Mark Rose]

Then it's time to PM James.

[preda]

All done already, James was very quick.

[preda]

I don't know, I didn't have access to FPGA. But if those FPGA have some OpenCL implementation, adapting shouldn't be too hard... Otherwise (no OpenCL) it may be more work.

[preda]

Although, it my opinion, GPU are more cost-effective than FPGA for the same computation power.