GpuOwl 7.x

[kriesel]

I don't know if this is a bug or feature. v7.2-53-ge27846f help output says (and nothing in my config.txt or tiny batch file says otherwise):

-save <N>          : specify the number of savefiles to keep (default 12).

Yet I have 20 and 23 .prp save files accumulated in each exponents' folders, dated after I began using v7.2-53. This is on Windows. Maybe the occasional implicit-delete-by-rename-atop-is-prevented-by-Windows-design bug surfaced again?

[kriesel]

What's next? (Though Mihai has certainly earned a break and our thanks!)

Originally Posted by kriesel

PRP Proof capability was correctly top priority
reduced P-1 net cost.
LL proof per R Gerbicz' proposed method might be good someday. such as for PROVING LL tests were correct for the next Mp found, or the previous ones.

Among the other things gpuowl does not currently support that may interest some potential users are:
TF
ECM
shifts
PRP-CF
prp certs accepted by PrimeNet
Longer fft lengths (P-1 for OBD and maybe large enough for F33)
Proth numbers
Fermat numbers (Ernst may tackle this someday)
Wagstaff numbers
General repdigit numbers
APIs other than OpenCL (OpenCL 1.1/1.2 on old gpus)
OS other than Windows or Linux
P-1 bounds extension(V7.x allows P-1 B2 extension during the PRP run)
Substantial P-1 error detection/correction
UI or control flow changes (continue on to next worktodo item instead of quit on serious or repeated error)
v7.0 is dropping LL support. Some will want it back, preferably with Jacobi and shift (for DC or prime confirmations).

Other wishlist items:
Persistent save files every 10M or 20M iterations, for salvaging part of long runs that go wrong along the way despite the Jacobi check or whatever.
Perhaps occasional light maintenance of the V6.11 branch, for things V7 does not include.
Documentation in detail.
Return of supersafe for large exponent testing / large new prime verification.

Or a variant that consults a prepared list of interim iteration / residue pairs from a previous run, and retries a limited number of times from the last matching point.
Very simply formatted, file named <exponent>-compare.txt, iteration numbers must be multiple of -log interval; single header record for consistency check during the current run. (Following Fan Ming DC interim residues have been confirmed by my TC in progress.)

332194529 LL seed 4
1000000 6f3e8db940a2da46
5000000 af733e924b0ff14d
10000000 3445b03b12e7e63c
15000000 7f89336ab4c99ccf
20000000 7cc12f9bc5569c53
24000000 2a0843590d428092

Similar could also be useful at times during PRP DC. Header record something like

332194529 PRP type 1 base 3

Building something like that into CUDALucas, Gpuowl, Mlucas, prime95 could be useful during future "confirm this prime discovery" exercises.

[kriesel]

help output excerpt (note the zero values for upper limits) from gpuowl-win V7.2-86

FFT Configurations (specify with -fft <width>:<middle>:<height> from the set below):
FFT  256K [  1.57M -    0.00M]  256:2:256
FFT  384K [  2.36M -    7.72M]  256:3:256
FFT  512K [  3.15M -    0.00M]  256:4:256 256:2:512 512:2:256

These are probably ~5.14M and 10.3M respectively.

[joejoefla]

Getting an unexpected crash with 7.2-86.

I could ran DC PRP with Proof on the previous version. With this version I'm getting this

20211130 23:22:23 NVIDIA GeForce GTX 1080-0 Exception gpu_error: INVALID_KERNEL_ARGS fftMiddleIn at clwrap.cpp:324 run