| Author |
Message |
|---|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
 
|
|
gcwsieve is now part of the mtsieve framework. You can read about the update here: http://www.mersenneforum.org/showpost.php?p=490541&postcount=57. This version of gcwsieve supports mixed CPU and GPU sieving. In other words, gcwsievecl has been merged into gcwsieve.
fpsieve is now called fsieve (factorial sieve) is also in the framework and supports CPU and GPU. Once I get the AVX logic into fsieve, the CPU threads should get a nice speed bump over the current CPU only fpsieve.
The primorial sieve piece of fpsieve is not in the framework yet, but should be easy to add. When it is added it will have AVX logic out of the box. psieve will not have GPU support. I tried porting a few years ago but the performance is terrible.
The mtsieve framework is described here: http://mersenneforum.org/rogue/mtsieve.html.
If you have any questions, concerns, comments, etc, please post here or e-mail me. |
|
|
|
|
|
So, from what I gathered this version now supports multithreading and GPU computing, nice!
Is it possible to start testing it through anonymous platform or are there too many changes?
____________
|
|
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 437,259,124
RAC: 877,171
                              
|
Is it possible to start testing it through anonymous platform or are there too many changes?
I assume you are referring to gcw sieve? That project is close to being suspended since we're approaching optimal sieving depth, as such, we currently have no plans for Mark's new sieve apps.
We have not tested it.
Mark didn't say anything about it being compatible with BOINC.
There's a few other hoops that need to be jumped through before it could be integrated here.
So, no, you can't use it with app_info.xml at this time. (As far as I know, anyway.)
That being said...
A GPU version of the GCW sieve should be a lot faster than the CPU sieve, right? That will mean the "optimal sieving depth", when using a GPU sieve, will probably be much higher. That, in turn means that we've still got a lot more sieving to do on GCW. It's not unreasonable to think that we'll take a look into this at some point. We haven't even started talking about it yet, however. We only found out about it yesterday.
____________
My lucky number is 75898524288+1 |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I have incorporated a primorial sieve (called psieve) into the framework. It supports AVX out of the box and is about 30% faster than fpsieve. |
|
|
|
|
I have incorporated a primorial sieve (called psieve) into the framework. It supports AVX out of the box and is about 30% faster than fpsieve.
I take it you haven't yet included a binary for psieve in the mtsieve_1.7.7z archive, is that correct?
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I have incorporated a primorial sieve (called psieve) into the framework. It supports AVX out of the box and is about 30% faster than fpsieve.
I take it you haven't yet included a binary for psieve in the mtsieve_1.7.7z archive, is that correct?
Oops. psieve was not included when building all of the other binaries. I have updated the 7z file. |
|
|
|
|
|
psieve instantly crashes when running it with:
D:\sieving\psieve>psieve.exe -p 8151000000000 -P 8152000000000 -n 2500009 -N 9999991
psieve v1.1, a program to find factors of primorial
Sieve started: 8151e9 < p < 8152e9 with 963014 terms
D:\sieving\psieve>
And exits without running when specifying the psieve_9T.txt file:
D:\sieving\psieve>psieve.exe -p 8151000000000 -P 8152000000000 -i psieve_9T.txt
psieve v1.1, a program to find factors of primorial
Fatal Error: Line 1 is malformed in input file psieve_9T.txt
D:\sieving\psieve>
specs: AMD FX-6300, Windows 10 1607
____________
|
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,270,184
RAC: 110,123
            
|
psieve instantly crashes when running it with:
D:\sieving\psieve>psieve.exe -p 8151000000000 -P 8152000000000 -n 2500009 -N 9999991
psieve v1.1, a program to find factors of primorial
Sieve started: 8151e9 < p < 8152e9 with 963014 terms
D:\sieving\psieve>
And exits without running when specifying the psieve_9T.txt file:
Exactly. How do you expect it to run without a sieve file when you didn't specify a min and max n value? Where is it supposed to search?
D:\sieving\psieve>psieve.exe -p 8151000000000 -P 8152000000000 -i psieve_9T.txt
psieve v1.1, a program to find factors of primorial
Fatal Error: Line 1 is malformed in input file psieve_9T.txt
D:\sieving\psieve>
specs: AMD FX-6300, Windows 10 1607
It doesn't like the sieve file. Remove the "+" from the header line. It'll probably crash in a few seconds after that (it did for me) but at least it'll read the sieve file properly. And running an already-searched range should never find anything - those candidates were already removed from the sieve. I've emailed rogue with specifics on the crash I got. |
|
|
|
|
Exactly. How do you expect it to run without a sieve file when you didn't specify a min and max n value? Where is it supposed to search?
I thought the -n and -N parameters are what specify the min and max n values, at least that's what I understood from the documentation. How do I specify those values instead?
It doesn't like the sieve file. Remove the "+" from the header line. It'll probably crash in a few seconds after that (it did for me) but at least it'll read the sieve file properly. And running an already-searched range should never find anything - those candidates were already removed from the sieve. I've emailed rogue with specifics on the crash I got.
Ok, I did that and now it does seem to have read the sieve file, outputting "Sieve started: 8151e9 < p < 8152e9 with 502790 terms". However, it does still crash right after that. As per running an already sieved range, it was just some values I found on the manual sieving page that I knew would (should) already work with fpsieve.
____________
|
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,270,184
RAC: 110,123
|
I thought the -n and -N parameters are what specify the min and max n values, at least that's what I understood from the documentation. How do I specify those values instead?
My mistake. I completely missed the -n and -N on the command line. I ran a short initial sieve here and it worked for me, but it was done in less than a second. That's how I knew what it wanted in the sieve file (I had it output a new sieve file). |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I will address the use of the "+" on the ABC line. It will be optional in the next build. It is completely unnecessary as long as each subsequent line has a "+" or "-" on it.
Note that psieve uses scientific notation for numeric inputs. -p1e6 is the same as -p1000000. It does not support the G or T suffix on input which were supported by fpsieve. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I will address the use of the "+" on the ABC line. It will be optional in the next build. It is completely unnecessary as long as each subsequent line has a "+" or "-" on it.
Note that psieve uses scientific notation for numeric inputs. -p1e6 is the same as -p1000000. It does not support the G or T suffix on input which were supported by fpsieve.
Sorry I was mistaken. It does support a G or T suffix on those parameters. I missed that because it was really old code and I never use those suffixes.
I'm working with Jim to find the cause behind a psieve crash on Windows 7. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I fixed the crash that Jim ran into. It only happens if running on a machine that doesn't support either AVX or FMA instructions. |
|
|
|
|
|
I get an error when trying to run psieve (mtsieve 1.7.3) on a laptop with no dedicated graphics chip. i5-3317U, Windows 10 (1607). fpsieve works fine on this machine. 
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I get an error when trying to run psieve (mtsieve 1.7.3) on a laptop with no dedicated graphics chip. i5-3317U, Windows 10 (1607). fpsieve works fine on this machine.
Please send me (via e-mail) the command line arguments used, the output it writes to the command line, and any input files you are using. Please use 7-zip to zip up input files and attach them. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I have posted mtsieve 1.7.4. The most notable improvement is that mfsieve now supports AVX (but not AVX512 yet) and is about 40% faster than the previous release of mfsieve. Note that is only comparing CPU-only code.
I have not been provided the files I have asked for, so the issue reported by Gabriel probably still exists. |
|
|
|
|
I have not been provided the files I have asked for, so the issue reported by Gabriel probably still exists.
I searched for your email address for a bit but couldn't find it. The matter is that just running psieve.exe with no arguments or with whichever arguments makes it crash in the way shown in the picture https://i.imgur.com/7u3nxH5.png, with no output to the command line what so ever. I am away from that machine right now, but I am pretty confident the issue still exists with 1.7.4, since it has existed since 1.7.
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I have not been provided the files I have asked for, so the issue reported by Gabriel probably still exists.
I searched for your email address for a bit but couldn't find it. The matter is that just running psieve.exe with no arguments or with whichever arguments makes it crash in the way shown in the picture https://i.imgur.com/7u3nxH5.png, with no output to the command line what so ever. I am away from that machine right now, but I am pretty confident the issue still exists with 1.7.4, since it has existed since 1.7.
I thought my e-mail was on the mtsieve page. You can contact me at rogue at wi.rr.com.
That error is odd. It seems to be failing trying to determine if you have GPU support. Do you get errors when you try to run it properly? |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
Actually I think the issue is that you have an outdated video driver. Please install the latest video driver to see if that solves the problem. |
|
|
