Posts by [B@P] Daniel

1) Message boards : Number crunching : Very long WUs? (Message 549) Posted 2 Jul 2017 by [B@P] Daniel Post: You can change preferences to use max. 4 threads per WU. They'll take more time to complete, but the total CPU load will always be good. Thanks, looks that I will have to do this. Although it would be a better idea if server could send tasks in more smart way and try to sent such tasks to hosts which have smaller number of CPUs, which matches actual WU limit.
2) Message boards : Number crunching : Very long WUs? (Message 545) Posted 2 Jul 2017 by [B@P] Daniel Post: Most new 10^20 tasks on my 32 core Xeon machine completes in about 0.5 hour, but there are sone significantly longer, for example this one (link below) took almost 7 hours! Recently I also aborted one which after 7.5h was at 75%, assuming that it hung and won't complete. However when I looked on completed task list, I found that it might complete successfully. Why some tasks are significantly longer that others? Completed task: https://sech.me/boinc/Amicable/result.php?resultid=4799087 Aborted task: https://sech.me/boinc/Amicable/result.php?resultid=4799109 Edit: CPU time for this task is about 7 hours, so average CPU usage calculated as run time divided by CPU time is 26%. This is very low. I checked some other task which competed in about 0.5h (4798397) and got 63%. Does it mean that thread synchronization was so bad? Should I lower number of CPUs for WUs and run few in parallel? Edit2: I have one more tasks in progress, now for almost 5h. "top" command now shows 100%, about minute ago it was 200%. 1 to 2 out of 32 CPUs used is very bad result.
3) Message boards : News : New application for double checking current search (Message 480) Posted 16 Jun 2017 by [B@P] Daniel Post: What about amicable numbers found by this new app? Are they also counted as from normal app? And are we searching range where there are plenty of them, or one where only handful of them are present?
4) Message boards : Number crunching : 64-bit ARM support (Message 366) Posted 24 Mar 2017 by [B@P] Daniel Post: Few other projects have Android apps and many people uses them (I crunch WUs from Moo! Wrapper, Collatz, theSkyNet POGS, Rosetta, GoofyxGrid and WuProp). Android version of BOINC is specifically designed for smartphones - by default it uses half of available cores, do not crunch when on battery or temperature is over 40°C, and transfers data when WiFi is available. Android also can be installed on devices like Odroid which are better suited for crunching. So Android app version would be a nice addition to your project.
5) Message boards : Number crunching : 64-bit ARM support (Message 364) Posted 24 Mar 2017 by [B@P] Daniel Post: Could you prepare 64-bit Android app version too? You will need Standalone Toolchain for this, configured to support API level 21. In order to compile BOINC libs, you will have to to use command like this: ./configure --prefix=/cygdrive/c/somedir/android/_boinc64/ --host=aarch64-linux-android --disable-server --disable-client --enable-libraries --enable-static --disable-shared --disable-manager --disable-boinczip CC=/cygdrive/c/somedir/android/_arm64/bin/aarch64-linux-android-gcc CXX=/cygdrive/c/somedir/android/_arm64/bin/aarch64-linux-android-g++ CFLAGS="--sysroot=c:/somedir/android/_arm64/sysroot/ -DANDROID -DANDROID_64 -DDECLARE_TIMEZONE -O3 -fomit-frame-pointer -fPIE" CXXFLAGS="--sysroot=c:/somedir/android/_arm64/sysroot -DANDROID -DANDROID_64 -funroll-loops -fexceptions -O3 -fomit-frame-pointer -fPIE" LDFLAGS="--sysroot=c:/somedir/android/_arm64/sysroot -fPIE -pie"
6) Message boards : Random stuff : Are Badges On the Way? (Message 321) Posted 8 Mar 2017 by [B@P] Daniel Post: This project exports them and some sites like signature.statseb.fr already shows them. You need to contact admin of FreeDC and ask him to add them.
7) Message boards : Number crunching : GPU app for Mali T-628? (Message 315) Posted 6 Mar 2017 by [B@P] Daniel Post: Yes, I have one. I have just tried to compile this app, but it was not possible - it turned out that on 32-bit ARM gcc does not provide __int128 type. I wonder if it would be worth to provide some class which will simulate this type, probably yes if most calculations are done in OpenCL code. GMP lib probably will be useful here. Or maybe it would be possible to remove these parts of code completely and use OpenCL instead? Edit: GMP provides C++ interface too in gmpxx.h, so it should be even easier to use it. https://gmplib.org/manual/C_002b_002b-Class-Interface.html#C_002b_002b-Class-Interface Edit 2: there are also other libs: https://en.wikipedia.org/wiki/List_of_arbitrary-precision_arithmetic_software
8) Message boards : Number crunching : GPU app for Mali T-628? (Message 313) Posted 6 Mar 2017 by [B@P] Daniel Post: Hi, Could you prepare OpenCL app for Mali T-628, used by Odroid XU4? This GPU has memory shared with CPU, plus every GPU thread has its own instruction counter, so probably you could take existing CPU app and replace thread and memory allocation API with OpenCL calls. Please also check this link for more details: https://github.com/BOINC/boinc/issues/1686.
9) Message boards : Number crunching : 64-bit ARM support (Message 303) Posted 5 Mar 2017 by [B@P] Daniel Post: 1st WU completed and validated on my C2, it took 10h 46min.
10) Message boards : Number crunching : 64-bit ARM support (Message 301) Posted 4 Mar 2017 by [B@P] Daniel Post: Thanks! I have successfully attached my Odroid C2 and downloaded few tasks. It has 2GB RAM, so I use default config without extra changes. It also uses 64-bit Linux by default. BOINC Client also was installed from repo, no compilation was needed. Much simpler for average user :) So far after 80 minutes it uses 824MB of RAM (this is for 4 threads). It is hard to tell how much time it will need to complete, probably a bit less than 12 hours.
11) Message boards : Random stuff : Are Badges On the Way? (Message 285) Posted 1 Mar 2017 by [B@P] Daniel Post: https://sech.me/boinc/Amicable/forum_thread.php?id=36 Thanks. Please pin it too, people will be looking often for it.
12) Message boards : Random stuff : Are Badges On the Way? (Message 282) Posted 1 Mar 2017 by [B@P] Daniel Post: FYI Sergei ... Good Job but the Badges still aren't showing up in the Signatures from the Signature Sites ... They're all exported to /stats folder in standard BOINC format, what else can I do? I know that admin of signature.statseb.fr has to add them manually. Admins of other similar sites may also have to it manually. It would be nice to have all badges listed somewhere, e.g. in dedicated thread here.
13) Message boards : Random stuff : Credit system. (Message 264) Posted 28 Feb 2017 by [B@P] Daniel Post: Daniel And? I haven't heard of AVX/AVX2/whatever being able to do 64x64->64 bit and 64x64->128 bit integer multiplications. These are the crucial part of my program. Hmm, I saw other int ops there so I thought that 64-bit multiplication will be there too, but it turned out that it is missing. There is _mm_mullo_epi64 which performs 64x64->64 multiplication, but it needs AVX512VL + AVX512DQ. The other closest thing is _mm256_mul_epu32 from AVX2, gcc used it to implement AVX vector multiplication 64x64->64. It used 8 instructions for this plus load/store. I do not have hardware to benchmark it so it is hard to tell if it will provide any performance gain. 64x64->128 probably cannot be optimized this way at all. I tried to play a bit with gcc. It was able to vectorize 128x128->128, but it needed many instructions for this (~30), including few loads and stores. Code used AVX instructions, AVX2 does not change anything.
14) Message boards : Random stuff : Credit system. (Message 262) Posted 28 Feb 2017 by [B@P] Daniel Post: In sieve Yes. If You Try LLR it would be half of it. BTW does Your aplication use AVX/FMA/AVX2 ? Half of it would be exactly what it is here now. My application uses only integer arithmetic. Floating point just can't represent 64-bit integers without losing precision. SSE2 and AVX2 have separate instructions for integer operations too. SSE versions above 2 also added some specialized instructions for integers. FMA is for float point only.
15) Message boards : Random stuff : Are Badges On the Way? (Message 258) Posted 28 Feb 2017 by [B@P] Daniel Post: I am trying, so far I have just 20, so still some more work to do to get to 100. After 150,000 points I have 20, so after 300,000 may have 40, after 1,200,000 I may have 80 and I may get 100 after 1,350,000 points They are not consistent with credits - it's the "luck of the draw", I suppose. After 357,000 credits I have found 377 pairs... Looks that this is the case. I have joined yesterday. Now I have ~340k credits and 15 pairs. I also suspect that it was much easier to find pairs in the past.