Posts by Tatsuo Yamagata

1) Message boards : Number crunching : Is source code available? (Message 850)
Posted 19 May 2018 by Tatsuo Yamagata
Post:
hmmmm....

I used the same task, https://sech.me/boinc/Amicable/workunit.php?wuid=5678963 in standalone mode.

The result is

Haswell (4790K, 4th gen. core i7) optimized: 41929.51 seconds.

Stock application: 42847.81 seconds.

Only difference of less than 2.5%. It can be within error!! Intel Compiler doesn't have little advantage in integer math. I will not make further and won't release optimized apps.

Sorry....
2) Message boards : Number crunching : Is source code available? (Message 849)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
Ok. I've done it!

Hint was in the name of work units' names. They begin with amicable_10_20, so I set it to the app_name, it worked!

But the number of cpus must be set as avg_ncpus in app_info.xml file manually. Otherwise threads of the number of cpu minus one were made.

Anyway it worked!!

Thank you for your help!

EDIT: oh! Sergei, thank you for mentioning it!

EDIT2: Intel Compiler version has only half of the original EST on my Haswell rig, but I don't know it works so nicely, because the compiler has much advantage in floating point calculations. I will test this for some days.
3) Message boards : Number crunching : Is source code available? (Message 847)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
It says "syntax error" because you have mismatched tags:
 <app_name>amicable</name>


Thank you!! Good point. Somehow emacs didn't notify it to me. But fixing it, I still have an error message.

Thu 17 May 2018 10:05:11 PM JST | Amicable Numbers | Message from server: Unknown app name amicable in app_info.xml
Thu 17 May 2018 10:05:11 PM JST | Amicable Numbers | No tasks sent
Thu 17 May 2018 10:05:11 PM JST | Amicable Numbers | Message from server: Your app_info.xml file doesn't have a usable version of Amicable Numbers up to 10^20.


So the app_name must include info on 10^20. What app_name is set on the server, Sergei?

EDIT: I need to edit Amicable/Amicable/Platform_GCC_Linux.h so that Intel Compiler can utilize __int64. "typedef long long __int64;" gives an error with Intel compiler, so I modified it so that Intel compiler should use "#define __int64 long long". Intel compiler uses a predefined macro __INTEL_COMPILER, so #ifndef/#else/#endif were used.

And I saw http://boinc.berkeley.edu/wiki/Anonymous_platform and changed version_num to 280, but the error message was the same.
4) Message boards : Number crunching : Is source code available? (Message 845)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
Hi all,

I need some help!!

I compile an optimized gcc version (optimized for haswell) of amicable number, but I don't know app_info.xml for amicable numbers. I tried some, but all failed with "Syntax error in app_info.xml", because (maybe) app_name and name were wrong. Mine is as follows. What's wrong?
<app_info>
  <app>
    <name>amicable</name>
  </app>
  <file_info>
    <name>amicable</name>
    <executable/>
  </file_info>
  <app_version>
    <app_name>amicable</name>
    <version_num>208</version_num>
    <api_version>7.11.0</api_version>
    <file_ref>
      <file_name>amicable</file_name>
      <main_program/>
    </file_ref>
  </app_version>
</app_info>


I will distribute optimized linux application for cpu when it's built.

Thanks in advance!
5) Message boards : Number crunching : Is source code available? (Message 844)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
Another small bug. I checked out boinc version and compiled. It failed to find config.h which was in boinc directory. So I added "-I ../boinc" to INCLUDES variable in Makefile.
6) Message boards : Number crunching : Is source code available? (Message 843)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
I don't have any issues running stock applications. I just want to optimize it with Intel compiler.

I found errors compiling with gcc 7.3.1 coming with Fedora 27 (current latest version is gcc 8.1.1 on Fedora 28).

It's in switch statement in Amicable/primesieve/src/primesieve/EratSmall.cpp, many fallthrough were seen; each "case:" statement doesn't end with "break;" and go through next "case:" statement. It is forbidden since gcc/g++ 7. In order to disable the compiler to check, I added "-Wimplicit-fallthrough=0" to WARNINGS and made a new makefile called Makefile. I don't know it's allowed by previous gcc/g++.

There is another workaround for c++11 modifying the source file, but it will fail with Visual Studio.

As for this fallthrough problem, see https://developers.redhat.com/blog/2017/03/10/wimplicit-fallthrough-in-gcc-7/.
7) Message boards : Number crunching : Is source code available? (Message 841)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
https://github.com/SChernykh/Amicable/tree/boinc-version-128-bit


Thank you for quick response!!! It'll be help me a lot!
8) Message boards : Number crunching : Is source code available? (Message 839)
Posted 17 May 2018 by Tatsuo Yamagata
Post:
Hi,

I'm wondering source code for recent Amicable Numbers version v2.08 is available or not for Linux 64-bit. I found old source repository in github which supports up to 10^13.

Thanks in advance!!



©2024 Sergei Chernykh