- Boost-users - lists.boost.org

boost MPI about finalize problem
by Jack Bryan 25 Oct '10

25 Oct '10

Hi I got a problem of MPI. I donot know whether it can be resolved in boost MPI. My program has 5 processes. All of them can run MPI_Finalize() and return 0. But, the whole program cannot be completed. In the MPI cluster job queue, it is still in running status. If I use 1 process to run it, no problem. Why ? My program: int main (int argc, char **argv) { MPI_Init(&argc, &argv); MPI_Comm_rank(MPI_COMM_WORLD, &myRank); MPI_Comm_size(MPI_COMM_WORLD, &mySize); MPI_Comm world; world = MPI_COMM_WORLD; if (myRank == 0) { do some things. } if (myRank != 0) { do some things. } cout << " I am rank " << rank << " I am before MPI_Finalize()" << endl; MPI_Finalize(); cout << " I am rank " << rank << " I am after MPI_Finalize()" << endl; return 0; } I can get the output " I am rank 0 (1, 2, ....) I am before MPI_Finalize() ". and " I am rank 0 I am after MPI_Finalize() " But, other processes do not printed out "I am rank ... I am after MPI_Finalize()" . It is weird. The process has reached the point just before MPI_Finalize(), why they are hanged there ? Are there other better ways to check this ? Any help is appreciated. And, some output files get wrong codes, which can not be readible. In 1-process case, the program can print correct results to these output files . Any help is appreciated. thanks Jack Oct. 24 2010

2 1

Variadic Max Function
by Hossein Haeri 25 Oct '10

25 Oct '10

Dear all, Is there any variadic max function in Boost? I mean something which returns the max of all the arguments it takes -- no matter how many they are. (Feel free to correct me if this use of 'variadic' is not the correct terminology.) Please note that I'm not after working on ranges of arguments. I'm rather looking for a max function which is applicable in all the examples below: int m1 = max(a1, a2); int m2 = max(a1, a2, a3); ... int mk = max(a1, a2, ..., ak); TIA, --Hossein

4 5

order of headers with Boost.Serialization and Boost.MPI
by Riccardo Murri 25 Oct '10

25 Oct '10

Hello, I'm writing a code that uses Boost.MPI to send some GMP (http://gmplib.org/) datatypes around. Since I do not want to change GMP, I'm using the "non-intrusive" serialization support (code in attached file "gmpxx_boost_serialization.hpp"). However, I'm running into a strange include-order mess: * If I include first my "gmpxx_boost_serialization.hpp" and _then_ "boost/mpi.h", then I get a slew of errors (attached file "errors.txt") * If I reverse the include order, and include "boost/mpi.h" _first_, then everything compiles fine. Is this to be expected, or am I doing something wrong in my code? For the record, I'm compiling on Ubuntu 10.10, gcc 4.4.5, arch. amd64. Thanks for any hint! Best regards, Riccardo Murri -- Riccardo Murri, Hadlaubstr. 150, 8006 Zürich (CH)

2 7

General question about C++ [slightly off topic, sorry]
by Julien Martin 25 Oct '10

25 Oct '10

Hello, I am a beginner in C++. I learn using the Linux platform. What is the best place to look for general help about C++ please? Thanks in advance, J.

2 2

Changes breaking behavior of operators /, /= and << in filesystem v3?
by Roel Vanhout 25 Oct '10

25 Oct '10

Hello, I enabled filesystem v3 after upgrading to 1.44. The api changes that caused the compiler to signal errors were rather easy to fix. However when running my program, it behaves differently from the way it used to. I narrowed it down to the following: char buf[MAX_PATH]; HRESULT res = ::SHGetFolderPath(NULL, CSIDL_APPDATA, NULL, SHGFP_TYPE_CURRENT, buf); boost::filesystem::path result(buf); // 'result' is after this "C:\Documents and Settings\rvanhout\Application Data" result /= "MyCompany"; At this point, 'result' *still* contains "C:\Documents and Settings\rvanhout\Application Data" whereas with v2 it used to contain "C:\Documents and Settings\rvanhout\Application Data\MyCompany\MyProduct". (environment: Windows (obviously) with VS 2008 (vc9)) There seem to be two way to get the desired behaviour: - initialize the path with a trailing slash, like this: boost::filesystem::path result(buf + std::string("/")); - convert 'buf' into a std::string first: std::string str(buf); boost::filesystem::path result(str); So I guess my code broke because the path is not initialized with the platform's native path encoding (although std::string is char also and it does work). Is this by design? It seems quite unintuitive to me. Second thing I was bit by is that the io streaming operator now seems to add quotes to a path with spaces in it when streaming a path object to an ostringstream. The string that is returned with .string() isn't quoted, of course, so my 'fix' is to not stream path objects anymore but add the .string() call. What's more important, is there a way to let the compiler signal these situations in my code? The way it is now, everything compiles but I have no way to know if it will actually work as it used to/should. Otherwise, it might be an idea to add #pragma warning statements to those parts for a release or two, to give people a heads up. Thanks. regards, Roel Vanhout

2 2

32 bit time stamp to date and time
by Lloyd 25 Oct '10

25 Oct '10

Hi, I am writing a simple NTP client. NTP retuns the time as a 32bit value, as the seconds since 1900. using boost date time library is it possible to convert it to a date and time (year,month,day of week,day,hour,minute,second) ? Thanks, Lloyd ______________________________________ Scanned and protected by Email scanner

4 8

this_thread::sleep hangs on local clock change
by Goldshmidt, Mark 25 Oct '10

25 Oct '10

Hello, We're using boost::thread library in a consumer application. Part of the requirements for the application is to continue operation when the user changes local time. Our problem is that if local time is set backwards X minutes during a call to this_thread::sleep(duration) the thread sleeps for X minutes+duration. This happens on MacOS in release build, but not in debug. On Windows behavior is correct. I didn't have a chance to check other platforms. I believe that the intention of this_thread::sleep(duration) is to suspend thread execution for <duration> period regardless of local clock setting, and provide consistent behavior across all platforms and build configurations. Looking into the code of libs/thread/src/pthread/thread.cpp I see that problematic flow may occur in case of detail::get_current_thread_data() is returning null. Apparently this is what causes different behavior in debug and release. The "offendinng" code tries to make sure that absolute point in time (current time + requested delay) is reached using 5 attempts and calculating necessary sleep duration each time. So if during the first sleep local clock is set backwards, second sleep will try to "compensate" for the new difference. The simplest workaround for the problem I could think of is to replace the first statement of the sleep(...) function: detail::thread_data_base* const thread_info= detail::get_current_thread_data(); with this one: detail::thread_data_base* const thread_info= get_or_make_current_thread_data(); Would it create any unwanted side effects? Is there a better way to deal with is problem? Thanks in advance, Mark This message is confidential and intended only for the addressee. If you have received this message in error, please immediately notify the postmaster(a)nds.com and delete it from your system as well as any copies. The content of e-mails as well as traffic data may be monitored by NDS for employment and security purposes. To protect the environment please do not print this e-mail unless necessary. An NDS Group Limited company. www.nds.com

1 0

extents
by Alle Meije Wink 25 Oct '10

25 Oct '10

Dear all, I am trying to load data from a file, where they are (obviously) stored as a 1D contiguous block, to a N-dimensional array. For the type of files I use N could be 1 through 8. I have code to for a 1..8-dimensional container (as a class template), but how to load in the data from the file? The multi_array_ref way to do it looked ideal, eg for a 3D float-image something like im = image_ref< float, 2 >( image->data, boost::extents[512][512] ) might work. But to code the different dimensionalities in a neat way would mean to store the extents in a variable (or something), like boost_extents_type e; switch ndim { // ndim=dim[0] 1: int dim1=dim[1]; e=boost::extents[dim1]; break; 2: int dim1=dim[1],dim2=dim[2]; e=boost::extents[dim1][dim2]; //etc } Is something like this possible with extents? Many thanks, Alle Meije

1 1

[program_options] How to use "notifier" in options_description with member functions?
by Carlos Franke 25 Oct '10

25 Oct '10

Hello! I am looking for advice about the use of the "notifier" function in program_options::options_description with class methods. My problem seems very basic to mee, yet I wasn't able to find the (any) related documentation. Here is a minimal example of what I cannot get to work: struct test { void function(int i) {std::cout << i << std::endl;} void does_not_compile() { po::options_description desc; desc.add_options() ("option", po::value<int>()->notifier(&test::function), "Option") ; } }; Compilation fails with: […] test.cpp:62:82: instantiated from here /usr/include/boost/function/function_template.hpp:225:11: error: no match for call to '(boost::_mfi::mf1<void, test, int>) (const int&)' /usr/include/boost/bind/mem_fn_template.hpp:163:7: note: candidates are: R boost::_mfi::mf1<R, T, A1>::operator()(T*, A1) const [with R = void, T = test, A1 = int] /usr/include/boost/bind/mem_fn_template.hpp:184:7: note: R boost::_mfi::mf1<R, T, A1>::operator()(T&, A1) const [with R = void, T = test, A1 = int] Sorry if this is trivial. Your help is all the more appreciated! Carlos

2 2

[asio] udp sendto performance bottleneck
by Hajo Kirchhoff 25 Oct '10

25 Oct '10

Hi all, I am porting an application from linux (suse 11) to windows. This application is sending 75.000 UDP Pakets per second, 1416 bytes each. It uses sendto. IOW: the application is calling sendto approx 75.000 times per second. The GBit LAN utilization is in the 80-90% range. This works fine under linux, but the sendto call under windows takes way longer than under linux. The overhead is so high that throughput drops from about 100MB/s to less than 20MB/s. This seems to be a known problem with the windows sendto call, as cygwin has a rather complex sendto implementation/emulation and does not suffer from this problem. They are using WSASendMsg internally and pass multiple buffers with one system call. WSASendMsg passes an array of buffers to the networking stack and my impression is that it allows sending several packets with one system call. My question: Is it possible to send multiple udp pakets with one asio::sendto call? asio::sendto accepts multiple buffers but seems to combine all them into one udp paket. Is there a way to prevent that? Specifically I want to do the following: vector<array<char, 1416> > b; b.push_back(paket1); b.push_back(paket2); b.push_back(paket3); ... m_socket.sendto(b); This should send three udp pakets with 1416 bytes each instead of one udp paket with 3*1416 bytes. How can I do that? Is this possible? Regards Hajo (Sidenote to all people with dropped udp pakets under windows Vista and 7: Have you set the network throttling index to 0xffffffff? Its a registry key and without it you'll never reach maximum throughput. Another important aspect is to set the receive buffer count in your network driver to maximum - usually 512 or more - and to use a good network card. It _is_ possible to receive close to one Gigabit/sec UDP pakets if done right. Thanks to asio it is also rather easy :-)

1 0