Files
MPIonCell/documentation/developer/html/d8/dd7/doxygen__main_8doxygen-source.html
2017-10-26 19:57:24 +02:00

267 lines
24 KiB
HTML
Executable File

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html><head><meta http-equiv="Content-Type" content="text/html;charset=UTF-8">
<title>MPI On Cell (Documentation): include/doxygen_main.doxygen Source File</title>
<link href="../../doxygen.css" rel="stylesheet" type="text/css">
<link href="../../tabs.css" rel="stylesheet" type="text/css">
</head><body>
<!-- Generated by Doxygen 1.5.6 -->
<div class="navigation" id="top">
<div class="tabs">
<ul>
<li><a href="../../main.html"><span>Main&nbsp;Page</span></a></li>
<li><a href="../../modules.html"><span>Modules</span></a></li>
<li><a href="../../annotated.html"><span>Data&nbsp;Structures</span></a></li>
<li class="current"><a href="../../files.html"><span>Files</span></a></li>
</ul>
</div>
<h1>include/doxygen_main.doxygen</h1><a href="../../da/d41/doxygen__main_8doxygen.html">Go to the documentation of this file.</a><div class="fragment"><pre class="fragment"><a name="l00001"></a>00001 <span class="preprocessor">#ifndef MOC_DOXYGEN_MAIN_H_</span>
<a name="l00002"></a>00002 <span class="preprocessor"></span><span class="preprocessor">#define MOC_DOXYGEN_MAIN_H_</span>
<a name="l00003"></a>00003 <span class="preprocessor"></span><span class="comment"></span>
<a name="l00004"></a>00004 <span class="comment">/**</span>
<a name="l00005"></a>00005 <span class="comment"> * \mainpage MPI On Cell</span>
<a name="l00006"></a>00006 <span class="comment"> *</span>
<a name="l00007"></a>00007 <span class="comment"> * The idea of this project is to implement a message passing</span>
<a name="l00008"></a>00008 <span class="comment"> * library for work on the Cell BE. The main idea behind such</span>
<a name="l00009"></a>00009 <span class="comment"> * an implementation would be to help the programmer by adding</span>
<a name="l00010"></a>00010 <span class="comment"> * an layer of abstraction between the very low level libraries</span>
<a name="l00011"></a>00011 <span class="comment"> * by IBM and give him the possibility to work with an interface</span>
<a name="l00012"></a>00012 <span class="comment"> * he is more familiar with. To achieve this we decided to try</span>
<a name="l00013"></a>00013 <span class="comment"> * to keep very closely to the standard provided by the Message</span>
<a name="l00014"></a>00014 <span class="comment"> * Passing Interface (MPI), so that programmers knowing how to</span>
<a name="l00015"></a>00015 <span class="comment"> * write MPI programmers can easily work with our library.</span>
<a name="l00016"></a>00016 <span class="comment"> * Because of the very special nature of the Cell, it might not</span>
<a name="l00017"></a>00017 <span class="comment"> * be possible to keep 100% true to the standard, but we try to</span>
<a name="l00018"></a>00018 <span class="comment"> * introduce changes only when absolutely necessary and of great</span>
<a name="l00019"></a>00019 <span class="comment"> * benefit, because of some specialties of the Cell's hardware.</span>
<a name="l00020"></a>00020 <span class="comment"> *</span>
<a name="l00021"></a>00021 <span class="comment"> *</span>
<a name="l00022"></a>00022 <span class="comment"> * \par Table of Content</span>
<a name="l00023"></a>00023 <span class="comment"> * \par</span>
<a name="l00024"></a>00024 <span class="comment"> * \ref end_user</span>
<a name="l00025"></a>00025 <span class="comment"> * \par</span>
<a name="l00026"></a>00026 <span class="comment"> * \subpage restrictions</span>
<a name="l00027"></a>00027 <span class="comment"> * \par</span>
<a name="l00028"></a>00028 <span class="comment"> * \subpage violations</span>
<a name="l00029"></a>00029 <span class="comment"> * \par</span>
<a name="l00030"></a>00030 <span class="comment"> * \subpage performance</span>
<a name="l00031"></a>00031 <span class="comment"> * \par</span>
<a name="l00032"></a>00032 <span class="comment"> * \subpage conclusion</span>
<a name="l00033"></a>00033 <span class="comment"> * \if full</span>
<a name="l00034"></a>00034 <span class="comment"> * \par</span>
<a name="l00035"></a>00035 <span class="comment"> * \par</span>
<a name="l00036"></a>00036 <span class="comment"> * \ref internal_main</span>
<a name="l00037"></a>00037 <span class="comment"> * \endif</span>
<a name="l00038"></a>00038 <span class="comment"> **/</span>
<a name="l00039"></a>00039
<a name="l00040"></a>00040 <span class="comment"></span>
<a name="l00041"></a>00041 <span class="comment">/**</span>
<a name="l00042"></a>00042 <span class="comment"> * \defgroup end_user End-User Documentation</span>
<a name="l00043"></a>00043 <span class="comment"> *</span>
<a name="l00044"></a>00044 <span class="comment"> * This section describes the public API provided by the MOC-library.</span>
<a name="l00045"></a>00045 <span class="comment"> *</span>
<a name="l00046"></a>00046 <span class="comment"> * If some things remain unclear, consult the official MPI-standard (Version 1.3).</span>
<a name="l00047"></a>00047 <span class="comment"> *</span>
<a name="l00048"></a>00048 <span class="comment"> * This library uses Direct Memory Access (DMA) provided by the Cell Broadband Engine.</span>
<a name="l00049"></a>00049 <span class="comment"> * So the use of other DMA operations than provided of this library is discouraged,</span>
<a name="l00050"></a>00050 <span class="comment"> * because it may interfere with this library. One exceptions are signals.</span>
<a name="l00051"></a>00051 <span class="comment"> * We don't use signals at all at the moment, so one can savely use signal.</span>
<a name="l00052"></a>00052 <span class="comment"> **/</span>
<a name="l00053"></a>00053
<a name="l00054"></a>00054 <span class="comment"></span>
<a name="l00055"></a>00055 <span class="comment">/**</span>
<a name="l00056"></a>00056 <span class="comment"> * \defgroup environment_main Environmental Management</span>
<a name="l00057"></a>00057 <span class="comment"> * \ingroup end_user</span>
<a name="l00058"></a>00058 <span class="comment"> *</span>
<a name="l00059"></a>00059 <span class="comment"> * This chapter discusses routines for getting and, where appropriate, setting various parameters</span>
<a name="l00060"></a>00060 <span class="comment"> * that relate to the MPI implementation and the execution environment (such as error handling,</span>
<a name="l00061"></a>00061 <span class="comment"> * startup&amp;shutdown and Time).</span>
<a name="l00062"></a>00062 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00063"></a>00063 <span class="comment">/**</span>
<a name="l00064"></a>00064 <span class="comment"> * \defgroup environment_startup Startup &amp; Shutdown</span>
<a name="l00065"></a>00065 <span class="comment"> * \ingroup environment_main</span>
<a name="l00066"></a>00066 <span class="comment"> *</span>
<a name="l00067"></a>00067 <span class="comment"> * Functions to start and terminate the MOC-execution environment.</span>
<a name="l00068"></a>00068 <span class="comment"> *</span>
<a name="l00069"></a>00069 <span class="comment"> * To be able to communicate with other processes every MOC-process needs to setup his</span>
<a name="l00070"></a>00070 <span class="comment"> * MOC-Environment before any MOC-function can be called (there are some exceptions though).</span>
<a name="l00071"></a>00071 <span class="comment"> *</span>
<a name="l00072"></a>00072 <span class="comment"> * The PPU-process has a special role, because he is the one who starts the SPU-processes.</span>
<a name="l00073"></a>00073 <span class="comment"> * SPU-processes can be added to the MOC-environment by calling \ref MOC_Add_spu_thread with</span>
<a name="l00074"></a>00074 <span class="comment"> * the appropriate SPU-thread-handle (This implementation only supports the embedded binary</span>
<a name="l00075"></a>00075 <span class="comment"> * concept where the SPU-binary is embedded into the PPU-binary and accessed via a handle).</span>
<a name="l00076"></a>00076 <span class="comment"> * The order of calling determines the rank of the SPU-processes. To get the number of total</span>
<a name="l00077"></a>00077 <span class="comment"> * and available SPUs the functions \ref MOC_Get_avail_spu_count and \ref MOC_Get_max_spu_count</span>
<a name="l00078"></a>00078 <span class="comment"> * can be used.</span>
<a name="l00079"></a>00079 <span class="comment"> *</span>
<a name="l00080"></a>00080 <span class="comment"> * The buffer-size and the size of the message-queues can be set before calling \ref MOC_Init</span>
<a name="l00081"></a>00081 <span class="comment"> * on the PPU using \ref MOC_Set_buffer_size and \ref MOC_Set_queue_size.</span>
<a name="l00082"></a>00082 <span class="comment"> *</span>
<a name="l00083"></a>00083 <span class="comment"> * MOC_Init only returns after all processes have called this function and have set up their</span>
<a name="l00084"></a>00084 <span class="comment"> * environment.</span>
<a name="l00085"></a>00085 <span class="comment"> *</span>
<a name="l00086"></a>00086 <span class="comment"> * If the MOC-environment isn't needed anymore, \ref MOC_Finalize should be called by all</span>
<a name="l00087"></a>00087 <span class="comment"> * MOC-processes. MOC_Finalize only returns after all MOC-processes have called this functions</span>
<a name="l00088"></a>00088 <span class="comment"> * and have shutdown their environments.</span>
<a name="l00089"></a>00089 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00090"></a>00090 <span class="comment">/**</span>
<a name="l00091"></a>00091 <span class="comment"> * \defgroup environment_information Implementation Information</span>
<a name="l00092"></a>00092 <span class="comment"> * \ingroup environment_main</span>
<a name="l00093"></a>00093 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00094"></a>00094 <span class="comment">/**</span>
<a name="l00095"></a>00095 <span class="comment"> * \defgroup environment_error Error-Handling</span>
<a name="l00096"></a>00096 <span class="comment"> * \ingroup environment_main</span>
<a name="l00097"></a>00097 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00098"></a>00098 <span class="comment">/**</span>
<a name="l00099"></a>00099 <span class="comment"> * \defgroup environment_timer Timers and Synchronization</span>
<a name="l00100"></a>00100 <span class="comment"> * \ingroup environment_main</span>
<a name="l00101"></a>00101 <span class="comment"> **/</span>
<a name="l00102"></a>00102 <span class="comment"></span>
<a name="l00103"></a>00103 <span class="comment">/**</span>
<a name="l00104"></a>00104 <span class="comment"> * \defgroup p2p_main Point-to-Point Communication</span>
<a name="l00105"></a>00105 <span class="comment"> * \ingroup end_user</span>
<a name="l00106"></a>00106 <span class="comment"> *</span>
<a name="l00107"></a>00107 <span class="comment"> * Sending and receiving of messages by processes is the basic MPI communication mechanism.</span>
<a name="l00108"></a>00108 <span class="comment"> * The basic point-to-point communication operations are send and receive.</span>
<a name="l00109"></a>00109 <span class="comment"> *</span>
<a name="l00110"></a>00110 <span class="comment"> * \ref MOC_Send uses either buffered or synchronous send depending on the message-size</span>
<a name="l00111"></a>00111 <span class="comment"> * (Short messages are transmitted using buffered send and long messages are transmitted using</span>
<a name="l00112"></a>00112 <span class="comment"> * synchronous send).</span>
<a name="l00113"></a>00113 <span class="comment"> *</span>
<a name="l00114"></a>00114 <span class="comment"> * \ref MOC_Recv receives all messages irrespective wether buffered or synchronous</span>
<a name="l00115"></a>00115 <span class="comment"> * communication is used.</span>
<a name="l00116"></a>00116 <span class="comment"> *</span>
<a name="l00117"></a>00117 <span class="comment"> * \note Because there are separate message-queues for each remote process and no distinct</span>
<a name="l00118"></a>00118 <span class="comment"> * global ordering of messages, there may be some situation where a message from a</span>
<a name="l00119"></a>00119 <span class="comment"> * process may never be received. This may occur when MOC_Recv, MOC_Probe and MOC_IProbe</span>
<a name="l00120"></a>00120 <span class="comment"> * are used with MOC_ANY_SOURCE, because then there is the possibility that new messages</span>
<a name="l00121"></a>00121 <span class="comment"> * are received before a specific message from a process with a higher rank than the other</span>
<a name="l00122"></a>00122 <span class="comment"> * senders. This may also occur with lower possibility when using MOC_ANY_TAG between</span>
<a name="l00123"></a>00123 <span class="comment"> * messages from the same sender but with other tags.</span>
<a name="l00124"></a>00124 <span class="comment"> *</span>
<a name="l00125"></a>00125 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00126"></a>00126 <span class="comment">/**</span>
<a name="l00127"></a>00127 <span class="comment"> * \defgroup p2p_status Return Status</span>
<a name="l00128"></a>00128 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00129"></a>00129 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00130"></a>00130 <span class="comment">/**</span>
<a name="l00131"></a>00131 <span class="comment"> * \defgroup p2p_datatype Datatypes</span>
<a name="l00132"></a>00132 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00133"></a>00133 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00134"></a>00134 <span class="comment">/**</span>
<a name="l00135"></a>00135 <span class="comment"> * \defgroup p2p_synch Synchronous Communication</span>
<a name="l00136"></a>00136 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00137"></a>00137 <span class="comment"> *</span>
<a name="l00138"></a>00138 <span class="comment"> * Direct blocking sending of messages without any intermediate buffer.</span>
<a name="l00139"></a>00139 <span class="comment"> *</span>
<a name="l00140"></a>00140 <span class="comment"> * The blocking synchronous send-operation returns only when the receiver has completely</span>
<a name="l00141"></a>00141 <span class="comment"> * received the message.</span>
<a name="l00142"></a>00142 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00143"></a>00143 <span class="comment">/**</span>
<a name="l00144"></a>00144 <span class="comment"> * \defgroup p2p_buffered Buffered Communication</span>
<a name="l00145"></a>00145 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00146"></a>00146 <span class="comment"> *</span>
<a name="l00147"></a>00147 <span class="comment"> * Blocking sending of messages with the use of an intermediate buffer.</span>
<a name="l00148"></a>00148 <span class="comment"> *</span>
<a name="l00149"></a>00149 <span class="comment"> * The blocking buffered send-operation returns if the send-buffer can be safely reused</span>
<a name="l00150"></a>00150 <span class="comment"> * (after the data has been copied into the intermediate buffer).</span>
<a name="l00151"></a>00151 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00152"></a>00152 <span class="comment">/**</span>
<a name="l00153"></a>00153 <span class="comment"> * \defgroup p2p_sendrecv Send-Receive</span>
<a name="l00154"></a>00154 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00155"></a>00155 <span class="comment"> *</span>
<a name="l00156"></a>00156 <span class="comment"> * The send-receive operations combine in one call the sending of a message</span>
<a name="l00157"></a>00157 <span class="comment"> * to one destination and the receiving of another message, from another process.</span>
<a name="l00158"></a>00158 <span class="comment"> *</span>
<a name="l00159"></a>00159 <span class="comment"> * A message sent by a send-receive operation can be received by a regular receive</span>
<a name="l00160"></a>00160 <span class="comment"> * operation or probed by a probe operation; a send-receive operation can receive</span>
<a name="l00161"></a>00161 <span class="comment"> * a message sent by a regular send operation.</span>
<a name="l00162"></a>00162 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00163"></a>00163 <span class="comment">/**</span>
<a name="l00164"></a>00164 <span class="comment"> * \defgroup p2p_imm Non-Blocking Communication</span>
<a name="l00165"></a>00165 <span class="comment"> * \ingroup p2p_main</span>
<a name="l00166"></a>00166 <span class="comment"> *</span>
<a name="l00167"></a>00167 <span class="comment"> * Non-blocking sending and receiving messages.</span>
<a name="l00168"></a>00168 <span class="comment"> *</span>
<a name="l00169"></a>00169 <span class="comment"> * Because of restrictions of the Cell Broadband Engine non-blocking communication can</span>
<a name="l00170"></a>00170 <span class="comment"> * only be provided with a lot of limitations. In fact the only operation where</span>
<a name="l00171"></a>00171 <span class="comment"> * non-blocking communication make sense is synchronous send.</span>
<a name="l00172"></a>00172 <span class="comment"> *</span>
<a name="l00173"></a>00173 <span class="comment"> * \ref MOC_ISend uses blocking buffered send or non-blocking synchronous send according to</span>
<a name="l00174"></a>00174 <span class="comment"> * the message size.</span>
<a name="l00175"></a>00175 <span class="comment"> *</span>
<a name="l00176"></a>00176 <span class="comment"> * \note Every non-blocking communication operation has to be finished using \ref MOC_Test</span>
<a name="l00177"></a>00177 <span class="comment"> * or \ref MOC_Wait, otherwise resources are no freed and some data are not copied correctly.</span>
<a name="l00178"></a>00178 <span class="comment"> **/</span>
<a name="l00179"></a>00179
<a name="l00180"></a>00180 <span class="comment"></span>
<a name="l00181"></a>00181 <span class="comment">/**</span>
<a name="l00182"></a>00182 <span class="comment"> * \defgroup broadcast Collective Communication</span>
<a name="l00183"></a>00183 <span class="comment"> * \ingroup end_user</span>
<a name="l00184"></a>00184 <span class="comment"> *</span>
<a name="l00185"></a>00185 <span class="comment"> * Communication that involves a group of processes.</span>
<a name="l00186"></a>00186 <span class="comment"> *</span>
<a name="l00187"></a>00187 <span class="comment"> * A collective operation is normally executed by having all processes in the group call</span>
<a name="l00188"></a>00188 <span class="comment"> * the communication routine, with matching arguments. Since there are no means for creating</span>
<a name="l00189"></a>00189 <span class="comment"> * groups currently implemented, the only available groups are MOC_COMM_WORLD and</span>
<a name="l00190"></a>00190 <span class="comment"> * MOC_COMM_SELF.</span>
<a name="l00191"></a>00191 <span class="comment"> **/</span>
<a name="l00192"></a>00192
<a name="l00193"></a>00193 <span class="comment"></span>
<a name="l00194"></a>00194 <span class="comment">/**</span>
<a name="l00195"></a>00195 <span class="comment"> * \defgroup dma_access Direct Memory Access</span>
<a name="l00196"></a>00196 <span class="comment"> * \ingroup end_user</span>
<a name="l00197"></a>00197 <span class="comment"> *</span>
<a name="l00198"></a>00198 <span class="comment"> * Functions which use DMA to transfer Data.</span>
<a name="l00199"></a>00199 <span class="comment"> *</span>
<a name="l00200"></a>00200 <span class="comment"> * The operations defined in this section provide simple means to use Direct Memory</span>
<a name="l00201"></a>00201 <span class="comment"> * Access (DMA).</span>
<a name="l00202"></a>00202 <span class="comment"> *</span>
<a name="l00203"></a>00203 <span class="comment"> * The MPI-standard assumes a distributed memory architecture, where</span>
<a name="l00204"></a>00204 <span class="comment"> * active participation of the sending and the receiving processes is required. But</span>
<a name="l00205"></a>00205 <span class="comment"> * the Cell Broadband Engine is a distributed shared memory architecture which provides</span>
<a name="l00206"></a>00206 <span class="comment"> * a common address space and DMA-Operations enabling processes to write data directly</span>
<a name="l00207"></a>00207 <span class="comment"> * into the memory of an remote process without active participation of the receiving</span>
<a name="l00208"></a>00208 <span class="comment"> * process.</span>
<a name="l00209"></a>00209 <span class="comment"> * DMA is faster than Send-Receive but more complicated to handle, and so there are some</span>
<a name="l00210"></a>00210 <span class="comment"> * functions provided to enable the advantages of DMA and to hide as much complexity as</span>
<a name="l00211"></a>00211 <span class="comment"> * possible. So the user doesn't have to care about the upper-limit of an DMA-transfer</span>
<a name="l00212"></a>00212 <span class="comment"> * and in get-operations the data-size doesn't have to be a multiple of 16-byte. However</span>
<a name="l00213"></a>00213 <span class="comment"> * the data-address should be at least 16-byte aligned and in put-operations the data-size</span>
<a name="l00214"></a>00214 <span class="comment"> * has to be a multiple of 16-byte.</span>
<a name="l00215"></a>00215 <span class="comment"> *</span>
<a name="l00216"></a>00216 <span class="comment"> * There are two kind of addresses. Local-store addresses are used on the SPUs and are only</span>
<a name="l00217"></a>00217 <span class="comment"> * valid in context of a specific SPU. Effective addresses define a common address-space and</span>
<a name="l00218"></a>00218 <span class="comment"> * are used on the PPU. Functions to translate between those address types are provided.</span>
<a name="l00219"></a>00219 <span class="comment"> *</span>
<a name="l00220"></a>00220 <span class="comment"> * DMA-operations can be blocking or non-blocking. The status of non-blocking DMA-calls can</span>
<a name="l00221"></a>00221 <span class="comment"> * be requested by using \ref MOC_Test and \ref MOC_Wait.</span>
<a name="l00222"></a>00222 <span class="comment"> *</span>
<a name="l00223"></a>00223 <span class="comment"> * \note Every non-blocking DMA operation has to be finished using \ref MOC_Test</span>
<a name="l00224"></a>00224 <span class="comment"> * or \ref MOC_Wait, otherwise resources are no freed and some data are not copied correctly.</span>
<a name="l00225"></a>00225 <span class="comment"> **/</span>
<a name="l00226"></a>00226 <span class="comment"></span>
<a name="l00227"></a>00227 <span class="comment">/**</span>
<a name="l00228"></a>00228 <span class="comment"> * \defgroup comm_main Groups, Context and Communicators</span>
<a name="l00229"></a>00229 <span class="comment"> * \ingroup end_user</span>
<a name="l00230"></a>00230 <span class="comment"> *</span>
<a name="l00231"></a>00231 <span class="comment"> * The MOC communicator specifies a group of processes inside which</span>
<a name="l00232"></a>00232 <span class="comment"> * a communication occurs.</span>
<a name="l00233"></a>00233 <span class="comment"> * Initially there is a single communicator, MPI_COMM_WORLD, which</span>
<a name="l00234"></a>00234 <span class="comment"> * contains all of the processes involved in the computation.</span>
<a name="l00235"></a>00235 <span class="comment"> **/</span><span class="comment"></span>
<a name="l00236"></a>00236 <span class="comment">/**</span>
<a name="l00237"></a>00237 <span class="comment"> * \defgroup comm_mgt Communicator Management</span>
<a name="l00238"></a>00238 <span class="comment"> * \ingroup comm_main</span>
<a name="l00239"></a>00239 <span class="comment"> **/</span>
<a name="l00240"></a>00240
<a name="l00241"></a>00241
<a name="l00242"></a>00242
<a name="l00243"></a>00243
<a name="l00244"></a>00244 <span class="preprocessor">#endif </span><span class="comment">/*MOC_DOXYGEN_MAIN_H_*/</span>
</pre></div></div>
<hr size="1"><address style="text-align: right;"><small>Generated on Thu Feb 12 13:49:47 2009 for MPI On Cell (Documentation) by&nbsp;
<a href="http://www.doxygen.org/index.html">
<img src="../../doxygen.png" alt="doxygen" align="middle" border="0"></a> 1.5.6 </small></address>
</body>
</html>