\@writefile{toc}{\author{Stephane Vialle}{}}
\@writefile{toc}{\author{Jens Gustedt}{}}
\@writefile{loa}{\addvspace {10\p@ }}
-\@writefile{toc}{\contentsline {chapter}{\numberline {6}Development methodologies for GPU and cluster of GPUs}{81}}
+\@writefile{toc}{\contentsline {chapter}{\numberline {6}Development methodologies for GPU and cluster of GPUs}{87}}
\@writefile{lof}{\addvspace {10\p@ }}
\@writefile{lot}{\addvspace {10\p@ }}
-\@writefile{toc}{\contentsline {section}{\numberline {6.1}Introduction}{82}}
-\newlabel{ch6:intro}{{6.1}{82}}
-\@writefile{toc}{\contentsline {section}{\numberline {6.2}General scheme of synchronous code with computation/communication overlapping in GPU clusters}{82}}
-\newlabel{ch6:part1}{{6.2}{82}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.1}Synchronous parallel algorithms on GPU clusters}{82}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.1}{\ignorespaces Native overlap of internode CPU communications with GPU computations.\relax }}{84}}
-\newlabel{fig:ch6p1overlapnative}{{6.1}{84}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.2}Native overlap of CPU communications and GPU computations}{84}}
-\newlabel{algo:ch6p1overlapnative}{{6.1}{85}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.1}Generic scheme implicitly overlapping MPI communications with CUDA GPU computations}{85}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.2}{\ignorespaces Overlap of internode CPU communications with a sequence of CPU/GPU data transfers and GPU computations.\relax }}{86}}
-\newlabel{fig:ch6p1overlapseqsequence}{{6.2}{86}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.3}Overlapping with sequences of transfers and computations}{86}}
-\newlabel{algo:ch6p1overlapseqsequence}{{6.2}{87}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.2}Generic scheme explicitly overlapping MPI communications with sequences of CUDA CPU/GPU transfers and CUDA GPU computations}{87}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.3}{\ignorespaces Overlap of internode CPU communications with a streamed sequence of CPU/GPU data transfers and GPU computations.\relax }}{88}}
-\newlabel{fig:ch6p1overlapstreamsequence}{{6.3}{88}}
-\newlabel{algo:ch6p1overlapstreamsequence}{{6.3}{89}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.3}Generic scheme explicitly overlapping MPI communications with streamed sequences of CUDA CPU/GPU transfers and CUDA GPU computations}{89}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.4}{\ignorespaces Complete overlap of internode CPU communications, CPU/GPU data transfers and GPU computations, interleaving computation-communication iterations\relax }}{91}}
-\newlabel{fig:ch6p1overlapinterleaved}{{6.4}{91}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.4}Interleaved communications-transfers-computations overlapping}{91}}
-\newlabel{algo:ch6p1overlapinterleaved}{{6.4}{92}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.4}Generic scheme explicitly overlapping MPI communications, CUDA CPU/GPU transfers and CUDA GPU computations, interleaving computation-communication iterations}{92}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.5}Experimental validation}{94}}
-\newlabel{ch6:p1expes}{{6.2.5}{94}}
-\newlabel{ch6:p1block-cyclic}{{6.2.5}{94}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.5}{\ignorespaces Experimental performances of different synchronous algorithms computing a dense matrix product\relax }}{95}}
-\newlabel{fig:ch6p1syncexpematrixprod}{{6.5}{95}}
-\@writefile{toc}{\contentsline {section}{\numberline {6.3}General scheme of asynchronous parallel code with computation/communication overlapping}{96}}
-\newlabel{ch6:part2}{{6.3}{96}}
-\@writefile{loa}{\contentsline {algorithm}{\numberline {3}{\ignorespaces Synchronous iterative scheme\relax }}{96}}
-\newlabel{algo:ch6p2sync}{{3}{96}}
-\@writefile{loa}{\contentsline {algorithm}{\numberline {4}{\ignorespaces Asynchronous iterative scheme\relax }}{96}}
-\newlabel{algo:ch6p2async}{{4}{96}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.1}A basic asynchronous scheme}{98}}
-\newlabel{ch6:p2BasicAsync}{{6.3.1}{98}}
-\newlabel{algo:ch6p2BasicAsync}{{6.5}{98}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.5}Initialization of the basic asynchronous scheme}{98}}
-\newlabel{algo:ch6p2BasicAsyncComp}{{6.6}{99}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.6}Computing function in the basic asynchronous scheme}{99}}
-\newlabel{algo:ch6p2BasicAsyncSendings}{{6.7}{100}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.7}Sending function in the basic asynchronous scheme}{100}}
-\newlabel{algo:ch6p2BasicAsyncReceptions}{{6.8}{101}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.8}Reception function in the basic asynchronous scheme}{101}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.2}Synchronization of the asynchronous scheme}{102}}
-\newlabel{ch6:p2SsyncOverAsync}{{6.3.2}{102}}
-\newlabel{algo:ch6p2Sync}{{6.9}{103}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.9}Initialization of the synchronized scheme}{103}}
-\newlabel{algo:ch6p2SyncComp}{{6.10}{104}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.10}Computing function in the synchronized scheme}{104}}
-\newlabel{algo:ch6p2SyncReceptions}{{6.11}{105}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.11}Reception function in the synchronized scheme}{105}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.3}Asynchronous scheme using MPI, OpenMP and CUDA}{106}}
-\newlabel{ch6:p2GPUAsync}{{6.3.3}{106}}
-\newlabel{algo:ch6p2AsyncSyncComp}{{6.12}{108}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.12}Computing function in the final asynchronous scheme}{108}}
-\newlabel{algo:ch6p2syncGPU}{{6.13}{109}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.13}Computing function in the final asynchronous scheme}{109}}
-\newlabel{algo:ch6p2FullOverAsyncMain}{{6.14}{111}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.14}Initialization of the main process of complete overlap with asynchronism}{111}}
-\newlabel{algo:ch6p2FullOverAsyncComp1}{{6.15}{112}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.15}Computing function in the final asynchronous scheme with CPU/GPU overlap}{112}}
-\newlabel{algo:ch6p2FullOverAsyncComp2}{{6.16}{113}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.16}Auxiliary computing function in the final asynchronous scheme with CPU/GPU overlap}{113}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.4}Experimental validation}{114}}
-\newlabel{sec:ch6p2expes}{{6.3.4}{114}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.6}{\ignorespaces Computation times of the test application in synchronous and asynchronous modes.\relax }}{115}}
-\newlabel{fig:ch6p2syncasync}{{6.6}{115}}
-\@writefile{lof}{\contentsline {figure}{\numberline {6.7}{\ignorespaces Computation times with or without overlap of Jacobian updatings in asynchronous mode.\relax }}{116}}
-\newlabel{fig:ch6p2aux}{{6.7}{116}}
-\@writefile{toc}{\contentsline {section}{\numberline {6.4}Perspective: A unifying programming model}{117}}
-\newlabel{sec:ch6p3unify}{{6.4}{117}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.1}Resources}{117}}
-\newlabel{sec:ch6p3resources}{{6.4.1}{117}}
-\newlabel{algo:ch6p3ORWLresources}{{6.17}{118}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.17}Declaration of ORWL resources for a block-cyclic matrix multiplication}{118}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.2}Control}{118}}
-\newlabel{sec:ch6p3ORWLcontrol}{{6.4.2}{118}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.3}Example: block-cyclic matrix multiplication (MM)}{119}}
-\newlabel{sec:ch6p3ORWLMM}{{6.4.3}{119}}
-\newlabel{algo:ch6p3ORWLBCCMM}{{6.18}{119}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.18}Block-cyclic matrix multiplication, high level per task view}{119}}
-\newlabel{algo:ch6p3ORWLlcopy}{{6.19}{120}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.19}An iterative local copy operation}{120}}
-\newlabel{algo:ch6p3ORWLrcopy}{{6.20}{120}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.20}An iterative remote copy operation as part of a block cyclic matrix multiplication task}{120}}
-\newlabel{algo:ch6p3ORWLtrans}{{6.21}{120}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.21}An iterative GPU transfer and compute operation as part of a block cyclic matrix multiplication task}{120}}
-\newlabel{algo:ch6p3ORWLdecl}{{6.22}{121}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.22}Dynamic declaration of handles to represent the resources}{121}}
-\newlabel{algo:ch6p3ORWLinit}{{6.23}{122}}
-\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.23}Dynamic initialization of access mode and priorities}{122}}
-\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.4}Tasks and operations}{122}}
-\newlabel{sec:ch6p3tasks}{{6.4.4}{122}}
-\@writefile{toc}{\contentsline {section}{\numberline {6.5}Conclusion}{123}}
-\newlabel{ch6:conclu}{{6.5}{123}}
-\@writefile{toc}{\contentsline {section}{\numberline {6.6}Glossary}{123}}
-\@writefile{toc}{\contentsline {section}{Bibliography}{124}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.1}Introduction}{88}}
+\newlabel{ch6:intro}{{6.1}{88}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.2}General scheme of synchronous code with computation/communication overlapping in GPU clusters}{88}}
+\newlabel{ch6:part1}{{6.2}{88}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.1}Synchronous parallel algorithms on GPU clusters}{88}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.1}{\ignorespaces Native overlap of internode CPU communications with GPU computations.\relax }}{90}}
+\newlabel{fig:ch6p1overlapnative}{{6.1}{90}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.2}Native overlap of CPU communications and GPU computations}{90}}
+\newlabel{algo:ch6p1overlapnative}{{6.1}{91}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.1}Generic scheme implicitly overlapping MPI communications with CUDA GPU computations}{91}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.2}{\ignorespaces Overlap of internode CPU communications with a sequence of CPU/GPU data transfers and GPU computations.\relax }}{92}}
+\newlabel{fig:ch6p1overlapseqsequence}{{6.2}{92}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.3}Overlapping with sequences of transfers and computations}{92}}
+\newlabel{algo:ch6p1overlapseqsequence}{{6.2}{93}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.2}Generic scheme explicitly overlapping MPI communications with sequences of CUDA CPU/GPU transfers and CUDA GPU computations}{93}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.3}{\ignorespaces Overlap of internode CPU communications with a streamed sequence of CPU/GPU data transfers and GPU computations.\relax }}{95}}
+\newlabel{fig:ch6p1overlapstreamsequence}{{6.3}{95}}
+\newlabel{algo:ch6p1overlapstreamsequence}{{6.3}{95}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.3}Generic scheme explicitly overlapping MPI communications with streamed sequences of CUDA CPU/GPU transfers and CUDA GPU computations}{95}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.4}Interleaved communications-transfers-computations overlapping}{97}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.4}{\ignorespaces Complete overlap of internode CPU communications, CPU/GPU data transfers and GPU computations, interleaving computation-communication iterations\relax }}{98}}
+\newlabel{fig:ch6p1overlapinterleaved}{{6.4}{98}}
+\newlabel{algo:ch6p1overlapinterleaved}{{6.4}{98}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.4}Generic scheme explicitly overlapping MPI communications, CUDA CPU/GPU transfers and CUDA GPU computations, interleaving computation-communication iterations}{98}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.2.5}Experimental validation}{100}}
+\newlabel{ch6:p1expes}{{6.2.5}{100}}
+\newlabel{ch6:p1block-cyclic}{{6.2.5}{100}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.5}{\ignorespaces Experimental performances of different synchronous algorithms computing a dense matrix product\relax }}{101}}
+\newlabel{fig:ch6p1syncexpematrixprod}{{6.5}{101}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.3}General scheme of asynchronous parallel code with computation/communication overlapping}{102}}
+\newlabel{ch6:part2}{{6.3}{102}}
+\@writefile{loa}{\contentsline {algocf}{\numberline {3}{\ignorespaces Synchronous iterative scheme\relax }}{102}}
+\newlabel{algo:ch6p2sync}{{3}{102}}
+\@writefile{loa}{\contentsline {algocf}{\numberline {4}{\ignorespaces Asynchronous iterative scheme\relax }}{103}}
+\newlabel{algo:ch6p2async}{{4}{103}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.1}A basic asynchronous scheme}{104}}
+\newlabel{ch6:p2BasicAsync}{{6.3.1}{104}}
+\newlabel{algo:ch6p2BasicAsync}{{6.5}{104}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.5}Initialization of the basic asynchronous scheme}{104}}
+\newlabel{algo:ch6p2BasicAsyncComp}{{6.6}{105}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.6}Computing function in the basic asynchronous scheme}{105}}
+\newlabel{algo:ch6p2BasicAsyncSendings}{{6.7}{107}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.7}Sending function in the basic asynchronous scheme}{107}}
+\newlabel{algo:ch6p2BasicAsyncReceptions}{{6.8}{107}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.8}Reception function in the basic asynchronous scheme}{107}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.2}Synchronization of the asynchronous scheme}{109}}
+\newlabel{ch6:p2SsyncOverAsync}{{6.3.2}{109}}
+\newlabel{algo:ch6p2Sync}{{6.9}{109}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.9}Initialization of the synchronized scheme}{109}}
+\newlabel{algo:ch6p2SyncComp}{{6.10}{110}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.10}Computing function in the synchronized scheme}{110}}
+\newlabel{algo:ch6p2SyncReceptions}{{6.11}{112}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.11}Reception function in the synchronized scheme}{112}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.3}Asynchronous scheme using MPI, OpenMP and CUDA}{113}}
+\newlabel{ch6:p2GPUAsync}{{6.3.3}{113}}
+\newlabel{algo:ch6p2AsyncSyncComp}{{6.12}{114}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.12}Computing function in the final asynchronous scheme}{114}}
+\newlabel{algo:ch6p2syncGPU}{{6.13}{115}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.13}Computing function in the final asynchronous scheme}{115}}
+\newlabel{algo:ch6p2FullOverAsyncMain}{{6.14}{118}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.14}Initialization of the main process of complete overlap with asynchronism}{118}}
+\newlabel{algo:ch6p2FullOverAsyncComp1}{{6.15}{119}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.15}Computing function in the final asynchronous scheme with CPU/GPU overlap}{119}}
+\newlabel{algo:ch6p2FullOverAsyncComp2}{{6.16}{120}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.16}Auxiliary computing function in the final asynchronous scheme with CPU/GPU overlap}{120}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.3.4}Experimental validation}{121}}
+\newlabel{sec:ch6p2expes}{{6.3.4}{121}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.6}{\ignorespaces Computation times of the test application in synchronous and asynchronous modes.\relax }}{122}}
+\newlabel{fig:ch6p2syncasync}{{6.6}{122}}
+\@writefile{lof}{\contentsline {figure}{\numberline {6.7}{\ignorespaces Computation times with or without overlap of Jacobian updatings in asynchronous mode.\relax }}{123}}
+\newlabel{fig:ch6p2aux}{{6.7}{123}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.4}Perspective: A unifying programming model}{124}}
+\newlabel{sec:ch6p3unify}{{6.4}{124}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.1}Resources}{124}}
+\newlabel{sec:ch6p3resources}{{6.4.1}{124}}
+\newlabel{algo:ch6p3ORWLresources}{{6.17}{125}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.17}Declaration of ORWL resources for a block-cyclic matrix multiplication}{125}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.2}Control}{125}}
+\newlabel{sec:ch6p3ORWLcontrol}{{6.4.2}{125}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.3}Example: block-cyclic matrix multiplication (MM)}{126}}
+\newlabel{sec:ch6p3ORWLMM}{{6.4.3}{126}}
+\newlabel{algo:ch6p3ORWLBCCMM}{{6.18}{126}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.18}Block-cyclic matrix multiplication, high level per task view}{126}}
+\newlabel{algo:ch6p3ORWLlcopy}{{6.19}{127}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.19}An iterative local copy operation}{127}}
+\newlabel{algo:ch6p3ORWLrcopy}{{6.20}{127}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.20}An iterative remote copy operation as part of a block cyclic matrix multiplication task}{127}}
+\newlabel{algo:ch6p3ORWLtrans}{{6.21}{127}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.21}An iterative GPU transfer and compute operation as part of a block cyclic matrix multiplication task}{127}}
+\newlabel{algo:ch6p3ORWLdecl}{{6.22}{128}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.22}Dynamic declaration of handles to represent the resources}{128}}
+\newlabel{algo:ch6p3ORWLinit}{{6.23}{128}}
+\@writefile{lol}{\contentsline {lstlisting}{\numberline {6.23}Dynamic initialization of access mode and priorities}{128}}
+\@writefile{toc}{\contentsline {subsection}{\numberline {6.4.4}Tasks and operations}{129}}
+\newlabel{sec:ch6p3tasks}{{6.4.4}{129}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.5}Conclusion}{130}}
+\newlabel{ch6:conclu}{{6.5}{130}}
+\@writefile{toc}{\contentsline {section}{\numberline {6.6}Glossary}{130}}
+\@writefile{toc}{\contentsline {section}{Bibliography}{131}}
\@setckpt{Chapters/chapter6/ch6}{
-\setcounter{page}{126}
+\setcounter{page}{133}
\setcounter{equation}{0}
\setcounter{enumi}{4}
\setcounter{enumii}{0}
\setcounter{enumiv}{21}
\setcounter{footnote}{0}
\setcounter{mpfootnote}{0}
-\setcounter{part}{1}
+\setcounter{part}{3}
\setcounter{chapter}{6}
\setcounter{section}{6}
\setcounter{subsection}{0}
\setcounter{lotdepth}{1}
\setcounter{lstnumber}{17}
\setcounter{ContinuedFloat}{0}
-\setcounter{float@type}{16}
-\setcounter{algorithm}{4}
-\setcounter{ALC@unique}{0}
-\setcounter{ALC@line}{0}
-\setcounter{ALC@rem}{0}
-\setcounter{ALC@depth}{0}
-\setcounter{AlgoLine}{0}
-\setcounter{algocfline}{0}
-\setcounter{algocfproc}{0}
-\setcounter{algocf}{0}
+\setcounter{AlgoLine}{6}
+\setcounter{algocfline}{4}
+\setcounter{algocfproc}{4}
+\setcounter{algocf}{4}
+\setcounter{nprt@mantissa@digitsbefore}{0}
+\setcounter{nprt@mantissa@digitsafter}{0}
+\setcounter{nprt@exponent@digitsbefore}{0}
+\setcounter{nprt@exponent@digitsafter}{0}
+\setcounter{nprt@digitsfirstblock}{0}
+\setcounter{nprt@blockcnt}{0}
+\setcounter{nprt@cntprint}{0}
\setcounter{proposition}{0}
\setcounter{theorem}{0}
\setcounter{exercise}{0}
