Mercurial > hg > Papers > 2014 > masakoha-thesis > final

annotate slide/index.html @ 13:ff5c2b4ca18a

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
add slide
author Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
date Tue, 28 Jan 2014 00:48:21 +0900
parents
children 0e7c972b5ca1
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
13
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
1 <!DOCTYPE html>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
2
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
3 <html>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
4 <head>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5 <title>Presentation</title>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 <meta charset='utf-8'>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 <script
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 src='./slides.js'></script>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 </head>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
12 <style>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
13 /* Your individual styles here, or just use inline styles if that’s
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
14 what you want. */
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
15
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
16 </style>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 <body style='display: none'>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 <section class='slides layout-regular template-default'>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
21 <!--
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
22 Your slides (<article>s) go here. Delete or comment out the
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 slides below.
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 <article >
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 <h1>Cerium TaskManagerのGPGPUのサポート</h1>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27 <h3 class="title">Yuhi TOMARI　　21 Feb 2013</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28 <div align="right">担当教官 : 河野 真治</div>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 <article >
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 <h3>研究概要</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
32 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
33 当研究室ではCellおよびLinux、
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
34 Mac OSX上で動く並列プログラミングフレームワーク、
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 Ceriumの開発・改良を行っている
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
36 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 <img src='images/cerium_resource.png' style="height:300px">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
38 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
39 <font color="#ff3300">本研究では新たにGPU上での並列実行に対応</font>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 これによりヘテロジニアス(異種混合)環境下<br>でのプログラミングをサポートする
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
44
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
45 <!--
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
47 CeriumにOpenCLを組み込む事でGpu上での並列実行にも対応すると、
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
48 CPU・GPUそれぞれにtaskを割り振ることが可能となる。</p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
49 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
50 これにより、ヘテロジニアス(異種混合)な環境で
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
51 それぞれのアーキテクチャに適したプログラミングをサポートする。
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
52 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
53 -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
54
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
55 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
56 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
57 <h3>OpenCLによるGPGPUへの対応</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
58 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
59 OpenCLは
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60 ヘテロジニアス環境での並列計算を支援するフレームワーク
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
61 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
62 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
63 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
64 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 <td><img src='images/opencl.png' style="height:120px"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 <td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
67 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
68 OpenCLのtaskはkernelと呼ばれ、OpenCL Cという
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
69 Cに似た独自の言語で記述される
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71 </td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
73 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
74 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
75 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
76 <h3 class='yellow'>CPU/Cell上で実行する場合とGPU上で実行する場合、<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
77 それぞれほぼ同じ形式でkernelを記述できる</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
78 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
79 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
80 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81 <h3>Cerium Task Manager</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
82 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 <td><img src='images/createtask.png' style="height:350px"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
86 <td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
87 <ol>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88 <li>Taskを生成</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89 <li>依存関係のチェック</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90 <li>Schedulerに転送</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
91 <li>並列実行</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92 </ol>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93 </td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
95 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
96 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
97 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
98 CpuThreads、Schedulerに対応させる形でGpuThreadsとGpuSchedulerを作成した
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
99 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
100 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
101 <article class="nobackground">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
102 <h3>GPUのアーキテクチャ</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
103 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
104 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
105 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
106 <td><img src='images/gpu_arch.png' style="height:400px"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
107 <td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
108 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
109 GPU上のローカルメモリはCPUとメモリ空間が異なるのでメモリの共有ができない。
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
110 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
111 </td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
112 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
113 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
114 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
115 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
116 メモリのコピー処理やTaskを順にCommand Queueに
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
117 enqueueする。enqueueされた処理は並列実行される
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
118 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
119 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
120 <h3>GPU Task実行の流れ</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
121 <br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
122 <h3 class="yellow">kernel fileの記述</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
123 <pre>__kernel void // kernel.cl(kernel file)
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
124 twice(__global int *input_data,__global int *output_data) {
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
125 long count = (long)data_count[0];
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
126 for (int i = 0; i&lt count; i++) {
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
127 output_data[i] = input_data[i] * 2;;
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
128 }
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
129 }
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
130 </pre>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
131 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
132 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
133 <h3>GPU Task 実行の流れ</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
134 <br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
135 <h3 class="yellow">kernelをTaskとしてCeriumに登録</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
136 <pre>void
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
137 task_init(void) { // task_init.cc
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
138 GpuSchedRegister(Twice, "./twice.cl", "twice");
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
139 }</pre>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
140 <table border="2" style="font-size:18pt;">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
141 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
142 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
143 <td> 第1引数<br>Twice</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
144 <td >Taskのid。enumで定義されている<br>TaskManagerはこの値でTaskを識別する</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
145 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
146 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
147 <td> 第2引数<br>"./twice.cl"</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
148 <td>OpenCLが処理するkernelが記述されているファイルのパス</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
149 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
150 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
151 <td> 第3引数<br>"twice"</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
152 <td >関数の指定。kernel file内にある、実行する関数名を指定<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
153 Taskにあたる部分</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
154 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
155 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
156 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
157 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
158 <article class="nobackground">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
159 <h3>GPU Task 実行の流れ</h3><br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
160 <h3 class="yellow">GPU Task生成</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
161 <pre>// main.cc
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
162 HTaskPtr twice = manager->create_task(Twice);
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
163 twice->set_inData(0, indata, sizeof (int)*length);
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
164 // twice->set_inData(1, indata2, sizeof (int)*length);
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
165 twice->set_outData(0, outdata, sizeof (int)*length);
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
166 twice->set_cpu(GPU);
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
167 twice->spawn();
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
168 </pre>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
169 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
170
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
171
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
172 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
173 <h3>Cerium OpenCL API比較</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
174 <img src='images/api.png' style="height:500px">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
175 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
176 <article class="nobackground">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
177 <h3>ベンチマーク</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
178 <table >
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
179 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
180 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
181 <td> <!-- benchmark -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
182 <table border="2" style="font-size:18pt;">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
183 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
184 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
185 <td bgcolor="#8091B5"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
186 <td style="text-align: center;">Time</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
187 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
188 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
189 <td style="text-align: center;">1 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
190 <td style="text-align: right;"> 796 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
191 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
192 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
193 <td style="text-align: center;">2 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
194 <td style="text-align: right;"> 439 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
195 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
196 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
197 <td style="text-align: center;">6 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
198 <td style="text-align: right;"> 153 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
199 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
200 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
201 <td style="text-align: center;">12 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
202 <td style="text-align: right;"> 96 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
203 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
204 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
205 <td style="text-align: center;">24 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
206 <td style="text-align: right;"> 89 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
207 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
208 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
209 <td style="text-align: center;" bgcolor="#ffffcc">GPU(改良前)</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
210 <td style="text-align: right; " bgcolor="#ffffcc"> 330752 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
211 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
212 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
213 <td style="text-align: center;" bgcolor="#ffffcc">GPU(改良後)</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
214 <td style="text-align: right;" bgcolor="#ffffcc"> 5306 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
215 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
216 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
217 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
218 </td> <!-- /benchmark -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
219 <td> <!-- system env -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
220 <h3 class="yellow">10万入力によるBitonic Sort</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
221 <font size="5">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
222 <p>実験環境</p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
223 OS : MacOS 10.8.2<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
224 CPU : 2*2.66GHz 6-CoreIntel Xeon<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
225 Memory : 16GB<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
226 Compiler : Apple clang version 4.1 <br>　　　　　(based on LLVM 3.1svn)<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
227 GPU : AMD ATI Radeon HD 5870 1024MB<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
228 </font>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
229 <h3 class="yellow">結果</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
230 1coreのCPUよりも10倍遅い
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
231 </td> <!--system env -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
232 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
233 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
234 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
235 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
236 充分な性能が出なかったため、一度に送信する
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
237 data のサイズを増やす改善を行ったところ、
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
238 <font color="red">約60倍</font>程実行速度が向上した
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
239 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
240 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
241 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
242 <h3>考察</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
243 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
244 性能向上は見られたが、CPUと比べると未だ差が開いている
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
245 GPU向けに適切なチューニングが今後の課題となる
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
246 </p><br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
247 <h3 class="yellow">改善案</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
248 <ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
249 <li>データ並列によるkernelの実行</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
250 <li>同期機構の見直し</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
251 </ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
252 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
253 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
254 <h3>データ並列</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
255 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
256 データを2、3次元に分割し、分割した部分に対して並列処理する並列化手法。
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
257 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
258 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
259 OpenCL ではin/outするデータ郡をWork Itemと呼ぶ。
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
260 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
261 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
262 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
263 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
264 <td><img src='images/ndrange_arch.png' style="height:350px"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
265 <td>各Work Item のサイズを指定するとOpenCLがデータ並列で実行する。</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
266 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
267 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
268 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
269 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
270 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
271 <h3>同期機構</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
272 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
273 GpuSchedulerはCommand Queueの内部でパイプライン的に実行を行っている。
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
274 パイプラインを構成するには処理にwaitをかける必要がある。現在はclWaitForEvent APIを使用
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
275 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
276 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
277 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
278 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
279 <td bgcolor="#8091B5"><font color="white">API</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
280 <td>機能</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
281 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
282 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
283 <td bgcolor="#8091B5"><font color="white">clFlush()</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
284 <td>Command Queueに投入したTask全てをDeviceで実行する</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
285 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
286 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
287 <td bgcolor="#8091B5"><font color="white">clWaitForEvent()</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
288 <td>特定の処理の終了を待つ</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
289 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
290 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
291 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
292 <p>clFlushは実行は保証するが、<font color="red">終了は保証しない</font>仕様になっている</p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
293 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
294 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
295 <h3>新しい同期</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
296 <table border="0" cellpadding="0" cellspacing="0">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
297 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
298 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
299 <td bgcolor="#8091B5"><font color="white">FrameWork</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
300 <td>Dependency</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
301 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
302 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
303 <td bgcolor="#8091B5"><font color="white">Cerium</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
304 <td>Task Dependency</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
305 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
306 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
307 <td bgcolor="#8091B5"><font color="white">OpenCL</font></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
308 <td>Data Dependency</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
309 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
310 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
311 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
312 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
313 Task Dependency:Schedulerで依存関係が決定<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
314 Data Dependency:GPUに読み込まれた時に決定
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
315 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
316 <p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
317 GPGPUはなるべくGPU内部で処理を行う方が高速なため、性能向上が見込める
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
318 </p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
319 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
320 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
321 <h3>まとめ</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
322 <ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
323 <li>Cerium Task ManagerをGPGPUに対応</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
324 <li>同期機構の実装</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
325 <li>マルチコア実行とGPU実行のベンチマーク</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
326 </ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
327 <h3 class="yellow">今後の課題</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
328 <ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
329 <li>データ並列による実行のサポート</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
330 <li>同期機構の見直し</li>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
331 </ul>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
332 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
333 <article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
334 <h3>ベンチマーク</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
335 <table >
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
336 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
337 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
338 <td> <!-- benchmark -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
339 <table border="2" style="font-size:18pt;">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
340 <tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
341 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
342 <td bgcolor="#8091B5"></td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
343 <td style="text-align: center;">Time</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
344 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
345 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
346 <td style="text-align: center;">1 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
347 <td style="text-align: right;"> 67 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
348 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
349 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
350 <td style="text-align: center;">2 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
351 <td style="text-align: right;"> 34 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
352 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
353 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
354 <td style="text-align: center;">6 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
355 <td style="text-align: right;"> 12 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
356 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
357 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
358 <td style="text-align: center;">12 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
359 <td style="text-align: right;"> 9 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
360 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
361 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
362 <td style="text-align: center;">24 CPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
363 <td style="text-align: right;"> 6 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
364 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
365 <tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
366 <td style="text-align: center;" bgcolor="#ffffcc">GPU</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
367 <td style="text-align: right;" bgcolor="#ffffcc"> 10201 ms</td>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
368 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
369 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
370 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
371 </td> <!-- /benchmark -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
372 <td> <!-- system env -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
373 <h3 class="yellow">word count</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
374 <font size="5">
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
375 <p>10MBのテキストファイルを分割<br>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
376 各Taskがcountしていく</p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
377 <p>スペースと改行区切りでword countしていく</p>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
378 </font>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
379 <h3 class="yellow">結果</h3>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
380 CPUの方が150倍早い
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
381 </td> <!--system env -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
382 </tr>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
383 </tbody>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
384 </table>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
385 </article>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
386 <!--- <img src='images/flow_chart.jpg' width="300" height="500"> -->
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
387 </body>
ff5c2b4ca18a add slide
Masataka Kohagura <e085726@ie.u-ryukyu.ac.jp>
parents:
diff changeset
388 </html>