source: trunk/anuga_work/anuga_cuda/readme.html @ 9342

Last change on this file since 9342 was 9017, checked in by steve, 11 years ago

Adding in Zhe (John) Weng's anuga_cuda code as obtained from googlecode ​https://code.google.com/p/anuga-cuda
File size: 7.6 KB
RevLine 
[9017]1<!doctype html>
2<html>
3<head>
4<meta charset="utf-8">
5<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
6<style>
7h1,
8h2,
9h3,
10h4,
11h5,
12h6,
13p,
14blockquote {
15    margin: 0;
16    padding: 0;
17}
18body {
19    font-family: "Helvetica Neue", Helvetica, "Hiragino Sans GB", Arial, sans-serif;
20    font-size: 13px;
21    line-height: 18px;
22    color: #737373;
23    background-color: white;
24    margin: 10px 13px 10px 13px;
25}
26table {
27        margin: 10px 0 15px 0;
28        border-collapse: collapse;
29}
30td,th { 
31        border: 1px solid #ddd;
32        padding: 3px 10px;
33}
34th {
35        padding: 5px 10px;     
36}
37
38a {
39    color: #0069d6;
40}
41a:hover {
42    color: #0050a3;
43    text-decoration: none;
44}
45a img {
46    border: none;
47}
48p {
49    margin-bottom: 9px;
50}
51h1,
52h2,
53h3,
54h4,
55h5,
56h6 {
57    color: #404040;
58    line-height: 36px;
59}
60h1 {
61    margin-bottom: 18px;
62    font-size: 30px;
63}
64h2 {
65    font-size: 24px;
66}
67h3 {
68    font-size: 18px;
69}
70h4 {
71    font-size: 16px;
72}
73h5 {
74    font-size: 14px;
75}
76h6 {
77    font-size: 13px;
78}
79hr {
80    margin: 0 0 19px;
81    border: 0;
82    border-bottom: 1px solid #ccc;
83}
84blockquote {
85    padding: 13px 13px 21px 15px;
86    margin-bottom: 18px;
87    font-family:georgia,serif;
88    font-style: italic;
89}
90blockquote:before {
91    content:"\201C";
92    font-size:40px;
93    margin-left:-10px;
94    font-family:georgia,serif;
95    color:#eee;
96}
97blockquote p {
98    font-size: 14px;
99    font-weight: 300;
100    line-height: 18px;
101    margin-bottom: 0;
102    font-style: italic;
103}
104code, pre {
105    font-family: Monaco, Andale Mono, Courier New, monospace;
106}
107code {
108    background-color: #fee9cc;
109    color: rgba(0, 0, 0, 0.75);
110    padding: 1px 3px;
111    font-size: 12px;
112    -webkit-border-radius: 3px;
113    -moz-border-radius: 3px;
114    border-radius: 3px;
115}
116pre {
117    display: block;
118    padding: 14px;
119    margin: 0 0 18px;
120    line-height: 16px;
121    font-size: 11px;
122    border: 1px solid #d9d9d9;
123    white-space: pre-wrap;
124    word-wrap: break-word;
125}
126pre code {
127    background-color: #fff;
128    color:#737373;
129    font-size: 11px;
130    padding: 0;
131}
132sup {
133    font-size: 0.83em;
134    vertical-align: super;
135    line-height: 0;
136}
137* {
138        -webkit-print-color-adjust: exact;
139}
140@media screen and (min-width: 914px) {
141    body {
142        width: 854px;
143        margin:10px auto;
144    }
145}
146@media print {
147        body,code,pre code,h1,h2,h3,h4,h5,h6 {
148                color: black;
149        }
150        table, pre {
151                page-break-inside: avoid;
152        }
153}
154</style>
155<title>This is the GPU based ANUGA
156===========================
157
158
159
160Documentation
161-------------
162
163Documentation is under doc directory, and the html</title>
164<script type="text/x-mathjax-config">MathJax.Hub.Config({tex2jax:{inlineMath:[['$$$','$$$']]}});</script><script src="http://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
165</head>
166<body>
167<h1>This is the GPU based ANUGA </h1>
168
169<h2>Documentation</h2>
170
171<p>Documentation is under doc directory, and the html version generated by <strong>Sphinx</strong> under the directory <em>doc/sphinx/build/html/</em></p>
172
173<h2>Install Guide</h2>
174
175<ol>
176<li>Original <a href="anuga.anu.edu.au">ANUGA</a> is required</li>
177<li>For CUDA version, <a href="documen.tician.de/pycuda">PyCUDA</a> is required</li>
178<li>For OpenHMPP version, current implementation is based on the <a href="www.caps-entreprise.com">CAPS OpenHMPP Compiler</a>
179
180<ul>
181<li>When compiling the code with <strong>Makefile</strong>, the NVIDIA device architecture and compute capability need to be specified
182
183<ul>
184<li>For example, GTX480 with 2.0 compute capability </br>
185<code>HMPP_FLAGS13 = -e --nvcc-options -Xptxas=-v,-arch=sm_20 -c --force</code></li>
186<li>For GTX680 with 3.0 compute capability </br>
187<code>HMPP_FLAGS13 = -e --nvcc-options -Xptxas=-v,-arch=sm_30 -c --force</code></li>
188</ul>
189</li>
190<li>Also the path for <strong>python</strong>, <strong>numpy</strong>, and <strong>ANUGA/utilities</strong> packages need to be specified</li>
191<li>Defining the macro <strong>USING_MIRROR_DATA</strong> in <strong>hmpp_fun.h</strong> </br>
192<code>#define USING_MIRROR_DATA</code>
193</br>will enable the advanced version, which uses OpenHMPP Mirrored Data technology so that data transmission costs can be effectively cut down, otherwise basic version is enabled.</li>
194</ul>
195</li>
196</ol>
197
198
199<h2>Environment Vars</h2>
200
201<p>Please add following vars to you <strong>.bashrc</strong> or <strong>.bash_profile</strong> file</p>
202
203<pre><code>export ANUGA_CUDA=/where_the_anuga-cuda/src
204
205export $PYTHONPATH=$PYTHONPATH:$ANUGA_CUDA
206</code></pre>
207
208<h2>Basic Code Structure</h2>
209
210<p><img src="docs/codeStructure.pdf" title="anuga-cuda code structure" alt="Code Structure" /></p>
211
212<ul>
213<li><strong>README</strong> (What you are reading)</li>
214<li><strong>docs/</strong> Documentation directory
215
216<ul>
217<li><strong>codeStructure.pdf</strong> The diagram above</li>
218<li><strong>Evolve workflow.pdf</strong> The overall workflow of the evolve procedure. This includes all the function dependency and function interfaces, which is helpful to understand the evolve procedure of ANUGA</li>
219<li><strong>device_spe/</strong> Some device specifications of our working station</li>
220<li><strong>profiling/</strong> Profiling results
221
222<ul>
223<li><strong>CUDA/</strong> All the profiling results on CUDA implementation</li>
224</ul>
225</li>
226<li><strong>sphinx/</strong> The <a href="http://sphinx-doc.org/">Sphinx</a> generated documentation
227
228<ul>
229<li><strong>source/</strong> The source files for Sphinx based documents</li>
230<li><strong>build/</strong> The generated documents
231
232<ul>
233<li><strong>html/</strong> HTML version documentation</li>
234</ul>
235</li>
236</ul>
237</li>
238</ul>
239</li>
240<li><strong>src/</strong> Source code directory
241
242<ul>
243<li><strong>anuga_cuda/</strong> The CUDA implementation
244
245<ul>
246<li><strong>config.py</strong> Detail configuration for the CUDA implementation, including the path for all the kernel functions, optimal CUDA thread block configuration, etc.</li>
247<li><strong>gpu_domain_advanced.py</strong> Python Class for CUDA implementation in advanced version</li>
248<li><strong>gpu_domain_basic.py</strong> Python Class for CUDA implementation in basic version</li>
249</ul>
250</li>
251<li><strong>anuga_HMPP/</strong> The OpenHMPP implementation
252
253<ul>
254<li><strong>Makefile</strong> The Makefile</li>
255<li><strong>hmpp_dimain.py</strong> Python Class for OpenHMPP implementation</li>
256<li><strong>hmpp_python_glue.c</strong> The Python/C API to set up communication between Python ANUGA and OpenHMPP.</li>
257<li><strong>sw_domain.h</strong> The C Struct type <strong>domain</strong> used in C implementation to access mesh information generated in Python ANUGA</li>
258<li><strong>sw_domain_fun.h</strong> Connect C Struct type <strong>domain</strong> to all mesh information</li>
259<li><strong>hmpp_fun.h</strong> All function declarations.</li>
260<li><strong>evolve.c</strong> The evolve procedure</li>
261</ul>
262</li>
263<li><strong>scripts/</strong> Some useful bash script</li>
264<li><strong>utilities/</strong> Utilities for sorting mesh information, checking results, etc.</li>
265</ul>
266</li>
267<li><strong>test/</strong> Testing cases
268
269<ul>
270<li><strong>CUDA/</strong> Testing cases for CUDA implementation
271
272<ul>
273<li><strong>merimbula/</strong> Merimbula testing case directory
274
275<ul>
276<li><strong>merimbula.py</strong> Merimbula testing case</li>
277</ul>
278</li>
279</ul>
280</li>
281<li><strong>OpenHMPP/</strong> Testing cases for OpenHMPP implementation
282
283<ul>
284<li><strong>merimbula.py</strong> Merimbula testing case</li>
285</ul>
286</li>
287</ul>
288</li>
289</ul>
290
291
292<h2>Examples</h2>
293
294<p>Running Merimbula model with CUDA implementation.</p>
295
296<p><code>$ python merimbula.py -gpu</code></p>
297
298<p>With pair-testing.</p>
299
300<p><code>$ python merimbula.py -gpu -test</code></p>
301
302<p>With rearranged mesh information.</p>
303
304<p><code>$ python merimbula.py -gpu -rg</code></p>
305
306<h2>Author</h2>
307
308<p>Mail to <a href="wengcsyz@gmail.com">Zhe Weng (John)</a></p>
309</body>
310</html>
Note: See TracBrowser for help on using the repository browser.