File: node65.html

package info (click to toggle)
lapack 3.0.20000531a-28
  • links: PTS
  • area: main
  • in suites: sarge
  • size: 61,920 kB
  • ctags: 46,200
  • sloc: fortran: 584,835; perl: 8,226; makefile: 2,331; awk: 71; sh: 45
file content (323 lines) | stat: -rw-r--r-- 9,278 bytes parent folder | download | duplicates (4)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2 Final//EN">
<!--Converted with LaTeX2HTML 98.2 beta6 (August 14th, 1998)
original version by:  Nikos Drakos, CBLU, University of Leeds
* revised and updated by:  Marcus Hennecke, Ross Moore, Herb Swan
* with significant contributions from:
  Jens Lippmann, Marek Rouchal, Martin Wilck and others -->
<HTML>
<HEAD>
<TITLE>The BLAS as the Key to Portability</TITLE>
<META NAME="description" CONTENT="The BLAS as the Key to Portability">
<META NAME="keywords" CONTENT="lug_l2h">
<META NAME="resource-type" CONTENT="document">
<META NAME="distribution" CONTENT="global">
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
<LINK REL="STYLESHEET" HREF="lug_l2h.css">
<LINK REL="next" HREF="node66.html">
<LINK REL="previous" HREF="node61.html">
<LINK REL="up" HREF="node60.html">
<LINK REL="next" HREF="node66.html">
</HEAD>
<BODY >
<!--Navigation Panel-->
<A NAME="tex2html5070"
 HREF="node66.html">
<IMG WIDTH="37" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="next"
 SRC="next_motif.png"></A> 
<A NAME="tex2html5064"
 HREF="node60.html">
<IMG WIDTH="26" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="up"
 SRC="up_motif.png"></A> 
<A NAME="tex2html5058"
 HREF="node64.html">
<IMG WIDTH="63" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="previous"
 SRC="previous_motif.png"></A> 
<A NAME="tex2html5066"
 HREF="node1.html">
<IMG WIDTH="65" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="contents"
 SRC="contents_motif.png"></A> 
<A NAME="tex2html5068"
 HREF="node152.html">
<IMG WIDTH="43" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="index"
 SRC="index_motif.png"></A> 
<BR>
<B> Next:</B> <A NAME="tex2html5071"
 HREF="node66.html">Block Algorithms and their</A>
<B> Up:</B> <A NAME="tex2html5065"
 HREF="node60.html">Performance of LAPACK</A>
<B> Previous:</B> <A NAME="tex2html5059"
 HREF="node64.html">Parallelism</A>
 &nbsp <B>  <A NAME="tex2html5067"
 HREF="node1.html">Contents</A></B> 
 &nbsp <B>  <A NAME="tex2html5069"
 HREF="node152.html">Index</A></B> 
<BR>
<BR>
<!--End of Navigation Panel-->

<H1><A NAME="SECTION03320000000000000000"></A><A NAME="secblasport"></A><A NAME="7605"></A>
<BR>
The BLAS as the Key to Portability
</H1>

<P>
How then can we hope to be able to achieve sufficient control over
vectorization, data movement, and parallelism in portable Fortran code,
to obtain the levels of performance that machines can offer?

<P>
The LAPACK strategy for combining
efficiency with portability<A NAME="7606"></A> is to construct the software
as much as possible
out of calls to the BLAS (Basic Linear Algebra Subprograms); the BLAS are
used
as building blocks.

<P>
The efficiency<A NAME="7607"></A> of LAPACK software depends on efficient
implementations
of the BLAS being provided by computer vendors (or others) for their
machines.
Thus the BLAS form a low-level interface between LAPACK software and
different machine architectures.
Above this level, almost all of the LAPACK software is truly portable.

<P>
There are now three levels of BLAS:

<P>
<DL>
<DT><STRONG>Level 1 BLAS [<A
 HREF="node151.html#blas1">78</A>]:</STRONG>
<DD>for vector operations, such as

<!-- MATH
 $y \leftarrow \alpha x + y$
 -->
<IMG
 WIDTH="92" HEIGHT="30" ALIGN="MIDDLE" BORDER="0"
 SRC="img207.png"
 ALT="$y \leftarrow \alpha x + y$">
<P>
<DT><STRONG>Level 2 BLAS [<A
 HREF="node151.html#blas2">42</A>]:</STRONG>
<DD>for matrix-vector operations, such as

<!-- MATH
 $y \leftarrow \alpha A x + \beta y$
 -->
<IMG
 WIDTH="116" HEIGHT="32" ALIGN="MIDDLE" BORDER="0"
 SRC="img208.png"
 ALT="$y \leftarrow \alpha A x + \beta y$">

<P>
<DT><STRONG>Level 3 BLAS [<A
 HREF="node151.html#blas3">40</A>]:</STRONG>
<DD>for matrix-matrix operations, such as

<!-- MATH
 $C \leftarrow \alpha A B + \beta C$
 -->
<IMG
 WIDTH="129" HEIGHT="32" ALIGN="MIDDLE" BORDER="0"
 SRC="img209.png"
 ALT="$C \leftarrow \alpha A B + \beta C$">

<P>
</DL>

<P>
Here, <B><I>A</I></B>, <B><I>B</I></B> and <B><I>C</I></B> are matrices, <B><I>x</I></B> and <B><I>y</I></B> are vectors, and <IMG
 WIDTH="16" HEIGHT="16" ALIGN="BOTTOM" BORDER="0"
 SRC="img49.png"
 ALT="$\alpha$">
and
<IMG
 WIDTH="15" HEIGHT="32" ALIGN="MIDDLE" BORDER="0"
 SRC="img50.png"
 ALT="$\beta$">
are scalars.

<P>
The Level 1 BLAS<A NAME="7613"></A> are used in LAPACK,
but for convenience rather than
for performance: they perform an insignificant fraction of the computation,
and they cannot achieve high efficiency on most modern supercomputers.

<P>
The Level 2 BLAS<A NAME="7614"></A> can achieve near-peak performance
on many vector processors,
such as a single processor of a CRAY Y-MP, CRAY C90, or CONVEX C4 machine.
However on other vector processors, such as a CRAY&nbsp;2,
or a RISC workstation or PC with one more levels of cache,
their performance is limited by the rate of data movement between different
levels of memory.

<P>
This limitation is overcome by the Level 3 BLAS<A NAME="7615"></A>,
which perform <B><I>O</I>(<I>n</I><SUP>3</SUP>)</B>
floating-point operations on <B><I>O</I>(<I>n</I><SUP>2</SUP>)</B> data, whereas the Level 2 BLAS
perform only <B><I>O</I>(<I>n</I><SUP>2</SUP>)</B> operations on <B><I>O</I>(<I>n</I><SUP>2</SUP>)</B> data.

<P>
The BLAS also allow us to exploit parallelism in a way that is transparent
to the software that calls them.
Even the Level 2 BLAS offer some scope for exploiting parallelism, but
greater
scope is provided by the Level 3 BLAS, as Table <A HREF="node65.html#tabsgiblas">3.1</A>
illustrates.

<P>
<BR>
<DIV ALIGN="CENTER">

<A NAME="tabsgiblas"></A>
<P>
<DIV ALIGN="CENTER">
(all matrices are of order 1000; <B><I>U</I></B> is upper triangular)
</DIV>
<P>
<DIV ALIGN="CENTER">
<BR>
</DIV>
<P>
<DIV ALIGN="CENTER"><A NAME="7618"></A>
<TABLE CELLPADDING=3 BORDER="1">
<CAPTION><STRONG>Table 3.1:</STRONG>
Speed in megaflops of Level 2 and Level 3 BLAS operations on an
SGI Origin 2000</CAPTION>
<TR><TD ALIGN="LEFT">Number of processors:</TD>
<TD ALIGN="RIGHT">1</TD>
<TD ALIGN="RIGHT">2</TD>
<TD ALIGN="RIGHT">4</TD>
<TD ALIGN="RIGHT">8</TD>
<TD ALIGN="RIGHT">16</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 2: 
<!-- MATH
 $y \leftarrow \alpha A x + \beta y$
 -->
<IMG
 WIDTH="116" HEIGHT="32" ALIGN="MIDDLE" BORDER="0"
 SRC="img208.png"
 ALT="$y \leftarrow \alpha A x + \beta y$"></TD>
<TD ALIGN="RIGHT">210</TD>
<TD ALIGN="RIGHT">154</TD>
<TD ALIGN="RIGHT">385</TD>
<TD ALIGN="RIGHT">493</TD>
<TD ALIGN="RIGHT">163</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 3: 
<!-- MATH
 $C \leftarrow \alpha A B + \beta C$
 -->
<IMG
 WIDTH="129" HEIGHT="32" ALIGN="MIDDLE" BORDER="0"
 SRC="img209.png"
 ALT="$C \leftarrow \alpha A B + \beta C$"></TD>
<TD ALIGN="RIGHT">555</TD>
<TD ALIGN="RIGHT">1021</TD>
<TD ALIGN="RIGHT">1864</TD>
<TD ALIGN="RIGHT">3758</TD>
<TD ALIGN="RIGHT">4200</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 2: 
<!-- MATH
 $x \leftarrow U x$
 -->
<IMG
 WIDTH="65" HEIGHT="16" ALIGN="BOTTOM" BORDER="0"
 SRC="img210.png"
 ALT="$x \leftarrow U x$"></TD>
<TD ALIGN="RIGHT">162</TD>
<TD ALIGN="RIGHT">152</TD>
<TD ALIGN="RIGHT">148</TD>
<TD ALIGN="RIGHT">153</TD>
<TD ALIGN="RIGHT">137</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 3: 
<!-- MATH
 $B \leftarrow U B$
 -->
<IMG
 WIDTH="74" HEIGHT="16" ALIGN="BOTTOM" BORDER="0"
 SRC="img211.png"
 ALT="$B \leftarrow U B$"></TD>
<TD ALIGN="RIGHT">518</TD>
<TD ALIGN="RIGHT">910</TD>
<TD ALIGN="RIGHT">1842</TD>
<TD ALIGN="RIGHT">3519</TD>
<TD ALIGN="RIGHT">5171</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 2: 
<!-- MATH
 $x \leftarrow U^{-1} x$
 -->
<IMG
 WIDTH="83" HEIGHT="19" ALIGN="BOTTOM" BORDER="0"
 SRC="img212.png"
 ALT="$x \leftarrow U^{-1} x$"></TD>
<TD ALIGN="RIGHT">172</TD>
<TD ALIGN="RIGHT">80</TD>
<TD ALIGN="RIGHT">140</TD>
<TD ALIGN="RIGHT">168</TD>
<TD ALIGN="RIGHT">17</TD>
</TR>
<TR><TD ALIGN="LEFT">Level 3: 
<!-- MATH
 $B \leftarrow U^{-1} B$
 -->
<IMG
 WIDTH="92" HEIGHT="19" ALIGN="BOTTOM" BORDER="0"
 SRC="img213.png"
 ALT="$B \leftarrow U^{-1} B$"></TD>
<TD ALIGN="RIGHT">519</TD>
<TD ALIGN="RIGHT">967</TD>
<TD ALIGN="RIGHT">1859</TD>
<TD ALIGN="RIGHT">3507</TD>
<TD ALIGN="RIGHT">5078</TD>
</TR>
</TABLE> 
</DIV>
</DIV>
<BR><HR>
<!--Navigation Panel-->
<A NAME="tex2html5070"
 HREF="node66.html">
<IMG WIDTH="37" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="next"
 SRC="next_motif.png"></A> 
<A NAME="tex2html5064"
 HREF="node60.html">
<IMG WIDTH="26" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="up"
 SRC="up_motif.png"></A> 
<A NAME="tex2html5058"
 HREF="node64.html">
<IMG WIDTH="63" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="previous"
 SRC="previous_motif.png"></A> 
<A NAME="tex2html5066"
 HREF="node1.html">
<IMG WIDTH="65" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="contents"
 SRC="contents_motif.png"></A> 
<A NAME="tex2html5068"
 HREF="node152.html">
<IMG WIDTH="43" HEIGHT="24" ALIGN="BOTTOM" BORDER="0" ALT="index"
 SRC="index_motif.png"></A> 
<BR>
<B> Next:</B> <A NAME="tex2html5071"
 HREF="node66.html">Block Algorithms and their</A>
<B> Up:</B> <A NAME="tex2html5065"
 HREF="node60.html">Performance of LAPACK</A>
<B> Previous:</B> <A NAME="tex2html5059"
 HREF="node64.html">Parallelism</A>
 &nbsp <B>  <A NAME="tex2html5067"
 HREF="node1.html">Contents</A></B> 
 &nbsp <B>  <A NAME="tex2html5069"
 HREF="node152.html">Index</A></B> 
<!--End of Navigation Panel-->
<ADDRESS>
<I>Susan Blackford</I>
<BR><I>1999-10-01</I>
</ADDRESS>
</BODY>
</HTML>