blob: 5ce4e9980cde32d8a954402a74c3af8a6629b5e1 [file] [log] [blame]
Ted Kremenek797a2472009-04-08 05:07:30 +00001<html>
2 <head>
3 <title>Pretokenized Headers (PTH)</title>
4 <link type="text/css" rel="stylesheet" href="../menu.css" />
5 <link type="text/css" rel="stylesheet" href="../content.css" />
6 <style type="text/css">
7 td {
8 vertical-align: top;
9 }
10 </style>
11</head>
12<body>
13
14<!--#include virtual="../menu.html.incl"-->
15
16<div id="content">
17<h1>Pretokenized Headers</h1>
18
19<p> <a href="http://en.wikipedia.org/wiki/Precompiled_header">Precompiled
Chris Lattnercdf59da2009-04-08 05:50:25 +000020headers</a> are a general approach employed by many compilers to reduce
21compilation time. The underlying motivation of the approach is that it is
22common for the same (and often large) header files to be included by
Ted Kremenek797a2472009-04-08 05:07:30 +000023multiple source files. Consequently, compile times can often be greatly improved
24by caching some of the (redundant) work done by a compiler to process headers.
Chris Lattnercdf59da2009-04-08 05:50:25 +000025Precompiled header files, which represent one of many ways to implement
Ted Kremenek797a2472009-04-08 05:07:30 +000026this optimization, are literally files that represent an on-disk cache that
27contains the vital information necessary to reduce some (or all) of the work
28needed to process a corresponding header file. While details of precompiled
29headers vary between compilers, precompiled headers have been shown to be a
30highly effective at speeding up program compilation on systems with very large
Chris Lattnercdf59da2009-04-08 05:50:25 +000031system headers (e.g., Mac OS/X).</p>
Ted Kremenek797a2472009-04-08 05:07:30 +000032
33<p>Clang supports an implementation of precompiled headers known as
34<em>pre-tokenized headers</em> (PTH). Clang's pre-tokenized headers support most
35of same interfaces as GCC's pre-compiled headers (as well as others) but are
Chris Lattnercdf59da2009-04-08 05:50:25 +000036completely different in their implementation. This first describes the
Ted Kremenek797a2472009-04-08 05:07:30 +000037interface for using PTH and then briefly elaborates on their design and
38implementation.</p>
39
40
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000041<h2>Using Pretokenized Headers with <tt>clang</tt></h2>
Ted Kremenek797a2472009-04-08 05:07:30 +000042
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000043<p>The high-level <tt>clang</tt> driver supports an interface to use PTH files
44that is similar to GCC's interface for precompiled headers.</p>
45
46<h3>Generating a PTH File</h3>
47
48<p>To generate a PTH file using <tt>clang</tt>, one invokes <tt>clang</tt> using
Ted Kremenek8e164082009-04-09 18:20:08 +000049the <b><tt>-x <i>&lt;language&gt;</i>-header</tt></b> option. This mirrors the
50interface in GCC for generating PCH files:</p>
Ted Kremenek797a2472009-04-08 05:07:30 +000051
52<pre>
53 $ gcc -x c-header test.h -o test.h.gch
54 $ clang -x c-header test.h -o test.h.pth
55</pre>
56
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000057<h3>Using a PTH File</h3>
58
59<p>A PTH file can then be used as a prefix header when a
60<b><tt>-include</tt></b> option is passed to <tt>clang</tt>:</p>
Ted Kremenek797a2472009-04-08 05:07:30 +000061
62<pre>
63 $ clang -include test.h test.c -o test
64</pre>
65
66<p>The <tt>clang</tt> driver will first check if a PTH file for <tt>test.h</tt>
67is available; if so, the contents of <tt>test.h</tt> (and the files it includes)
68will be processed from the PTH file. Otherwise, <tt>clang</tt> falls back to
69directly processing the content of <tt>test.h</tt>. This mirrors the behavior of
70GCC.</p>
71
72<p><b>NOTE:</b> <tt>clang</tt> does <em>not</em> automatically used PTH files
73for headers that are directly included within a source file. For example:</p>
74
75<pre>
76 $ clang -x c-header test.h -o test.h.pth
77 $ cat test.c
78 #include "test.h"
79 $ clang test.c -o test
80</pre>
81
82<p>In this example, <tt>clang</tt> will not automatically use the PTH file for
83<tt>test.h</tt> since <tt>test.h</tt> was included directly in the source file
84and not specified on the command line using <tt>-include</tt>.</p>
85
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000086<h2>Using Pretokenized Headers with <tt>clang-cc</tt> (Low-level Interface)</h2>
Ted Kremenek797a2472009-04-08 05:07:30 +000087
Chris Lattnercdf59da2009-04-08 05:50:25 +000088<p>The low-level Clang compiler tool, <tt>clang-cc</tt>, supports three command
89line options for generating and using PTH files.<p>
Ted Kremenek797a2472009-04-08 05:07:30 +000090
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000091<p>To generate PTH files using <tt>clang-cc</tt>, use the option
92<b><tt>-emit-pth</tt></b>:
93
94<pre> $ clang-cc test.h -emit-pth -o test.h.pth </pre>
Ted Kremenek797a2472009-04-08 05:07:30 +000095
96<p>This option is transparently used by <tt>clang</tt> when generating PTH
Ted Kremenekb7fd6b02009-04-09 18:17:39 +000097files. Similarly, PTH files can be used as prefix headers using the
98<b><tt>-include-pth</tt></b> option:</p>
Ted Kremenek797a2472009-04-08 05:07:30 +000099
100<pre>
101 $ clang-cc -include-pth test.h.pth test.c -o test.s
102</pre>
103
104<p>Alternatively, Clang's PTH files can be used as a raw &quot;token-cache&quot;
105(or &quot;content&quot; cache) of the source included by the original header
106file. This means that the contents of the PTH file are searched as substitutes
107for <em>any</em> source files that are used by <tt>clang-cc</tt> to process a
Ted Kremenekb7fd6b02009-04-09 18:17:39 +0000108source file. This is done by specifying the <b><tt>-token-cache</tt></b>
109option:</p>
Ted Kremenek797a2472009-04-08 05:07:30 +0000110
111<pre>
112 $ cat test.h
Chris Lattner0a069992009-04-08 06:00:32 +0000113 #include &lt;stdio.h&gt;
Ted Kremenek797a2472009-04-08 05:07:30 +0000114 $ clang-cc -emit-pth test.h -o test.h.pth
115 $ cat test.c
116 #include "test.h"
117 $ clang-cc test.c -o test -token-cache test.h.pth
118</pre>
119
120<p>In this example the contents of <tt>stdio.h</tt> (and the files it includes)
121will be retrieved from <tt>test.h.pth</tt>, as the PTH file is being used in
122this case as a raw cache of the contents of <tt>test.h</tt>. This is a low-level
123interface used to both implement the high-level PTH interface as well as to
124provide alternative means to use PTH-style caching.</p>
125
126<h2>PTH Design and Implementation</h2>
127
128<p>Unlike GCC's precompiled headers, which cache the full ASTs and preprocessor
129state of a header file, Clang's pretokenized header files mainly cache the raw
130lexer <em>tokens</em> that are needed to segment the stream of characters in a
131source file into keywords, identifiers, and operators. Consequently, PTH serves
132to mainly directly speed up the lexing and preprocessing of a source file, while
133parsing and type-checking must be completely redone every time a PTH file is
134used.</p>
135
136<h3>Basic Design Tradeoffs</h3>
137
138<p>In the long term there are plans to provide an alternate PCH implementation
139for Clang that also caches the work for parsing and type checking the contents
140of header files. The current implementation of PCH in Clang as pretokenized
141header files was motivated by the following factors:<p>
142
143<ul>
Ted Kremenek07f08d22009-04-09 18:03:21 +0000144
145<li><p><em>Language independence</em>: PTH files work with any language that
146Clang's lexer can handle, including C, Objective-C, and (in the early stages)
147C++. This means development on language features at the parsing level or above
148(which is basically almost all interesting pieces) does not require PTH to be
149modified.</p></li>
Ted Kremenek797a2472009-04-08 05:07:30 +0000150
151<li><em>Simple design</em>: Relatively speaking, PTH has a simple design and
152implementation, making it easy to test. Further, because the machinery for PTH
153resides at the lower-levels of the Clang library stack it is fairly
154straightforward to profile and optimize.</li>
155</ul>
156
157<p>Further, compared to GCC's PCH implementation (which is the dominate
158precompiled header file implementation that Clang can be directly compared
159against) the PTH design in Clang yields several attractive features:</p>
160
161<ul>
162
163<li><p><em>Architecture independence</em>: In contrast to GCC's PCH files (and
164those of several other compilers), Clang's PTH files are architecture
165independent, requiring only a single PTH file when building an program for
166multiple architectures.</p>
167
168<p>For example, on Mac OS X one may wish to
169compile a &quot;universal binary&quot; that runs on PowerPC, 32-bit Intel
170(i386), and 64-bit Intel architectures. In contrast, GCC requires a PCH file for
171each architecture, as the definitions of types in the AST are
172architecture-specific. Since a Clang PTH file essentially represents a lexical
173cache of header files, a single PTH file can be safely used when compiling for
174multiple architectures. This can also reduce compile times because only a single
175PTH file needs to be generated during a build instead of several.</p></li>
176
177<li><p><em>Reduced memory pressure</em>: Similar to GCC,
178Clang reads PTH files via the use of memory mapping (i.e., <tt>mmap</tt>).
179Clang, however, memory maps PTH files as read-only, meaning that multiple
180invocations of <tt>clang-cc</tt> can share the same pages in memory from a
181memory-mapped PTH file. In comparison, GCC also memory maps its PCH files but
182also modifies those pages in memory, incurring the copy-on-write costs. The
183read-only nature of PTH can greatly reduce memory pressure for builds involving
184multiple cores, thus improving overall scalability.</p></li>
185
Ted Kremenekb7fd6b02009-04-09 18:17:39 +0000186<li><p><em>Fast generation</em>: PTH files can be generated in a small fraction
Ted Kremenek07f08d22009-04-09 18:03:21 +0000187of the time needed to generate GCC's PCH files. Since PTH/PCH generation is a
188serial operation that typically blocks progress during a build, faster
189generation time leads to improved processor utilization with parallel builds on
190multicore machines.</p></li>
191
Ted Kremenek797a2472009-04-08 05:07:30 +0000192</ul>
193
194<p>Despite these strengths, PTH's simple design suffers some algorithmic
195handicaps compared to other PCH strategies such as those used by GCC. While PTH
196can greatly speed up the processing time of a header file, the amount of work
197required to process a header file is still roughly linear in the size of the
198header file. In contrast, the amount of work done by GCC to process a
199precompiled header is (theoretically) constant (the ASTs for the header are
200literally memory mapped into the compiler). This means that only the pieces of
201the header file that are referenced by the source file including the header are
202the only ones the compiler needs to process during actual compilation. While
203GCC's particular implementation of PCH mitigates some of these algorithmic
204strengths via the use of copy-on-write pages, the approach itself can
205fundamentally dominate at an algorithmic level, especially when one considers
206header files of arbitrary size.</p>
207
Ted Kremenek07f08d22009-04-09 18:03:21 +0000208<p>There are plans to potentially implement an complementary PCH implementation
209for Clang based on the lazy deserialization of ASTs. This approach would
210theoretically have the same constant-time algorithmic advantages just mentioned
211but would also retain some of the strengths of PTH such as reduced memory
212pressure (ideal for multi-core builds).</p>
Ted Kremenek797a2472009-04-08 05:07:30 +0000213
214<h3>Internal PTH Optimizations</h3>
215
216<p>While the main optimization employed by PTH is to reduce lexing time of
217header files by caching pre-lexed tokens, PTH also employs several other
218optimizations to speed up the processing of header files:</p>
219
220<ul>
221
222<li><p><em><tt>stat</tt> caching</em>: PTH files cache information obtained via
223calls to <tt>stat</tt> that <tt>clang-cc</tt> uses to resolve which files are
224included by <tt>#include</tt> directives. This greatly reduces the overhead
225involved in context-switching to the kernel to resolve included files.</p></li>
226
227<li><p><em>Fasting skipping of <tt>#ifdef</tt>...<tt>#endif</tt> chains</em>:
228PTH files record the basic structure of nested preprocessor blocks. When the
229condition of the preprocessor block is false, all of its tokens are immediately
230skipped instead of requiring them to be handled by Clang's
231preprocessor.</p></li>
232
233</ul>
234
235</div>
236</body>
237</html>