1 Customizing LLVMC: Reference Manual
2 ===================================
4 LLVMC is a generic compiler driver, designed to be customizable and
5 extensible. It plays the same role for LLVM as the ``gcc`` program
6 does for GCC - LLVMC's job is essentially to transform a set of input
7 files into a set of targets depending on configuration rules and user
8 options. What makes LLVMC different is that these transformation rules
9 are completely customizable - in fact, LLVMC knows nothing about the
10 specifics of transformation (even the command-line options are mostly
11 not hard-coded) and regards the transformation structure as an
12 abstract graph. This makes it possible to adapt LLVMC for other
13 purposes - for example, as a build tool for game resources.
15 Because LLVMC employs TableGen [1]_ as its configuration language, you
16 need to be familiar with it to customize LLVMC.
21 LLVMC tries hard to be as compatible with ``gcc`` as possible,
22 although there are some small differences. Most of the time, however,
23 you shouldn't be able to notice them::
25 $ # This works as expected:
26 $ llvmc2 -O3 -Wall hello.cpp
30 One nice feature of LLVMC is that one doesn't have to distinguish
31 between different compilers for different languages (think ``g++`` and
32 ``gcc``) - the right toolchain is chosen automatically based on input
33 language names (which are, in turn, determined from file
34 extensions). If you want to force files ending with ".c" to compile as
35 C++, use the ``-x`` option, just like you would do it with ``gcc``::
37 $ llvmc2 -x c hello.cpp
38 $ # hello.cpp is really a C file
42 On the other hand, when using LLVMC as a linker to combine several C++
43 object files you should provide the ``--linker`` option since it's
44 impossible for LLVMC to choose the right linker in that case::
48 [A lot of link-time errors skipped]
49 $ llvmc2 --linker=c++ hello.o
54 Customizing LLVMC: the compilation graph
55 ----------------------------------------
57 At the time of writing LLVMC does not support on-the-fly reloading of
58 configuration, so to customize LLVMC you'll have to recompile the
59 source code (which lives under ``$LLVM_DIR/tools/llvmc2``). The
60 default configuration files are ``Common.td`` (contains common
61 definitions, don't forget to ``include`` it in your configuration
62 files), ``Tools.td`` (tool descriptions) and ``Graph.td`` (compilation
65 To compile LLVMC with your own configuration file (say,``MyGraph.td``),
66 run ``make`` like this::
68 $ cd $LLVM_DIR/tools/llvmc2
69 $ make GRAPH=MyGraph.td TOOLNAME=my_llvmc
71 This will build an executable named ``my_llvmc``. There are also
72 several sample configuration files in the ``llvmc2/examples``
73 subdirectory that should help to get you started.
75 Internally, LLVMC stores information about possible source
76 transformations in form of a graph. Nodes in this graph represent
77 tools, and edges between two nodes represent a transformation path. A
78 special "root" node is used to mark entry points for the
79 transformations. LLVMC also assigns a weight to each edge (more on
80 this later) to choose between several alternative edges.
82 The definition of the compilation graph (see file ``Graph.td``) is
83 just a list of edges::
85 def CompilationGraph : CompilationGraph<[
86 Edge<root, llvm_gcc_c>,
87 Edge<root, llvm_gcc_assembler>,
90 Edge<llvm_gcc_c, llc>,
91 Edge<llvm_gcc_cpp, llc>,
94 OptionalEdge<llvm_gcc_c, opt, [(switch_on "opt")]>,
95 OptionalEdge<llvm_gcc_cpp, opt, [(switch_on "opt")]>,
98 OptionalEdge<llvm_gcc_assembler, llvm_gcc_cpp_linker,
99 (case (input_languages_contain "c++"), (inc_weight),
100 (or (parameter_equals "linker", "g++"),
101 (parameter_equals "linker", "c++")), (inc_weight))>,
106 As you can see, the edges can be either default or optional, where
107 optional edges are differentiated by sporting a ``case`` expression
108 used to calculate the edge's weight.
110 The default edges are assigned a weight of 1, and optional edges get a
111 weight of 0 + 2*N where N is the number of tests that evaluated to
112 true in the ``case`` expression. It is also possible to provide an
113 integer parameter to ``inc_weight`` and ``dec_weight`` - in this case,
114 the weight is increased (or decreased) by the provided value instead
117 When passing an input file through the graph, LLVMC picks the edge
118 with the maximum weight. To avoid ambiguity, there should be only one
119 default edge between two nodes (with the exception of the root node,
120 which gets a special treatment - there you are allowed to specify one
121 default edge *per language*).
123 To get a visual representation of the compilation graph (useful for
124 debugging), run ``llvmc2 --view-graph``. You will need ``dot`` and
125 ``gsview`` installed for this to work properly.
128 The 'case' expression
129 ---------------------
131 The 'case' construct can be used to calculate weights of the optional
132 edges and to choose between several alternative command line strings
133 in the ``cmd_line`` tool property. It is designed after the
134 similarly-named construct in functional languages and takes the form
135 ``(case (test_1), statement_1, (test_2), statement_2, ... (test_N),
136 statement_N)``. The statements are evaluated only if the corresponding
137 tests evaluate to true.
141 // Increases edge weight by 5 if "-A" is provided on the
142 // command-line, and by 5 more if "-B" is also provided.
144 (switch_on "A"), (inc_weight 5),
145 (switch_on "B"), (inc_weight 5))
147 // Evaluates to "cmdline1" if option "-A" is provided on the
148 // command line, otherwise to "cmdline2"
150 (switch_on "A"), ("cmdline1"),
151 (default), ("cmdline2"))
154 * Possible tests are:
156 - ``switch_on`` - Returns true if a given command-line option is
157 provided by the user. Example: ``(switch_on "opt")``. Note that
158 you have to define all possible command-line options separately in
159 the tool descriptions. See the next section for the discussion of
160 different kinds of command-line options.
162 - ``parameter_equals`` - Returns true if a command-line parameter equals
163 a given value. Example: ``(parameter_equals "W", "all")``.
165 - ``element_in_list`` - Returns true if a command-line parameter list
166 includes a given value. Example: ``(parameter_in_list "l", "pthread")``.
168 - ``input_languages_contain`` - Returns true if a given language
169 belongs to the current input language set. Example:
170 ```(input_languages_contain "c++")``.
172 - ``default`` - Always evaluates to true. Should always be the last
173 test in the ``case`` expression.
175 - ``and`` - A standard logical combinator that returns true iff all
176 of its arguments return true. Used like this: ``(and (test1),
177 (test2), ... (testN))``. Nesting of ``and`` and ``or`` is allowed,
180 - ``or`` - Another logical combinator that returns true only if any
181 one of its arguments returns true. Example: ``(or (test1),
182 (test2), ... (testN))``.
185 Writing a tool description
186 --------------------------
188 As was said earlier, nodes in the compilation graph represent tools,
189 which are described separately. A tool definition looks like this
190 (taken from the ``Tools.td`` file)::
192 def llvm_gcc_cpp : Tool<[
194 (out_language "llvm-assembler"),
195 (output_suffix "bc"),
196 (cmd_line "llvm-g++ -c $INFILE -o $OUTFILE -emit-llvm"),
200 This defines a new tool called ``llvm_gcc_cpp``, which is an alias for
201 ``llvm-g++``. As you can see, a tool definition is just a list of
202 properties; most of them should be self-explanatory. The ``sink``
203 property means that this tool should be passed all command-line
204 options that lack explicit descriptions.
206 The complete list of the currently implemented tool properties follows:
208 * Possible tool properties:
210 - ``in_language`` - input language name.
212 - ``out_language`` - output language name.
214 - ``output_suffix`` - output file suffix.
216 - ``cmd_line`` - the actual command used to run the tool. You can
217 use ``$INFILE`` and ``$OUTFILE`` variables, output redirection
218 with ``>``, hook invocations (``$CALL``), environment variables
219 (via ``$ENV``) and the ``case`` construct (more on this below).
221 - ``join`` - this tool is a "join node" in the graph, i.e. it gets a
222 list of input files and joins them together. Used for linkers.
224 - ``sink`` - all command-line options that are not handled by other
225 tools are passed to this tool.
227 The next tool definition is slightly more complex::
229 def llvm_gcc_linker : Tool<[
230 (in_language "object-code"),
231 (out_language "executable"),
232 (output_suffix "out"),
233 (cmd_line "llvm-gcc $INFILE -o $OUTFILE"),
235 (prefix_list_option "L", (forward),
236 (help "add a directory to link path")),
237 (prefix_list_option "l", (forward),
238 (help "search a library when linking")),
239 (prefix_list_option "Wl", (unpack_values),
240 (help "pass options to linker"))
243 This tool has a "join" property, which means that it behaves like a
244 linker. This tool also defines several command-line options: ``-l``,
245 ``-L`` and ``-Wl`` which have their usual meaning. An option has two
246 attributes: a name and a (possibly empty) list of properties. All
247 currently implemented option types and properties are described below:
249 * Possible option types:
251 - ``switch_option`` - a simple boolean switch, for example ``-time``.
253 - ``parameter_option`` - option that takes an argument, for example
256 - ``parameter_list_option`` - same as the above, but more than one
257 occurence of the option is allowed.
259 - ``prefix_option`` - same as the parameter_option, but the option name
260 and parameter value are not separated.
262 - ``prefix_list_option`` - same as the above, but more than one
263 occurence of the option is allowed; example: ``-lm -lpthread``.
266 * Possible option properties:
268 - ``append_cmd`` - append a string to the tool invocation command.
270 - ``forward`` - forward this option unchanged.
272 - ``output_suffix`` - modify the output suffix of this
273 tool. Example : ``(switch "E", (output_suffix "i")``.
275 - ``stop_compilation`` - stop compilation after this phase.
277 - ``unpack_values`` - used for for splitting and forwarding
278 comma-separated lists of options, e.g. ``-Wa,-foo=bar,-baz`` is
279 converted to ``-foo=bar -baz`` and appended to the tool invocation
282 - ``help`` - help string associated with this option. Used for
285 - ``required`` - this option is obligatory.
288 Hooks and environment variables
289 -------------------------------
291 Normally, LLVMC executes programs from the system ``PATH``. Sometimes,
292 this is not sufficient: for example, we may want to specify tool names
293 in the configuration file. This can be achieved via the mechanism of
294 hooks - to compile LLVMC with your hooks, just drop a .cpp file into
295 ``tools/llvmc2`` directory. Hooks should live in the ``hooks``
296 namespace and have the signature ``std::string hooks::MyHookName
297 (void)``. They can be used from the ``cmd_line`` tool property::
299 (cmd_line "$CALL(MyHook)/path/to/file -o $CALL(AnotherHook)")
301 It is also possible to use environment variables in the same manner::
303 (cmd_line "$ENV(VAR1)/path/to/file -o $ENV(VAR2)")
305 To change the command line string based on user-provided options use
306 the ``case`` expression (which we have already seen before)::
311 "llvm-g++ -E -x c $INFILE -o $OUTFILE",
313 "llvm-g++ -c -x c $INFILE -o $OUTFILE -emit-llvm"))
319 One last thing that you will need to modify when adding support for a
320 new language to LLVMC is the language map, which defines mappings from
321 file extensions to language names. It is used to choose the proper
322 toolchain(s) for a given input file set. Language map definition is
323 located in the file ``Tools.td`` and looks like this::
325 def LanguageMap : LanguageMap<
326 [LangToSuffixes<"c++", ["cc", "cp", "cxx", "cpp", "CPP", "c++", "C"]>,
327 LangToSuffixes<"c", ["c"]>,
335 .. [1] TableGen Fundamentals
336 http://llvm.cs.uiuc.edu/docs/TableGenFundamentals.html