4 <nidx>GHC, using</nidx>
7 GHC is a command-line compiler: in order to compile a Haskell program,
8 GHC must be invoked on the source file(s) by typing a command to the
9 shell. The steps involved in compiling a program can be automated
10 using the @make@ tool (this is especially useful if the program
11 consists of multiple source files which depend on each other). This
12 section describes how to use GHC from the command-line.
14 %************************************************************************
16 <sect1> Overall command-line structure
17 <label id="command-line-structure">
19 <nidx>structure, command-line</nidx>
20 <nidx>command-line structure</nidx>
22 %************************************************************************
24 An invocation of GHC takes the following form:
30 Command-line arguments are either options or file names.
32 Command-line options begin with @-@. They may <em>not</em> be
33 grouped: @-vO@ is different from @-v -O@. Options need not
34 precede filenames: e.g., @ghc *.o -o foo@. All options are
35 processed and then applied to all files; you cannot, for example, invoke
36 @ghc -c -O1 Foo.hs -O2 Bar.hs@ to apply different optimisation
37 levels to the files @Foo.hs@ and @Bar.hs@. For conflicting
38 options, e.g., @-c -S@, we reserve the right to do anything we
39 want. (Usually, the last one applies.)
41 %************************************************************************
43 <sect1>Meaningful file suffixes
44 <label id="file-suffixes">
46 <nidx>suffixes, file</nidx>
47 <nidx>file suffixes for GHC</nidx>
49 %************************************************************************
51 File names with ``meaningful'' suffixes (e.g., @.lhs@ or @.o@)
52 cause the ``right thing'' to happen to those files.
56 <nidx>lhs suffix</nidx>
57 A ``literate Haskell'' module.
60 A not-so-literate Haskell module.
63 A Haskell interface file, probably compiler-generated.
66 Intermediate C file produced by the Haskell compiler.
69 A C~file not produced by the Haskell compiler.
72 % C code after it has be preprocessed by the C compiler (using the
76 An assembly-language source file, usually
77 produced by the compiler.
80 An object file, produced by an assembler.
83 Files with other suffixes (or without suffixes) are passed straight
86 %************************************************************************
88 <sect1>Help and verbosity options
89 <label id="options-help">
91 <nidx>help options (GHC)</nidx>
92 <nidx>verbose option (GHC)</nidx>
94 %************************************************************************
96 A good option to start with is the @-help@ (or @-?@) option.
97 <nidx>-help option</nidx>
98 <nidx>-? option</nidx>
99 GHC spews a long message to standard output and then exits.
101 The @-v@<nidx>-v option</nidx> option makes GHC <em>verbose</em>: it
102 reports its version number and shows (on stderr) exactly how it invokes each
103 phase of the compilation system. Moreover, it passes
104 the @-v@ flag to most phases; each reports
105 its version number (and possibly some other information).
107 Please, oh please, use the @-v@ option when reporting bugs!
108 Knowing that you ran the right bits in the right order is always the
109 first thing we want to verify.
111 If you're just interested in the compiler version number, the
112 @--version@<nidx>--version option</nidx> option prints out a
113 one-line string containing the requested info.
115 %************************************************************************
117 <sect1>Running the right phases in the right order
118 <label id="options-order">
120 <nidx>order of passes in GHC</nidx>
121 <nidx>pass ordering in GHC</nidx>
123 %************************************************************************
125 The basic task of the @ghc@ driver is to run each input file
126 through the right phases (compiling, linking, etc.).
128 The first phase to run is determined by the input-file suffix, and the
129 last phase is determined by a flag. If no relevant flag is present,
130 then go all the way through linking. This table summarises:
133 Phase of the | Suffix saying | Flag saying | (suffix of) @@
134 compilation system | ``start here''| ``stop after''| output file @@
136 literate pre-processor | .lhs | - | - @@
137 C pre-processor (opt.) | - | - | - @@
138 Haskell compiler | .hs | -C, -S | .hc, .s @@
139 C compiler (opt.) | .hc or .c | -S | .s @@
140 assembler | .s | -c | .o @@
141 linker | other | - | a.out @@
143 <nidx>-C option</nidx>
144 <nidx>-S option</nidx>
145 <nidx>-c option</nidx>
147 Thus, a common invocation would be: @ghc -c Foo.hs@
149 Note: What the Haskell compiler proper produces depends on whether a
150 native-code generator is used (producing assembly language) or not
153 The option @-cpp@<nidx>-cpp option</nidx> must be given for the C
154 pre-processor phase to be run, that is, the pre-processor will be run
155 over your Haskell source file before continuing.
157 The option @-E@<nidx>-E option</nidx> runs just the pre-processing
158 passes of the compiler, outputting the result on stdout before
159 stopping. If used in conjunction with -cpp, the output is the
160 code blocks of the original (literal) source after having put it
161 through the grinder that is the C pre-processor. Sans @-cpp@, the
162 output is the de-litted version of the original source.
164 The option @-optcpp-E@<nidx>-optcpp-E option</nidx> runs just the
165 pre-processing stage of the C-compiling phase, sending the result to
166 stdout. (For debugging or obfuscation contests, usually.)
168 %************************************************************************
170 <sect1>Re-directing the compilation output(s)
171 <label id="options-output">
173 <nidx>output-directing options</nidx>
175 %************************************************************************
177 GHC's compiled output normally goes into a @.hc@, @.o@, etc., file,
178 depending on the last-run compilation phase. The option @-o
179 foo@<nidx>-o option</nidx> re-directs the output of that last-run
182 Note: this ``feature'' can be counterintuitive:
183 @ghc -C -o foo.o foo.hs@ will put the intermediate C code in the
184 file @foo.o@, name notwithstanding!
186 EXOTICA: But the @-o@ option isn't of much use if you have
187 <em>several</em> input files... Non-interface output files are
188 normally put in the same directory as their corresponding input file
189 came from. You may specify that they be put in another directory
190 using the @-odir <dir>@<nidx>-odir <dir> option</nidx> (the
191 ``Oh, dear'' option). For example:
194 % ghc -c parse/Foo.hs parse/Bar.hs gurgle/Bumble.hs -odir `arch`
197 The output files, @Foo.o@, @Bar.o@, and @Bumble.o@ would be
198 put into a subdirectory named after the architecture of the executing
199 machine (@sun4@, @mips@, etc). The directory must already
200 exist; it won't be created.
202 Note that the @-odir@ option does <em>not</em> affect where the
203 interface files are put. In the above example, they would still be
204 put in @parse/Foo.hi@, @parse/Bar.hi@, and @gurgle/Bumble.hi@.
206 MORE EXOTICA: The @-osuf <suffix>@<nidx>-osuf <suffix>
207 option</nidx> will change the @.o@ file suffix for object files to
208 whatever you specify. (We use this in compiling the prelude.).
209 Similarly, the @-hisuf <suffix>@<nidx>-hisuf <suffix>
210 option</nidx> will change the @.hi@ file suffix for non-system
211 interface files (see Section <ref name="Other options related to
212 interface files" id="hi-options">).
214 The @-hisuf@/@-osuf@ game is useful if you want to compile a program
215 with both GHC and HBC (say) in the same directory. Let HBC use the
216 standard @.hi@/@.o@ suffixes; add @-hisuf g_hi -osuf g_o@ to your
217 @make@ rule for GHC compiling...
219 FURTHER EXOTICA: If you are doing a normal @.hs@-to-@.o@ compilation
220 but would like to hang onto the intermediate @.hc@ C file, just
221 throw in a @-keep-hc-file-too@ option<nidx>-keep-hc-file-too option</nidx>.
222 If you would like to look at the assembler output, toss in a
223 @-keep-s-file-too@,<nidx>-keep-s-file-too option</nidx> too.
225 <sect2> Saving GHC's standard error output
226 <label id="saving-ghc-stderr">
228 <nidx>standard error, saving</nidx>
230 Sometimes, you may cause GHC to be rather chatty on standard error;
231 with @-v@, for example. You can instruct GHC to <em>append</em> this
232 output to a particular log file with a @-odump <blah>@<nidx>-odump
233 <blah> option</nidx> option.
235 <sect2> Redirecting temporary files
236 <label id="temp-files">
238 <nidx>temporary files, redirecting</nidx>
240 If you have trouble because of running out of space in @/tmp@ (or
241 wherever your installation thinks temporary files should go), you may
242 use the @-tmpdir <dir>@<nidx>-tmpdir <dir> option</nidx> option
243 to specify an alternate directory. For example, @-tmpdir .@ says to
244 put temporary files in the current working directory.
246 Alternatively, use your @TMPDIR@ environment variable.<nidx>TMPDIR
247 environment variable</nidx> Set it to the name of the directory where
248 temporary files should be put. GCC and other programs will honour the
249 @TMPDIR@ variable as well.
251 Even better idea: Set the @TMPDIR@ variable when building GHC, and
252 never worry about @TMPDIR@ again. (see the build documentation).
254 %************************************************************************
256 <sect1>Warnings and sanity-checking
257 <label id="options-sanity">
259 <nidx>sanity-checking options</nidx>
260 <nidx>warnings</nidx>
262 %************************************************************************
264 GHC has a number of options that select which types of non-fatal error
265 messages, otherwise known as warnings, can be generated during
266 compilation. By default, you get a standard set of warnings which are
267 generally likely to indicate bugs in your program. These are:
268 @-fwarn-overlpapping-patterns@, @-fwarn-duplicate-exports@, and
269 @-fwarn-missing-methods@. The following flags are simple ways to
270 select standard ``packages'' of warnings:
275 <nidx>-Wnot option</nidx>
277 Turns off all warnings, including the standard ones.
280 <nidx>-w option</nidx>
285 <nidx>-W option</nidx>
287 Provides the standard warnings plus @-fwarn-incomplete-patterns@,
288 @-fwarn-unused-imports@ and @-fwarn-unused-binds@.
291 <nidx>-Wall option</nidx>
293 Turns on all warning options.
297 The full set of warning options is described below. To turn off any
298 warning, simply give the corresponding @-fno-warn-...@ option on
303 <tag>@-fwarn-name-shadowing@:</tag>
304 <nidx>-fwarn-name-shadowing option</nidx>
305 <nidx>shadowing, warning</nidx>
307 This option causes a warning to be emitted whenever an inner-scope
308 value has the same name as an outer-scope value, i.e. the inner value
309 shadows the outer one. This can catch typographical errors that turn
310 into hard-to-find bugs, e.g., in the inadvertent cyclic definition
311 @let x = ... x ... in@.
313 Consequently, this option does <em>not</em> allow cyclic recursive
316 <tag>@-fwarn-overlapping-patterns@:</tag>
317 <nidx>-fwarn-overlapping-patterns option</nidx>
318 <nidx>overlapping patterns, warning</nidx>
319 <nidx>patterns, overlapping</nidx>
321 By default, the compiler will warn you if a set of patterns are
331 where the last pattern match in @f@ won't ever be reached, as the
332 second pattern overlaps it. More often than not, redundant patterns
333 is a programmer mistake/error, so this option is enabled by default.
335 <tag>@-fwarn-incomplete-patterns@:</tag>
336 <nidx>-fwarn-incomplete-patterns option</nidx>
337 <nidx>incomplete patterns, warning</nidx>
338 <nidx>patterns, incomplete</nidx>
340 Similarly for incomplete patterns, the function @g@ below will fail
341 when applied to non-empty lists, so the compiler will emit a warning
342 about this when @-fwarn-incomplete-patterns@ is enabled.
348 This option isn't enabled be default because it can be a bit noisy,
349 and it doesn't always indicate a bug in the program. However, it's
350 generally considered good practice to cover all the cases in your
353 <tag>@-fwarn-missing-methods@:</tag>
354 <nidx>-fwarn-missing-methods option</nidx>
355 <nidx>missing methods, warning</nidx>
356 <nidx>methods, missing</nidx>
358 This option is on by default, and warns you whenever an instance
359 declaration is missing one or more methods, and the corresponding
360 class declaration has no default declaration for them.
362 <tag>@-fwarn-unused-imports@:</tag>
363 <nidx>-fwarn-unused-imports option</nidx>
364 <nidx>unused imports, warning</nidx>
365 <nidx>imports, unused</nidx>
367 Report any objects that are explicitly imported but never used.
369 <tag>@-fwarn-unused-binds@:</tag>
370 <nidx>-fwarn-unused-binds option</nidx>
371 <nidx>unused binds, warning</nidx>
372 <nidx>binds, unused</nidx>
374 Report any function definitions (and local bindings) which are unused.
375 For top-level functions, the warning is only given if the binding is
378 <tag>@-fwarn-unused-matches@:</tag>
379 <nidx>-fwarn-unused-matches option</nidx>
380 <nidx>unused matches, warning</nidx>
381 <nidx>matches, unused</nidx>
383 Report all unused variables which arise from pattern matches,
384 including patterns consisting of a single variable. For instance @f x
385 y = []@ would report @x@ and @y@ as unused. To eliminate the warning,
386 all unused variables can be replaced with wildcards.
388 <tag>@-fwarn-duplicate-exports@:</tag>
389 <nidx>-fwarn-duplicate-exports option</nidx>
390 <nidx>duplicate exports, warning</nidx>
391 <nidx>export lists, duplicates</nidx>
393 Have the compiler warn about duplicate entries in export lists. This
394 is useful information if you maintain large export lists, and want to
395 avoid the continued export of a definition after you've deleted (one)
396 mention of it in the export list.
398 This option is on by default.
400 <tag><tt>-fwarn-type-defaults</tt>:</tag>
401 <nidx>-fwarn-type-defaults option</nidx>
402 <nidx>defaulting mechanism, warning</nidx>
404 Have the compiler warn/inform you where in your source the Haskell
405 defaulting mechanism for numeric types kicks in. This is useful
406 information when converting code from a context that assumed one
407 default into one with another, e.g., the 'default default' for Haskell
408 1.4 caused the otherwise unconstrained value <tt>1</tt> to be given
409 the type <tt>Int</tt>, whereas Haskell 98 defaults it to
410 <tt>Integer</tt>. This may lead to differences in performance and
411 behaviour, hence the usefulness of being non-silent about this.
413 This warning is off by default.
415 <tag>@-fwarn-missing-signatures@:</tag>
416 <nidx>-fwarn-missing-signatures option</nidx>
417 <nidx>type signatures, missing</nidx>
419 If you would like GHC to check that every top-level function/value has
420 a type signature, use the @-fwarn-missing-signatures@ option. This
421 option is off by default.
425 If you're feeling really paranoid, the @-dcore-lint@
426 option<nidx>-dcore-lint option</nidx> is a good choice. It turns on
427 heavyweight intra-pass sanity-checking within GHC. (It checks GHC's
430 %************************************************************************
432 <sect1>Separate compilation
433 <label id="separate-compilation">
435 <nidx>separate compilation</nidx>
436 <nidx>recompilation checker</nidx>
437 <nidx>make and recompilation</nidx>
439 %************************************************************************
441 This section describes how GHC supports separate compilation.
443 <sect2>Interface files
444 <label id="hi-files">
446 <nidx>interface files</nidx>
447 <nidx>.hi files</nidx>
449 When GHC compiles a source file @F@ which contains a module @A@, say,
450 it generates an object @F.o@, <em>and</em> a companion <em>interface
451 file</em> @A.hi@. The interface file is not intended for human
452 consumption, as you'll see if you take a look at one. It's merely
453 there to help the compiler compile other modules in the same program.
455 NOTE: Having the name of the interface file follow the module name and
456 not the file name, means that working with tools such as @make(1)@
457 become harder. @make@ implicitly assumes that any output files
458 produced by processing a translation unit will have file names that
459 can be derived from the file name of the translation unit. For
460 instance, pattern rules becomes unusable. For this reason, we
461 recommend you stick to using the same file name as the module name.
463 The interface file for @A@ contains information needed by the compiler
464 when it compiles any module @B@ that imports @A@, whether directly or
465 indirectly. When compiling @B@, GHC will read @A.hi@ to find the
466 details that it needs to know about things defined in @A@.
468 Furthermore, when compiling module @C@ which imports @B@, GHC may
469 decide that it needs to know something about @A@ --- for example, @B@
470 might export a function that involves a type defined in @A@. In this
471 case, GHC will go and read @A.hi@ even though @C@ does not explicitly
474 The interface file may contain all sorts of things that aren't
475 explicitly exported from @A@ by the programmer. For example, even
476 though a data type is exported abstractly, @A.hi@ will contain the
477 full data type definition. For small function definitions, @A.hi@
478 will contain the complete definition of the function. For bigger
479 functions, @A.hi@ will contain strictness information about the
480 function. And so on. GHC puts much more information into @.hi@ files
481 when optimisation is turned on with the @-O@ flag. Without @-O@ it
482 puts in just the minimum; with @-O@ it lobs in a whole pile of stuff.
483 <nidx>optimsation, effect on .hi files</nidx>
485 @A.hi@ should really be thought of as a compiler-readable version of
486 @A.o@. If you use a @.hi@ file that wasn't generated by the same
487 compilation run that generates the @.o@ file the compiler may assume
488 all sorts of incorrect things about @A@, resulting in core dumps and
489 other unpleasant happenings.
491 %************************************************************************
493 <sect2>Finding interface files
494 <label id="options-finding-imports">
496 <nidx>interface files, finding them</nidx>
497 <nidx>finding interface files</nidx>
499 %************************************************************************
501 In your program, you import a module @Foo@ by saying
502 @import Foo@. GHC goes looking for an interface file, @Foo.hi@.
503 It has a builtin list of directories (notably including @.@) where
508 <tag>@-i<dirs>@</tag><nidx>-i<dirs> option</nidx> This flag
509 prepends a colon-separated list of @dirs@ to the ``import
511 See also Section <ref id="recomp"> for the significance of using
512 relative and absolute pathnames in the @-i@ list.
514 <tag>@-i@</tag> resets the ``import directories'' list back to nothing.
516 <tag>@-fno-implicit-prelude@</tag>
517 <nidx>-fno-implicit-prelude option</nidx>
518 GHC normally imports @Prelude.hi@ files for you. If you'd rather it
519 didn't, then give it a @-fno-implicit-prelude@ option. You are
520 unlikely to get very far without a Prelude, but, hey, it's a free
523 <tag>@-syslib <lib>@</tag>
524 <nidx>-syslib <lib> option</nidx>
526 If you are using a system-supplied non-Prelude library (e.g., the
527 POSIX library), just use a @-syslib posix@ option (for example). The
528 right interface files should then be available. Section <ref
529 name="The GHC Prelude and Libraries" id="ghc-prelude"> lists the
530 libraries available by this mechanism.
533 <nidx>-I<dir> option</nidx>
535 Once a Haskell module has been compiled to C (@.hc@ file), you may
536 wish to specify where GHC tells the C compiler to look for @.h@ files.
537 (Or, if you are using the @-cpp@ option<nidx>-cpp option</nidx>, where
538 it tells the C pre-processor to look...) For this purpose, use a @-I@
539 option in the usual C-ish way.
543 %************************************************************************
545 <sect2>Other options related to interface files
546 <label id="hi-options">
548 <nidx>interface files, options</nidx>
550 %************************************************************************
552 The interface output may be directed to another file
553 @bar2/Wurble.iface@ with the option @-ohi bar2/Wurble.iface@<nidx>-ohi
554 <file> option</nidx> (not recommended).
556 To avoid generating an interface file at all, use a @-nohi@
557 option.<nidx>-nohi option</nidx>
559 The compiler does not overwrite an existing @.hi@ interface file if
560 the new one is byte-for-byte the same as the old one; this is friendly
561 to @make@. When an interface does change, it is often enlightening to
562 be informed. The @-hi-diffs@<nidx>-hi-diffs option</nidx> option will
563 make @ghc@ run @diff@ on the old and new @.hi@ files. You can also
564 record the difference in the interface file itself, the
565 @-keep-hi-diffs@<nidx>-keep-hi-diffs</nidx> option takes care of that.
567 The @.hi@ files from GHC contain ``usage'' information which changes
568 often and uninterestingly. If you really want to see these changes
569 reported, you need to use the
570 @-hi-diffs-with-usages@<nidx>-hi-diffs-with-usages option</nidx>
573 Interface files are normally jammed full of compiler-produced
574 <em>pragmas</em>, which record arities, strictness info, etc. If you
575 think these pragmas are messing you up (or you are doing some kind of
576 weird experiment), you can tell GHC to ignore them with the
577 @-fignore-interface-pragmas@<nidx>-fignore-interface-pragmas
578 option</nidx> option.
580 When compiling without optimisations on, the compiler is extra-careful
581 about not slurping in data constructors and instance declarations that
582 it will not need. If you believe it is getting it wrong and not
583 importing stuff which you think it should, this optimisation can be
584 turned off with @-fno-prune-tydecls@ and @-fno-prune-instdecls@.
585 <nidx>-fno-prune-tydecls option</nidx><nidx>-fno-prune-instdecls
588 See also Section <ref name="Linking and consistency-checking"
589 id="options-linker">, which describes how the linker finds standard
592 %************************************************************************
594 <sect2>The recompilation checker
597 <nidx>recompilation checker</nidx>
599 %************************************************************************
601 In the olden days, GHC compared the newly-generated @.hi@ file with
602 the previous version; if they were identical, it left the old one
603 alone and didn't change its modification date. In consequence,
604 importers of a module with an unchanged output @.hi@ file were not
607 This doesn't work any more. In our earlier example, module @C@ does
608 not import module @A@ directly, yet changes to @A.hi@ should force a
609 recompilation of @C@. And some changes to @A@ (changing the
610 definition of a function that appears in an inlining of a function
611 exported by @B@, say) may conceivably not change @B.hi@ one jot. So
614 GHC keeps a version number on each interface file, and on each type
615 signature within the interface file. It also keeps in every interface
616 file a list of the version numbers of everything it used when it last
617 compiled the file. If the source file's modification date is earlier
618 than the @.o@ file's date (i.e. the source hasn't changed since the
619 file was last compiled), GHC will be clever. It compares the version
620 numbers on the things it needs this time with the version numbers on
621 the things it needed last time (gleaned from the interface file of the
622 module being compiled); if they are all the same it stops compiling
623 rather early in the process saying ``Compilation IS NOT required''.
624 What a beautiful sight!
626 GHC <em>only</em> keeps detailed dependency information for ``user'' modules,
627 not for ``library'' modules. It distinguishes the two by a hack: a module
628 whose @.hi@ file has an absolute path name is considered a library module,
629 while a relative path name indicates a user module. So if you have a
630 multi-directory application, use <em>relative</em> path names in your
631 @-i@ path, to force GHC to record detailed dependency information.
632 Use absolute path names only for directories containing slowly-changing
635 Patrick Sansom had a workshop paper about how all this is done (though
636 the details have changed quite a bit). Ask
637 him (email: <htmlurl name="sansom@@dcs.gla.ac.uk"
638 url="mailto:sansom@@dcs.gla.ac.uk">) if you want a copy.
640 %************************************************************************
643 <label id="using-make">
647 %************************************************************************
649 It is reasonably straightforward to set up a @Makefile@ to use with
650 GHC, assuming you name your source files the same as your modules.
655 HC_OPTS = -cpp $(EXTRA_HC_OPTS)
657 SRCS = Main.lhs Foo.lhs Bar.lhs
658 OBJS = Main.o Foo.o Bar.o
660 .SUFFIXES : .o .hi .lhs .hc .s
664 $(HC) -o $@ $(HC_OPTS) $(OBJS)
666 # Standard suffix rules
671 $(HC) -c $< $(HC_OPTS)
674 $(HC) -c $< $(HC_OPTS)
676 # Inter-module dependencies
677 Foo.o Foo.hc Foo.s : Baz.hi # Foo imports Baz
678 Main.o Main.hc Main.s : Foo.hi Baz.hi # Main imports Foo and Baz
681 (Sophisticated @make@ variants may achieve some of the above more
682 elegantly. Notably, @gmake@'s pattern rules let you write the more
687 $(HC) -c $< $(HC_OPTS)
690 What we've shown should work with any @make@.)
692 Note the cheesy @.o.hi@ rule: It records the dependency of the
693 interface (@.hi@) file on the source. The rule says a @.hi@ file can
694 be made from a @.o@ file by doing... nothing. Which is true.
696 Note the inter-module dependencies at the end of the Makefile, which
700 Foo.o Foo.hc Foo.s : Baz.hi # Foo imports Baz
703 They tell @make@ that if any of @Foo.o@, @Foo.hc@ or @Foo.s@ have an
704 earlier modification date than @Baz.hi@, then the out-of-date file
705 must be brought up to date. To bring it up to date, @make@ looks for
706 a rule to do so; one of the preceding suffix rules does the job
709 Putting inter-dependencies of the form @Foo.o : Bar.hi@ into your
710 @Makefile@ by hand is rather error-prone. Don't worry---never fear,
711 @mkdependHS@ is here! (and is distributed as part of GHC) Add the
712 following to your @Makefile@:
716 mkdependHS -- $(HC_OPTS) -- $(SRCS)
719 Now, before you start compiling, and any time you change the @imports@
720 in your program, do @make depend@ before you do @make cool_pgm@.
721 @mkdependHS@ will append the needed dependencies to your @Makefile@.
722 @mkdependHS@ is fully describe in Section <ref name="Makefile
723 dependencies in Haskell: using mkdependHS" id="mkdependHS">.
725 A few caveats about this simple scheme:
729 <item> You may need to compile some modules explicitly to create their
730 interfaces in the first place (e.g., @make Bar.o@ to create @Bar.hi@).
732 <item> You may have to type @make@ more than once for the dependencies
733 to have full effect. However, a @make@ run that does nothing
734 <em>does</em> mean ``everything's up-to-date.''
736 <item> This scheme will work with mutually-recursive modules but,
737 again, it may take multiple iterations to ``settle.''
741 %************************************************************************
743 <sect2>How to compile mutually recursive modules
744 <label id="mutual-recursion">
746 <nidx>module system, recursion</nidx>
747 <nidx>recursion, between modules</nidx>
749 %************************************************************************
751 Currently, the compiler does not have proper support for dealing with
752 mutually recursive modules:
759 newtype TA = MkTA Int
774 When compiling either module A and B, the compiler will try (in vain)
775 to look for the interface file of the other. So, to get mutually
776 recursive modules off the ground, you need to hand write an interface
777 file for A or B, so as to break the loop. These hand-written
778 interface files are called @hi-boot@ files, and are placed in a file
779 called @<module>.hi-boot@. To import from an @hi-boot@ file instead
780 of the standard @.hi@ file, use the following syntax in the importing module:
781 <nidx>hi-boot files</nidx>
782 <nidx>importing, hi-boot files</nidx>
785 import {-# SOURCE #-} A
788 The hand-written interface need only contain the bare minimum of
789 information needed to get the bootstrapping process started. For
790 example, it doesn't need to contain declarations for <em/everything/
791 that module @A@ exports, only the things required by the module that
792 imports @A@ recursively.
794 For the example at hand, the boot interface file for A would look like
798 __interface A 1 404 where
799 __export A TA{MkTA} ;
800 1 newtype TA = MkTA PrelBase.Int ;
803 The syntax is essentially the same as a normal @.hi@ file
804 (unfortunately), but you can usually tailor an existing @.hi@ file to
805 make a @.hi-boot@ file.
807 Notice that we only put the declaration for the newtype @TA@ in the
808 @hi-boot@ file, not the signature for @f@, since @f@ isn't used by
811 The number ``1'' after ``__interface A'' gives the version number of module A;
812 it is incremented whenever anything in A's interface file changes. The ``404'' is
813 the version number of the interface file <em>syntax</em>; we change it when
814 we change the syntax of interface files so that you get a better error message when
815 you try to read an old-format file with a new-format compiler.
817 The number ``1'' at the beginning of a declaration is the <em>version
818 number</em> of that declaration: for the purposes of @.hi-boot@ files
819 these can all be set to 1. All names must be fully qualified with the
820 <em/original/ module that an object comes from: for example, the
821 reference to @Int@ in the interface for @A@ comes from @PrelBase@,
822 which is a module internal to GHC's prelude. It's a pain, but that's
825 If you want an hi-boot file to export a data type, but you don't want to give its constructors
826 (because the constructors aren't used by the SOURCE-importing module), you can write simply:
829 __interface A 1 404 where
834 (You must write all the type parameters, but leave out the '=' and everything that follows it.)
836 <bf>Note:</bf> This is all a temporary solution, a version of the
837 compiler that handles mutually recursive properly without the manual
838 construction of interface files, is (allegedly) in the works.
840 %************************************************************************
842 <sect1>Optimisation (code improvement)
843 <label id="options-optimise">
845 <nidx>optimisation (GHC)</nidx>
846 <nidx>improvement, code (GHC)</nidx>
848 %************************************************************************
850 The @-O*@ options specify convenient ``packages'' of optimisation
851 flags; the @-f*@ options described later on specify
852 <em>individual</em> optimisations to be turned on/off; the @-m*@
853 options specify <em>machine-specific</em> optimisations to be turned
856 %----------------------------------------------------------------------
857 <sect2>@-O*@: convenient ``packages'' of optimisation flags.
858 <label id="optimise-pkgs">
860 <nidx>-O options</nidx>
862 There are <em>many</em> options that affect the quality of code
863 produced by GHC. Most people only have a general goal, something like
864 ``Compile quickly'' or ``Make my program run like greased lightning.''
865 The following ``packages'' of optimisations (or lack thereof) should
868 Once you choose a @-O*@ ``package,'' stick with it---don't chop and
869 change. Modules' interfaces <em>will</em> change with a shift to a new
870 @-O*@ option, and you may have to recompile a large chunk of all
871 importing modules before your program can again be run
872 safely (see Section <ref name="The recompilation checker" id="recomp">).
875 <tag>No @-O*@-type option specified:</tag>
876 <nidx>-O* not specified</nidx>
877 This is taken to mean: ``Please compile quickly; I'm not over-bothered
878 about compiled-code quality.'' So, for example: @ghc -c Foo.hs@
880 <tag>@-O@ or @-O1@:</tag>
881 <nidx>-O option</nidx>
882 <nidx>-O1 option</nidx>
883 <nidx>optimise normally</nidx>
884 Means: ``Generate good-quality code without taking too long about it.''
885 Thus, for example: @ghc -c -O Main.lhs@
888 <nidx>-O2 option</nidx>
889 <nidx>optimise aggressively</nidx>
890 Means: ``Apply every non-dangerous optimisation, even if it means
891 significantly longer compile times.''
893 The avoided ``dangerous'' optimisations are those that can make
894 runtime or space <em>worse</em> if you're unlucky. They are
895 normally turned on or off individually.
897 At the moment, @-O2@ is <em>unlikely</em> to produce
898 better code than @-O@.
900 <tag>@-O2-for-C@:</tag>
901 <nidx>-O2-for-C option</nidx>
902 <nidx>gcc, invoking with -O2</nidx>
904 Says to run GCC with @-O2@, which may be worth a few percent in
905 execution speed. Don't forget @-fvia-C@, lest you use the native-code
906 generator and bypass GCC altogether!
909 <nidx>-Onot option</nidx>
910 <nidx>optimising, reset</nidx>
912 This option will make GHC ``forget'' any @-O@ish options it has seen so
913 far. Sometimes useful; for example: @make all EXTRA_HC_OPTS=-Onot@.
915 <tag>@-Ofile <file>@:</tag>
916 <nidx>-Ofile <file> option</nidx>
917 <nidx>optimising, customised</nidx>
919 For those who need <em>absolute</em> control over <em>exactly</em>
920 what options are used (e.g., compiler writers, sometimes :-), a list
921 of options can be put in a file and then slurped in with @-Ofile@.
923 In that file, comments are of the @#@-to-end-of-line variety; blank
924 lines and most whitespace is ignored.
926 Please ask if you are baffled and would like an example of @-Ofile@!
929 At Glasgow, we don't use a @-O*@ flag for day-to-day work. We use
930 @-O@ to get respectable speed; e.g., when we want to measure
931 something. When we want to go for broke, we tend to use @-O -fvia-C
932 -O2-for-C@ (and we go for lots of coffee breaks).
934 The easiest way to see what @-O@ (etc.) ``really mean'' is to run with
935 @-v@, then stand back in amazement. Alternatively, just look at the
936 @HsC_minus<blah>@ lists in the @ghc@ driver script.
938 %----------------------------------------------------------------------
939 <sect2>@-f*@: platform-independent flags
941 <nidx>-f* options (GHC)</nidx>
942 <nidx>-fno-* options (GHC)</nidx>
944 Flags can be turned <em>off</em> individually. (NB: I hope you have a
945 good reason for doing this....) To turn off the @-ffoo@ flag, just use
946 the @-fno-foo@ flag.<nidx>-fno-<opt> anti-option</nidx> So, for
947 example, you can say @-O2 -fno-strictness@, which will then drop out
948 any running of the strictness analyser.
950 The options you are most likely to want to turn off are:
953 @-fno-strictness@<nidx>-fno-strictness option</nidx> (strictness
954 analyser [because it is sometimes slow]),
956 @-fno-specialise@<nidx>-fno-specialise option</nidx> (automatic
957 specialisation of overloaded functions [because it makes your code
958 bigger]) [US spelling also accepted], and
960 @-fno-cpr-analyse@<nidx>-fno-cpr-analyse option</nidx> switches off the CPR (constructed product
964 Should you wish to turn individual flags <em>on</em>, you are advised
965 to use the @-Ofile@ option, described above. Because the order in
966 which optimisation passes are run is sometimes crucial, it's quite
967 hard to do with command-line options.
969 Here are some ``dangerous'' optimisations you <em>might</em> want to try:
971 %------------------------------------------------------------------
972 <tag>@-fvia-C@:</tag>
973 <nidx>-fvia-C option</nidx>
974 <nidx>native code generator, turning off</nidx>
976 Compile via C, and don't use the native-code generator. (There are
977 many cases when GHC does this on its own.) You might pick up a little
978 bit of speed by compiling via C. If you use @_ccall_gc_@s or
979 @_casm_@s, you probably <em>have to</em> use @-fvia-C@.
981 The lower-case incantation, @-fvia-c@, is synonymous.
983 Compiling via C will probably be slower (in compilation time) than
984 using GHC's native code generator.
986 <tag>@-funfolding-interface-threshold<n>@:</tag>
987 <nidx>-funfolding-interface-threshold option</nidx>
988 <nidx>inlining, controlling</nidx>
989 <nidx>unfolding, controlling</nidx>
990 (Default: 30) By raising or lowering this number, you can raise or
991 lower the amount of pragmatic junk that gets spewed into interface
992 files. (An unfolding has a ``size'' that reflects the cost in terms
993 of ``code bloat'' of expanding that unfolding in another module. A
994 bigger function would be assigned a bigger cost.)
996 <tag>@-funfolding-creation-threshold<n>@:</tag>
997 <nidx>-funfolding-creation-threshold option</nidx>
998 <nidx>inlining, controlling</nidx>
999 <nidx>unfolding, controlling</nidx>
1000 (Default: 30) This option is similar to
1001 @-funfolding-interface-threshold@, except that it governs unfoldings
1002 within a single module. Increasing this figure is more likely to
1003 result in longer compile times than faster code. The next option is
1006 <tag>@-funfolding-use-threshold<n>@:</tag>
1007 <nidx>-funfolding-use-threshold option</nidx>
1008 <nidx>inlining, controlling</nidx>
1009 <nidx>unfolding, controlling</nidx>
1010 (Default: 8) This is the magic cut-off figure for unfolding: below
1011 this size, a function definition will be unfolded at the call-site,
1012 any bigger and it won't. The size computed for a function depends on
1013 two things: the actual size of the expression minus any discounts that
1014 apply (see @-funfolding-con-discount@).
1016 <tag>@-funfolding-con-discount<n>@:</tag>
1017 <nidx>-funfolding-con-discount option</nidx>
1018 <nidx>inlining, controlling</nidx>
1019 <nidx>unfolding, controlling</nidx>
1020 (Default: 2) If the compiler decides that it can eliminate some
1021 computation by performing an unfolding, then this is a discount factor
1022 that it applies to the funciton size before deciding whether to unfold
1025 OK, folks, these magic numbers `30', `8', and '2' are mildly
1026 arbitrary; they are of the ``seem to be OK'' variety. The `8' is the
1027 more critical one; it's what determines how eager GHC is about
1028 expanding unfoldings.
1030 <tag>@-funbox-strict-fields@:</tag>
1031 <nidx>-funbox-strict-fields option</nidx>
1032 <nidx>strict constructor fields</nidx>
1033 <nidx>constructor fields, strict</nidx>
1035 This option causes all constructor fields which are marked strict
1036 (i.e. ``!'') to be unboxed or unpacked if possible. For example:
1039 data T = T !Float !Float
1042 will create a constructor @T@ containing two unboxed floats if the
1043 @-funbox-strict-fields@ flag is given. This may not always be an
1044 optimisation: if the @T@ constructor is scrutinised and the floats
1045 passed to a non-strict function for example, they will have to be
1046 reboxed (this is done automatically by the compiler).
1048 This option should only be used in conjunction with @-O@, in order to
1049 expose unfoldings to the compiler so the reboxing can be removed as
1050 often as possible. For example:
1054 f (T f1 f2) = f1 + f2
1057 The compiler will avoid reboxing @f1@ and @f2@ by inlining @+@ on
1058 floats, but only when @-O@ is on.
1060 Any single-constructor data is eligible for unpacking; for example
1063 data T = T !(Int,Int)
1066 will store the two @Int@s directly in the @T@ constructor, by flattening
1067 the pair. Multi-level unpacking is also supported:
1071 data S = S !Int !Int
1074 will store two unboxed @Int#@s directly in the @T@ constructor.
1076 <tag>@-fsemi-tagging@:</tag>
1077 This option (which <em>does not work</em> with the native-code generator)
1078 tells the compiler to add extra code to test for already-evaluated
1079 values. You win if you have lots of such values during a run of your
1080 program, you lose otherwise. (And you pay in extra code space.)
1082 We have not played with @-fsemi-tagging@ enough to recommend it.
1083 (For all we know, it doesn't even work anymore... Sigh.)
1086 %----------------------------------------------------------------------
1087 <sect2>@-m*@: platform-specific flags
1089 <nidx>-m* options (GHC)</nidx>
1090 <nidx>platform-specific options</nidx>
1091 <nidx>machine-specific options</nidx>
1093 Some flags only make sense for particular target platforms.
1097 (SPARC machines)<nidx>-mv8 option (SPARC only)</nidx>
1098 Means to pass the like-named option to GCC; it says to use the
1099 Version 8 SPARC instructions, notably integer multiply and divide.
1100 The similiar @-m*@ GCC options for SPARC also work, actually.
1102 <tag>@-mlong-calls@:</tag>
1103 (HPPA machines)<nidx>-mlong-calls option (HPPA only)</nidx>
1104 Means to pass the like-named option to GCC. Required for Very Big
1105 modules, maybe. (Probably means you're in trouble...)
1107 <tag>@-monly-[32]-regs@:</tag>
1108 (iX86 machines)<nidx>-monly-N-regs option (iX86 only)</nidx>
1109 GHC tries to ``steal'' four registers from GCC, for performance
1110 reasons; it almost always works. However, when GCC is compiling some
1111 modules with four stolen registers, it will crash, probably saying:
1113 Foo.hc:533: fixed or forbidden register was spilled.
1114 This may be due to a compiler bug or to impossible asm
1115 statements or clauses.
1117 Just give some registers back with @-monly-N-regs@. Try `3' first,
1118 then `2'. If `2' doesn't work, please report the bug to us.
1121 %----------------------------------------------------------------------
1122 <sect2>Code improvement by the C compiler.
1123 <label id="optimise-C-compiler">
1125 <nidx>optimisation by GCC</nidx>
1126 <nidx>GCC optimisation</nidx>
1128 The C~compiler (GCC) is run with @-O@ turned on. (It has
1131 If you want to run GCC with @-O2@---which may be worth a few
1132 percent in execution speed---you can give a
1133 @-O2-for-C@<nidx>-O2-for-C option</nidx> option.
1135 %************************************************************************
1137 <sect1>Options related to a particular phase
1138 <label id="options-phases">
1141 %************************************************************************
1143 <sect2> The C pre-processor
1144 <label id="c-pre-processor">
1146 <nidx>pre-processing: cpp</nidx>
1147 <nidx>C pre-processor options</nidx>
1148 <nidx>cpp, pre-processing with</nidx>
1150 The C pre-processor @cpp@ is run over your Haskell code only if the
1151 @-cpp@ option <nidx>-cpp option</nidx> is given. Unless you are
1152 building a large system with significant doses of conditional
1153 compilation, you really shouldn't need it.
1155 <tag>@-D<foo>@:</tag>
1156 <nidx>-D<name> option</nidx>
1157 Define macro @<foo>@ in the usual way. NB: does <em>not</em> affect
1158 @-D@ macros passed to the C~compiler when compiling via C! For those,
1159 use the @-optc-Dfoo@ hack... (see Section <ref name="Forcing options
1160 to a particular phase." id="forcing-options-through">).
1162 <tag>@-U<foo>@:</tag>
1163 <nidx>-U<name> option</nidx>
1164 Undefine macro @<foo>@ in the usual way.
1166 <tag>@-I<dir>@:</tag>
1167 <nidx>-I<dir> option</nidx>
1168 Specify a directory in which to look for @#include@ files, in
1172 The @ghc@ driver pre-defines several macros when processing Haskell
1173 source code (@.hs@ or @.lhs@ files):
1176 <tag>@__HASKELL98__@:</tag>
1177 <nidx>__HASKELL98__</nidx>
1178 If defined, this means that GHC supports the language defined by the
1181 <tag>@__HASKELL__=98@:</tag>
1182 <nidx>__HASKELL__</nidx>
1183 In GHC 4.04 and later, the @__HASKELL__@ macro is defined as having
1186 <tag>@__HASKELL1__@:</tag>
1187 <nidx>__HASKELL1__ macro</nidx>
1188 If defined to <em/n/, that means GHC supports the Haskell language
1189 defined in the Haskell report version <em/1.n/. Currently 5. This
1190 macro is deprecated, and will probably disappear in future versions.
1192 <tag>@__GLASGOW_HASKELL__@:</tag>
1193 <nidx>__GLASGOW_HASKELL__ macro</nidx>
1194 For version <em/n/ of the GHC system, this will be @#define@d to
1195 <em/100n/. So, for version 4.00, it is 400.
1197 With any luck, @__GLASGOW_HASKELL__@ will be undefined in all other
1198 implementations that support C-style pre-processing.
1200 (For reference: the comparable symbols for other systems are:
1201 @__HUGS__@ for Hugs and @__HBC__@ for Chalmers.)
1203 NB. This macro is set when pre-processing both Haskell source and C
1204 source, including the C source generated from a Haskell module
1205 (ie. @.hs@, @.lhs@, @.c@ and @.hc@ files).
1207 <tag>@__CONCURRENT_HASKELL__@:</tag>
1208 <nidx>__CONCURRENT_HASKELL__ macro</nidx>
1209 This symbol is defined when pre-processing Haskell (input) and
1210 pre-processing C (GHC output). Since GHC from verion 4.00 now
1211 supports concurrent haskell by default, this symbol is always defined.
1213 <tag>@__PARALLEL_HASKELL__@:</tag>
1214 <nidx>__PARALLEL_HASKELL__ macro</nidx>
1215 Only defined when @-parallel@ is in use! This symbol is defined when
1216 pre-processing Haskell (input) and pre-processing C (GHC output).
1219 Options other than the above can be forced through to the C
1220 pre-processor with the @-opt@ flags (see
1221 Section <ref name="Forcing options to a particular phase." id="forcing-options-through">).
1223 A small word of warning: @-cpp@ is not friendly to ``string
1224 gaps''.<nidx>-cpp vs string gaps</nidx><nidx>string gaps vs
1225 -cpp</nidx>. In other words, strings such as the following:
1233 don't work with @-cpp@; @/usr/bin/cpp@ elides the
1234 backslash-newline pairs.
1236 However, it appears that if you add a space at the end of the line,
1237 then @cpp@ (at least GNU @cpp@ and possibly other @cpp@s)
1238 leaves the backslash-space pairs alone and the string gap works as
1241 %************************************************************************
1243 <sect2>Options affecting the C compiler (if applicable)
1244 <label id="options-C-compiler">
1246 <nidx>include-file options</nidx>
1247 <nidx>C compiler options</nidx>
1248 <nidx>GCC options</nidx>
1250 %************************************************************************
1252 At the moment, quite a few common C-compiler options are passed on
1253 quietly to the C compilation of Haskell-compiler-generated C files.
1254 THIS MAY CHANGE. Meanwhile, options so sent are:
1257 @-ansi@ | do ANSI C (not K&R) @@
1258 @-pedantic@ | be so@@
1259 @-dgcc-lint@ | (hack) short for ``make GCC very paranoid''@@
1261 <nidx>-ansi option (for GCC)</nidx>
1262 <nidx>-pedantic option (for GCC)</nidx>
1263 <nidx>-dgcc-lint option (GCC paranoia)</nidx>
1265 If you are compiling with lots of @ccalls@, etc., you may need to
1266 tell the C~compiler about some @#include@ files. There is no real
1267 pretty way to do this, but you can use this hack from the
1271 % ghc -c '-#include <X/Xlib.h>' Xstuff.lhs
1275 %************************************************************************
1277 <sect2>Linking and consistency-checking
1278 <label id="options-linker">
1280 <nidx>linker options</nidx>
1281 <nidx>ld options</nidx>
1283 %************************************************************************
1285 GHC has to link your code with various libraries, possibly including:
1286 user-supplied, GHC-supplied, and system-supplied (@-lm@ math
1287 library, for example).
1290 <tag>@-l<FOO>@:</tag>
1291 <nidx>-l<lib> option</nidx>
1292 Link in a library named @lib<FOO>.a@ which resides somewhere on the
1293 library directories path.
1295 Because of the sad state of most UNIX linkers, the order of such
1296 options does matter. Thus: @ghc -lbar *.o@ is almost certainly
1297 wrong, because it will search @libbar.a@ <em>before</em> it has
1298 collected unresolved symbols from the @*.o@ files.
1299 @ghc *.o -lbar@ is probably better.
1301 The linker will of course be informed about some GHC-supplied
1302 libraries automatically; these are:
1305 <bf>-l equivalent</bf> | <bf>description</bf> @@
1307 @-lHSrts,-lHSclib@ | basic runtime libraries @@
1308 @-lHS@ | standard Prelude library @@
1309 @-lHS_cbits@ | C support code for standard Prelude library @@
1310 @-lgmp@ | GNU multi-precision library (for Integers)@@
1313 <nidx>-lHS library</nidx>
1314 <nidx>-lHS_cbits library</nidx>
1315 <nidx>-lHSrts library</nidx>
1316 <nidx>-lgmp library</nidx>
1318 <tag>@-syslib <name>@:</tag>
1319 <nidx>-syslib <name> option</nidx>
1321 If you are using a Haskell ``system library'' (e.g., the POSIX
1322 library), just use the @-syslib posix@ option, and the correct code
1323 should be linked in.
1325 <tag>@-L<dir>@:</tag>
1326 <nidx>-L<dir> option</nidx>
1327 Where to find user-supplied libraries... Prepend the directory
1328 @<dir>@ to the library directories path.
1330 <tag>@-static@:</tag>
1331 <nidx>-static option</nidx>
1332 Tell the linker to avoid shared libraries.
1334 <tag>@-no-link-chk@ and @-link-chk@:</tag>
1335 <nidx>-no-link-chk option</nidx>
1336 <nidx>-link-chk option</nidx>
1337 <nidx>consistency checking of executables</nidx>
1338 By default, immediately after linking an executable, GHC verifies that
1339 the pieces that went into it were compiled with compatible flags; a
1340 ``consistency check''.
1341 (This is to avoid mysterious failures caused by non-meshing of
1342 incompatibly-compiled programs; e.g., if one @.o@ file was compiled
1343 for a parallel machine and the others weren't.) You may turn off this
1344 check with @-no-link-chk@. You can turn it (back) on with
1345 @-link-chk@ (the default).
1347 <tag><tt>-no-hs-main</tt>:</tag>
1348 <nidx>-no-hs-main option</nidx>
1349 <nidx>linking Haskell libraries with foreign code</nidx>
1351 In the event you want to include ghc-compiled code as part of another
1352 (non-Haskell) program, the RTS will not be supplying its definition of
1353 <tt/main()/ at link-time, you will have to. To signal that to the
1354 driver script when linking, use <tt/-no-hs-main/.
1356 Notice that since the command-line passed to the linker is rather
1357 involved, you probably want to use the ghc driver script to do the
1358 final link of your `mixed-language' application. This is not a
1359 requirement though, just try linking once with <tt/-v/ on to see what
1360 options the driver passes through to the linker.
1364 %************************************************************************
1366 <sect1>Using Concurrent Haskell
1368 <nidx>Concurrent Haskell---use</nidx>
1370 %************************************************************************
1372 GHC (as of version 4.00) supports Concurrent Haskell by default,
1373 without requiring a special option or libraries compiled in a certain
1374 way. To get access to the support libraries for Concurrent Haskell
1375 (ie. @Concurrent@ and friends), use the @-syslib concurrent@ option.
1377 Three RTS options are provided for modifying the behaviour of the
1378 threaded runtime system. See the descriptions of @-C[<us>]@, @-q@,
1379 and @-t<num>@ in Section <ref name="RTS options for
1380 Concurrent/Parallel Haskell" id="parallel-rts-opts">.
1382 Concurrent Haskell is described in more detail in Section <ref
1383 name="Concurrent and Parallel Haskell" id="concurrent-and-parallel">.
1385 %************************************************************************
1387 <sect1>Using Parallel Haskell
1389 <nidx>Parallel Haskell---use</nidx>
1391 %************************************************************************
1393 [You won't be able to execute parallel Haskell programs unless PVM3
1394 (Parallel Virtual Machine, version 3) is installed at your site.]
1396 To compile a Haskell program for parallel execution under PVM, use the
1397 @-parallel@ option,<nidx>-parallel option</nidx> both when compiling
1398 <em>and linking</em>. You will probably want to @import Parallel@
1399 into your Haskell modules.
1401 To run your parallel program, once PVM is going, just invoke it ``as
1402 normal''. The main extra RTS option is @-N<n>@, to say how many
1403 PVM ``processors'' your program to run on. (For more details of
1404 all relevant RTS options, please see Section <ref name="RTS options for Concurrent/Parallel Haskell" id="parallel-rts-opts">.)
1406 In truth, running Parallel Haskell programs and getting information
1407 out of them (e.g., parallelism profiles) is a battle with the vagaries of
1408 PVM, detailed in the following sections.
1410 %************************************************************************
1412 <sect2>Dummy's guide to using PVM
1414 <nidx>PVM, how to use</nidx>
1415 <nidx>Parallel Haskell---PVM use</nidx>
1417 %************************************************************************
1419 Before you can run a parallel program under PVM, you must set the
1420 required environment variables (PVM's idea, not ours); something like,
1421 probably in your @.cshrc@ or equivalent:
1423 setenv PVM_ROOT /wherever/you/put/it
1424 setenv PVM_ARCH `$PVM_ROOT/lib/pvmgetarch`
1425 setenv PVM_DPATH $PVM_ROOT/lib/pvmd
1428 Creating and/or controlling your ``parallel machine'' is a purely-PVM
1429 business; nothing specific to Parallel Haskell.
1431 You use the @pvm@<nidx>pvm command</nidx> command to start PVM on your
1432 machine. You can then do various things to control/monitor your
1433 ``parallel machine;'' the most useful being:
1436 @Control-D@ & exit @pvm@, leaving it running \\
1437 @halt@ & kill off this ``parallel machine'' \& exit \\
1438 @add <host>@ & add @<host>@ as a processor \\
1439 @delete <host>@ & delete @<host>@ \\
1440 @reset@ & kill what's going, but leave PVM up \\
1441 @conf@ & list the current configuration \\
1442 @ps@ & report processes' status \\
1443 @pstat <pid>@ & status of a particular process \\
1446 The PVM documentation can tell you much, much more about @pvm@!
1448 %************************************************************************
1450 <sect2>Parallelism profiles
1452 <nidx>parallelism profiles</nidx>
1453 <nidx>profiles, parallelism</nidx>
1454 <nidx>visualisation tools</nidx>
1456 %************************************************************************
1458 With Parallel Haskell programs, we usually don't care about the
1459 results---only with ``how parallel'' it was! We want pretty pictures.
1461 Parallelism profiles (\`a la @hbcpp@) can be generated with the
1462 @-q@<nidx>-q RTS option (concurrent, parallel)</nidx> RTS option. The
1463 per-processor profiling info is dumped into files named
1464 @<full-path><program>.gr@. These are then munged into a PostScript picture,
1465 which you can then display. For example, to run your program
1466 @a.out@ on 8 processors, then view the parallelism profile, do:
1469 % ./a.out +RTS -N8 -q
1470 % grs2gr *.???.gr > temp.gr # combine the 8 .gr files into one
1471 % gr2ps -O temp.gr # cvt to .ps; output in temp.ps
1472 % ghostview -seascape temp.ps # look at it!
1475 The scripts for processing the parallelism profiles are distributed
1476 in @ghc/utils/parallel/@.
1478 %************************************************************************
1480 <sect2>Other useful info about running parallel programs
1483 %************************************************************************
1485 The ``garbage-collection statistics'' RTS options can be useful for
1486 seeing what parallel programs are doing. If you do either
1487 @+RTS -Sstderr@<nidx>-Sstderr RTS option</nidx> or @+RTS -sstderr@, then
1488 you'll get mutator, garbage-collection, etc., times on standard
1489 error. The standard error of all PE's other than the `main thread'
1490 appears in @/tmp/pvml.nnn@, courtesy of PVM.
1492 Whether doing @+RTS -Sstderr@ or not, a handy way to watch
1493 what's happening overall is: @tail -f /tmp/pvml.nnn@.
1495 %************************************************************************
1497 <sect2>RTS options for Concurrent/Parallel Haskell
1498 <label id="parallel-rts-opts">
1500 <nidx>RTS options, concurrent</nidx>
1501 <nidx>RTS options, parallel</nidx>
1502 <nidx>Concurrent Haskell---RTS options</nidx>
1503 <nidx>Parallel Haskell---RTS options</nidx>
1505 %************************************************************************
1507 Besides the usual runtime system (RTS) options
1508 (Section <ref name="Running a compiled program" id="runtime-control">), there are a few options particularly
1509 for concurrent/parallel execution.
1513 <nidx>-N<N> RTS option (parallel)</nidx>
1514 (PARALLEL ONLY) Use @<N>@ PVM processors to run this program;
1517 <tag>@-C[<us>]@:</tag>
1518 <nidx>-C<us> RTS option</nidx>
1519 Sets the context switch interval to @<us>@ microseconds. A context
1520 switch will occur at the next heap allocation after the timer expires.
1521 With @-C0@ or @-C@, context switches will occur as often as
1522 possible (at every heap allocation). By default, context switches
1523 occur every 10 milliseconds. Note that many interval timers are only
1524 capable of 10 millisecond granularity, so the default setting may be
1525 the finest granularity possible, short of a context switch at every
1528 [NOTE: this option currently has no effect (version 4.00). Context
1529 switches happen when the current heap block is full, i.e. every 4k of
1533 <nidx>-q RTS option</nidx>
1534 (PARALLEL ONLY) Produce a quasi-parallel profile of thread activity,
1535 in the file @<program>.qp@. In the style of @hbcpp@, this profile
1536 records the movement of threads between the green (runnable) and red
1537 (blocked) queues. If you specify the verbose suboption (@-qv@), the
1538 green queue is split into green (for the currently running thread
1539 only) and amber (for other runnable threads). We do not recommend
1540 that you use the verbose suboption if you are planning to use the
1541 @hbcpp@ profiling tools or if you are context switching at every heap
1544 <tag>@-t<num>@:</tag>
1545 <nidx>-t<num> RTS option</nidx>
1546 (PARALLEL ONLY) Limit the number of concurrent threads per processor
1547 to @<num>@. The default is 32. Each thread requires slightly over 1K
1548 <em>words</em> in the heap for thread state and stack objects. (For
1549 32-bit machines, this translates to 4K bytes, and for 64-bit machines,
1553 <nidx>-d RTS option (parallel)</nidx>
1554 (PARALLEL ONLY) Turn on debugging. It pops up one xterm (or GDB, or
1555 something...) per PVM processor. We use the standard @debugger@
1556 script that comes with PVM3, but we sometimes meddle with the
1557 @debugger2@ script. We include ours in the GHC distribution,
1558 in @ghc/utils/pvm/@.
1560 <tag>@-e<num>@:</tag>
1561 <nidx>-e<num> RTS option (parallel)</nidx>
1562 (PARALLEL ONLY) Limit the number of pending sparks per processor to
1563 @<num>@. The default is 100. A larger number may be appropriate if
1564 your program generates large amounts of parallelism initially.
1566 <tag>@-Q<num>@:</tag>
1567 <nidx>-Q<num> RTS option (parallel)</nidx>
1568 (PARALLEL ONLY) Set the size of packets transmitted between processors
1569 to @<num>@. The default is 1024 words. A larger number may be
1570 appropriate if your machine has a high communication cost relative to