]> gcc.gnu.org Git - gcc.git/blame - gcc/doc/tm.texi.in
c-common.c (c_common_reswords): Add __is_literal_type.
[gcc.git] / gcc / doc / tm.texi.in
CommitLineData
38f8b050
JR
1@c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2@c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3@c Free Software Foundation, Inc.
4@c This is part of the GCC manual.
5@c For copying conditions, see the file gcc.texi.
6
7@node Target Macros
8@chapter Target Description Macros and Functions
9@cindex machine description macros
10@cindex target description macros
11@cindex macros, target description
12@cindex @file{tm.h} macros
13
14In addition to the file @file{@var{machine}.md}, a machine description
15includes a C header file conventionally given the name
16@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
17The header file defines numerous macros that convey the information
18about the target machine that does not fit into the scheme of the
19@file{.md} file. The file @file{tm.h} should be a link to
20@file{@var{machine}.h}. The header file @file{config.h} includes
21@file{tm.h} and most compiler source files include @file{config.h}. The
22source file defines a variable @code{targetm}, which is a structure
23containing pointers to functions and data relating to the target
24machine. @file{@var{machine}.c} should also contain their definitions,
25if they are not defined elsewhere in GCC, and other functions called
26through the macros defined in the @file{.h} file.
27
28@menu
29* Target Structure:: The @code{targetm} variable.
30* Driver:: Controlling how the driver runs the compilation passes.
31* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
32* Per-Function Data:: Defining data structures for per-function information.
33* Storage Layout:: Defining sizes and alignments of data.
34* Type Layout:: Defining sizes and properties of basic user data types.
35* Registers:: Naming and describing the hardware registers.
36* Register Classes:: Defining the classes of hardware registers.
37* Old Constraints:: The old way to define machine-specific constraints.
38* Stack and Calling:: Defining which way the stack grows and by how much.
39* Varargs:: Defining the varargs macros.
40* Trampolines:: Code set up at run time to enter a nested function.
41* Library Calls:: Controlling how library routines are implicitly called.
42* Addressing Modes:: Defining addressing modes valid for memory operands.
43* Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
44* Condition Code:: Defining how insns update the condition code.
45* Costs:: Defining relative costs of different operations.
46* Scheduling:: Adjusting the behavior of the instruction scheduler.
47* Sections:: Dividing storage into text, data, and other sections.
48* PIC:: Macros for position independent code.
49* Assembler Format:: Defining how to write insns and pseudo-ops to output.
50* Debugging Info:: Defining the format of debugging output.
51* Floating Point:: Handling floating point for cross-compilers.
52* Mode Switching:: Insertion of mode-switching instructions.
53* Target Attributes:: Defining target-specific uses of @code{__attribute__}.
54* Emulated TLS:: Emulated TLS support.
55* MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
56* PCH Target:: Validity checking for precompiled headers.
57* C++ ABI:: Controlling C++ ABI changes.
58* Named Address Spaces:: Adding support for named address spaces
59* Misc:: Everything else.
60@end menu
61
62@node Target Structure
63@section The Global @code{targetm} Variable
64@cindex target hooks
65@cindex target functions
66
67@deftypevar {struct gcc_target} targetm
68The target @file{.c} file must define the global @code{targetm} variable
69which contains pointers to functions and data relating to the target
70machine. The variable is declared in @file{target.h};
71@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
72used to initialize the variable, and macros for the default initializers
73for elements of the structure. The @file{.c} file should override those
74macros for which the default definition is inappropriate. For example:
75@smallexample
76#include "target.h"
77#include "target-def.h"
78
79/* @r{Initialize the GCC target structure.} */
80
81#undef TARGET_COMP_TYPE_ATTRIBUTES
82#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
83
84struct gcc_target targetm = TARGET_INITIALIZER;
85@end smallexample
86@end deftypevar
87
88Where a macro should be defined in the @file{.c} file in this manner to
89form part of the @code{targetm} structure, it is documented below as a
90``Target Hook'' with a prototype. Many macros will change in future
91from being defined in the @file{.h} file to being part of the
92@code{targetm} structure.
93
94@node Driver
95@section Controlling the Compilation Driver, @file{gcc}
96@cindex driver
97@cindex controlling the compilation driver
98
99@c prevent bad page break with this line
100You can control the compilation driver.
101
102@defmac SWITCH_TAKES_ARG (@var{char})
103A C expression which determines whether the option @option{-@var{char}}
104takes arguments. The value should be the number of arguments that
105option takes--zero, for many options.
106
107By default, this macro is defined as
108@code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
109properly. You need not define @code{SWITCH_TAKES_ARG} unless you
110wish to add additional options which take arguments. Any redefinition
111should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
112additional options.
113@end defmac
114
115@defmac WORD_SWITCH_TAKES_ARG (@var{name})
116A C expression which determines whether the option @option{-@var{name}}
117takes arguments. The value should be the number of arguments that
118option takes--zero, for many options. This macro rather than
119@code{SWITCH_TAKES_ARG} is used for multi-character option names.
120
121By default, this macro is defined as
122@code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
123properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
124wish to add additional options which take arguments. Any redefinition
125should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
126additional options.
127@end defmac
128
38f8b050
JR
129@defmac TARGET_OPTION_TRANSLATE_TABLE
130If defined, a list of pairs of strings, the first of which is a
131potential command line target to the @file{gcc} driver program, and the
132second of which is a space-separated (tabs and other whitespace are not
133supported) list of options with which to replace the first option. The
134target defining this list is responsible for assuring that the results
135are valid. Replacement options may not be the @code{--opt} style, they
136must be the @code{-opt} style. It is the intention of this macro to
137provide a mechanism for substitution that affects the multilibs chosen,
138such as one option that enables many options, some of which select
139multilibs. Example nonsensical definition, where @option{-malt-abi},
140@option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
141
142@smallexample
143#define TARGET_OPTION_TRANSLATE_TABLE \
144@{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
145@{ "-compat", "-EB -malign=4 -mspoo" @}
146@end smallexample
147@end defmac
148
149@defmac DRIVER_SELF_SPECS
150A list of specs for the driver itself. It should be a suitable
151initializer for an array of strings, with no surrounding braces.
152
153The driver applies these specs to its own command line between loading
154default @file{specs} files (but not command-line specified ones) and
155choosing the multilib directory or running any subcommands. It
156applies them in the order given, so each spec can depend on the
157options added by earlier ones. It is also possible to remove options
158using @samp{%<@var{option}} in the usual way.
159
160This macro can be useful when a port has several interdependent target
161options. It provides a way of standardizing the command line so
162that the other specs are easier to write.
163
164Do not define this macro if it does not need to do anything.
165@end defmac
166
167@defmac OPTION_DEFAULT_SPECS
168A list of specs used to support configure-time default options (i.e.@:
169@option{--with} options) in the driver. It should be a suitable initializer
170for an array of structures, each containing two strings, without the
171outermost pair of surrounding braces.
172
173The first item in the pair is the name of the default. This must match
174the code in @file{config.gcc} for the target. The second item is a spec
175to apply if a default with this name was specified. The string
176@samp{%(VALUE)} in the spec will be replaced by the value of the default
177everywhere it occurs.
178
179The driver will apply these specs to its own command line between loading
180default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
181the same mechanism as @code{DRIVER_SELF_SPECS}.
182
183Do not define this macro if it does not need to do anything.
184@end defmac
185
186@defmac CPP_SPEC
187A C string constant that tells the GCC driver program options to
188pass to CPP@. It can also specify how to translate options you
189give to GCC into options for GCC to pass to the CPP@.
190
191Do not define this macro if it does not need to do anything.
192@end defmac
193
194@defmac CPLUSPLUS_CPP_SPEC
195This macro is just like @code{CPP_SPEC}, but is used for C++, rather
196than C@. If you do not define this macro, then the value of
197@code{CPP_SPEC} (if any) will be used instead.
198@end defmac
199
200@defmac CC1_SPEC
201A C string constant that tells the GCC driver program options to
202pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
203front ends.
204It can also specify how to translate options you give to GCC into options
205for GCC to pass to front ends.
206
207Do not define this macro if it does not need to do anything.
208@end defmac
209
210@defmac CC1PLUS_SPEC
211A C string constant that tells the GCC driver program options to
212pass to @code{cc1plus}. It can also specify how to translate options you
213give to GCC into options for GCC to pass to the @code{cc1plus}.
214
215Do not define this macro if it does not need to do anything.
216Note that everything defined in CC1_SPEC is already passed to
217@code{cc1plus} so there is no need to duplicate the contents of
218CC1_SPEC in CC1PLUS_SPEC@.
219@end defmac
220
221@defmac ASM_SPEC
222A C string constant that tells the GCC driver program options to
223pass to the assembler. It can also specify how to translate options
224you give to GCC into options for GCC to pass to the assembler.
225See the file @file{sun3.h} for an example of this.
226
227Do not define this macro if it does not need to do anything.
228@end defmac
229
230@defmac ASM_FINAL_SPEC
231A C string constant that tells the GCC driver program how to
232run any programs which cleanup after the normal assembler.
233Normally, this is not needed. See the file @file{mips.h} for
234an example of this.
235
236Do not define this macro if it does not need to do anything.
237@end defmac
238
239@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
240Define this macro, with no value, if the driver should give the assembler
241an argument consisting of a single dash, @option{-}, to instruct it to
242read from its standard input (which will be a pipe connected to the
243output of the compiler proper). This argument is given after any
244@option{-o} option specifying the name of the output file.
245
246If you do not define this macro, the assembler is assumed to read its
247standard input if given no non-option arguments. If your assembler
248cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
249see @file{mips.h} for instance.
250@end defmac
251
252@defmac LINK_SPEC
253A C string constant that tells the GCC driver program options to
254pass to the linker. It can also specify how to translate options you
255give to GCC into options for GCC to pass to the linker.
256
257Do not define this macro if it does not need to do anything.
258@end defmac
259
260@defmac LIB_SPEC
261Another C string constant used much like @code{LINK_SPEC}. The difference
262between the two is that @code{LIB_SPEC} is used at the end of the
263command given to the linker.
264
265If this macro is not defined, a default is provided that
266loads the standard C library from the usual place. See @file{gcc.c}.
267@end defmac
268
269@defmac LIBGCC_SPEC
270Another C string constant that tells the GCC driver program
271how and when to place a reference to @file{libgcc.a} into the
272linker command line. This constant is placed both before and after
273the value of @code{LIB_SPEC}.
274
275If this macro is not defined, the GCC driver provides a default that
276passes the string @option{-lgcc} to the linker.
277@end defmac
278
279@defmac REAL_LIBGCC_SPEC
280By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
281@code{LIBGCC_SPEC} is not directly used by the driver program but is
282instead modified to refer to different versions of @file{libgcc.a}
283depending on the values of the command line flags @option{-static},
284@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
285targets where these modifications are inappropriate, define
286@code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
287driver how to place a reference to @file{libgcc} on the link command
288line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
289@end defmac
290
291@defmac USE_LD_AS_NEEDED
292A macro that controls the modifications to @code{LIBGCC_SPEC}
293mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
294generated that uses --as-needed and the shared libgcc in place of the
295static exception handler library, when linking without any of
296@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
297@end defmac
298
299@defmac LINK_EH_SPEC
300If defined, this C string constant is added to @code{LINK_SPEC}.
301When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
302the modifications to @code{LIBGCC_SPEC} mentioned in
303@code{REAL_LIBGCC_SPEC}.
304@end defmac
305
306@defmac STARTFILE_SPEC
307Another C string constant used much like @code{LINK_SPEC}. The
308difference between the two is that @code{STARTFILE_SPEC} is used at
309the very beginning of the command given to the linker.
310
311If this macro is not defined, a default is provided that loads the
312standard C startup file from the usual place. See @file{gcc.c}.
313@end defmac
314
315@defmac ENDFILE_SPEC
316Another C string constant used much like @code{LINK_SPEC}. The
317difference between the two is that @code{ENDFILE_SPEC} is used at
318the very end of the command given to the linker.
319
320Do not define this macro if it does not need to do anything.
321@end defmac
322
323@defmac THREAD_MODEL_SPEC
324GCC @code{-v} will print the thread model GCC was configured to use.
325However, this doesn't work on platforms that are multilibbed on thread
326models, such as AIX 4.3. On such platforms, define
327@code{THREAD_MODEL_SPEC} such that it evaluates to a string without
328blanks that names one of the recognized thread models. @code{%*}, the
329default value of this macro, will expand to the value of
330@code{thread_file} set in @file{config.gcc}.
331@end defmac
332
333@defmac SYSROOT_SUFFIX_SPEC
334Define this macro to add a suffix to the target sysroot when GCC is
335configured with a sysroot. This will cause GCC to search for usr/lib,
336et al, within sysroot+suffix.
337@end defmac
338
339@defmac SYSROOT_HEADERS_SUFFIX_SPEC
340Define this macro to add a headers_suffix to the target sysroot when
341GCC is configured with a sysroot. This will cause GCC to pass the
342updated sysroot+headers_suffix to CPP, causing it to search for
343usr/include, et al, within sysroot+headers_suffix.
344@end defmac
345
346@defmac EXTRA_SPECS
347Define this macro to provide additional specifications to put in the
348@file{specs} file that can be used in various specifications like
349@code{CC1_SPEC}.
350
351The definition should be an initializer for an array of structures,
352containing a string constant, that defines the specification name, and a
353string constant that provides the specification.
354
355Do not define this macro if it does not need to do anything.
356
357@code{EXTRA_SPECS} is useful when an architecture contains several
358related targets, which have various @code{@dots{}_SPECS} which are similar
359to each other, and the maintainer would like one central place to keep
360these definitions.
361
362For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
363define either @code{_CALL_SYSV} when the System V calling sequence is
364used or @code{_CALL_AIX} when the older AIX-based calling sequence is
365used.
366
367The @file{config/rs6000/rs6000.h} target file defines:
368
369@smallexample
370#define EXTRA_SPECS \
371 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
372
373#define CPP_SYS_DEFAULT ""
374@end smallexample
375
376The @file{config/rs6000/sysv.h} target file defines:
377@smallexample
378#undef CPP_SPEC
379#define CPP_SPEC \
380"%@{posix: -D_POSIX_SOURCE @} \
381%@{mcall-sysv: -D_CALL_SYSV @} \
382%@{!mcall-sysv: %(cpp_sysv_default) @} \
383%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
384
385#undef CPP_SYSV_DEFAULT
386#define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
387@end smallexample
388
389while the @file{config/rs6000/eabiaix.h} target file defines
390@code{CPP_SYSV_DEFAULT} as:
391
392@smallexample
393#undef CPP_SYSV_DEFAULT
394#define CPP_SYSV_DEFAULT "-D_CALL_AIX"
395@end smallexample
396@end defmac
397
398@defmac LINK_LIBGCC_SPECIAL_1
399Define this macro if the driver program should find the library
400@file{libgcc.a}. If you do not define this macro, the driver program will pass
401the argument @option{-lgcc} to tell the linker to do the search.
402@end defmac
403
404@defmac LINK_GCC_C_SEQUENCE_SPEC
405The sequence in which libgcc and libc are specified to the linker.
406By default this is @code{%G %L %G}.
407@end defmac
408
409@defmac LINK_COMMAND_SPEC
410A C string constant giving the complete command line need to execute the
411linker. When you do this, you will need to update your port each time a
412change is made to the link command line within @file{gcc.c}. Therefore,
413define this macro only if you need to completely redefine the command
414line for invoking the linker and there is no other way to accomplish
415the effect you need. Overriding this macro may be avoidable by overriding
416@code{LINK_GCC_C_SEQUENCE_SPEC} instead.
417@end defmac
418
419@defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
420A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
421directories from linking commands. Do not give it a nonzero value if
422removing duplicate search directories changes the linker's semantics.
423@end defmac
424
425@defmac MULTILIB_DEFAULTS
426Define this macro as a C expression for the initializer of an array of
427string to tell the driver program which options are defaults for this
428target and thus do not need to be handled specially when using
429@code{MULTILIB_OPTIONS}.
430
431Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
432the target makefile fragment or if none of the options listed in
433@code{MULTILIB_OPTIONS} are set by default.
434@xref{Target Fragment}.
435@end defmac
436
437@defmac RELATIVE_PREFIX_NOT_LINKDIR
438Define this macro to tell @command{gcc} that it should only translate
439a @option{-B} prefix into a @option{-L} linker option if the prefix
440indicates an absolute file name.
441@end defmac
442
443@defmac MD_EXEC_PREFIX
444If defined, this macro is an additional prefix to try after
445@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
446when the compiler is built as a cross
447compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
448to the list of directories used to find the assembler in @file{configure.in}.
449@end defmac
450
451@defmac STANDARD_STARTFILE_PREFIX
452Define this macro as a C string constant if you wish to override the
453standard choice of @code{libdir} as the default prefix to
454try when searching for startup files such as @file{crt0.o}.
455@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
456is built as a cross compiler.
457@end defmac
458
459@defmac STANDARD_STARTFILE_PREFIX_1
460Define this macro as a C string constant if you wish to override the
461standard choice of @code{/lib} as a prefix to try after the default prefix
462when searching for startup files such as @file{crt0.o}.
463@code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
464is built as a cross compiler.
465@end defmac
466
467@defmac STANDARD_STARTFILE_PREFIX_2
468Define this macro as a C string constant if you wish to override the
469standard choice of @code{/lib} as yet another prefix to try after the
470default prefix when searching for startup files such as @file{crt0.o}.
471@code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
472is built as a cross compiler.
473@end defmac
474
475@defmac MD_STARTFILE_PREFIX
476If defined, this macro supplies an additional prefix to try after the
477standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
478compiler is built as a cross compiler.
479@end defmac
480
481@defmac MD_STARTFILE_PREFIX_1
482If defined, this macro supplies yet another prefix to try after the
483standard prefixes. It is not searched when the compiler is built as a
484cross compiler.
485@end defmac
486
487@defmac INIT_ENVIRONMENT
488Define this macro as a C string constant if you wish to set environment
489variables for programs called by the driver, such as the assembler and
490loader. The driver passes the value of this macro to @code{putenv} to
491initialize the necessary environment variables.
492@end defmac
493
494@defmac LOCAL_INCLUDE_DIR
495Define this macro as a C string constant if you wish to override the
496standard choice of @file{/usr/local/include} as the default prefix to
497try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
498comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
499
500Cross compilers do not search either @file{/usr/local/include} or its
501replacement.
502@end defmac
503
504@defmac SYSTEM_INCLUDE_DIR
505Define this macro as a C string constant if you wish to specify a
506system-specific directory to search for header files before the standard
507directory. @code{SYSTEM_INCLUDE_DIR} comes before
508@code{STANDARD_INCLUDE_DIR} in the search order.
509
510Cross compilers do not use this macro and do not search the directory
511specified.
512@end defmac
513
514@defmac STANDARD_INCLUDE_DIR
515Define this macro as a C string constant if you wish to override the
516standard choice of @file{/usr/include} as the default prefix to
517try when searching for header files.
518
519Cross compilers ignore this macro and do not search either
520@file{/usr/include} or its replacement.
521@end defmac
522
523@defmac STANDARD_INCLUDE_COMPONENT
524The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
525See @code{INCLUDE_DEFAULTS}, below, for the description of components.
526If you do not define this macro, no component is used.
527@end defmac
528
529@defmac INCLUDE_DEFAULTS
530Define this macro if you wish to override the entire default search path
531for include files. For a native compiler, the default search path
532usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
533@code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
534@code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
535and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
536and specify private search areas for GCC@. The directory
537@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
538
539The definition should be an initializer for an array of structures.
540Each array element should have four elements: the directory name (a
541string constant), the component name (also a string constant), a flag
542for C++-only directories,
543and a flag showing that the includes in the directory don't need to be
544wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
545the array with a null element.
546
547The component name denotes what GNU package the include file is part of,
548if any, in all uppercase letters. For example, it might be @samp{GCC}
549or @samp{BINUTILS}. If the package is part of a vendor-supplied
550operating system, code the component name as @samp{0}.
551
552For example, here is the definition used for VAX/VMS:
553
554@smallexample
555#define INCLUDE_DEFAULTS \
556@{ \
557 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
558 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
559 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
560 @{ ".", 0, 0, 0@}, \
561 @{ 0, 0, 0, 0@} \
562@}
563@end smallexample
564@end defmac
565
566Here is the order of prefixes tried for exec files:
567
568@enumerate
569@item
570Any prefixes specified by the user with @option{-B}.
571
572@item
573The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
574is not set and the compiler has not been installed in the configure-time
575@var{prefix}, the location in which the compiler has actually been installed.
576
577@item
578The directories specified by the environment variable @code{COMPILER_PATH}.
579
580@item
581The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
582in the configured-time @var{prefix}.
583
584@item
585The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
586
587@item
588The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
589
590@item
591The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
592compiler.
593@end enumerate
594
595Here is the order of prefixes tried for startfiles:
596
597@enumerate
598@item
599Any prefixes specified by the user with @option{-B}.
600
601@item
602The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
603value based on the installed toolchain location.
604
605@item
606The directories specified by the environment variable @code{LIBRARY_PATH}
607(or port-specific name; native only, cross compilers do not use this).
608
609@item
610The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
611in the configured @var{prefix} or this is a native compiler.
612
613@item
614The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
615
616@item
617The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
618compiler.
619
620@item
621The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
622native compiler, or we have a target system root.
623
624@item
625The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
626native compiler, or we have a target system root.
627
628@item
629The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
630If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
631the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
632
633@item
634The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
635compiler, or we have a target system root. The default for this macro is
636@file{/lib/}.
637
638@item
639The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
640compiler, or we have a target system root. The default for this macro is
641@file{/usr/lib/}.
642@end enumerate
643
644@node Run-time Target
645@section Run-time Target Specification
646@cindex run-time target specification
647@cindex predefined macros
648@cindex target specifications
649
650@c prevent bad page break with this line
651Here are run-time target specifications.
652
653@defmac TARGET_CPU_CPP_BUILTINS ()
654This function-like macro expands to a block of code that defines
655built-in preprocessor macros and assertions for the target CPU, using
656the functions @code{builtin_define}, @code{builtin_define_std} and
657@code{builtin_assert}. When the front end
658calls this macro it provides a trailing semicolon, and since it has
659finished command line option processing your code can use those
660results freely.
661
662@code{builtin_assert} takes a string in the form you pass to the
663command-line option @option{-A}, such as @code{cpu=mips}, and creates
664the assertion. @code{builtin_define} takes a string in the form
665accepted by option @option{-D} and unconditionally defines the macro.
666
667@code{builtin_define_std} takes a string representing the name of an
668object-like macro. If it doesn't lie in the user's namespace,
669@code{builtin_define_std} defines it unconditionally. Otherwise, it
670defines a version with two leading underscores, and another version
671with two leading and trailing underscores, and defines the original
672only if an ISO standard was not requested on the command line. For
673example, passing @code{unix} defines @code{__unix}, @code{__unix__}
674and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
675@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
676defines only @code{_ABI64}.
677
678You can also test for the C dialect being compiled. The variable
679@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
680or @code{clk_objective_c}. Note that if we are preprocessing
681assembler, this variable will be @code{clk_c} but the function-like
682macro @code{preprocessing_asm_p()} will return true, so you might want
683to check for that first. If you need to check for strict ANSI, the
684variable @code{flag_iso} can be used. The function-like macro
685@code{preprocessing_trad_p()} can be used to check for traditional
686preprocessing.
687@end defmac
688
689@defmac TARGET_OS_CPP_BUILTINS ()
690Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
691and is used for the target operating system instead.
692@end defmac
693
694@defmac TARGET_OBJFMT_CPP_BUILTINS ()
695Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
696and is used for the target object format. @file{elfos.h} uses this
697macro to define @code{__ELF__}, so you probably do not need to define
698it yourself.
699@end defmac
700
701@deftypevar {extern int} target_flags
702This variable is declared in @file{options.h}, which is included before
703any target-specific headers.
704@end deftypevar
705
706@hook TARGET_DEFAULT_TARGET_FLAGS
707This variable specifies the initial value of @code{target_flags}.
708Its default setting is 0.
709@end deftypevr
710
711@cindex optional hardware or system features
712@cindex features, optional, in system conventions
713
714@hook TARGET_HANDLE_OPTION
715This hook is called whenever the user specifies one of the
716target-specific options described by the @file{.opt} definition files
717(@pxref{Options}). It has the opportunity to do some option-specific
718processing and should return true if the option is valid. The default
719definition does nothing but return true.
720
721@var{code} specifies the @code{OPT_@var{name}} enumeration value
722associated with the selected option; @var{name} is just a rendering of
723the option name in which non-alphanumeric characters are replaced by
724underscores. @var{arg} specifies the string argument and is null if
725no argument was given. If the option is flagged as a @code{UInteger}
726(@pxref{Option properties}), @var{value} is the numeric value of the
727argument. Otherwise @var{value} is 1 if the positive form of the
728option was used and 0 if the ``no-'' form was.
729@end deftypefn
730
731@hook TARGET_HANDLE_C_OPTION
732This target hook is called whenever the user specifies one of the
733target-specific C language family options described by the @file{.opt}
734definition files(@pxref{Options}). It has the opportunity to do some
735option-specific processing and should return true if the option is
736valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
737default definition does nothing but return false.
738
739In general, you should use @code{TARGET_HANDLE_OPTION} to handle
740options. However, if processing an option requires routines that are
741only available in the C (and related language) front ends, then you
742should use @code{TARGET_HANDLE_C_OPTION} instead.
743@end deftypefn
744
26705988
IS
745@hook TARGET_OBJC_CONSTRUCT_STRING
746
38f8b050
JR
747@defmac TARGET_VERSION
748This macro is a C statement to print on @code{stderr} a string
749describing the particular machine description choice. Every machine
750description should define @code{TARGET_VERSION}. For example:
751
752@smallexample
753#ifdef MOTOROLA
754#define TARGET_VERSION \
755 fprintf (stderr, " (68k, Motorola syntax)");
756#else
757#define TARGET_VERSION \
758 fprintf (stderr, " (68k, MIT syntax)");
759#endif
760@end smallexample
761@end defmac
762
38f8b050
JR
763@hook TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE
764This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
765but is called when the optimize level is changed via an attribute or
766pragma or when it is reset at the end of the code affected by the
767attribute or pragma. It is not called at the beginning of compilation
768when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
769actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
770@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
771@end deftypefn
772
773@defmac C_COMMON_OVERRIDE_OPTIONS
c5387660
JM
774This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
775but is only used in the C
38f8b050
JR
776language frontends (C, Objective-C, C++, Objective-C++) and so can be
777used to alter option flag variables which only exist in those
778frontends.
779@end defmac
780
3020190e 781@hook TARGET_OPTION_OPTIMIZATION_TABLE
38f8b050 782Some machines may desire to change what optimizations are performed for
3020190e
JM
783various optimization levels. This variable, if defined, describes
784options to enable at particular sets of optimization levels. These
785options are processed once
38f8b050 786just after the optimization level is determined and before the remainder
3020190e
JM
787of the command options have been parsed, so may be overridden by other
788options passed explicily.
38f8b050 789
3020190e 790This processing is run once at program startup and when the optimization
38f8b050
JR
791options are changed via @code{#pragma GCC optimize} or by using the
792@code{optimize} attribute.
3020190e 793@end deftypevr
38f8b050 794
7e4aae92
JM
795@hook TARGET_OPTION_INIT_STRUCT
796
128dc8e2
JM
797@hook TARGET_OPTION_DEFAULT_PARAMS
798
38f8b050
JR
799@hook TARGET_HELP
800This hook is called in response to the user invoking
801@option{--target-help} on the command line. It gives the target a
802chance to display extra information on the target specific command
803line options found in its @file{.opt} file.
804@end deftypefn
805
3bd36029
RS
806@defmac SWITCHABLE_TARGET
807Some targets need to switch between substantially different subtargets
808during compilation. For example, the MIPS target has one subtarget for
809the traditional MIPS architecture and another for MIPS16. Source code
810can switch between these two subarchitectures using the @code{mips16}
811and @code{nomips16} attributes.
812
813Such subtargets can differ in things like the set of available
814registers, the set of available instructions, the costs of various
815operations, and so on. GCC caches a lot of this type of information
816in global variables, and recomputing them for each subtarget takes a
817significant amount of time. The compiler therefore provides a facility
818for maintaining several versions of the global variables and quickly
819switching between them; see @file{target-globals.h} for details.
820
821Define this macro to 1 if your target needs this facility. The default
822is 0.
823@end defmac
824
38f8b050
JR
825@node Per-Function Data
826@section Defining data structures for per-function information.
827@cindex per-function data
828@cindex data structures
829
830If the target needs to store information on a per-function basis, GCC
831provides a macro and a couple of variables to allow this. Note, just
832using statics to store the information is a bad idea, since GCC supports
833nested functions, so you can be halfway through encoding one function
834when another one comes along.
835
836GCC defines a data structure called @code{struct function} which
837contains all of the data specific to an individual function. This
838structure contains a field called @code{machine} whose type is
839@code{struct machine_function *}, which can be used by targets to point
840to their own specific data.
841
842If a target needs per-function specific data it should define the type
843@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
844This macro should be used to initialize the function pointer
845@code{init_machine_status}. This pointer is explained below.
846
847One typical use of per-function, target specific data is to create an
848RTX to hold the register containing the function's return address. This
849RTX can then be used to implement the @code{__builtin_return_address}
850function, for level 0.
851
852Note---earlier implementations of GCC used a single data area to hold
853all of the per-function information. Thus when processing of a nested
854function began the old per-function data had to be pushed onto a
855stack, and when the processing was finished, it had to be popped off the
856stack. GCC used to provide function pointers called
857@code{save_machine_status} and @code{restore_machine_status} to handle
858the saving and restoring of the target specific information. Since the
859single data area approach is no longer used, these pointers are no
860longer supported.
861
862@defmac INIT_EXPANDERS
863Macro called to initialize any target specific information. This macro
864is called once per function, before generation of any RTL has begun.
865The intention of this macro is to allow the initialization of the
866function pointer @code{init_machine_status}.
867@end defmac
868
869@deftypevar {void (*)(struct function *)} init_machine_status
870If this function pointer is non-@code{NULL} it will be called once per
871function, before function compilation starts, in order to allow the
872target to perform any target specific initialization of the
873@code{struct function} structure. It is intended that this would be
874used to initialize the @code{machine} of that structure.
875
876@code{struct machine_function} structures are expected to be freed by GC@.
877Generally, any memory that they reference must be allocated by using
878GC allocation, including the structure itself.
879@end deftypevar
880
881@node Storage Layout
882@section Storage Layout
883@cindex storage layout
884
885Note that the definitions of the macros in this table which are sizes or
886alignments measured in bits do not need to be constant. They can be C
887expressions that refer to static variables, such as the @code{target_flags}.
888@xref{Run-time Target}.
889
890@defmac BITS_BIG_ENDIAN
891Define this macro to have the value 1 if the most significant bit in a
892byte has the lowest number; otherwise define it to have the value zero.
893This means that bit-field instructions count from the most significant
894bit. If the machine has no bit-field instructions, then this must still
895be defined, but it doesn't matter which value it is defined to. This
896macro need not be a constant.
897
898This macro does not affect the way structure fields are packed into
899bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
900@end defmac
901
902@defmac BYTES_BIG_ENDIAN
903Define this macro to have the value 1 if the most significant byte in a
904word has the lowest number. This macro need not be a constant.
905@end defmac
906
907@defmac WORDS_BIG_ENDIAN
908Define this macro to have the value 1 if, in a multiword object, the
909most significant word has the lowest number. This applies to both
910memory locations and registers; GCC fundamentally assumes that the
911order of words in memory is the same as the order in registers. This
912macro need not be a constant.
913@end defmac
914
38f8b050
JR
915@defmac FLOAT_WORDS_BIG_ENDIAN
916Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
917@code{TFmode} floating point numbers are stored in memory with the word
918containing the sign bit at the lowest address; otherwise define it to
919have the value 0. This macro need not be a constant.
920
921You need not define this macro if the ordering is the same as for
922multi-word integers.
923@end defmac
924
925@defmac BITS_PER_UNIT
926Define this macro to be the number of bits in an addressable storage
927unit (byte). If you do not define this macro the default is 8.
928@end defmac
929
930@defmac BITS_PER_WORD
931Number of bits in a word. If you do not define this macro, the default
932is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
933@end defmac
934
935@defmac MAX_BITS_PER_WORD
936Maximum number of bits in a word. If this is undefined, the default is
937@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
938largest value that @code{BITS_PER_WORD} can have at run-time.
939@end defmac
940
941@defmac UNITS_PER_WORD
942Number of storage units in a word; normally the size of a general-purpose
943register, a power of two from 1 or 8.
944@end defmac
945
946@defmac MIN_UNITS_PER_WORD
947Minimum number of units in a word. If this is undefined, the default is
948@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
949smallest value that @code{UNITS_PER_WORD} can have at run-time.
950@end defmac
951
38f8b050
JR
952@defmac POINTER_SIZE
953Width of a pointer, in bits. You must specify a value no wider than the
954width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
955you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
956a value the default is @code{BITS_PER_WORD}.
957@end defmac
958
959@defmac POINTERS_EXTEND_UNSIGNED
960A C expression that determines how pointers should be extended from
961@code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
962greater than zero if pointers should be zero-extended, zero if they
963should be sign-extended, and negative if some other sort of conversion
964is needed. In the last case, the extension is done by the target's
965@code{ptr_extend} instruction.
966
967You need not define this macro if the @code{ptr_mode}, @code{Pmode}
968and @code{word_mode} are all the same width.
969@end defmac
970
971@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
972A macro to update @var{m} and @var{unsignedp} when an object whose type
973is @var{type} and which has the specified mode and signedness is to be
974stored in a register. This macro is only called when @var{type} is a
975scalar type.
976
977On most RISC machines, which only have operations that operate on a full
978register, define this macro to set @var{m} to @code{word_mode} if
979@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
980cases, only integer modes should be widened because wider-precision
981floating-point operations are usually more expensive than their narrower
982counterparts.
983
984For most machines, the macro definition does not change @var{unsignedp}.
985However, some machines, have instructions that preferentially handle
986either signed or unsigned quantities of certain modes. For example, on
987the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
988sign-extend the result to 64 bits. On such machines, set
989@var{unsignedp} according to which kind of extension is more efficient.
990
991Do not define this macro if it would never modify @var{m}.
992@end defmac
993
994@hook TARGET_PROMOTE_FUNCTION_MODE
995Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
996function return values. The target hook should return the new mode
997and possibly change @code{*@var{punsignedp}} if the promotion should
998change signedness. This function is called only for scalar @emph{or
999pointer} types.
1000
1001@var{for_return} allows to distinguish the promotion of arguments and
1002return values. If it is @code{1}, a return value is being promoted and
1003@code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
1004If it is @code{2}, the returned mode should be that of the register in
1005which an incoming parameter is copied, or the outgoing result is computed;
1006then the hook should return the same mode as @code{promote_mode}, though
1007the signedness may be different.
1008
1009The default is to not promote arguments and return values. You can
1010also define the hook to @code{default_promote_function_mode_always_promote}
1011if you would like to apply the same rules given by @code{PROMOTE_MODE}.
1012@end deftypefn
1013
1014@defmac PARM_BOUNDARY
1015Normal alignment required for function parameters on the stack, in
1016bits. All stack parameters receive at least this much alignment
1017regardless of data type. On most machines, this is the same as the
1018size of an integer.
1019@end defmac
1020
1021@defmac STACK_BOUNDARY
1022Define this macro to the minimum alignment enforced by hardware for the
1023stack pointer on this machine. The definition is a C expression for the
1024desired alignment (measured in bits). This value is used as a default
1025if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1026this should be the same as @code{PARM_BOUNDARY}.
1027@end defmac
1028
1029@defmac PREFERRED_STACK_BOUNDARY
1030Define this macro if you wish to preserve a certain alignment for the
1031stack pointer, greater than what the hardware enforces. The definition
1032is a C expression for the desired alignment (measured in bits). This
1033macro must evaluate to a value equal to or larger than
1034@code{STACK_BOUNDARY}.
1035@end defmac
1036
1037@defmac INCOMING_STACK_BOUNDARY
1038Define this macro if the incoming stack boundary may be different
1039from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
1040to a value equal to or larger than @code{STACK_BOUNDARY}.
1041@end defmac
1042
1043@defmac FUNCTION_BOUNDARY
1044Alignment required for a function entry point, in bits.
1045@end defmac
1046
1047@defmac BIGGEST_ALIGNMENT
1048Biggest alignment that any data type can require on this machine, in
1049bits. Note that this is not the biggest alignment that is supported,
1050just the biggest alignment that, when violated, may cause a fault.
1051@end defmac
1052
1053@defmac MALLOC_ABI_ALIGNMENT
1054Alignment, in bits, a C conformant malloc implementation has to
1055provide. If not defined, the default value is @code{BITS_PER_WORD}.
1056@end defmac
1057
1058@defmac ATTRIBUTE_ALIGNED_VALUE
1059Alignment used by the @code{__attribute__ ((aligned))} construct. If
1060not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1061@end defmac
1062
1063@defmac MINIMUM_ATOMIC_ALIGNMENT
1064If defined, the smallest alignment, in bits, that can be given to an
1065object that can be referenced in one operation, without disturbing any
1066nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1067on machines that don't have byte or half-word store operations.
1068@end defmac
1069
1070@defmac BIGGEST_FIELD_ALIGNMENT
1071Biggest alignment that any structure or union field can require on this
1072machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1073structure and union fields only, unless the field alignment has been set
1074by the @code{__attribute__ ((aligned (@var{n})))} construct.
1075@end defmac
1076
1077@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1078An expression for the alignment of a structure field @var{field} if the
1079alignment computed in the usual way (including applying of
1080@code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1081alignment) is @var{computed}. It overrides alignment only if the
1082field alignment has not been set by the
1083@code{__attribute__ ((aligned (@var{n})))} construct.
1084@end defmac
1085
1086@defmac MAX_STACK_ALIGNMENT
1087Biggest stack alignment guaranteed by the backend. Use this macro
1088to specify the maximum alignment of a variable on stack.
1089
1090If not defined, the default value is @code{STACK_BOUNDARY}.
1091
1092@c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1093@c But the fix for PR 32893 indicates that we can only guarantee
1094@c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1095@c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1096@end defmac
1097
1098@defmac MAX_OFILE_ALIGNMENT
1099Biggest alignment supported by the object file format of this machine.
1100Use this macro to limit the alignment which can be specified using the
1101@code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1102the default value is @code{BIGGEST_ALIGNMENT}.
1103
1104On systems that use ELF, the default (in @file{config/elfos.h}) is
1105the largest supported 32-bit ELF section alignment representable on
1106a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1107On 32-bit ELF the largest supported section alignment in bits is
1108@samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1109@end defmac
1110
1111@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1112If defined, a C expression to compute the alignment for a variable in
1113the static store. @var{type} is the data type, and @var{basic-align} is
1114the alignment that the object would ordinarily have. The value of this
1115macro is used instead of that alignment to align the object.
1116
1117If this macro is not defined, then @var{basic-align} is used.
1118
1119@findex strcpy
1120One use of this macro is to increase alignment of medium-size data to
1121make it all fit in fewer cache lines. Another is to cause character
1122arrays to be word-aligned so that @code{strcpy} calls that copy
1123constants to character arrays can be done inline.
1124@end defmac
1125
1126@defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1127If defined, a C expression to compute the alignment given to a constant
1128that is being placed in memory. @var{constant} is the constant and
1129@var{basic-align} is the alignment that the object would ordinarily
1130have. The value of this macro is used instead of that alignment to
1131align the object.
1132
1133If this macro is not defined, then @var{basic-align} is used.
1134
1135The typical use of this macro is to increase alignment for string
1136constants to be word aligned so that @code{strcpy} calls that copy
1137constants can be done inline.
1138@end defmac
1139
1140@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1141If defined, a C expression to compute the alignment for a variable in
1142the local store. @var{type} is the data type, and @var{basic-align} is
1143the alignment that the object would ordinarily have. The value of this
1144macro is used instead of that alignment to align the object.
1145
1146If this macro is not defined, then @var{basic-align} is used.
1147
1148One use of this macro is to increase alignment of medium-size data to
1149make it all fit in fewer cache lines.
1150@end defmac
1151
1152@defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1153If defined, a C expression to compute the alignment for stack slot.
1154@var{type} is the data type, @var{mode} is the widest mode available,
1155and @var{basic-align} is the alignment that the slot would ordinarily
1156have. The value of this macro is used instead of that alignment to
1157align the slot.
1158
1159If this macro is not defined, then @var{basic-align} is used when
1160@var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1161be used.
1162
1163This macro is to set alignment of stack slot to the maximum alignment
1164of all possible modes which the slot may have.
1165@end defmac
1166
1167@defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1168If defined, a C expression to compute the alignment for a local
1169variable @var{decl}.
1170
1171If this macro is not defined, then
1172@code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1173is used.
1174
1175One use of this macro is to increase alignment of medium-size data to
1176make it all fit in fewer cache lines.
1177@end defmac
1178
1179@defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1180If defined, a C expression to compute the minimum required alignment
1181for dynamic stack realignment purposes for @var{exp} (a type or decl),
1182@var{mode}, assuming normal alignment @var{align}.
1183
1184If this macro is not defined, then @var{align} will be used.
1185@end defmac
1186
1187@defmac EMPTY_FIELD_BOUNDARY
1188Alignment in bits to be given to a structure bit-field that follows an
1189empty field such as @code{int : 0;}.
1190
1191If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1192@end defmac
1193
1194@defmac STRUCTURE_SIZE_BOUNDARY
1195Number of bits which any structure or union's size must be a multiple of.
1196Each structure or union's size is rounded up to a multiple of this.
1197
1198If you do not define this macro, the default is the same as
1199@code{BITS_PER_UNIT}.
1200@end defmac
1201
1202@defmac STRICT_ALIGNMENT
1203Define this macro to be the value 1 if instructions will fail to work
1204if given data not on the nominal alignment. If instructions will merely
1205go slower in that case, define this macro as 0.
1206@end defmac
1207
1208@defmac PCC_BITFIELD_TYPE_MATTERS
1209Define this if you wish to imitate the way many other C compilers handle
1210alignment of bit-fields and the structures that contain them.
1211
1212The behavior is that the type written for a named bit-field (@code{int},
1213@code{short}, or other integer type) imposes an alignment for the entire
1214structure, as if the structure really did contain an ordinary field of
1215that type. In addition, the bit-field is placed within the structure so
1216that it would fit within such a field, not crossing a boundary for it.
1217
1218Thus, on most machines, a named bit-field whose type is written as
1219@code{int} would not cross a four-byte boundary, and would force
1220four-byte alignment for the whole structure. (The alignment used may
1221not be four bytes; it is controlled by the other alignment parameters.)
1222
1223An unnamed bit-field will not affect the alignment of the containing
1224structure.
1225
1226If the macro is defined, its definition should be a C expression;
1227a nonzero value for the expression enables this behavior.
1228
1229Note that if this macro is not defined, or its value is zero, some
1230bit-fields may cross more than one alignment boundary. The compiler can
1231support such references if there are @samp{insv}, @samp{extv}, and
1232@samp{extzv} insns that can directly reference memory.
1233
1234The other known way of making bit-fields work is to define
1235@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1236Then every structure can be accessed with fullwords.
1237
1238Unless the machine has bit-field instructions or you define
1239@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1240@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1241
1242If your aim is to make GCC use the same conventions for laying out
1243bit-fields as are used by another compiler, here is how to investigate
1244what the other compiler does. Compile and run this program:
1245
1246@smallexample
1247struct foo1
1248@{
1249 char x;
1250 char :0;
1251 char y;
1252@};
1253
1254struct foo2
1255@{
1256 char x;
1257 int :0;
1258 char y;
1259@};
1260
1261main ()
1262@{
1263 printf ("Size of foo1 is %d\n",
1264 sizeof (struct foo1));
1265 printf ("Size of foo2 is %d\n",
1266 sizeof (struct foo2));
1267 exit (0);
1268@}
1269@end smallexample
1270
1271If this prints 2 and 5, then the compiler's behavior is what you would
1272get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1273@end defmac
1274
1275@defmac BITFIELD_NBYTES_LIMITED
1276Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1277to aligning a bit-field within the structure.
1278@end defmac
1279
1280@hook TARGET_ALIGN_ANON_BITFIELD
1281When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1282whether unnamed bitfields affect the alignment of the containing
1283structure. The hook should return true if the structure should inherit
1284the alignment requirements of an unnamed bitfield's type.
1285@end deftypefn
1286
1287@hook TARGET_NARROW_VOLATILE_BITFIELD
1288This target hook should return @code{true} if accesses to volatile bitfields
1289should use the narrowest mode possible. It should return @code{false} if
1290these accesses should use the bitfield container type.
1291
1292The default is @code{!TARGET_STRICT_ALIGN}.
1293@end deftypefn
1294
1295@defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1296Return 1 if a structure or array containing @var{field} should be accessed using
1297@code{BLKMODE}.
1298
1299If @var{field} is the only field in the structure, @var{mode} is its
1300mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1301case where structures of one field would require the structure's mode to
1302retain the field's mode.
1303
1304Normally, this is not needed.
1305@end defmac
1306
1307@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1308Define this macro as an expression for the alignment of a type (given
1309by @var{type} as a tree node) if the alignment computed in the usual
1310way is @var{computed} and the alignment explicitly specified was
1311@var{specified}.
1312
1313The default is to use @var{specified} if it is larger; otherwise, use
1314the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1315@end defmac
1316
1317@defmac MAX_FIXED_MODE_SIZE
1318An integer expression for the size in bits of the largest integer
1319machine mode that should actually be used. All integer machine modes of
1320this size or smaller can be used for structures and unions with the
1321appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1322(DImode)} is assumed.
1323@end defmac
1324
1325@defmac STACK_SAVEAREA_MODE (@var{save_level})
1326If defined, an expression of type @code{enum machine_mode} that
1327specifies the mode of the save area operand of a
1328@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1329@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1330@code{SAVE_NONLOCAL} and selects which of the three named patterns is
1331having its mode specified.
1332
1333You need not define this macro if it always returns @code{Pmode}. You
1334would most commonly define this macro if the
1335@code{save_stack_@var{level}} patterns need to support both a 32- and a
133664-bit mode.
1337@end defmac
1338
1339@defmac STACK_SIZE_MODE
1340If defined, an expression of type @code{enum machine_mode} that
1341specifies the mode of the size increment operand of an
1342@code{allocate_stack} named pattern (@pxref{Standard Names}).
1343
1344You need not define this macro if it always returns @code{word_mode}.
1345You would most commonly define this macro if the @code{allocate_stack}
1346pattern needs to support both a 32- and a 64-bit mode.
1347@end defmac
1348
1349@hook TARGET_LIBGCC_CMP_RETURN_MODE
1350This target hook should return the mode to be used for the return value
1351of compare instructions expanded to libgcc calls. If not defined
1352@code{word_mode} is returned which is the right choice for a majority of
1353targets.
1354@end deftypefn
1355
1356@hook TARGET_LIBGCC_SHIFT_COUNT_MODE
1357This target hook should return the mode to be used for the shift count operand
1358of shift instructions expanded to libgcc calls. If not defined
1359@code{word_mode} is returned which is the right choice for a majority of
1360targets.
1361@end deftypefn
1362
1363@hook TARGET_UNWIND_WORD_MODE
1364Return machine mode to be used for @code{_Unwind_Word} type.
1365The default is to use @code{word_mode}.
1366@end deftypefn
1367
1368@defmac ROUND_TOWARDS_ZERO
1369If defined, this macro should be true if the prevailing rounding
1370mode is towards zero.
1371
1372Defining this macro only affects the way @file{libgcc.a} emulates
1373floating-point arithmetic.
1374
1375Not defining this macro is equivalent to returning zero.
1376@end defmac
1377
1378@defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1379This macro should return true if floats with @var{size}
1380bits do not have a NaN or infinity representation, but use the largest
1381exponent for normal numbers instead.
1382
1383Defining this macro only affects the way @file{libgcc.a} emulates
1384floating-point arithmetic.
1385
1386The default definition of this macro returns false for all sizes.
1387@end defmac
1388
1389@hook TARGET_MS_BITFIELD_LAYOUT_P
1390This target hook returns @code{true} if bit-fields in the given
1391@var{record_type} are to be laid out following the rules of Microsoft
1392Visual C/C++, namely: (i) a bit-field won't share the same storage
1393unit with the previous bit-field if their underlying types have
1394different sizes, and the bit-field will be aligned to the highest
1395alignment of the underlying types of itself and of the previous
1396bit-field; (ii) a zero-sized bit-field will affect the alignment of
1397the whole enclosing structure, even if it is unnamed; except that
1398(iii) a zero-sized bit-field will be disregarded unless it follows
1399another bit-field of nonzero size. If this hook returns @code{true},
1400other macros that control bit-field layout are ignored.
1401
1402When a bit-field is inserted into a packed record, the whole size
1403of the underlying type is used by one or more same-size adjacent
1404bit-fields (that is, if its long:3, 32 bits is used in the record,
1405and any additional adjacent long bit-fields are packed into the same
1406chunk of 32 bits. However, if the size changes, a new field of that
1407size is allocated). In an unpacked record, this is the same as using
1408alignment, but not equivalent when packing.
1409
1410If both MS bit-fields and @samp{__attribute__((packed))} are used,
1411the latter will take precedence. If @samp{__attribute__((packed))} is
1412used on a single field when MS bit-fields are in use, it will take
1413precedence for that field, but the alignment of the rest of the structure
1414may affect its placement.
1415@end deftypefn
1416
1417@hook TARGET_DECIMAL_FLOAT_SUPPORTED_P
1418Returns true if the target supports decimal floating point.
1419@end deftypefn
1420
1421@hook TARGET_FIXED_POINT_SUPPORTED_P
1422Returns true if the target supports fixed-point arithmetic.
1423@end deftypefn
1424
1425@hook TARGET_EXPAND_TO_RTL_HOOK
1426This hook is called just before expansion into rtl, allowing the target
1427to perform additional initializations or analysis before the expansion.
1428For example, the rs6000 port uses it to allocate a scratch stack slot
1429for use in copying SDmode values between memory and floating point
1430registers whenever the function being expanded has any SDmode
1431usage.
1432@end deftypefn
1433
1434@hook TARGET_INSTANTIATE_DECLS
1435This hook allows the backend to perform additional instantiations on rtl
1436that are not actually in any insns yet, but will be later.
1437@end deftypefn
1438
1439@hook TARGET_MANGLE_TYPE
1440If your target defines any fundamental types, or any types your target
1441uses should be mangled differently from the default, define this hook
1442to return the appropriate encoding for these types as part of a C++
1443mangled name. The @var{type} argument is the tree structure representing
1444the type to be mangled. The hook may be applied to trees which are
1445not target-specific fundamental types; it should return @code{NULL}
1446for all such types, as well as arguments it does not recognize. If the
1447return value is not @code{NULL}, it must point to a statically-allocated
1448string constant.
1449
1450Target-specific fundamental types might be new fundamental types or
1451qualified versions of ordinary fundamental types. Encode new
1452fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1453is the name used for the type in source code, and @var{n} is the
1454length of @var{name} in decimal. Encode qualified versions of
1455ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1456@var{name} is the name used for the type qualifier in source code,
1457@var{n} is the length of @var{name} as above, and @var{code} is the
1458code used to represent the unqualified version of this type. (See
1459@code{write_builtin_type} in @file{cp/mangle.c} for the list of
1460codes.) In both cases the spaces are for clarity; do not include any
1461spaces in your string.
1462
1463This hook is applied to types prior to typedef resolution. If the mangled
1464name for a particular type depends only on that type's main variant, you
1465can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1466before mangling.
1467
1468The default version of this hook always returns @code{NULL}, which is
1469appropriate for a target that does not define any new fundamental
1470types.
1471@end deftypefn
1472
1473@node Type Layout
1474@section Layout of Source Language Data Types
1475
1476These macros define the sizes and other characteristics of the standard
1477basic data types used in programs being compiled. Unlike the macros in
1478the previous section, these apply to specific features of C and related
1479languages, rather than to fundamental aspects of storage layout.
1480
1481@defmac INT_TYPE_SIZE
1482A C expression for the size in bits of the type @code{int} on the
1483target machine. If you don't define this, the default is one word.
1484@end defmac
1485
1486@defmac SHORT_TYPE_SIZE
1487A C expression for the size in bits of the type @code{short} on the
1488target machine. If you don't define this, the default is half a word.
1489(If this would be less than one storage unit, it is rounded up to one
1490unit.)
1491@end defmac
1492
1493@defmac LONG_TYPE_SIZE
1494A C expression for the size in bits of the type @code{long} on the
1495target machine. If you don't define this, the default is one word.
1496@end defmac
1497
1498@defmac ADA_LONG_TYPE_SIZE
1499On some machines, the size used for the Ada equivalent of the type
1500@code{long} by a native Ada compiler differs from that used by C@. In
1501that situation, define this macro to be a C expression to be used for
1502the size of that type. If you don't define this, the default is the
1503value of @code{LONG_TYPE_SIZE}.
1504@end defmac
1505
1506@defmac LONG_LONG_TYPE_SIZE
1507A C expression for the size in bits of the type @code{long long} on the
1508target machine. If you don't define this, the default is two
1509words. If you want to support GNU Ada on your machine, the value of this
1510macro must be at least 64.
1511@end defmac
1512
1513@defmac CHAR_TYPE_SIZE
1514A C expression for the size in bits of the type @code{char} on the
1515target machine. If you don't define this, the default is
1516@code{BITS_PER_UNIT}.
1517@end defmac
1518
1519@defmac BOOL_TYPE_SIZE
1520A C expression for the size in bits of the C++ type @code{bool} and
1521C99 type @code{_Bool} on the target machine. If you don't define
1522this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1523@end defmac
1524
1525@defmac FLOAT_TYPE_SIZE
1526A C expression for the size in bits of the type @code{float} on the
1527target machine. If you don't define this, the default is one word.
1528@end defmac
1529
1530@defmac DOUBLE_TYPE_SIZE
1531A C expression for the size in bits of the type @code{double} on the
1532target machine. If you don't define this, the default is two
1533words.
1534@end defmac
1535
1536@defmac LONG_DOUBLE_TYPE_SIZE
1537A C expression for the size in bits of the type @code{long double} on
1538the target machine. If you don't define this, the default is two
1539words.
1540@end defmac
1541
1542@defmac SHORT_FRACT_TYPE_SIZE
1543A C expression for the size in bits of the type @code{short _Fract} on
1544the target machine. If you don't define this, the default is
1545@code{BITS_PER_UNIT}.
1546@end defmac
1547
1548@defmac FRACT_TYPE_SIZE
1549A C expression for the size in bits of the type @code{_Fract} on
1550the target machine. If you don't define this, the default is
1551@code{BITS_PER_UNIT * 2}.
1552@end defmac
1553
1554@defmac LONG_FRACT_TYPE_SIZE
1555A C expression for the size in bits of the type @code{long _Fract} on
1556the target machine. If you don't define this, the default is
1557@code{BITS_PER_UNIT * 4}.
1558@end defmac
1559
1560@defmac LONG_LONG_FRACT_TYPE_SIZE
1561A C expression for the size in bits of the type @code{long long _Fract} on
1562the target machine. If you don't define this, the default is
1563@code{BITS_PER_UNIT * 8}.
1564@end defmac
1565
1566@defmac SHORT_ACCUM_TYPE_SIZE
1567A C expression for the size in bits of the type @code{short _Accum} on
1568the target machine. If you don't define this, the default is
1569@code{BITS_PER_UNIT * 2}.
1570@end defmac
1571
1572@defmac ACCUM_TYPE_SIZE
1573A C expression for the size in bits of the type @code{_Accum} on
1574the target machine. If you don't define this, the default is
1575@code{BITS_PER_UNIT * 4}.
1576@end defmac
1577
1578@defmac LONG_ACCUM_TYPE_SIZE
1579A C expression for the size in bits of the type @code{long _Accum} on
1580the target machine. If you don't define this, the default is
1581@code{BITS_PER_UNIT * 8}.
1582@end defmac
1583
1584@defmac LONG_LONG_ACCUM_TYPE_SIZE
1585A C expression for the size in bits of the type @code{long long _Accum} on
1586the target machine. If you don't define this, the default is
1587@code{BITS_PER_UNIT * 16}.
1588@end defmac
1589
1590@defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1591Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1592if you want routines in @file{libgcc2.a} for a size other than
1593@code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1594default is @code{LONG_DOUBLE_TYPE_SIZE}.
1595@end defmac
1596
1597@defmac LIBGCC2_HAS_DF_MODE
a18bdccd 1598Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor
38f8b050
JR
1599@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1600@code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
a18bdccd 1601anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE}
38f8b050
JR
1602or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1603otherwise it is 0.
1604@end defmac
1605
1606@defmac LIBGCC2_HAS_XF_MODE
1607Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1608@code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1609anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1610is 80 then the default is 1, otherwise it is 0.
1611@end defmac
1612
1613@defmac LIBGCC2_HAS_TF_MODE
1614Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1615@code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1616anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1617is 128 then the default is 1, otherwise it is 0.
1618@end defmac
1619
1620@defmac SF_SIZE
1621@defmacx DF_SIZE
1622@defmacx XF_SIZE
1623@defmacx TF_SIZE
1624Define these macros to be the size in bits of the mantissa of
1625@code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1626if the defaults in @file{libgcc2.h} are inappropriate. By default,
1627@code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1628for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1629@code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
a18bdccd 1630@code{DOUBLE_TYPE_SIZE} or
38f8b050
JR
1631@code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1632@end defmac
1633
1634@defmac TARGET_FLT_EVAL_METHOD
1635A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1636assuming, if applicable, that the floating-point control word is in its
1637default state. If you do not define this macro the value of
1638@code{FLT_EVAL_METHOD} will be zero.
1639@end defmac
1640
1641@defmac WIDEST_HARDWARE_FP_SIZE
1642A C expression for the size in bits of the widest floating-point format
1643supported by the hardware. If you define this macro, you must specify a
1644value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1645If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1646is the default.
1647@end defmac
1648
1649@defmac DEFAULT_SIGNED_CHAR
1650An expression whose value is 1 or 0, according to whether the type
1651@code{char} should be signed or unsigned by default. The user can
1652always override this default with the options @option{-fsigned-char}
1653and @option{-funsigned-char}.
1654@end defmac
1655
1656@hook TARGET_DEFAULT_SHORT_ENUMS
1657This target hook should return true if the compiler should give an
1658@code{enum} type only as many bytes as it takes to represent the range
1659of possible values of that type. It should return false if all
1660@code{enum} types should be allocated like @code{int}.
1661
1662The default is to return false.
1663@end deftypefn
1664
1665@defmac SIZE_TYPE
1666A C expression for a string describing the name of the data type to use
1667for size values. The typedef name @code{size_t} is defined using the
1668contents of the string.
1669
1670The string can contain more than one keyword. If so, separate them with
1671spaces, and write first any length keyword, then @code{unsigned} if
1672appropriate, and finally @code{int}. The string must exactly match one
1673of the data type names defined in the function
1674@code{init_decl_processing} in the file @file{c-decl.c}. You may not
1675omit @code{int} or change the order---that would cause the compiler to
1676crash on startup.
1677
1678If you don't define this macro, the default is @code{"long unsigned
1679int"}.
1680@end defmac
1681
1682@defmac PTRDIFF_TYPE
1683A C expression for a string describing the name of the data type to use
1684for the result of subtracting two pointers. The typedef name
1685@code{ptrdiff_t} is defined using the contents of the string. See
1686@code{SIZE_TYPE} above for more information.
1687
1688If you don't define this macro, the default is @code{"long int"}.
1689@end defmac
1690
1691@defmac WCHAR_TYPE
1692A C expression for a string describing the name of the data type to use
1693for wide characters. The typedef name @code{wchar_t} is defined using
1694the contents of the string. See @code{SIZE_TYPE} above for more
1695information.
1696
1697If you don't define this macro, the default is @code{"int"}.
1698@end defmac
1699
1700@defmac WCHAR_TYPE_SIZE
1701A C expression for the size in bits of the data type for wide
1702characters. This is used in @code{cpp}, which cannot make use of
1703@code{WCHAR_TYPE}.
1704@end defmac
1705
1706@defmac WINT_TYPE
1707A C expression for a string describing the name of the data type to
1708use for wide characters passed to @code{printf} and returned from
1709@code{getwc}. The typedef name @code{wint_t} is defined using the
1710contents of the string. See @code{SIZE_TYPE} above for more
1711information.
1712
1713If you don't define this macro, the default is @code{"unsigned int"}.
1714@end defmac
1715
1716@defmac INTMAX_TYPE
1717A C expression for a string describing the name of the data type that
1718can represent any value of any standard or extended signed integer type.
1719The typedef name @code{intmax_t} is defined using the contents of the
1720string. See @code{SIZE_TYPE} above for more information.
1721
1722If you don't define this macro, the default is the first of
1723@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1724much precision as @code{long long int}.
1725@end defmac
1726
1727@defmac UINTMAX_TYPE
1728A C expression for a string describing the name of the data type that
1729can represent any value of any standard or extended unsigned integer
1730type. The typedef name @code{uintmax_t} is defined using the contents
1731of the string. See @code{SIZE_TYPE} above for more information.
1732
1733If you don't define this macro, the default is the first of
1734@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1735unsigned int"} that has as much precision as @code{long long unsigned
1736int}.
1737@end defmac
1738
1739@defmac SIG_ATOMIC_TYPE
1740@defmacx INT8_TYPE
1741@defmacx INT16_TYPE
1742@defmacx INT32_TYPE
1743@defmacx INT64_TYPE
1744@defmacx UINT8_TYPE
1745@defmacx UINT16_TYPE
1746@defmacx UINT32_TYPE
1747@defmacx UINT64_TYPE
1748@defmacx INT_LEAST8_TYPE
1749@defmacx INT_LEAST16_TYPE
1750@defmacx INT_LEAST32_TYPE
1751@defmacx INT_LEAST64_TYPE
1752@defmacx UINT_LEAST8_TYPE
1753@defmacx UINT_LEAST16_TYPE
1754@defmacx UINT_LEAST32_TYPE
1755@defmacx UINT_LEAST64_TYPE
1756@defmacx INT_FAST8_TYPE
1757@defmacx INT_FAST16_TYPE
1758@defmacx INT_FAST32_TYPE
1759@defmacx INT_FAST64_TYPE
1760@defmacx UINT_FAST8_TYPE
1761@defmacx UINT_FAST16_TYPE
1762@defmacx UINT_FAST32_TYPE
1763@defmacx UINT_FAST64_TYPE
1764@defmacx INTPTR_TYPE
1765@defmacx UINTPTR_TYPE
1766C expressions for the standard types @code{sig_atomic_t},
1767@code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1768@code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1769@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1770@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1771@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1772@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1773@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1774@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1775@code{SIZE_TYPE} above for more information.
1776
1777If any of these macros evaluates to a null pointer, the corresponding
1778type is not supported; if GCC is configured to provide
1779@code{<stdint.h>} in such a case, the header provided may not conform
1780to C99, depending on the type in question. The defaults for all of
1781these macros are null pointers.
1782@end defmac
1783
1784@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1785The C++ compiler represents a pointer-to-member-function with a struct
1786that looks like:
1787
1788@smallexample
1789 struct @{
1790 union @{
1791 void (*fn)();
1792 ptrdiff_t vtable_index;
1793 @};
1794 ptrdiff_t delta;
1795 @};
1796@end smallexample
1797
1798@noindent
1799The C++ compiler must use one bit to indicate whether the function that
1800will be called through a pointer-to-member-function is virtual.
1801Normally, we assume that the low-order bit of a function pointer must
1802always be zero. Then, by ensuring that the vtable_index is odd, we can
1803distinguish which variant of the union is in use. But, on some
1804platforms function pointers can be odd, and so this doesn't work. In
1805that case, we use the low-order bit of the @code{delta} field, and shift
1806the remainder of the @code{delta} field to the left.
1807
1808GCC will automatically make the right selection about where to store
1809this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1810However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1811set such that functions always start at even addresses, but the lowest
1812bit of pointers to functions indicate whether the function at that
1813address is in ARM or Thumb mode. If this is the case of your
1814architecture, you should define this macro to
1815@code{ptrmemfunc_vbit_in_delta}.
1816
1817In general, you should not have to define this macro. On architectures
1818in which function addresses are always even, according to
1819@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1820@code{ptrmemfunc_vbit_in_pfn}.
1821@end defmac
1822
1823@defmac TARGET_VTABLE_USES_DESCRIPTORS
1824Normally, the C++ compiler uses function pointers in vtables. This
1825macro allows the target to change to use ``function descriptors''
1826instead. Function descriptors are found on targets for whom a
1827function pointer is actually a small data structure. Normally the
1828data structure consists of the actual code address plus a data
1829pointer to which the function's data is relative.
1830
1831If vtables are used, the value of this macro should be the number
1832of words that the function descriptor occupies.
1833@end defmac
1834
1835@defmac TARGET_VTABLE_ENTRY_ALIGN
1836By default, the vtable entries are void pointers, the so the alignment
1837is the same as pointer alignment. The value of this macro specifies
1838the alignment of the vtable entry in bits. It should be defined only
1839when special alignment is necessary. */
1840@end defmac
1841
1842@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1843There are a few non-descriptor entries in the vtable at offsets below
1844zero. If these entries must be padded (say, to preserve the alignment
1845specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1846of words in each data entry.
1847@end defmac
1848
1849@node Registers
1850@section Register Usage
1851@cindex register usage
1852
1853This section explains how to describe what registers the target machine
1854has, and how (in general) they can be used.
1855
1856The description of which registers a specific instruction can use is
1857done with register classes; see @ref{Register Classes}. For information
1858on using registers to access a stack frame, see @ref{Frame Registers}.
1859For passing values in registers, see @ref{Register Arguments}.
1860For returning values in registers, see @ref{Scalar Return}.
1861
1862@menu
1863* Register Basics:: Number and kinds of registers.
1864* Allocation Order:: Order in which registers are allocated.
1865* Values in Registers:: What kinds of values each reg can hold.
1866* Leaf Functions:: Renumbering registers for leaf functions.
1867* Stack Registers:: Handling a register stack such as 80387.
1868@end menu
1869
1870@node Register Basics
1871@subsection Basic Characteristics of Registers
1872
1873@c prevent bad page break with this line
1874Registers have various characteristics.
1875
1876@defmac FIRST_PSEUDO_REGISTER
1877Number of hardware registers known to the compiler. They receive
1878numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1879pseudo register's number really is assigned the number
1880@code{FIRST_PSEUDO_REGISTER}.
1881@end defmac
1882
1883@defmac FIXED_REGISTERS
1884@cindex fixed register
1885An initializer that says which registers are used for fixed purposes
1886all throughout the compiled code and are therefore not available for
1887general allocation. These would include the stack pointer, the frame
1888pointer (except on machines where that can be used as a general
1889register when no frame pointer is needed), the program counter on
1890machines where that is considered one of the addressable registers,
1891and any other numbered register with a standard use.
1892
1893This information is expressed as a sequence of numbers, separated by
1894commas and surrounded by braces. The @var{n}th number is 1 if
1895register @var{n} is fixed, 0 otherwise.
1896
1897The table initialized from this macro, and the table initialized by
1898the following one, may be overridden at run time either automatically,
1899by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1900the user with the command options @option{-ffixed-@var{reg}},
1901@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1902@end defmac
1903
1904@defmac CALL_USED_REGISTERS
1905@cindex call-used register
1906@cindex call-clobbered register
1907@cindex call-saved register
1908Like @code{FIXED_REGISTERS} but has 1 for each register that is
1909clobbered (in general) by function calls as well as for fixed
1910registers. This macro therefore identifies the registers that are not
1911available for general allocation of values that must live across
1912function calls.
1913
1914If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1915automatically saves it on function entry and restores it on function
1916exit, if the register is used within the function.
1917@end defmac
1918
1919@defmac CALL_REALLY_USED_REGISTERS
1920@cindex call-used register
1921@cindex call-clobbered register
1922@cindex call-saved register
1923Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1924that the entire set of @code{FIXED_REGISTERS} be included.
1925(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1926This macro is optional. If not specified, it defaults to the value
1927of @code{CALL_USED_REGISTERS}.
1928@end defmac
1929
1930@defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1931@cindex call-used register
1932@cindex call-clobbered register
1933@cindex call-saved register
1934A C expression that is nonzero if it is not permissible to store a
1935value of mode @var{mode} in hard register number @var{regno} across a
1936call without some part of it being clobbered. For most machines this
1937macro need not be defined. It is only required for machines that do not
1938preserve the entire contents of a register across a call.
1939@end defmac
1940
1941@findex fixed_regs
1942@findex call_used_regs
1943@findex global_regs
1944@findex reg_names
1945@findex reg_class_contents
1946@defmac CONDITIONAL_REGISTER_USAGE
1947Zero or more C statements that may conditionally modify five variables
1948@code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1949@code{reg_names}, and @code{reg_class_contents}, to take into account
1950any dependence of these register sets on target flags. The first three
1951of these are of type @code{char []} (interpreted as Boolean vectors).
1952@code{global_regs} is a @code{const char *[]}, and
1953@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1954called, @code{fixed_regs}, @code{call_used_regs},
1955@code{reg_class_contents}, and @code{reg_names} have been initialized
1956from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1957@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1958@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1959@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1960command options have been applied.
1961
1962You need not define this macro if it has no work to do.
1963
1964@cindex disabling certain registers
1965@cindex controlling register usage
1966If the usage of an entire class of registers depends on the target
1967flags, you may indicate this to GCC by using this macro to modify
1968@code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1969registers in the classes which should not be used by GCC@. Also define
1970the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1971to return @code{NO_REGS} if it
1972is called with a letter for a class that shouldn't be used.
1973
1974(However, if this class is not included in @code{GENERAL_REGS} and all
1975of the insn patterns whose constraints permit this class are
1976controlled by target switches, then GCC will automatically avoid using
1977these registers when the target switches are opposed to them.)
1978@end defmac
1979
1980@defmac INCOMING_REGNO (@var{out})
1981Define this macro if the target machine has register windows. This C
1982expression returns the register number as seen by the called function
1983corresponding to the register number @var{out} as seen by the calling
1984function. Return @var{out} if register number @var{out} is not an
1985outbound register.
1986@end defmac
1987
1988@defmac OUTGOING_REGNO (@var{in})
1989Define this macro if the target machine has register windows. This C
1990expression returns the register number as seen by the calling function
1991corresponding to the register number @var{in} as seen by the called
1992function. Return @var{in} if register number @var{in} is not an inbound
1993register.
1994@end defmac
1995
1996@defmac LOCAL_REGNO (@var{regno})
1997Define this macro if the target machine has register windows. This C
1998expression returns true if the register is call-saved but is in the
1999register window. Unlike most call-saved registers, such registers
2000need not be explicitly restored on function exit or during non-local
2001gotos.
2002@end defmac
2003
2004@defmac PC_REGNUM
2005If the program counter has a register number, define this as that
2006register number. Otherwise, do not define it.
2007@end defmac
2008
2009@node Allocation Order
2010@subsection Order of Allocation of Registers
2011@cindex order of register allocation
2012@cindex register allocation order
2013
2014@c prevent bad page break with this line
2015Registers are allocated in order.
2016
2017@defmac REG_ALLOC_ORDER
2018If defined, an initializer for a vector of integers, containing the
2019numbers of hard registers in the order in which GCC should prefer
2020to use them (from most preferred to least).
2021
2022If this macro is not defined, registers are used lowest numbered first
2023(all else being equal).
2024
2025One use of this macro is on machines where the highest numbered
2026registers must always be saved and the save-multiple-registers
2027instruction supports only sequences of consecutive registers. On such
2028machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2029the highest numbered allocable register first.
2030@end defmac
2031
2032@defmac ADJUST_REG_ALLOC_ORDER
2033A C statement (sans semicolon) to choose the order in which to allocate
2034hard registers for pseudo-registers local to a basic block.
2035
2036Store the desired register order in the array @code{reg_alloc_order}.
2037Element 0 should be the register to allocate first; element 1, the next
2038register; and so on.
2039
2040The macro body should not assume anything about the contents of
2041@code{reg_alloc_order} before execution of the macro.
2042
2043On most machines, it is not necessary to define this macro.
2044@end defmac
2045
2046@defmac HONOR_REG_ALLOC_ORDER
2047Normally, IRA tries to estimate the costs for saving a register in the
2048prologue and restoring it in the epilogue. This discourages it from
2049using call-saved registers. If a machine wants to ensure that IRA
2050allocates registers in the order given by REG_ALLOC_ORDER even if some
2051call-saved registers appear earlier than call-used ones, this macro
2052should be defined.
2053@end defmac
2054
2055@defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2056In some case register allocation order is not enough for the
2057Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2058If this macro is defined, it should return a floating point value
2059based on @var{regno}. The cost of using @var{regno} for a pseudo will
2060be increased by approximately the pseudo's usage frequency times the
2061value returned by this macro. Not defining this macro is equivalent
2062to having it always return @code{0.0}.
2063
2064On most machines, it is not necessary to define this macro.
2065@end defmac
2066
2067@node Values in Registers
2068@subsection How Values Fit in Registers
2069
2070This section discusses the macros that describe which kinds of values
2071(specifically, which machine modes) each register can hold, and how many
2072consecutive registers are needed for a given mode.
2073
2074@defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2075A C expression for the number of consecutive hard registers, starting
2076at register number @var{regno}, required to hold a value of mode
2077@var{mode}. This macro must never return zero, even if a register
2078cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2079and/or CANNOT_CHANGE_MODE_CLASS instead.
2080
2081On a machine where all registers are exactly one word, a suitable
2082definition of this macro is
2083
2084@smallexample
2085#define HARD_REGNO_NREGS(REGNO, MODE) \
2086 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2087 / UNITS_PER_WORD)
2088@end smallexample
2089@end defmac
2090
2091@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2092A C expression that is nonzero if a value of mode @var{mode}, stored
2093in memory, ends with padding that causes it to take up more space than
2094in registers starting at register number @var{regno} (as determined by
2095multiplying GCC's notion of the size of the register when containing
2096this mode by the number of registers returned by
2097@code{HARD_REGNO_NREGS}). By default this is zero.
2098
2099For example, if a floating-point value is stored in three 32-bit
2100registers but takes up 128 bits in memory, then this would be
2101nonzero.
2102
2103This macros only needs to be defined if there are cases where
2104@code{subreg_get_info}
2105would otherwise wrongly determine that a @code{subreg} can be
2106represented by an offset to the register number, when in fact such a
2107@code{subreg} would contain some of the padding not stored in
2108registers and so not be representable.
2109@end defmac
2110
2111@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2112For values of @var{regno} and @var{mode} for which
2113@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2114returning the greater number of registers required to hold the value
2115including any padding. In the example above, the value would be four.
2116@end defmac
2117
2118@defmac REGMODE_NATURAL_SIZE (@var{mode})
2119Define this macro if the natural size of registers that hold values
2120of mode @var{mode} is not the word size. It is a C expression that
2121should give the natural size in bytes for the specified mode. It is
2122used by the register allocator to try to optimize its results. This
2123happens for example on SPARC 64-bit where the natural size of
2124floating-point registers is still 32-bit.
2125@end defmac
2126
2127@defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2128A C expression that is nonzero if it is permissible to store a value
2129of mode @var{mode} in hard register number @var{regno} (or in several
2130registers starting with that one). For a machine where all registers
2131are equivalent, a suitable definition is
2132
2133@smallexample
2134#define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2135@end smallexample
2136
2137You need not include code to check for the numbers of fixed registers,
2138because the allocation mechanism considers them to be always occupied.
2139
2140@cindex register pairs
2141On some machines, double-precision values must be kept in even/odd
2142register pairs. You can implement that by defining this macro to reject
2143odd register numbers for such modes.
2144
2145The minimum requirement for a mode to be OK in a register is that the
2146@samp{mov@var{mode}} instruction pattern support moves between the
2147register and other hard register in the same class and that moving a
2148value into the register and back out not alter it.
2149
2150Since the same instruction used to move @code{word_mode} will work for
2151all narrower integer modes, it is not necessary on any machine for
2152@code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2153you define patterns @samp{movhi}, etc., to take advantage of this. This
2154is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2155and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2156to be tieable.
2157
2158Many machines have special registers for floating point arithmetic.
2159Often people assume that floating point machine modes are allowed only
2160in floating point registers. This is not true. Any registers that
2161can hold integers can safely @emph{hold} a floating point machine
2162mode, whether or not floating arithmetic can be done on it in those
2163registers. Integer move instructions can be used to move the values.
2164
2165On some machines, though, the converse is true: fixed-point machine
2166modes may not go in floating registers. This is true if the floating
2167registers normalize any value stored in them, because storing a
2168non-floating value there would garble it. In this case,
2169@code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2170floating registers. But if the floating registers do not automatically
2171normalize, if you can store any bit pattern in one and retrieve it
2172unchanged without a trap, then any machine mode may go in a floating
2173register, so you can define this macro to say so.
2174
2175The primary significance of special floating registers is rather that
2176they are the registers acceptable in floating point arithmetic
2177instructions. However, this is of no concern to
2178@code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2179constraints for those instructions.
2180
2181On some machines, the floating registers are especially slow to access,
2182so that it is better to store a value in a stack frame than in such a
2183register if floating point arithmetic is not being done. As long as the
2184floating registers are not in class @code{GENERAL_REGS}, they will not
2185be used unless some pattern's constraint asks for one.
2186@end defmac
2187
2188@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2189A C expression that is nonzero if it is OK to rename a hard register
2190@var{from} to another hard register @var{to}.
2191
2192One common use of this macro is to prevent renaming of a register to
2193another register that is not saved by a prologue in an interrupt
2194handler.
2195
2196The default is always nonzero.
2197@end defmac
2198
2199@defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2200A C expression that is nonzero if a value of mode
2201@var{mode1} is accessible in mode @var{mode2} without copying.
2202
2203If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2204@code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2205any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2206should be nonzero. If they differ for any @var{r}, you should define
2207this macro to return zero unless some other mechanism ensures the
2208accessibility of the value in a narrower mode.
2209
2210You should define this macro to return nonzero in as many cases as
2211possible since doing so will allow GCC to perform better register
2212allocation.
2213@end defmac
2214
2215@hook TARGET_HARD_REGNO_SCRATCH_OK
2216This target hook should return @code{true} if it is OK to use a hard register
2217@var{regno} as scratch reg in peephole2.
2218
2219One common use of this macro is to prevent using of a register that
2220is not saved by a prologue in an interrupt handler.
2221
2222The default version of this hook always returns @code{true}.
2223@end deftypefn
2224
2225@defmac AVOID_CCMODE_COPIES
2226Define this macro if the compiler should avoid copies to/from @code{CCmode}
2227registers. You should only define this macro if support for copying to/from
2228@code{CCmode} is incomplete.
2229@end defmac
2230
2231@node Leaf Functions
2232@subsection Handling Leaf Functions
2233
2234@cindex leaf functions
2235@cindex functions, leaf
2236On some machines, a leaf function (i.e., one which makes no calls) can run
2237more efficiently if it does not make its own register window. Often this
2238means it is required to receive its arguments in the registers where they
2239are passed by the caller, instead of the registers where they would
2240normally arrive.
2241
2242The special treatment for leaf functions generally applies only when
2243other conditions are met; for example, often they may use only those
2244registers for its own variables and temporaries. We use the term ``leaf
2245function'' to mean a function that is suitable for this special
2246handling, so that functions with no calls are not necessarily ``leaf
2247functions''.
2248
2249GCC assigns register numbers before it knows whether the function is
2250suitable for leaf function treatment. So it needs to renumber the
2251registers in order to output a leaf function. The following macros
2252accomplish this.
2253
2254@defmac LEAF_REGISTERS
2255Name of a char vector, indexed by hard register number, which
2256contains 1 for a register that is allowable in a candidate for leaf
2257function treatment.
2258
2259If leaf function treatment involves renumbering the registers, then the
2260registers marked here should be the ones before renumbering---those that
2261GCC would ordinarily allocate. The registers which will actually be
2262used in the assembler code, after renumbering, should not be marked with 1
2263in this vector.
2264
2265Define this macro only if the target machine offers a way to optimize
2266the treatment of leaf functions.
2267@end defmac
2268
2269@defmac LEAF_REG_REMAP (@var{regno})
2270A C expression whose value is the register number to which @var{regno}
2271should be renumbered, when a function is treated as a leaf function.
2272
2273If @var{regno} is a register number which should not appear in a leaf
2274function before renumbering, then the expression should yield @minus{}1, which
2275will cause the compiler to abort.
2276
2277Define this macro only if the target machine offers a way to optimize the
2278treatment of leaf functions, and registers need to be renumbered to do
2279this.
2280@end defmac
2281
2282@findex current_function_is_leaf
2283@findex current_function_uses_only_leaf_regs
2284@code{TARGET_ASM_FUNCTION_PROLOGUE} and
2285@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2286specially. They can test the C variable @code{current_function_is_leaf}
2287which is nonzero for leaf functions. @code{current_function_is_leaf} is
2288set prior to local register allocation and is valid for the remaining
2289compiler passes. They can also test the C variable
2290@code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2291functions which only use leaf registers.
2292@code{current_function_uses_only_leaf_regs} is valid after all passes
2293that modify the instructions have been run and is only useful if
2294@code{LEAF_REGISTERS} is defined.
2295@c changed this to fix overfull. ALSO: why the "it" at the beginning
2296@c of the next paragraph?! --mew 2feb93
2297
2298@node Stack Registers
2299@subsection Registers That Form a Stack
2300
2301There are special features to handle computers where some of the
2302``registers'' form a stack. Stack registers are normally written by
2303pushing onto the stack, and are numbered relative to the top of the
2304stack.
2305
2306Currently, GCC can only handle one group of stack-like registers, and
2307they must be consecutively numbered. Furthermore, the existing
2308support for stack-like registers is specific to the 80387 floating
2309point coprocessor. If you have a new architecture that uses
2310stack-like registers, you will need to do substantial work on
2311@file{reg-stack.c} and write your machine description to cooperate
2312with it, as well as defining these macros.
2313
2314@defmac STACK_REGS
2315Define this if the machine has any stack-like registers.
2316@end defmac
2317
2318@defmac STACK_REG_COVER_CLASS
2319This is a cover class containing the stack registers. Define this if
2320the machine has any stack-like registers.
2321@end defmac
2322
2323@defmac FIRST_STACK_REG
2324The number of the first stack-like register. This one is the top
2325of the stack.
2326@end defmac
2327
2328@defmac LAST_STACK_REG
2329The number of the last stack-like register. This one is the bottom of
2330the stack.
2331@end defmac
2332
2333@node Register Classes
2334@section Register Classes
2335@cindex register class definitions
2336@cindex class definitions, register
2337
2338On many machines, the numbered registers are not all equivalent.
2339For example, certain registers may not be allowed for indexed addressing;
2340certain registers may not be allowed in some instructions. These machine
2341restrictions are described to the compiler using @dfn{register classes}.
2342
2343You define a number of register classes, giving each one a name and saying
2344which of the registers belong to it. Then you can specify register classes
2345that are allowed as operands to particular instruction patterns.
2346
2347@findex ALL_REGS
2348@findex NO_REGS
2349In general, each register will belong to several classes. In fact, one
2350class must be named @code{ALL_REGS} and contain all the registers. Another
2351class must be named @code{NO_REGS} and contain no registers. Often the
2352union of two classes will be another class; however, this is not required.
2353
2354@findex GENERAL_REGS
2355One of the classes must be named @code{GENERAL_REGS}. There is nothing
2356terribly special about the name, but the operand constraint letters
2357@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2358the same as @code{ALL_REGS}, just define it as a macro which expands
2359to @code{ALL_REGS}.
2360
2361Order the classes so that if class @var{x} is contained in class @var{y}
2362then @var{x} has a lower class number than @var{y}.
2363
2364The way classes other than @code{GENERAL_REGS} are specified in operand
2365constraints is through machine-dependent operand constraint letters.
2366You can define such letters to correspond to various classes, then use
2367them in operand constraints.
2368
2369You should define a class for the union of two classes whenever some
2370instruction allows both classes. For example, if an instruction allows
2371either a floating point (coprocessor) register or a general register for a
2372certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2373which includes both of them. Otherwise you will get suboptimal code.
2374
2375You must also specify certain redundant information about the register
2376classes: for each class, which classes contain it and which ones are
2377contained in it; for each pair of classes, the largest class contained
2378in their union.
2379
2380When a value occupying several consecutive registers is expected in a
2381certain class, all the registers used must belong to that class.
2382Therefore, register classes cannot be used to enforce a requirement for
2383a register pair to start with an even-numbered register. The way to
2384specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2385
2386Register classes used for input-operands of bitwise-and or shift
2387instructions have a special requirement: each such class must have, for
2388each fixed-point machine mode, a subclass whose registers can transfer that
2389mode to or from memory. For example, on some machines, the operations for
2390single-byte values (@code{QImode}) are limited to certain registers. When
2391this is so, each register class that is used in a bitwise-and or shift
2392instruction must have a subclass consisting of registers from which
2393single-byte values can be loaded or stored. This is so that
2394@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2395
2396@deftp {Data type} {enum reg_class}
2397An enumerated type that must be defined with all the register class names
2398as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2399must be the last register class, followed by one more enumerated value,
2400@code{LIM_REG_CLASSES}, which is not a register class but rather
2401tells how many classes there are.
2402
2403Each register class has a number, which is the value of casting
2404the class name to type @code{int}. The number serves as an index
2405in many of the tables described below.
2406@end deftp
2407
2408@defmac N_REG_CLASSES
2409The number of distinct register classes, defined as follows:
2410
2411@smallexample
2412#define N_REG_CLASSES (int) LIM_REG_CLASSES
2413@end smallexample
2414@end defmac
2415
2416@defmac REG_CLASS_NAMES
2417An initializer containing the names of the register classes as C string
2418constants. These names are used in writing some of the debugging dumps.
2419@end defmac
2420
2421@defmac REG_CLASS_CONTENTS
2422An initializer containing the contents of the register classes, as integers
2423which are bit masks. The @var{n}th integer specifies the contents of class
2424@var{n}. The way the integer @var{mask} is interpreted is that
2425register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2426
2427When the machine has more than 32 registers, an integer does not suffice.
2428Then the integers are replaced by sub-initializers, braced groupings containing
2429several integers. Each sub-initializer must be suitable as an initializer
2430for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2431In this situation, the first integer in each sub-initializer corresponds to
2432registers 0 through 31, the second integer to registers 32 through 63, and
2433so on.
2434@end defmac
2435
2436@defmac REGNO_REG_CLASS (@var{regno})
2437A C expression whose value is a register class containing hard register
2438@var{regno}. In general there is more than one such class; choose a class
2439which is @dfn{minimal}, meaning that no smaller class also contains the
2440register.
2441@end defmac
2442
2443@defmac BASE_REG_CLASS
2444A macro whose definition is the name of the class to which a valid
2445base register must belong. A base register is one used in an address
2446which is the register value plus a displacement.
2447@end defmac
2448
2449@defmac MODE_BASE_REG_CLASS (@var{mode})
2450This is a variation of the @code{BASE_REG_CLASS} macro which allows
2451the selection of a base register in a mode dependent manner. If
2452@var{mode} is VOIDmode then it should return the same value as
2453@code{BASE_REG_CLASS}.
2454@end defmac
2455
2456@defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2457A C expression whose value is the register class to which a valid
2458base register must belong in order to be used in a base plus index
2459register address. You should define this macro if base plus index
2460addresses have different requirements than other base register uses.
2461@end defmac
2462
2463@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2464A C expression whose value is the register class to which a valid
2465base register must belong. @var{outer_code} and @var{index_code} define the
2466context in which the base register occurs. @var{outer_code} is the code of
2467the immediately enclosing expression (@code{MEM} for the top level of an
2468address, @code{ADDRESS} for something that occurs in an
2469@code{address_operand}). @var{index_code} is the code of the corresponding
2470index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2471@end defmac
2472
2473@defmac INDEX_REG_CLASS
2474A macro whose definition is the name of the class to which a valid
2475index register must belong. An index register is one used in an
2476address where its value is either multiplied by a scale factor or
2477added to another register (as well as added to a displacement).
2478@end defmac
2479
2480@defmac REGNO_OK_FOR_BASE_P (@var{num})
2481A C expression which is nonzero if register number @var{num} is
2482suitable for use as a base register in operand addresses.
2483Like @code{TARGET_LEGITIMATE_ADDRESS_P}, this macro should also
2484define a strict and a non-strict variant. Both variants behave
2485the same for hard register; for pseudos, the strict variant will
2486pass only those that have been allocated to a valid hard registers,
2487while the non-strict variant will pass all pseudos.
2488
2489@findex REG_OK_STRICT
2490Compiler source files that want to use the strict variant of this and
2491other macros define the macro @code{REG_OK_STRICT}. You should use an
2492@code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
2493that case and the non-strict variant otherwise.
2494@end defmac
2495
2496@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2497A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2498that expression may examine the mode of the memory reference in
2499@var{mode}. You should define this macro if the mode of the memory
2500reference affects whether a register may be used as a base register. If
2501you define this macro, the compiler will use it instead of
2502@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2503addresses that appear outside a @code{MEM}, i.e., as an
2504@code{address_operand}.
2505
2506This macro also has strict and non-strict variants.
2507@end defmac
2508
2509@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2510A C expression which is nonzero if register number @var{num} is suitable for
2511use as a base register in base plus index operand addresses, accessing
2512memory in mode @var{mode}. It may be either a suitable hard register or a
2513pseudo register that has been allocated such a hard register. You should
2514define this macro if base plus index addresses have different requirements
2515than other base register uses.
2516
2517Use of this macro is deprecated; please use the more general
2518@code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2519
2520This macro also has strict and non-strict variants.
2521@end defmac
2522
2523@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2524A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except
2525that that expression may examine the context in which the register
2526appears in the memory reference. @var{outer_code} is the code of the
2527immediately enclosing expression (@code{MEM} if at the top level of the
2528address, @code{ADDRESS} for something that occurs in an
2529@code{address_operand}). @var{index_code} is the code of the
2530corresponding index expression if @var{outer_code} is @code{PLUS};
2531@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2532that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2533
2534This macro also has strict and non-strict variants.
2535@end defmac
2536
2537@defmac REGNO_OK_FOR_INDEX_P (@var{num})
2538A C expression which is nonzero if register number @var{num} is
2539suitable for use as an index register in operand addresses. It may be
2540either a suitable hard register or a pseudo register that has been
2541allocated such a hard register.
2542
2543The difference between an index register and a base register is that
2544the index register may be scaled. If an address involves the sum of
2545two registers, neither one of them scaled, then either one may be
2546labeled the ``base'' and the other the ``index''; but whichever
2547labeling is used must fit the machine's constraints of which registers
2548may serve in each capacity. The compiler will try both labelings,
2549looking for one that is valid, and will reload one or both registers
2550only if neither labeling works.
2551
2552This macro also has strict and non-strict variants.
2553@end defmac
2554
fba42e24
AS
2555@hook TARGET_PREFERRED_RELOAD_CLASS
2556A target hook that places additional restrictions on the register class
2557to use when it is necessary to copy value @var{x} into a register in class
2558@var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2559another, smaller class.
2560
2561The default version of this hook always returns value of @code{rclass} argument.
2562
2563Sometimes returning a more restrictive class makes better code. For
2564example, on the 68000, when @var{x} is an integer constant that is in range
2565for a @samp{moveq} instruction, the value of this macro is always
2566@code{DATA_REGS} as long as @var{rclass} includes the data registers.
2567Requiring a data register guarantees that a @samp{moveq} will be used.
2568
2569One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2570@var{rclass} is if @var{x} is a legitimate constant which cannot be
2571loaded into some register class. By returning @code{NO_REGS} you can
2572force @var{x} into a memory location. For example, rs6000 can load
2573immediate values into general-purpose registers, but does not have an
2574instruction for loading an immediate value into a floating-point
2575register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2576@var{x} is a floating-point constant. If the constant can't be loaded
2577into any kind of register, code generation will be better if
2578@code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2579of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2580
2581If an insn has pseudos in it after register allocation, reload will go
2582through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2583to find the best one. Returning @code{NO_REGS}, in this case, makes
2584reload add a @code{!} in front of the constraint: the x86 back-end uses
2585this feature to discourage usage of 387 registers when math is done in
2586the SSE registers (and vice versa).
2587@end deftypefn
2588
38f8b050
JR
2589@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2590A C expression that places additional restrictions on the register class
2591to use when it is necessary to copy value @var{x} into a register in class
2592@var{class}. The value is a register class; perhaps @var{class}, or perhaps
2593another, smaller class. On many machines, the following definition is
2594safe:
2595
2596@smallexample
2597#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2598@end smallexample
2599
2600Sometimes returning a more restrictive class makes better code. For
2601example, on the 68000, when @var{x} is an integer constant that is in range
2602for a @samp{moveq} instruction, the value of this macro is always
2603@code{DATA_REGS} as long as @var{class} includes the data registers.
2604Requiring a data register guarantees that a @samp{moveq} will be used.
2605
2606One case where @code{PREFERRED_RELOAD_CLASS} must not return
2607@var{class} is if @var{x} is a legitimate constant which cannot be
2608loaded into some register class. By returning @code{NO_REGS} you can
2609force @var{x} into a memory location. For example, rs6000 can load
2610immediate values into general-purpose registers, but does not have an
2611instruction for loading an immediate value into a floating-point
2612register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2613@var{x} is a floating-point constant. If the constant can't be loaded
2614into any kind of register, code generation will be better if
2615@code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2616of using @code{PREFERRED_RELOAD_CLASS}.
2617
2618If an insn has pseudos in it after register allocation, reload will go
2619through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2620to find the best one. Returning @code{NO_REGS}, in this case, makes
2621reload add a @code{!} in front of the constraint: the x86 back-end uses
2622this feature to discourage usage of 387 registers when math is done in
2623the SSE registers (and vice versa).
2624@end defmac
2625
2626@defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2627Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2628input reloads. If you don't define this macro, the default is to use
2629@var{class}, unchanged.
2630
2631You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2632reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2633@end defmac
2634
abd26bfb
AS
2635@hook TARGET_PREFERRED_OUTPUT_RELOAD_CLASS
2636Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2637input reloads.
2638
2639The default version of this hook always returns value of @code{rclass}
2640argument.
2641
2642You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2643reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2644@end deftypefn
2645
38f8b050
JR
2646@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2647A C expression that places additional restrictions on the register class
2648to use when it is necessary to be able to hold a value of mode
2649@var{mode} in a reload register for which class @var{class} would
2650ordinarily be used.
2651
2652Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2653there are certain modes that simply can't go in certain reload classes.
2654
2655The value is a register class; perhaps @var{class}, or perhaps another,
2656smaller class.
2657
2658Don't define this macro unless the target machine has limitations which
2659require the macro to do something nontrivial.
2660@end defmac
2661
2662@hook TARGET_SECONDARY_RELOAD
2663Many machines have some registers that cannot be copied directly to or
2664from memory or even from other types of registers. An example is the
2665@samp{MQ} register, which on most machines, can only be copied to or
2666from general registers, but not memory. Below, we shall be using the
2667term 'intermediate register' when a move operation cannot be performed
2668directly, but has to be done by copying the source into the intermediate
2669register first, and then copying the intermediate register to the
2670destination. An intermediate register always has the same mode as
2671source and destination. Since it holds the actual value being copied,
2672reload might apply optimizations to re-use an intermediate register
2673and eliding the copy from the source when it can determine that the
2674intermediate register still holds the required value.
2675
2676Another kind of secondary reload is required on some machines which
2677allow copying all registers to and from memory, but require a scratch
2678register for stores to some memory locations (e.g., those with symbolic
2679address on the RT, and those with certain symbolic address on the SPARC
2680when compiling PIC)@. Scratch registers need not have the same mode
2681as the value being copied, and usually hold a different value than
2682that being copied. Special patterns in the md file are needed to
2683describe how the copy is performed with the help of the scratch register;
2684these patterns also describe the number, register class(es) and mode(s)
2685of the scratch register(s).
2686
2687In some cases, both an intermediate and a scratch register are required.
2688
2689For input reloads, this target hook is called with nonzero @var{in_p},
2690and @var{x} is an rtx that needs to be copied to a register of class
2691@var{reload_class} in @var{reload_mode}. For output reloads, this target
2692hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2693needs to be copied to rtx @var{x} in @var{reload_mode}.
2694
2695If copying a register of @var{reload_class} from/to @var{x} requires
2696an intermediate register, the hook @code{secondary_reload} should
2697return the register class required for this intermediate register.
2698If no intermediate register is required, it should return NO_REGS.
2699If more than one intermediate register is required, describe the one
2700that is closest in the copy chain to the reload register.
2701
2702If scratch registers are needed, you also have to describe how to
2703perform the copy from/to the reload register to/from this
2704closest intermediate register. Or if no intermediate register is
2705required, but still a scratch register is needed, describe the
2706copy from/to the reload register to/from the reload operand @var{x}.
2707
2708You do this by setting @code{sri->icode} to the instruction code of a pattern
2709in the md file which performs the move. Operands 0 and 1 are the output
2710and input of this copy, respectively. Operands from operand 2 onward are
2711for scratch operands. These scratch operands must have a mode, and a
2712single-register-class
2713@c [later: or memory]
2714output constraint.
2715
2716When an intermediate register is used, the @code{secondary_reload}
2717hook will be called again to determine how to copy the intermediate
2718register to/from the reload operand @var{x}, so your hook must also
2719have code to handle the register class of the intermediate operand.
2720
2721@c [For later: maybe we'll allow multi-alternative reload patterns -
2722@c the port maintainer could name a mov<mode> pattern that has clobbers -
2723@c and match the constraints of input and output to determine the required
2724@c alternative. A restriction would be that constraints used to match
2725@c against reloads registers would have to be written as register class
2726@c constraints, or we need a new target macro / hook that tells us if an
2727@c arbitrary constraint can match an unknown register of a given class.
2728@c Such a macro / hook would also be useful in other places.]
2729
2730
2731@var{x} might be a pseudo-register or a @code{subreg} of a
2732pseudo-register, which could either be in a hard register or in memory.
2733Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2734in memory and the hard register number if it is in a register.
2735
2736Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2737currently not supported. For the time being, you will have to continue
2738to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2739
2740@code{copy_cost} also uses this target hook to find out how values are
2741copied. If you want it to include some extra cost for the need to allocate
2742(a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2743Or if two dependent moves are supposed to have a lower cost than the sum
2744of the individual moves due to expected fortuitous scheduling and/or special
2745forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2746@end deftypefn
2747
2748@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2749@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2750@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2751These macros are obsolete, new ports should use the target hook
2752@code{TARGET_SECONDARY_RELOAD} instead.
2753
2754These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2755target hook. Older ports still define these macros to indicate to the
2756reload phase that it may
2757need to allocate at least one register for a reload in addition to the
2758register to contain the data. Specifically, if copying @var{x} to a
2759register @var{class} in @var{mode} requires an intermediate register,
2760you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2761largest register class all of whose registers can be used as
2762intermediate registers or scratch registers.
2763
2764If copying a register @var{class} in @var{mode} to @var{x} requires an
2765intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2766was supposed to be defined be defined to return the largest register
2767class required. If the
2768requirements for input and output reloads were the same, the macro
2769@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2770macros identically.
2771
2772The values returned by these macros are often @code{GENERAL_REGS}.
2773Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2774can be directly copied to or from a register of @var{class} in
2775@var{mode} without requiring a scratch register. Do not define this
2776macro if it would always return @code{NO_REGS}.
2777
2778If a scratch register is required (either with or without an
2779intermediate register), you were supposed to define patterns for
2780@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2781(@pxref{Standard Names}. These patterns, which were normally
2782implemented with a @code{define_expand}, should be similar to the
2783@samp{mov@var{m}} patterns, except that operand 2 is the scratch
2784register.
2785
2786These patterns need constraints for the reload register and scratch
2787register that
2788contain a single register class. If the original reload register (whose
2789class is @var{class}) can meet the constraint given in the pattern, the
2790value returned by these macros is used for the class of the scratch
2791register. Otherwise, two additional reload registers are required.
2792Their classes are obtained from the constraints in the insn pattern.
2793
2794@var{x} might be a pseudo-register or a @code{subreg} of a
2795pseudo-register, which could either be in a hard register or in memory.
2796Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2797in memory and the hard register number if it is in a register.
2798
2799These macros should not be used in the case where a particular class of
2800registers can only be copied to memory and not to another class of
2801registers. In that case, secondary reload registers are not needed and
2802would not be helpful. Instead, a stack location must be used to perform
2803the copy and the @code{mov@var{m}} pattern should use memory as an
2804intermediate storage. This case often occurs between floating-point and
2805general registers.
2806@end defmac
2807
2808@defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2809Certain machines have the property that some registers cannot be copied
2810to some other registers without using memory. Define this macro on
2811those machines to be a C expression that is nonzero if objects of mode
2812@var{m} in registers of @var{class1} can only be copied to registers of
2813class @var{class2} by storing a register of @var{class1} into memory
2814and loading that memory location into a register of @var{class2}.
2815
2816Do not define this macro if its value would always be zero.
2817@end defmac
2818
2819@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2820Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2821allocates a stack slot for a memory location needed for register copies.
2822If this macro is defined, the compiler instead uses the memory location
2823defined by this macro.
2824
2825Do not define this macro if you do not define
2826@code{SECONDARY_MEMORY_NEEDED}.
2827@end defmac
2828
2829@defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2830When the compiler needs a secondary memory location to copy between two
2831registers of mode @var{mode}, it normally allocates sufficient memory to
2832hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2833load operations in a mode that many bits wide and whose class is the
2834same as that of @var{mode}.
2835
2836This is right thing to do on most machines because it ensures that all
2837bits of the register are copied and prevents accesses to the registers
2838in a narrower mode, which some machines prohibit for floating-point
2839registers.
2840
2841However, this default behavior is not correct on some machines, such as
2842the DEC Alpha, that store short integers in floating-point registers
2843differently than in integer registers. On those machines, the default
2844widening will not work correctly and you must define this macro to
2845suppress that widening in some cases. See the file @file{alpha.h} for
2846details.
2847
2848Do not define this macro if you do not define
2849@code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2850is @code{BITS_PER_WORD} bits wide is correct for your machine.
2851@end defmac
2852
07b8f0a8
AS
2853@hook TARGET_CLASS_LIKELY_SPILLED_P
2854A target hook which returns @code{true} if pseudos that have been assigned
2855to registers of class @var{rclass} would likely be spilled because
2856registers of @var{rclass} are needed for spill registers.
2857
2858The default version of this target hook returns @code{true} if @var{rclass}
2859has exactly one register and @code{false} otherwise. On most machines, this
2860default should be used. Only use this target hook to some other expression
2861if pseudos allocated by @file{local-alloc.c} end up in memory because their
2862hard registers were needed for spill registers. If this target hook returns
2863@code{false} for those classes, those pseudos will only be allocated by
2864@file{global.c}, which knows how to reallocate the pseudo to another
2865register. If there would not be another register available for reallocation,
2866you should not change the implementation of this target hook since
2867the only effect of such implementation would be to slow down register
2868allocation.
2869@end deftypefn
2870
38f8b050
JR
2871@defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2872A C expression for the maximum number of consecutive registers
2873of class @var{class} needed to hold a value of mode @var{mode}.
2874
2875This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2876the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2877should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2878@var{mode})} for all @var{regno} values in the class @var{class}.
2879
2880This macro helps control the handling of multiple-word values
2881in the reload pass.
2882@end defmac
2883
2884@defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2885If defined, a C expression that returns nonzero for a @var{class} for which
2886a change from mode @var{from} to mode @var{to} is invalid.
2887
2888For the example, loading 32-bit integer or floating-point objects into
2889floating-point registers on the Alpha extends them to 64 bits.
2890Therefore loading a 64-bit object and then storing it as a 32-bit object
2891does not store the low-order 32 bits, as would be the case for a normal
2892register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2893as below:
2894
2895@smallexample
2896#define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2897 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2898 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2899@end smallexample
2900@end defmac
2901
2902@hook TARGET_IRA_COVER_CLASSES
2903Return an array of cover classes for the Integrated Register Allocator
2904(@acronym{IRA}). Cover classes are a set of non-intersecting register
2905classes covering all hard registers used for register allocation
2906purposes. If a move between two registers in the same cover class is
2907possible, it should be cheaper than a load or store of the registers.
2908The array is terminated by a @code{LIM_REG_CLASSES} element.
2909
2910The order of cover classes in the array is important. If two classes
2911have the same cost of usage for a pseudo, the class occurred first in
2912the array is chosen for the pseudo.
2913
2914This hook is called once at compiler startup, after the command-line
2915options have been processed. It is then re-examined by every call to
2916@code{target_reinit}.
2917
2918The default implementation returns @code{IRA_COVER_CLASSES}, if defined,
2919otherwise there is no default implementation. You must define either this
2920macro or @code{IRA_COVER_CLASSES} in order to use the integrated register
2921allocator with Chaitin-Briggs coloring. If the macro is not defined,
2922the only available coloring algorithm is Chow's priority coloring.
2923@end deftypefn
2924
2925@defmac IRA_COVER_CLASSES
2926See the documentation for @code{TARGET_IRA_COVER_CLASSES}.
2927@end defmac
2928
2929@node Old Constraints
2930@section Obsolete Macros for Defining Constraints
2931@cindex defining constraints, obsolete method
2932@cindex constraints, defining, obsolete method
2933
2934Machine-specific constraints can be defined with these macros instead
2935of the machine description constructs described in @ref{Define
2936Constraints}. This mechanism is obsolete. New ports should not use
2937it; old ports should convert to the new mechanism.
2938
2939@defmac CONSTRAINT_LEN (@var{char}, @var{str})
2940For the constraint at the start of @var{str}, which starts with the letter
2941@var{c}, return the length. This allows you to have register class /
2942constant / extra constraints that are longer than a single letter;
2943you don't need to define this macro if you can do with single-letter
2944constraints only. The definition of this macro should use
2945DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2946to handle specially.
2947There are some sanity checks in genoutput.c that check the constraint lengths
2948for the md file, so you can also use this macro to help you while you are
2949transitioning from a byzantine single-letter-constraint scheme: when you
2950return a negative length for a constraint you want to re-use, genoutput
2951will complain about every instance where it is used in the md file.
2952@end defmac
2953
2954@defmac REG_CLASS_FROM_LETTER (@var{char})
2955A C expression which defines the machine-dependent operand constraint
2956letters for register classes. If @var{char} is such a letter, the
2957value should be the register class corresponding to it. Otherwise,
2958the value should be @code{NO_REGS}. The register letter @samp{r},
2959corresponding to class @code{GENERAL_REGS}, will not be passed
2960to this macro; you do not need to handle it.
2961@end defmac
2962
2963@defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2964Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2965passed in @var{str}, so that you can use suffixes to distinguish between
2966different variants.
2967@end defmac
2968
2969@defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2970A C expression that defines the machine-dependent operand constraint
2971letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2972particular ranges of integer values. If @var{c} is one of those
2973letters, the expression should check that @var{value}, an integer, is in
2974the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2975not one of those letters, the value should be 0 regardless of
2976@var{value}.
2977@end defmac
2978
2979@defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2980Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2981string passed in @var{str}, so that you can use suffixes to distinguish
2982between different variants.
2983@end defmac
2984
2985@defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2986A C expression that defines the machine-dependent operand constraint
2987letters that specify particular ranges of @code{const_double} values
2988(@samp{G} or @samp{H}).
2989
2990If @var{c} is one of those letters, the expression should check that
2991@var{value}, an RTX of code @code{const_double}, is in the appropriate
2992range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2993letters, the value should be 0 regardless of @var{value}.
2994
2995@code{const_double} is used for all floating-point constants and for
2996@code{DImode} fixed-point constants. A given letter can accept either
2997or both kinds of values. It can use @code{GET_MODE} to distinguish
2998between these kinds.
2999@end defmac
3000
3001@defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
3002Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
3003string passed in @var{str}, so that you can use suffixes to distinguish
3004between different variants.
3005@end defmac
3006
3007@defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
3008A C expression that defines the optional machine-dependent constraint
3009letters that can be used to segregate specific types of operands, usually
3010memory references, for the target machine. Any letter that is not
3011elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
3012@code{REG_CLASS_FROM_CONSTRAINT}
3013may be used. Normally this macro will not be defined.
3014
3015If it is required for a particular target machine, it should return 1
3016if @var{value} corresponds to the operand type represented by the
3017constraint letter @var{c}. If @var{c} is not defined as an extra
3018constraint, the value returned should be 0 regardless of @var{value}.
3019
3020For example, on the ROMP, load instructions cannot have their output
3021in r0 if the memory reference contains a symbolic address. Constraint
3022letter @samp{Q} is defined as representing a memory address that does
3023@emph{not} contain a symbolic address. An alternative is specified with
3024a @samp{Q} constraint on the input and @samp{r} on the output. The next
3025alternative specifies @samp{m} on the input and a register class that
3026does not include r0 on the output.
3027@end defmac
3028
3029@defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
3030Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
3031in @var{str}, so that you can use suffixes to distinguish between different
3032variants.
3033@end defmac
3034
3035@defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
3036A C expression that defines the optional machine-dependent constraint
3037letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
3038be treated like memory constraints by the reload pass.
3039
3040It should return 1 if the operand type represented by the constraint
3041at the start of @var{str}, the first letter of which is the letter @var{c},
3042comprises a subset of all memory references including
3043all those whose address is simply a base register. This allows the reload
3044pass to reload an operand, if it does not directly correspond to the operand
3045type of @var{c}, by copying its address into a base register.
3046
3047For example, on the S/390, some instructions do not accept arbitrary
3048memory references, but only those that do not make use of an index
3049register. The constraint letter @samp{Q} is defined via
3050@code{EXTRA_CONSTRAINT} as representing a memory address of this type.
3051If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
3052a @samp{Q} constraint can handle any memory operand, because the
3053reload pass knows it can be reloaded by copying the memory address
3054into a base register if required. This is analogous to the way
3055an @samp{o} constraint can handle any memory operand.
3056@end defmac
3057
3058@defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
3059A C expression that defines the optional machine-dependent constraint
3060letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
3061@code{EXTRA_CONSTRAINT_STR}, that should
3062be treated like address constraints by the reload pass.
3063
3064It should return 1 if the operand type represented by the constraint
3065at the start of @var{str}, which starts with the letter @var{c}, comprises
3066a subset of all memory addresses including
3067all those that consist of just a base register. This allows the reload
3068pass to reload an operand, if it does not directly correspond to the operand
3069type of @var{str}, by copying it into a base register.
3070
3071Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3072be used with the @code{address_operand} predicate. It is treated
3073analogously to the @samp{p} constraint.
3074@end defmac
3075
3076@node Stack and Calling
3077@section Stack Layout and Calling Conventions
3078@cindex calling conventions
3079
3080@c prevent bad page break with this line
3081This describes the stack layout and calling conventions.
3082
3083@menu
3084* Frame Layout::
3085* Exception Handling::
3086* Stack Checking::
3087* Frame Registers::
3088* Elimination::
3089* Stack Arguments::
3090* Register Arguments::
3091* Scalar Return::
3092* Aggregate Return::
3093* Caller Saves::
3094* Function Entry::
3095* Profiling::
3096* Tail Calls::
3097* Stack Smashing Protection::
3098@end menu
3099
3100@node Frame Layout
3101@subsection Basic Stack Layout
3102@cindex stack frame layout
3103@cindex frame layout
3104
3105@c prevent bad page break with this line
3106Here is the basic stack layout.
3107
3108@defmac STACK_GROWS_DOWNWARD
3109Define this macro if pushing a word onto the stack moves the stack
3110pointer to a smaller address.
3111
3112When we say, ``define this macro if @dots{}'', it means that the
3113compiler checks this macro only with @code{#ifdef} so the precise
3114definition used does not matter.
3115@end defmac
3116
3117@defmac STACK_PUSH_CODE
3118This macro defines the operation used when something is pushed
3119on the stack. In RTL, a push operation will be
3120@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3121
3122The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3123and @code{POST_INC}. Which of these is correct depends on
3124the stack direction and on whether the stack pointer points
3125to the last item on the stack or whether it points to the
3126space for the next item on the stack.
3127
3128The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3129defined, which is almost always right, and @code{PRE_INC} otherwise,
3130which is often wrong.
3131@end defmac
3132
3133@defmac FRAME_GROWS_DOWNWARD
3134Define this macro to nonzero value if the addresses of local variable slots
3135are at negative offsets from the frame pointer.
3136@end defmac
3137
3138@defmac ARGS_GROW_DOWNWARD
3139Define this macro if successive arguments to a function occupy decreasing
3140addresses on the stack.
3141@end defmac
3142
3143@defmac STARTING_FRAME_OFFSET
3144Offset from the frame pointer to the first local variable slot to be allocated.
3145
3146If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3147subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3148Otherwise, it is found by adding the length of the first slot to the
3149value @code{STARTING_FRAME_OFFSET}.
3150@c i'm not sure if the above is still correct.. had to change it to get
3151@c rid of an overfull. --mew 2feb93
3152@end defmac
3153
3154@defmac STACK_ALIGNMENT_NEEDED
3155Define to zero to disable final alignment of the stack during reload.
3156The nonzero default for this macro is suitable for most ports.
3157
3158On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3159is a register save block following the local block that doesn't require
3160alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3161stack alignment and do it in the backend.
3162@end defmac
3163
3164@defmac STACK_POINTER_OFFSET
3165Offset from the stack pointer register to the first location at which
3166outgoing arguments are placed. If not specified, the default value of
3167zero is used. This is the proper value for most machines.
3168
3169If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3170the first location at which outgoing arguments are placed.
3171@end defmac
3172
3173@defmac FIRST_PARM_OFFSET (@var{fundecl})
3174Offset from the argument pointer register to the first argument's
3175address. On some machines it may depend on the data type of the
3176function.
3177
3178If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3179the first argument's address.
3180@end defmac
3181
3182@defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3183Offset from the stack pointer register to an item dynamically allocated
3184on the stack, e.g., by @code{alloca}.
3185
3186The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3187length of the outgoing arguments. The default is correct for most
3188machines. See @file{function.c} for details.
3189@end defmac
3190
3191@defmac INITIAL_FRAME_ADDRESS_RTX
3192A C expression whose value is RTL representing the address of the initial
3193stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3194@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3195default value will be used. Define this macro in order to make frame pointer
3196elimination work in the presence of @code{__builtin_frame_address (count)} and
3197@code{__builtin_return_address (count)} for @code{count} not equal to zero.
3198@end defmac
3199
3200@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3201A C expression whose value is RTL representing the address in a stack
3202frame where the pointer to the caller's frame is stored. Assume that
3203@var{frameaddr} is an RTL expression for the address of the stack frame
3204itself.
3205
3206If you don't define this macro, the default is to return the value
3207of @var{frameaddr}---that is, the stack frame address is also the
3208address of the stack word that points to the previous frame.
3209@end defmac
3210
3211@defmac SETUP_FRAME_ADDRESSES
3212If defined, a C expression that produces the machine-specific code to
3213setup the stack so that arbitrary frames can be accessed. For example,
3214on the SPARC, we must flush all of the register windows to the stack
3215before we can access arbitrary stack frames. You will seldom need to
3216define this macro.
3217@end defmac
3218
3219@hook TARGET_BUILTIN_SETJMP_FRAME_VALUE
3220This target hook should return an rtx that is used to store
3221the address of the current frame into the built in @code{setjmp} buffer.
3222The default value, @code{virtual_stack_vars_rtx}, is correct for most
3223machines. One reason you may need to define this target hook is if
3224@code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3225@end deftypefn
3226
3227@defmac FRAME_ADDR_RTX (@var{frameaddr})
3228A C expression whose value is RTL representing the value of the frame
3229address for the current frame. @var{frameaddr} is the frame pointer
3230of the current frame. This is used for __builtin_frame_address.
3231You need only define this macro if the frame address is not the same
3232as the frame pointer. Most machines do not need to define it.
3233@end defmac
3234
3235@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3236A C expression whose value is RTL representing the value of the return
3237address for the frame @var{count} steps up from the current frame, after
3238the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3239frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3240@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3241
3242The value of the expression must always be the correct address when
3243@var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3244determine the return address of other frames.
3245@end defmac
3246
3247@defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3248Define this if the return address of a particular stack frame is accessed
3249from the frame pointer of the previous stack frame.
3250@end defmac
3251
3252@defmac INCOMING_RETURN_ADDR_RTX
3253A C expression whose value is RTL representing the location of the
3254incoming return address at the beginning of any function, before the
3255prologue. This RTL is either a @code{REG}, indicating that the return
3256value is saved in @samp{REG}, or a @code{MEM} representing a location in
3257the stack.
3258
3259You only need to define this macro if you want to support call frame
3260debugging information like that provided by DWARF 2.
3261
3262If this RTL is a @code{REG}, you should also define
3263@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3264@end defmac
3265
3266@defmac DWARF_ALT_FRAME_RETURN_COLUMN
3267A C expression whose value is an integer giving a DWARF 2 column
3268number that may be used as an alternative return column. The column
3269must not correspond to any gcc hard register (that is, it must not
3270be in the range of @code{DWARF_FRAME_REGNUM}).
3271
3272This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3273general register, but an alternative column needs to be used for signal
3274frames. Some targets have also used different frame return columns
3275over time.
3276@end defmac
3277
3278@defmac DWARF_ZERO_REG
3279A C expression whose value is an integer giving a DWARF 2 register
3280number that is considered to always have the value zero. This should
3281only be defined if the target has an architected zero register, and
3282someone decided it was a good idea to use that register number to
3283terminate the stack backtrace. New ports should avoid this.
3284@end defmac
3285
3286@hook TARGET_DWARF_HANDLE_FRAME_UNSPEC
3287This target hook allows the backend to emit frame-related insns that
3288contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3289info engine will invoke it on insns of the form
3290@smallexample
3291(set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3292@end smallexample
3293and
3294@smallexample
3295(set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3296@end smallexample
3297to let the backend emit the call frame instructions. @var{label} is
3298the CFI label attached to the insn, @var{pattern} is the pattern of
3299the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3300@end deftypefn
3301
3302@defmac INCOMING_FRAME_SP_OFFSET
3303A C expression whose value is an integer giving the offset, in bytes,
3304from the value of the stack pointer register to the top of the stack
3305frame at the beginning of any function, before the prologue. The top of
3306the frame is defined to be the value of the stack pointer in the
3307previous frame, just before the call instruction.
3308
3309You only need to define this macro if you want to support call frame
3310debugging information like that provided by DWARF 2.
3311@end defmac
3312
3313@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3314A C expression whose value is an integer giving the offset, in bytes,
3315from the argument pointer to the canonical frame address (cfa). The
3316final value should coincide with that calculated by
3317@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3318during virtual register instantiation.
3319
3320The default value for this macro is
3321@code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3322which is correct for most machines; in general, the arguments are found
3323immediately before the stack frame. Note that this is not the case on
3324some targets that save registers into the caller's frame, such as SPARC
3325and rs6000, and so such targets need to define this macro.
3326
3327You only need to define this macro if the default is incorrect, and you
3328want to support call frame debugging information like that provided by
3329DWARF 2.
3330@end defmac
3331
3332@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3333If defined, a C expression whose value is an integer giving the offset
3334in bytes from the frame pointer to the canonical frame address (cfa).
3335The final value should coincide with that calculated by
3336@code{INCOMING_FRAME_SP_OFFSET}.
3337
3338Normally the CFA is calculated as an offset from the argument pointer,
3339via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3340variable due to the ABI, this may not be possible. If this macro is
3341defined, it implies that the virtual register instantiation should be
3342based on the frame pointer instead of the argument pointer. Only one
3343of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3344should be defined.
3345@end defmac
3346
3347@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3348If defined, a C expression whose value is an integer giving the offset
3349in bytes from the canonical frame address (cfa) to the frame base used
3350in DWARF 2 debug information. The default is zero. A different value
3351may reduce the size of debug information on some ports.
3352@end defmac
3353
3354@node Exception Handling
3355@subsection Exception Handling Support
3356@cindex exception handling
3357
3358@defmac EH_RETURN_DATA_REGNO (@var{N})
3359A C expression whose value is the @var{N}th register number used for
3360data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3361@var{N} registers are usable.
3362
3363The exception handling library routines communicate with the exception
3364handlers via a set of agreed upon registers. Ideally these registers
3365should be call-clobbered; it is possible to use call-saved registers,
3366but may negatively impact code size. The target must support at least
33672 data registers, but should define 4 if there are enough free registers.
3368
3369You must define this macro if you want to support call frame exception
3370handling like that provided by DWARF 2.
3371@end defmac
3372
3373@defmac EH_RETURN_STACKADJ_RTX
3374A C expression whose value is RTL representing a location in which
3375to store a stack adjustment to be applied before function return.
3376This is used to unwind the stack to an exception handler's call frame.
3377It will be assigned zero on code paths that return normally.
3378
3379Typically this is a call-clobbered hard register that is otherwise
3380untouched by the epilogue, but could also be a stack slot.
3381
3382Do not define this macro if the stack pointer is saved and restored
3383by the regular prolog and epilog code in the call frame itself; in
3384this case, the exception handling library routines will update the
3385stack location to be restored in place. Otherwise, you must define
3386this macro if you want to support call frame exception handling like
3387that provided by DWARF 2.
3388@end defmac
3389
3390@defmac EH_RETURN_HANDLER_RTX
3391A C expression whose value is RTL representing a location in which
3392to store the address of an exception handler to which we should
3393return. It will not be assigned on code paths that return normally.
3394
3395Typically this is the location in the call frame at which the normal
3396return address is stored. For targets that return by popping an
3397address off the stack, this might be a memory address just below
3398the @emph{target} call frame rather than inside the current call
3399frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3400been assigned, so it may be used to calculate the location of the
3401target call frame.
3402
3403Some targets have more complex requirements than storing to an
3404address calculable during initial code generation. In that case
3405the @code{eh_return} instruction pattern should be used instead.
3406
3407If you want to support call frame exception handling, you must
3408define either this macro or the @code{eh_return} instruction pattern.
3409@end defmac
3410
3411@defmac RETURN_ADDR_OFFSET
3412If defined, an integer-valued C expression for which rtl will be generated
3413to add it to the exception handler address before it is searched in the
3414exception handling tables, and to subtract it again from the address before
3415using it to return to the exception handler.
3416@end defmac
3417
3418@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3419This macro chooses the encoding of pointers embedded in the exception
3420handling sections. If at all possible, this should be defined such
3421that the exception handling section will not require dynamic relocations,
3422and so may be read-only.
3423
3424@var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3425@var{global} is true if the symbol may be affected by dynamic relocations.
3426The macro should return a combination of the @code{DW_EH_PE_*} defines
3427as found in @file{dwarf2.h}.
3428
3429If this macro is not defined, pointers will not be encoded but
3430represented directly.
3431@end defmac
3432
3433@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3434This macro allows the target to emit whatever special magic is required
3435to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3436Generic code takes care of pc-relative and indirect encodings; this must
3437be defined if the target uses text-relative or data-relative encodings.
3438
3439This is a C statement that branches to @var{done} if the format was
3440handled. @var{encoding} is the format chosen, @var{size} is the number
3441of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3442to be emitted.
3443@end defmac
3444
3445@defmac MD_UNWIND_SUPPORT
3446A string specifying a file to be #include'd in unwind-dw2.c. The file
3447so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3448@end defmac
3449
3450@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3451This macro allows the target to add CPU and operating system specific
3452code to the call-frame unwinder for use when there is no unwind data
3453available. The most common reason to implement this macro is to unwind
3454through signal frames.
3455
3456This macro is called from @code{uw_frame_state_for} in
3457@file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3458@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3459@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3460for the address of the code being executed and @code{context->cfa} for
3461the stack pointer value. If the frame can be decoded, the register
3462save addresses should be updated in @var{fs} and the macro should
3463evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3464the macro should evaluate to @code{_URC_END_OF_STACK}.
3465
3466For proper signal handling in Java this macro is accompanied by
3467@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3468@end defmac
3469
3470@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3471This macro allows the target to add operating system specific code to the
3472call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3473usually used for signal or interrupt frames.
3474
3475This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3476@var{context} is an @code{_Unwind_Context};
3477@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3478for the abi and context in the @code{.unwabi} directive. If the
3479@code{.unwabi} directive can be handled, the register save addresses should
3480be updated in @var{fs}.
3481@end defmac
3482
3483@defmac TARGET_USES_WEAK_UNWIND_INFO
3484A C expression that evaluates to true if the target requires unwind
3485info to be given comdat linkage. Define it to be @code{1} if comdat
3486linkage is necessary. The default is @code{0}.
3487@end defmac
3488
3489@node Stack Checking
3490@subsection Specifying How Stack Checking is Done
3491
3492GCC will check that stack references are within the boundaries of the
3493stack, if the option @option{-fstack-check} is specified, in one of
3494three ways:
3495
3496@enumerate
3497@item
3498If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3499will assume that you have arranged for full stack checking to be done
3500at appropriate places in the configuration files. GCC will not do
3501other special processing.
3502
3503@item
3504If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3505@code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3506that you have arranged for static stack checking (checking of the
3507static stack frame of functions) to be done at appropriate places
3508in the configuration files. GCC will only emit code to do dynamic
3509stack checking (checking on dynamic stack allocations) using the third
3510approach below.
3511
3512@item
3513If neither of the above are true, GCC will generate code to periodically
3514``probe'' the stack pointer using the values of the macros defined below.
3515@end enumerate
3516
3517If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3518GCC will change its allocation strategy for large objects if the option
3519@option{-fstack-check} is specified: they will always be allocated
3520dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3521
3522@defmac STACK_CHECK_BUILTIN
3523A nonzero value if stack checking is done by the configuration files in a
3524machine-dependent manner. You should define this macro if stack checking
3525is required by the ABI of your machine or if you would like to do stack
3526checking in some more efficient way than the generic approach. The default
3527value of this macro is zero.
3528@end defmac
3529
3530@defmac STACK_CHECK_STATIC_BUILTIN
3531A nonzero value if static stack checking is done by the configuration files
3532in a machine-dependent manner. You should define this macro if you would
3533like to do static stack checking in some more efficient way than the generic
3534approach. The default value of this macro is zero.
3535@end defmac
3536
3537@defmac STACK_CHECK_PROBE_INTERVAL_EXP
3538An integer specifying the interval at which GCC must generate stack probe
3539instructions, defined as 2 raised to this integer. You will normally
3540define this macro so that the interval be no larger than the size of
3541the ``guard pages'' at the end of a stack area. The default value
3542of 12 (4096-byte interval) is suitable for most systems.
3543@end defmac
3544
3545@defmac STACK_CHECK_MOVING_SP
3546An integer which is nonzero if GCC should move the stack pointer page by page
3547when doing probes. This can be necessary on systems where the stack pointer
3548contains the bottom address of the memory area accessible to the executing
3549thread at any point in time. In this situation an alternate signal stack
3550is required in order to be able to recover from a stack overflow. The
3551default value of this macro is zero.
3552@end defmac
3553
3554@defmac STACK_CHECK_PROTECT
3555The number of bytes of stack needed to recover from a stack overflow, for
3556languages where such a recovery is supported. The default value of 75 words
3557with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
35588192 bytes with other exception handling mechanisms should be adequate for
3559most machines.
3560@end defmac
3561
3562The following macros are relevant only if neither STACK_CHECK_BUILTIN
3563nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3564in the opposite case.
3565
3566@defmac STACK_CHECK_MAX_FRAME_SIZE
3567The maximum size of a stack frame, in bytes. GCC will generate probe
3568instructions in non-leaf functions to ensure at least this many bytes of
3569stack are available. If a stack frame is larger than this size, stack
3570checking will not be reliable and GCC will issue a warning. The
3571default is chosen so that GCC only generates one instruction on most
3572systems. You should normally not change the default value of this macro.
3573@end defmac
3574
3575@defmac STACK_CHECK_FIXED_FRAME_SIZE
3576GCC uses this value to generate the above warning message. It
3577represents the amount of fixed frame used by a function, not including
3578space for any callee-saved registers, temporaries and user variables.
3579You need only specify an upper bound for this amount and will normally
3580use the default of four words.
3581@end defmac
3582
3583@defmac STACK_CHECK_MAX_VAR_SIZE
3584The maximum size, in bytes, of an object that GCC will place in the
3585fixed area of the stack frame when the user specifies
3586@option{-fstack-check}.
3587GCC computed the default from the values of the above macros and you will
3588normally not need to override that default.
3589@end defmac
3590
3591@need 2000
3592@node Frame Registers
3593@subsection Registers That Address the Stack Frame
3594
3595@c prevent bad page break with this line
3596This discusses registers that address the stack frame.
3597
3598@defmac STACK_POINTER_REGNUM
3599The register number of the stack pointer register, which must also be a
3600fixed register according to @code{FIXED_REGISTERS}. On most machines,
3601the hardware determines which register this is.
3602@end defmac
3603
3604@defmac FRAME_POINTER_REGNUM
3605The register number of the frame pointer register, which is used to
3606access automatic variables in the stack frame. On some machines, the
3607hardware determines which register this is. On other machines, you can
3608choose any register you wish for this purpose.
3609@end defmac
3610
3611@defmac HARD_FRAME_POINTER_REGNUM
3612On some machines the offset between the frame pointer and starting
3613offset of the automatic variables is not known until after register
3614allocation has been done (for example, because the saved registers are
3615between these two locations). On those machines, define
3616@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3617be used internally until the offset is known, and define
3618@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3619used for the frame pointer.
3620
3621You should define this macro only in the very rare circumstances when it
3622is not possible to calculate the offset between the frame pointer and
3623the automatic variables until after register allocation has been
3624completed. When this macro is defined, you must also indicate in your
3625definition of @code{ELIMINABLE_REGS} how to eliminate
3626@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3627or @code{STACK_POINTER_REGNUM}.
3628
3629Do not define this macro if it would be the same as
3630@code{FRAME_POINTER_REGNUM}.
3631@end defmac
3632
3633@defmac ARG_POINTER_REGNUM
3634The register number of the arg pointer register, which is used to access
3635the function's argument list. On some machines, this is the same as the
3636frame pointer register. On some machines, the hardware determines which
3637register this is. On other machines, you can choose any register you
3638wish for this purpose. If this is not the same register as the frame
3639pointer register, then you must mark it as a fixed register according to
3640@code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3641(@pxref{Elimination}).
3642@end defmac
3643
e3339d0f
JM
3644@defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3645Define this to a preprocessor constant that is nonzero if
3646@code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3647the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3648== FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3649definition is not suitable for use in preprocessor conditionals.
3650@end defmac
3651
3652@defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3653Define this to a preprocessor constant that is nonzero if
3654@code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3655same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3656ARG_POINTER_REGNUM)}; you only need to define this macro if that
3657definition is not suitable for use in preprocessor conditionals.
3658@end defmac
3659
38f8b050
JR
3660@defmac RETURN_ADDRESS_POINTER_REGNUM
3661The register number of the return address pointer register, which is used to
3662access the current function's return address from the stack. On some
3663machines, the return address is not at a fixed offset from the frame
3664pointer or stack pointer or argument pointer. This register can be defined
3665to point to the return address on the stack, and then be converted by
3666@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3667
3668Do not define this macro unless there is no other way to get the return
3669address from the stack.
3670@end defmac
3671
3672@defmac STATIC_CHAIN_REGNUM
3673@defmacx STATIC_CHAIN_INCOMING_REGNUM
3674Register numbers used for passing a function's static chain pointer. If
3675register windows are used, the register number as seen by the called
3676function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3677number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3678these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3679not be defined.
3680
3681The static chain register need not be a fixed register.
3682
3683If the static chain is passed in memory, these macros should not be
3684defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3685@end defmac
3686
3687@hook TARGET_STATIC_CHAIN
3688This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3689targets that may use different static chain locations for different
3690nested functions. This may be required if the target has function
3691attributes that affect the calling conventions of the function and
3692those calling conventions use different static chain locations.
3693
3694The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3695
3696If the static chain is passed in memory, this hook should be used to
3697provide rtx giving @code{mem} expressions that denote where they are stored.
3698Often the @code{mem} expression as seen by the caller will be at an offset
3699from the stack pointer and the @code{mem} expression as seen by the callee
3700will be at an offset from the frame pointer.
3701@findex stack_pointer_rtx
3702@findex frame_pointer_rtx
3703@findex arg_pointer_rtx
3704The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3705@code{arg_pointer_rtx} will have been initialized and should be used
3706to refer to those items.
3707@end deftypefn
3708
3709@defmac DWARF_FRAME_REGISTERS
3710This macro specifies the maximum number of hard registers that can be
3711saved in a call frame. This is used to size data structures used in
3712DWARF2 exception handling.
3713
3714Prior to GCC 3.0, this macro was needed in order to establish a stable
3715exception handling ABI in the face of adding new hard registers for ISA
3716extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3717in the number of hard registers. Nevertheless, this macro can still be
3718used to reduce the runtime memory requirements of the exception handling
3719routines, which can be substantial if the ISA contains a lot of
3720registers that are not call-saved.
3721
3722If this macro is not defined, it defaults to
3723@code{FIRST_PSEUDO_REGISTER}.
3724@end defmac
3725
3726@defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3727
3728This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3729for backward compatibility in pre GCC 3.0 compiled code.
3730
3731If this macro is not defined, it defaults to
3732@code{DWARF_FRAME_REGISTERS}.
3733@end defmac
3734
3735@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3736
3737Define this macro if the target's representation for dwarf registers
3738is different than the internal representation for unwind column.
3739Given a dwarf register, this macro should return the internal unwind
3740column number to use instead.
3741
3742See the PowerPC's SPE target for an example.
3743@end defmac
3744
3745@defmac DWARF_FRAME_REGNUM (@var{regno})
3746
3747Define this macro if the target's representation for dwarf registers
3748used in .eh_frame or .debug_frame is different from that used in other
3749debug info sections. Given a GCC hard register number, this macro
3750should return the .eh_frame register number. The default is
3751@code{DBX_REGISTER_NUMBER (@var{regno})}.
3752
3753@end defmac
3754
3755@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3756
3757Define this macro to map register numbers held in the call frame info
3758that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3759should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3760.eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3761return @code{@var{regno}}.
3762
3763@end defmac
3764
3765@node Elimination
3766@subsection Eliminating Frame Pointer and Arg Pointer
3767
3768@c prevent bad page break with this line
3769This is about eliminating the frame pointer and arg pointer.
3770
3771@hook TARGET_FRAME_POINTER_REQUIRED
3772This target hook should return @code{true} if a function must have and use
3773a frame pointer. This target hook is called in the reload pass. If its return
3774value is @code{true} the function will have a frame pointer.
3775
3776This target hook can in principle examine the current function and decide
3777according to the facts, but on most machines the constant @code{false} or the
3778constant @code{true} suffices. Use @code{false} when the machine allows code
3779to be generated with no frame pointer, and doing so saves some time or space.
3780Use @code{true} when there is no possible advantage to avoiding a frame
3781pointer.
3782
3783In certain cases, the compiler does not know how to produce valid code
3784without a frame pointer. The compiler recognizes those cases and
3785automatically gives the function a frame pointer regardless of what
3786@code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3787them.
3788
3789In a function that does not require a frame pointer, the frame pointer
3790register can be allocated for ordinary usage, unless you mark it as a
3791fixed register. See @code{FIXED_REGISTERS} for more information.
3792
3793Default return value is @code{false}.
3794@end deftypefn
3795
3796@findex get_frame_size
3797@defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3798A C statement to store in the variable @var{depth-var} the difference
3799between the frame pointer and the stack pointer values immediately after
3800the function prologue. The value would be computed from information
3801such as the result of @code{get_frame_size ()} and the tables of
3802registers @code{regs_ever_live} and @code{call_used_regs}.
3803
3804If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3805need not be defined. Otherwise, it must be defined even if
3806@code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3807case, you may set @var{depth-var} to anything.
3808@end defmac
3809
3810@defmac ELIMINABLE_REGS
3811If defined, this macro specifies a table of register pairs used to
3812eliminate unneeded registers that point into the stack frame. If it is not
3813defined, the only elimination attempted by the compiler is to replace
3814references to the frame pointer with references to the stack pointer.
3815
3816The definition of this macro is a list of structure initializations, each
3817of which specifies an original and replacement register.
3818
3819On some machines, the position of the argument pointer is not known until
3820the compilation is completed. In such a case, a separate hard register
3821must be used for the argument pointer. This register can be eliminated by
3822replacing it with either the frame pointer or the argument pointer,
3823depending on whether or not the frame pointer has been eliminated.
3824
3825In this case, you might specify:
3826@smallexample
3827#define ELIMINABLE_REGS \
3828@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3829 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3830 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3831@end smallexample
3832
3833Note that the elimination of the argument pointer with the stack pointer is
3834specified first since that is the preferred elimination.
3835@end defmac
3836
3837@hook TARGET_CAN_ELIMINATE
3838This target hook should returns @code{true} if the compiler is allowed to
3839try to replace register number @var{from_reg} with register number
3840@var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3841is defined, and will usually be @code{true}, since most of the cases
3842preventing register elimination are things that the compiler already
3843knows about.
3844
3845Default return value is @code{true}.
3846@end deftypefn
3847
3848@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3849This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3850specifies the initial difference between the specified pair of
3851registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3852defined.
3853@end defmac
3854
3855@node Stack Arguments
3856@subsection Passing Function Arguments on the Stack
3857@cindex arguments on stack
3858@cindex stack arguments
3859
3860The macros in this section control how arguments are passed
3861on the stack. See the following section for other macros that
3862control passing certain arguments in registers.
3863
3864@hook TARGET_PROMOTE_PROTOTYPES
3865This target hook returns @code{true} if an argument declared in a
3866prototype as an integral type smaller than @code{int} should actually be
3867passed as an @code{int}. In addition to avoiding errors in certain
3868cases of mismatch, it also makes for better code on certain machines.
3869The default is to not promote prototypes.
3870@end deftypefn
3871
3872@defmac PUSH_ARGS
3873A C expression. If nonzero, push insns will be used to pass
3874outgoing arguments.
3875If the target machine does not have a push instruction, set it to zero.
3876That directs GCC to use an alternate strategy: to
3877allocate the entire argument block and then store the arguments into
3878it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3879@end defmac
3880
3881@defmac PUSH_ARGS_REVERSED
3882A C expression. If nonzero, function arguments will be evaluated from
3883last to first, rather than from first to last. If this macro is not
3884defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3885and args grow in opposite directions, and 0 otherwise.
3886@end defmac
3887
3888@defmac PUSH_ROUNDING (@var{npushed})
3889A C expression that is the number of bytes actually pushed onto the
3890stack when an instruction attempts to push @var{npushed} bytes.
3891
3892On some machines, the definition
3893
3894@smallexample
3895#define PUSH_ROUNDING(BYTES) (BYTES)
3896@end smallexample
3897
3898@noindent
3899will suffice. But on other machines, instructions that appear
3900to push one byte actually push two bytes in an attempt to maintain
3901alignment. Then the definition should be
3902
3903@smallexample
3904#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3905@end smallexample
3906@end defmac
3907
3908@findex current_function_outgoing_args_size
3909@defmac ACCUMULATE_OUTGOING_ARGS
3910A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3911will be computed and placed into the variable
3912@code{current_function_outgoing_args_size}. No space will be pushed
3913onto the stack for each call; instead, the function prologue should
3914increase the stack frame size by this amount.
3915
3916Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3917is not proper.
3918@end defmac
3919
3920@defmac REG_PARM_STACK_SPACE (@var{fndecl})
3921Define this macro if functions should assume that stack space has been
3922allocated for arguments even when their values are passed in
3923registers.
3924
3925The value of this macro is the size, in bytes, of the area reserved for
3926arguments passed in registers for the function represented by @var{fndecl},
3927which can be zero if GCC is calling a library function.
3928The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3929of the function.
3930
3931This space can be allocated by the caller, or be a part of the
3932machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3933which.
3934@end defmac
3935@c above is overfull. not sure what to do. --mew 5feb93 did
3936@c something, not sure if it looks good. --mew 10feb93
3937
3938@defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3939Define this to a nonzero value if it is the responsibility of the
3940caller to allocate the area reserved for arguments passed in registers
3941when calling a function of @var{fntype}. @var{fntype} may be NULL
3942if the function called is a library function.
3943
3944If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3945whether the space for these arguments counts in the value of
3946@code{current_function_outgoing_args_size}.
3947@end defmac
3948
3949@defmac STACK_PARMS_IN_REG_PARM_AREA
3950Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3951stack parameters don't skip the area specified by it.
3952@c i changed this, makes more sens and it should have taken care of the
3953@c overfull.. not as specific, tho. --mew 5feb93
3954
3955Normally, when a parameter is not passed in registers, it is placed on the
3956stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3957suppresses this behavior and causes the parameter to be passed on the
3958stack in its natural location.
3959@end defmac
3960
893d13d5 3961@hook TARGET_RETURN_POPS_ARGS
38f8b050
JR
3962This target hook returns the number of bytes of its own arguments that
3963a function pops on returning, or 0 if the function pops no arguments
3964and the caller must therefore pop them all after the function returns.
3965
3966@var{fundecl} is a C variable whose value is a tree node that describes
3967the function in question. Normally it is a node of type
3968@code{FUNCTION_DECL} that describes the declaration of the function.
3969From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3970
3971@var{funtype} is a C variable whose value is a tree node that
3972describes the function in question. Normally it is a node of type
3973@code{FUNCTION_TYPE} that describes the data type of the function.
3974From this it is possible to obtain the data types of the value and
3975arguments (if known).
3976
3977When a call to a library function is being considered, @var{fundecl}
3978will contain an identifier node for the library function. Thus, if
3979you need to distinguish among various library functions, you can do so
3980by their names. Note that ``library function'' in this context means
3981a function used to perform arithmetic, whose name is known specially
3982in the compiler and was not mentioned in the C code being compiled.
3983
893d13d5 3984@var{size} is the number of bytes of arguments passed on the
38f8b050
JR
3985stack. If a variable number of bytes is passed, it is zero, and
3986argument popping will always be the responsibility of the calling function.
3987
3988On the VAX, all functions always pop their arguments, so the definition
893d13d5 3989of this macro is @var{size}. On the 68000, using the standard
38f8b050
JR
3990calling convention, no functions pop their arguments, so the value of
3991the macro is always 0 in this case. But an alternative calling
3992convention is available in which functions that take a fixed number of
3993arguments pop them but other functions (such as @code{printf}) pop
3994nothing (the caller pops all). When this convention is in use,
3995@var{funtype} is examined to determine whether a function takes a fixed
3996number of arguments.
3997@end deftypefn
3998
3999@defmac CALL_POPS_ARGS (@var{cum})
4000A C expression that should indicate the number of bytes a call sequence
4001pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
4002when compiling a function call.
4003
4004@var{cum} is the variable in which all arguments to the called function
4005have been accumulated.
4006
4007On certain architectures, such as the SH5, a call trampoline is used
4008that pops certain registers off the stack, depending on the arguments
4009that have been passed to the function. Since this is a property of the
4010call site, not of the called function, @code{RETURN_POPS_ARGS} is not
4011appropriate.
4012@end defmac
4013
4014@node Register Arguments
4015@subsection Passing Arguments in Registers
4016@cindex arguments in registers
4017@cindex registers arguments
4018
4019This section describes the macros which let you control how various
4020types of arguments are passed in registers or how they are arranged in
4021the stack.
4022
4023@defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4024A C expression that controls whether a function argument is passed
4025in a register, and which register.
4026
4027The arguments are @var{cum}, which summarizes all the previous
4028arguments; @var{mode}, the machine mode of the argument; @var{type},
4029the data type of the argument as a tree node or 0 if that is not known
4030(which happens for C support library functions); and @var{named},
4031which is 1 for an ordinary argument and 0 for nameless arguments that
4032correspond to @samp{@dots{}} in the called function's prototype.
4033@var{type} can be an incomplete type if a syntax error has previously
4034occurred.
4035
4036The value of the expression is usually either a @code{reg} RTX for the
4037hard register in which to pass the argument, or zero to pass the
4038argument on the stack.
4039
4040For machines like the VAX and 68000, where normally all arguments are
4041pushed, zero suffices as a definition.
4042
4043The value of the expression can also be a @code{parallel} RTX@. This is
4044used when an argument is passed in multiple locations. The mode of the
4045@code{parallel} should be the mode of the entire argument. The
4046@code{parallel} holds any number of @code{expr_list} pairs; each one
4047describes where part of the argument is passed. In each
4048@code{expr_list} the first operand must be a @code{reg} RTX for the hard
4049register in which to pass this part of the argument, and the mode of the
4050register RTX indicates how large this part of the argument is. The
4051second operand of the @code{expr_list} is a @code{const_int} which gives
4052the offset in bytes into the entire argument of where this part starts.
4053As a special exception the first @code{expr_list} in the @code{parallel}
4054RTX may have a first operand of zero. This indicates that the entire
4055argument is also stored on the stack.
4056
4057The last time this macro is called, it is called with @code{MODE ==
4058VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4059pattern as operands 2 and 3 respectively.
4060
4061@cindex @file{stdarg.h} and register arguments
4062The usual way to make the ISO library @file{stdarg.h} work on a machine
4063where some arguments are usually passed in registers, is to cause
4064nameless arguments to be passed on the stack instead. This is done
4065by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
4066
4067@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
4068@cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
4069You may use the hook @code{targetm.calls.must_pass_in_stack}
4070in the definition of this macro to determine if this argument is of a
4071type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4072is not defined and @code{FUNCTION_ARG} returns nonzero for such an
4073argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4074defined, the argument will be computed in the stack and then loaded into
4075a register.
4076@end defmac
4077
4078@hook TARGET_MUST_PASS_IN_STACK
4079This target hook should return @code{true} if we should not pass @var{type}
4080solely in registers. The file @file{expr.h} defines a
4081definition that is usually appropriate, refer to @file{expr.h} for additional
4082documentation.
4083@end deftypefn
4084
4085@defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4086Define this macro if the target machine has ``register windows'', so
4087that the register in which a function sees an arguments is not
4088necessarily the same as the one in which the caller passed the
4089argument.
4090
4091For such machines, @code{FUNCTION_ARG} computes the register in which
4092the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
4093be defined in a similar fashion to tell the function being called
4094where the arguments will arrive.
4095
4096If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
4097serves both purposes.
4098@end defmac
4099
4100@hook TARGET_ARG_PARTIAL_BYTES
4101This target hook returns the number of bytes at the beginning of an
4102argument that must be put in registers. The value must be zero for
4103arguments that are passed entirely in registers or that are entirely
4104pushed on the stack.
4105
4106On some machines, certain arguments must be passed partially in
4107registers and partially in memory. On these machines, typically the
4108first few words of arguments are passed in registers, and the rest
4109on the stack. If a multi-word argument (a @code{double} or a
4110structure) crosses that boundary, its first few words must be passed
4111in registers and the rest must be pushed. This macro tells the
4112compiler when this occurs, and how many bytes should go in registers.
4113
4114@code{FUNCTION_ARG} for these arguments should return the first
4115register to be used by the caller for this argument; likewise
4116@code{FUNCTION_INCOMING_ARG}, for the called function.
4117@end deftypefn
4118
ec9f85e5 4119@hook TARGET_PASS_BY_REFERENCE
38f8b050
JR
4120This target hook should return @code{true} if an argument at the
4121position indicated by @var{cum} should be passed by reference. This
4122predicate is queried after target independent reasons for being
4123passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4124
4125If the hook returns true, a copy of that argument is made in memory and a
4126pointer to the argument is passed instead of the argument itself.
4127The pointer is passed in whatever way is appropriate for passing a pointer
4128to that type.
4129@end deftypefn
4130
4131@hook TARGET_CALLEE_COPIES
4132The function argument described by the parameters to this hook is
4133known to be passed by reference. The hook should return true if the
4134function argument should be copied by the callee instead of copied
4135by the caller.
4136
4137For any argument for which the hook returns true, if it can be
4138determined that the argument is not modified, then a copy need
4139not be generated.
4140
4141The default version of this hook always returns false.
4142@end deftypefn
4143
4144@defmac CUMULATIVE_ARGS
4145A C type for declaring a variable that is used as the first argument of
4146@code{FUNCTION_ARG} and other related values. For some target machines,
4147the type @code{int} suffices and can hold the number of bytes of
4148argument so far.
4149
4150There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4151arguments that have been passed on the stack. The compiler has other
4152variables to keep track of that. For target machines on which all
4153arguments are passed on the stack, there is no need to store anything in
4154@code{CUMULATIVE_ARGS}; however, the data structure must exist and
4155should not be empty, so use @code{int}.
4156@end defmac
4157
4158@defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4159If defined, this macro is called before generating any code for a
4160function, but after the @var{cfun} descriptor for the function has been
4161created. The back end may use this macro to update @var{cfun} to
4162reflect an ABI other than that which would normally be used by default.
4163If the compiler is generating code for a compiler-generated function,
4164@var{fndecl} may be @code{NULL}.
4165@end defmac
4166
4167@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4168A C statement (sans semicolon) for initializing the variable
4169@var{cum} for the state at the beginning of the argument list. The
4170variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4171is the tree node for the data type of the function which will receive
4172the args, or 0 if the args are to a compiler support library function.
4173For direct calls that are not libcalls, @var{fndecl} contain the
4174declaration node of the function. @var{fndecl} is also set when
4175@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4176being compiled. @var{n_named_args} is set to the number of named
4177arguments, including a structure return address if it is passed as a
4178parameter, when making a call. When processing incoming arguments,
4179@var{n_named_args} is set to @minus{}1.
4180
4181When processing a call to a compiler support library function,
4182@var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4183contains the name of the function, as a string. @var{libname} is 0 when
4184an ordinary C function call is being processed. Thus, each time this
4185macro is called, either @var{libname} or @var{fntype} is nonzero, but
4186never both of them at once.
4187@end defmac
4188
4189@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4190Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4191it gets a @code{MODE} argument instead of @var{fntype}, that would be
4192@code{NULL}. @var{indirect} would always be zero, too. If this macro
4193is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
41940)} is used instead.
4195@end defmac
4196
4197@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4198Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4199finding the arguments for the function being compiled. If this macro is
4200undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4201
4202The value passed for @var{libname} is always 0, since library routines
4203with special calling conventions are never compiled with GCC@. The
4204argument @var{libname} exists for symmetry with
4205@code{INIT_CUMULATIVE_ARGS}.
4206@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4207@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4208@end defmac
4209
4210@defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
4211A C statement (sans semicolon) to update the summarizer variable
4212@var{cum} to advance past an argument in the argument list. The
4213values @var{mode}, @var{type} and @var{named} describe that argument.
4214Once this is done, the variable @var{cum} is suitable for analyzing
4215the @emph{following} argument with @code{FUNCTION_ARG}, etc.
4216
4217This macro need not do anything if the argument in question was passed
4218on the stack. The compiler knows how to track the amount of stack space
4219used for arguments without any special help.
4220@end defmac
4221
4222@defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4223If defined, a C expression that is the number of bytes to add to the
4224offset of the argument passed in memory. This is needed for the SPU,
4225which passes @code{char} and @code{short} arguments in the preferred
4226slot that is in the middle of the quad word instead of starting at the
4227top.
4228@end defmac
4229
4230@defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4231If defined, a C expression which determines whether, and in which direction,
4232to pad out an argument with extra space. The value should be of type
4233@code{enum direction}: either @code{upward} to pad above the argument,
4234@code{downward} to pad below, or @code{none} to inhibit padding.
4235
4236The @emph{amount} of padding is always just enough to reach the next
4237multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
4238it.
4239
4240This macro has a default definition which is right for most systems.
4241For little-endian machines, the default is to pad upward. For
4242big-endian machines, the default is to pad downward for an argument of
4243constant size shorter than an @code{int}, and upward otherwise.
4244@end defmac
4245
4246@defmac PAD_VARARGS_DOWN
4247If defined, a C expression which determines whether the default
4248implementation of va_arg will attempt to pad down before reading the
4249next argument, if that argument is smaller than its aligned space as
4250controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4251arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4252@end defmac
4253
4254@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4255Specify padding for the last element of a block move between registers and
4256memory. @var{first} is nonzero if this is the only element. Defining this
4257macro allows better control of register function parameters on big-endian
4258machines, without using @code{PARALLEL} rtl. In particular,
4259@code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4260registers, as there is no longer a "wrong" part of a register; For example,
4261a three byte aggregate may be passed in the high part of a register if so
4262required.
4263@end defmac
4264
4265@defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
4266If defined, a C expression that gives the alignment boundary, in bits,
4267of an argument with the specified mode and type. If it is not defined,
4268@code{PARM_BOUNDARY} is used for all arguments.
4269@end defmac
4270
4271@defmac FUNCTION_ARG_REGNO_P (@var{regno})
4272A C expression that is nonzero if @var{regno} is the number of a hard
4273register in which function arguments are sometimes passed. This does
4274@emph{not} include implicit arguments such as the static chain and
4275the structure-value address. On many machines, no registers can be
4276used for this purpose since all function arguments are pushed on the
4277stack.
4278@end defmac
4279
4280@hook TARGET_SPLIT_COMPLEX_ARG
4281This hook should return true if parameter of type @var{type} are passed
4282as two scalar parameters. By default, GCC will attempt to pack complex
4283arguments into the target's word size. Some ABIs require complex arguments
4284to be split and treated as their individual components. For example, on
4285AIX64, complex floats should be passed in a pair of floating point
4286registers, even though a complex float would fit in one 64-bit floating
4287point register.
4288
4289The default value of this hook is @code{NULL}, which is treated as always
4290false.
4291@end deftypefn
4292
4293@hook TARGET_BUILD_BUILTIN_VA_LIST
4294This hook returns a type node for @code{va_list} for the target.
4295The default version of the hook returns @code{void*}.
4296@end deftypefn
4297
07a5b2bc 4298@hook TARGET_ENUM_VA_LIST_P
38f8b050
JR
4299This target hook is used in function @code{c_common_nodes_and_builtins}
4300to iterate through the target specific builtin types for va_list. The
4301variable @var{idx} is used as iterator. @var{pname} has to be a pointer
07a5b2bc 4302to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
38f8b050 4303variable.
07a5b2bc 4304The arguments @var{pname} and @var{ptree} are used to store the result of
38f8b050
JR
4305this macro and are set to the name of the va_list builtin type and its
4306internal type.
4307If the return value of this macro is zero, then there is no more element.
4308Otherwise the @var{IDX} should be increased for the next call of this
4309macro to iterate through all types.
4310@end deftypefn
4311
4312@hook TARGET_FN_ABI_VA_LIST
4313This hook returns the va_list type of the calling convention specified by
4314@var{fndecl}.
4315The default version of this hook returns @code{va_list_type_node}.
4316@end deftypefn
4317
4318@hook TARGET_CANONICAL_VA_LIST_TYPE
4319This hook returns the va_list type of the calling convention specified by the
4320type of @var{type}. If @var{type} is not a valid va_list type, it returns
4321@code{NULL_TREE}.
4322@end deftypefn
4323
4324@hook TARGET_GIMPLIFY_VA_ARG_EXPR
4325This hook performs target-specific gimplification of
4326@code{VA_ARG_EXPR}. The first two parameters correspond to the
4327arguments to @code{va_arg}; the latter two are as in
4328@code{gimplify.c:gimplify_expr}.
4329@end deftypefn
4330
4331@hook TARGET_VALID_POINTER_MODE
4332Define this to return nonzero if the port can handle pointers
4333with machine mode @var{mode}. The default version of this
4334hook returns true for both @code{ptr_mode} and @code{Pmode}.
4335@end deftypefn
4336
4337@hook TARGET_SCALAR_MODE_SUPPORTED_P
4338Define this to return nonzero if the port is prepared to handle
4339insns involving scalar mode @var{mode}. For a scalar mode to be
4340considered supported, all the basic arithmetic and comparisons
4341must work.
4342
4343The default version of this hook returns true for any mode
4344required to handle the basic C types (as defined by the port).
4345Included here are the double-word arithmetic supported by the
4346code in @file{optabs.c}.
4347@end deftypefn
4348
4349@hook TARGET_VECTOR_MODE_SUPPORTED_P
4350Define this to return nonzero if the port is prepared to handle
4351insns involving vector mode @var{mode}. At the very least, it
4352must have move patterns for this mode.
4353@end deftypefn
4354
4355@hook TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P
4356Define this to return nonzero for machine modes for which the port has
4357small register classes. If this target hook returns nonzero for a given
4358@var{mode}, the compiler will try to minimize the lifetime of registers
4359in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4360In this case, the hook is expected to return nonzero if it returns nonzero
4361for any mode.
4362
4363On some machines, it is risky to let hard registers live across arbitrary
4364insns. Typically, these machines have instructions that require values
4365to be in specific registers (like an accumulator), and reload will fail
4366if the required hard register is used for another purpose across such an
4367insn.
4368
4369Passes before reload do not know which hard registers will be used
4370in an instruction, but the machine modes of the registers set or used in
4371the instruction are already known. And for some machines, register
4372classes are small for, say, integer registers but not for floating point
4373registers. For example, the AMD x86-64 architecture requires specific
4374registers for the legacy x86 integer instructions, but there are many
4375SSE registers for floating point operations. On such targets, a good
4376strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4377machine modes but zero for the SSE register classes.
4378
4379The default version of this hook retuns false for any mode. It is always
4380safe to redefine this hook to return with a nonzero value. But if you
4381unnecessarily define it, you will reduce the amount of optimizations
4382that can be performed in some cases. If you do not define this hook
4383to return a nonzero value when it is required, the compiler will run out
4384of spill registers and print a fatal error message.
4385@end deftypefn
4386
4387@node Scalar Return
4388@subsection How Scalar Function Values Are Returned
4389@cindex return values in registers
4390@cindex values, returned by functions
4391@cindex scalars, returned as values
4392
4393This section discusses the macros that control returning scalars as
4394values---values that can fit in registers.
4395
4396@hook TARGET_FUNCTION_VALUE
4397
4398Define this to return an RTX representing the place where a function
4399returns or receives a value of data type @var{ret_type}, a tree node
4400representing a data type. @var{fn_decl_or_type} is a tree node
4401representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4402function being called. If @var{outgoing} is false, the hook should
4403compute the register in which the caller will see the return value.
4404Otherwise, the hook should return an RTX representing the place where
4405a function returns a value.
4406
4407On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4408(Actually, on most machines, scalar values are returned in the same
4409place regardless of mode.) The value of the expression is usually a
4410@code{reg} RTX for the hard register where the return value is stored.
4411The value can also be a @code{parallel} RTX, if the return value is in
4412multiple places. See @code{FUNCTION_ARG} for an explanation of the
4413@code{parallel} form. Note that the callee will populate every
4414location specified in the @code{parallel}, but if the first element of
4415the @code{parallel} contains the whole return value, callers will use
4416that element as the canonical location and ignore the others. The m68k
4417port uses this type of @code{parallel} to return pointers in both
4418@samp{%a0} (the canonical location) and @samp{%d0}.
4419
4420If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4421the same promotion rules specified in @code{PROMOTE_MODE} if
4422@var{valtype} is a scalar type.
4423
4424If the precise function being called is known, @var{func} is a tree
4425node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4426pointer. This makes it possible to use a different value-returning
4427convention for specific functions when all their calls are
4428known.
4429
4430Some target machines have ``register windows'' so that the register in
4431which a function returns its value is not the same as the one in which
4432the caller sees the value. For such machines, you should return
4433different RTX depending on @var{outgoing}.
4434
4435@code{TARGET_FUNCTION_VALUE} is not used for return values with
4436aggregate data types, because these are returned in another way. See
4437@code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4438@end deftypefn
4439
4440@defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4441This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4442a new target instead.
4443@end defmac
4444
4445@defmac LIBCALL_VALUE (@var{mode})
4446A C expression to create an RTX representing the place where a library
4447function returns a value of mode @var{mode}.
4448
4449Note that ``library function'' in this context means a compiler
4450support routine, used to perform arithmetic, whose name is known
4451specially by the compiler and was not mentioned in the C code being
4452compiled.
4453@end defmac
4454
4455@hook TARGET_LIBCALL_VALUE
4456Define this hook if the back-end needs to know the name of the libcall
4457function in order to determine where the result should be returned.
4458
4459The mode of the result is given by @var{mode} and the name of the called
4460library function is given by @var{fun}. The hook should return an RTX
4461representing the place where the library function result will be returned.
4462
4463If this hook is not defined, then LIBCALL_VALUE will be used.
4464@end deftypefn
4465
4466@defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4467A C expression that is nonzero if @var{regno} is the number of a hard
4468register in which the values of called function may come back.
4469
4470A register whose use for returning values is limited to serving as the
4471second of a pair (for a value of type @code{double}, say) need not be
4472recognized by this macro. So for most machines, this definition
4473suffices:
4474
4475@smallexample
4476#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4477@end smallexample
4478
4479If the machine has register windows, so that the caller and the called
4480function use different registers for the return value, this macro
4481should recognize only the caller's register numbers.
4482
4483This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4484for a new target instead.
4485@end defmac
4486
4487@hook TARGET_FUNCTION_VALUE_REGNO_P
4488A target hook that return @code{true} if @var{regno} is the number of a hard
4489register in which the values of called function may come back.
4490
4491A register whose use for returning values is limited to serving as the
4492second of a pair (for a value of type @code{double}, say) need not be
4493recognized by this target hook.
4494
4495If the machine has register windows, so that the caller and the called
4496function use different registers for the return value, this target hook
4497should recognize only the caller's register numbers.
4498
4499If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4500@end deftypefn
4501
4502@defmac APPLY_RESULT_SIZE
4503Define this macro if @samp{untyped_call} and @samp{untyped_return}
4504need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4505saving and restoring an arbitrary return value.
4506@end defmac
4507
4508@hook TARGET_RETURN_IN_MSB
4509This hook should return true if values of type @var{type} are returned
4510at the most significant end of a register (in other words, if they are
4511padded at the least significant end). You can assume that @var{type}
4512is returned in a register; the caller is required to check this.
4513
4514Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4515be able to hold the complete return value. For example, if a 1-, 2-
4516or 3-byte structure is returned at the most significant end of a
45174-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4518@code{SImode} rtx.
4519@end deftypefn
4520
4521@node Aggregate Return
4522@subsection How Large Values Are Returned
4523@cindex aggregates as return values
4524@cindex large return values
4525@cindex returning aggregate values
4526@cindex structure value address
4527
4528When a function value's mode is @code{BLKmode} (and in some other
4529cases), the value is not returned according to
4530@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4531caller passes the address of a block of memory in which the value
4532should be stored. This address is called the @dfn{structure value
4533address}.
4534
4535This section describes how to control returning structure values in
4536memory.
4537
4538@hook TARGET_RETURN_IN_MEMORY
4539This target hook should return a nonzero value to say to return the
4540function value in memory, just as large structures are always returned.
4541Here @var{type} will be the data type of the value, and @var{fntype}
4542will be the type of the function doing the returning, or @code{NULL} for
4543libcalls.
4544
4545Note that values of mode @code{BLKmode} must be explicitly handled
4546by this function. Also, the option @option{-fpcc-struct-return}
4547takes effect regardless of this macro. On most systems, it is
4548possible to leave the hook undefined; this causes a default
4549definition to be used, whose value is the constant 1 for @code{BLKmode}
4550values, and 0 otherwise.
4551
4552Do not use this hook to indicate that structures and unions should always
4553be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4554to indicate this.
4555@end deftypefn
4556
4557@defmac DEFAULT_PCC_STRUCT_RETURN
4558Define this macro to be 1 if all structure and union return values must be
4559in memory. Since this results in slower code, this should be defined
4560only if needed for compatibility with other compilers or with an ABI@.
4561If you define this macro to be 0, then the conventions used for structure
4562and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4563target hook.
4564
4565If not defined, this defaults to the value 1.
4566@end defmac
4567
4568@hook TARGET_STRUCT_VALUE_RTX
4569This target hook should return the location of the structure value
4570address (normally a @code{mem} or @code{reg}), or 0 if the address is
4571passed as an ``invisible'' first argument. Note that @var{fndecl} may
4572be @code{NULL}, for libcalls. You do not need to define this target
4573hook if the address is always passed as an ``invisible'' first
4574argument.
4575
4576On some architectures the place where the structure value address
4577is found by the called function is not the same place that the
4578caller put it. This can be due to register windows, or it could
4579be because the function prologue moves it to a different place.
4580@var{incoming} is @code{1} or @code{2} when the location is needed in
4581the context of the called function, and @code{0} in the context of
4582the caller.
4583
4584If @var{incoming} is nonzero and the address is to be found on the
4585stack, return a @code{mem} which refers to the frame pointer. If
4586@var{incoming} is @code{2}, the result is being used to fetch the
4587structure value address at the beginning of a function. If you need
4588to emit adjusting code, you should do it at this point.
4589@end deftypefn
4590
4591@defmac PCC_STATIC_STRUCT_RETURN
4592Define this macro if the usual system convention on the target machine
4593for returning structures and unions is for the called function to return
4594the address of a static variable containing the value.
4595
4596Do not define this if the usual system convention is for the caller to
4597pass an address to the subroutine.
4598
4599This macro has effect in @option{-fpcc-struct-return} mode, but it does
4600nothing when you use @option{-freg-struct-return} mode.
4601@end defmac
4602
4603@node Caller Saves
4604@subsection Caller-Saves Register Allocation
4605
4606If you enable it, GCC can save registers around function calls. This
4607makes it possible to use call-clobbered registers to hold variables that
4608must live across calls.
4609
4610@defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4611A C expression to determine whether it is worthwhile to consider placing
4612a pseudo-register in a call-clobbered hard register and saving and
4613restoring it around each function call. The expression should be 1 when
4614this is worth doing, and 0 otherwise.
4615
4616If you don't define this macro, a default is used which is good on most
4617machines: @code{4 * @var{calls} < @var{refs}}.
4618@end defmac
4619
4620@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4621A C expression specifying which mode is required for saving @var{nregs}
4622of a pseudo-register in call-clobbered hard register @var{regno}. If
4623@var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4624returned. For most machines this macro need not be defined since GCC
4625will select the smallest suitable mode.
4626@end defmac
4627
4628@node Function Entry
4629@subsection Function Entry and Exit
4630@cindex function entry and exit
4631@cindex prologue
4632@cindex epilogue
4633
4634This section describes the macros that output function entry
4635(@dfn{prologue}) and exit (@dfn{epilogue}) code.
4636
4637@hook TARGET_ASM_FUNCTION_PROLOGUE
4638If defined, a function that outputs the assembler code for entry to a
4639function. The prologue is responsible for setting up the stack frame,
4640initializing the frame pointer register, saving registers that must be
4641saved, and allocating @var{size} additional bytes of storage for the
4642local variables. @var{size} is an integer. @var{file} is a stdio
4643stream to which the assembler code should be output.
4644
4645The label for the beginning of the function need not be output by this
4646macro. That has already been done when the macro is run.
4647
4648@findex regs_ever_live
4649To determine which registers to save, the macro can refer to the array
4650@code{regs_ever_live}: element @var{r} is nonzero if hard register
4651@var{r} is used anywhere within the function. This implies the function
4652prologue should save register @var{r}, provided it is not one of the
4653call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4654@code{regs_ever_live}.)
4655
4656On machines that have ``register windows'', the function entry code does
4657not save on the stack the registers that are in the windows, even if
4658they are supposed to be preserved by function calls; instead it takes
4659appropriate steps to ``push'' the register stack, if any non-call-used
4660registers are used in the function.
4661
4662@findex frame_pointer_needed
4663On machines where functions may or may not have frame-pointers, the
4664function entry code must vary accordingly; it must set up the frame
4665pointer if one is wanted, and not otherwise. To determine whether a
4666frame pointer is in wanted, the macro can refer to the variable
4667@code{frame_pointer_needed}. The variable's value will be 1 at run
4668time in a function that needs a frame pointer. @xref{Elimination}.
4669
4670The function entry code is responsible for allocating any stack space
4671required for the function. This stack space consists of the regions
4672listed below. In most cases, these regions are allocated in the
4673order listed, with the last listed region closest to the top of the
4674stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4675the highest address if it is not defined). You can use a different order
4676for a machine if doing so is more convenient or required for
4677compatibility reasons. Except in cases where required by standard
4678or by a debugger, there is no reason why the stack layout used by GCC
4679need agree with that used by other compilers for a machine.
4680@end deftypefn
4681
4682@hook TARGET_ASM_FUNCTION_END_PROLOGUE
4683If defined, a function that outputs assembler code at the end of a
4684prologue. This should be used when the function prologue is being
4685emitted as RTL, and you have some extra assembler that needs to be
4686emitted. @xref{prologue instruction pattern}.
4687@end deftypefn
4688
4689@hook TARGET_ASM_FUNCTION_BEGIN_EPILOGUE
4690If defined, a function that outputs assembler code at the start of an
4691epilogue. This should be used when the function epilogue is being
4692emitted as RTL, and you have some extra assembler that needs to be
4693emitted. @xref{epilogue instruction pattern}.
4694@end deftypefn
4695
4696@hook TARGET_ASM_FUNCTION_EPILOGUE
4697If defined, a function that outputs the assembler code for exit from a
4698function. The epilogue is responsible for restoring the saved
4699registers and stack pointer to their values when the function was
4700called, and returning control to the caller. This macro takes the
4701same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4702registers to restore are determined from @code{regs_ever_live} and
4703@code{CALL_USED_REGISTERS} in the same way.
4704
4705On some machines, there is a single instruction that does all the work
4706of returning from the function. On these machines, give that
4707instruction the name @samp{return} and do not define the macro
4708@code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4709
4710Do not define a pattern named @samp{return} if you want the
4711@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4712switches to control whether return instructions or epilogues are used,
4713define a @samp{return} pattern with a validity condition that tests the
4714target switches appropriately. If the @samp{return} pattern's validity
4715condition is false, epilogues will be used.
4716
4717On machines where functions may or may not have frame-pointers, the
4718function exit code must vary accordingly. Sometimes the code for these
4719two cases is completely different. To determine whether a frame pointer
4720is wanted, the macro can refer to the variable
4721@code{frame_pointer_needed}. The variable's value will be 1 when compiling
4722a function that needs a frame pointer.
4723
4724Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4725@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4726The C variable @code{current_function_is_leaf} is nonzero for such a
4727function. @xref{Leaf Functions}.
4728
4729On some machines, some functions pop their arguments on exit while
4730others leave that for the caller to do. For example, the 68020 when
4731given @option{-mrtd} pops arguments in functions that take a fixed
4732number of arguments.
4733
4734@findex current_function_pops_args
4735Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4736functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4737needs to know what was decided. The number of bytes of the current
4738function's arguments that this function should pop is available in
4739@code{crtl->args.pops_args}. @xref{Scalar Return}.
4740@end deftypefn
4741
4742@itemize @bullet
4743@item
4744@findex current_function_pretend_args_size
4745A region of @code{current_function_pretend_args_size} bytes of
4746uninitialized space just underneath the first argument arriving on the
4747stack. (This may not be at the very start of the allocated stack region
4748if the calling sequence has pushed anything else since pushing the stack
4749arguments. But usually, on such machines, nothing else has been pushed
4750yet, because the function prologue itself does all the pushing.) This
4751region is used on machines where an argument may be passed partly in
4752registers and partly in memory, and, in some cases to support the
4753features in @code{<stdarg.h>}.
4754
4755@item
4756An area of memory used to save certain registers used by the function.
4757The size of this area, which may also include space for such things as
4758the return address and pointers to previous stack frames, is
4759machine-specific and usually depends on which registers have been used
4760in the function. Machines with register windows often do not require
4761a save area.
4762
4763@item
4764A region of at least @var{size} bytes, possibly rounded up to an allocation
4765boundary, to contain the local variables of the function. On some machines,
4766this region and the save area may occur in the opposite order, with the
4767save area closer to the top of the stack.
4768
4769@item
4770@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4771Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4772@code{current_function_outgoing_args_size} bytes to be used for outgoing
4773argument lists of the function. @xref{Stack Arguments}.
4774@end itemize
4775
4776@defmac EXIT_IGNORE_STACK
4777Define this macro as a C expression that is nonzero if the return
4778instruction or the function epilogue ignores the value of the stack
4779pointer; in other words, if it is safe to delete an instruction to
4780adjust the stack pointer before a return from the function. The
4781default is 0.
4782
4783Note that this macro's value is relevant only for functions for which
4784frame pointers are maintained. It is never safe to delete a final
4785stack adjustment in a function that has no frame pointer, and the
4786compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4787@end defmac
4788
4789@defmac EPILOGUE_USES (@var{regno})
4790Define this macro as a C expression that is nonzero for registers that are
4791used by the epilogue or the @samp{return} pattern. The stack and frame
4792pointer registers are already assumed to be used as needed.
4793@end defmac
4794
4795@defmac EH_USES (@var{regno})
4796Define this macro as a C expression that is nonzero for registers that are
4797used by the exception handling mechanism, and so should be considered live
4798on entry to an exception edge.
4799@end defmac
4800
4801@defmac DELAY_SLOTS_FOR_EPILOGUE
4802Define this macro if the function epilogue contains delay slots to which
4803instructions from the rest of the function can be ``moved''. The
4804definition should be a C expression whose value is an integer
4805representing the number of delay slots there.
4806@end defmac
4807
4808@defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4809A C expression that returns 1 if @var{insn} can be placed in delay
4810slot number @var{n} of the epilogue.
4811
4812The argument @var{n} is an integer which identifies the delay slot now
4813being considered (since different slots may have different rules of
4814eligibility). It is never negative and is always less than the number
4815of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4816If you reject a particular insn for a given delay slot, in principle, it
4817may be reconsidered for a subsequent delay slot. Also, other insns may
4818(at least in principle) be considered for the so far unfilled delay
4819slot.
4820
4821@findex current_function_epilogue_delay_list
4822@findex final_scan_insn
4823The insns accepted to fill the epilogue delay slots are put in an RTL
4824list made with @code{insn_list} objects, stored in the variable
4825@code{current_function_epilogue_delay_list}. The insn for the first
4826delay slot comes first in the list. Your definition of the macro
4827@code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4828outputting the insns in this list, usually by calling
4829@code{final_scan_insn}.
4830
4831You need not define this macro if you did not define
4832@code{DELAY_SLOTS_FOR_EPILOGUE}.
4833@end defmac
4834
4835@hook TARGET_ASM_OUTPUT_MI_THUNK
4836A function that outputs the assembler code for a thunk
4837function, used to implement C++ virtual function calls with multiple
4838inheritance. The thunk acts as a wrapper around a virtual function,
4839adjusting the implicit object parameter before handing control off to
4840the real function.
4841
4842First, emit code to add the integer @var{delta} to the location that
4843contains the incoming first argument. Assume that this argument
4844contains a pointer, and is the one used to pass the @code{this} pointer
4845in C++. This is the incoming argument @emph{before} the function prologue,
4846e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4847all other incoming arguments.
4848
4849Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4850made after adding @code{delta}. In particular, if @var{p} is the
4851adjusted pointer, the following adjustment should be made:
4852
4853@smallexample
4854p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4855@end smallexample
4856
4857After the additions, emit code to jump to @var{function}, which is a
4858@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4859not touch the return address. Hence returning from @var{FUNCTION} will
4860return to whoever called the current @samp{thunk}.
4861
4862The effect must be as if @var{function} had been called directly with
4863the adjusted first argument. This macro is responsible for emitting all
4864of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4865and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4866
4867The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4868have already been extracted from it.) It might possibly be useful on
4869some targets, but probably not.
4870
4871If you do not define this macro, the target-independent code in the C++
4872front end will generate a less efficient heavyweight thunk that calls
4873@var{function} instead of jumping to it. The generic approach does
4874not support varargs.
4875@end deftypefn
4876
4877@hook TARGET_ASM_CAN_OUTPUT_MI_THUNK
4878A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4879to output the assembler code for the thunk function specified by the
4880arguments it is passed, and false otherwise. In the latter case, the
4881generic approach will be used by the C++ front end, with the limitations
4882previously exposed.
4883@end deftypefn
4884
4885@node Profiling
4886@subsection Generating Code for Profiling
4887@cindex profiling, code generation
4888
4889These macros will help you generate code for profiling.
4890
4891@defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4892A C statement or compound statement to output to @var{file} some
4893assembler code to call the profiling subroutine @code{mcount}.
4894
4895@findex mcount
4896The details of how @code{mcount} expects to be called are determined by
4897your operating system environment, not by GCC@. To figure them out,
4898compile a small program for profiling using the system's installed C
4899compiler and look at the assembler code that results.
4900
4901Older implementations of @code{mcount} expect the address of a counter
4902variable to be loaded into some register. The name of this variable is
4903@samp{LP} followed by the number @var{labelno}, so you would generate
4904the name using @samp{LP%d} in a @code{fprintf}.
4905@end defmac
4906
4907@defmac PROFILE_HOOK
4908A C statement or compound statement to output to @var{file} some assembly
4909code to call the profiling subroutine @code{mcount} even the target does
4910not support profiling.
4911@end defmac
4912
4913@defmac NO_PROFILE_COUNTERS
4914Define this macro to be an expression with a nonzero value if the
4915@code{mcount} subroutine on your system does not need a counter variable
4916allocated for each function. This is true for almost all modern
4917implementations. If you define this macro, you must not use the
4918@var{labelno} argument to @code{FUNCTION_PROFILER}.
4919@end defmac
4920
4921@defmac PROFILE_BEFORE_PROLOGUE
4922Define this macro if the code for function profiling should come before
4923the function prologue. Normally, the profiling code comes after.
4924@end defmac
4925
4926@node Tail Calls
4927@subsection Permitting tail calls
4928@cindex tail calls
4929
4930@hook TARGET_FUNCTION_OK_FOR_SIBCALL
4931True if it is ok to do sibling call optimization for the specified
4932call expression @var{exp}. @var{decl} will be the called function,
4933or @code{NULL} if this is an indirect call.
4934
4935It is not uncommon for limitations of calling conventions to prevent
4936tail calls to functions outside the current unit of translation, or
4937during PIC compilation. The hook is used to enforce these restrictions,
4938as the @code{sibcall} md pattern can not fail, or fall over to a
4939``normal'' call. The criteria for successful sibling call optimization
4940may vary greatly between different architectures.
4941@end deftypefn
4942
4943@hook TARGET_EXTRA_LIVE_ON_ENTRY
4944Add any hard registers to @var{regs} that are live on entry to the
4945function. This hook only needs to be defined to provide registers that
4946cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4947registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4948TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4949FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4950@end deftypefn
4951
4952@node Stack Smashing Protection
4953@subsection Stack smashing protection
4954@cindex stack smashing protection
4955
4956@hook TARGET_STACK_PROTECT_GUARD
4957This hook returns a @code{DECL} node for the external variable to use
4958for the stack protection guard. This variable is initialized by the
4959runtime to some random value and is used to initialize the guard value
4960that is placed at the top of the local stack frame. The type of this
4961variable must be @code{ptr_type_node}.
4962
4963The default version of this hook creates a variable called
4964@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4965@end deftypefn
4966
4967@hook TARGET_STACK_PROTECT_FAIL
4968This hook returns a tree expression that alerts the runtime that the
4969stack protect guard variable has been modified. This expression should
4970involve a call to a @code{noreturn} function.
4971
4972The default version of this hook invokes a function called
4973@samp{__stack_chk_fail}, taking no arguments. This function is
4974normally defined in @file{libgcc2.c}.
4975@end deftypefn
4976
7458026b
ILT
4977@hook TARGET_SUPPORTS_SPLIT_STACK
4978
38f8b050
JR
4979@node Varargs
4980@section Implementing the Varargs Macros
4981@cindex varargs implementation
4982
4983GCC comes with an implementation of @code{<varargs.h>} and
4984@code{<stdarg.h>} that work without change on machines that pass arguments
4985on the stack. Other machines require their own implementations of
4986varargs, and the two machine independent header files must have
4987conditionals to include it.
4988
4989ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4990the calling convention for @code{va_start}. The traditional
4991implementation takes just one argument, which is the variable in which
4992to store the argument pointer. The ISO implementation of
4993@code{va_start} takes an additional second argument. The user is
4994supposed to write the last named argument of the function here.
4995
4996However, @code{va_start} should not use this argument. The way to find
4997the end of the named arguments is with the built-in functions described
4998below.
4999
5000@defmac __builtin_saveregs ()
5001Use this built-in function to save the argument registers in memory so
5002that the varargs mechanism can access them. Both ISO and traditional
5003versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5004you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5005
5006On some machines, @code{__builtin_saveregs} is open-coded under the
5007control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5008other machines, it calls a routine written in assembler language,
5009found in @file{libgcc2.c}.
5010
5011Code generated for the call to @code{__builtin_saveregs} appears at the
5012beginning of the function, as opposed to where the call to
5013@code{__builtin_saveregs} is written, regardless of what the code is.
5014This is because the registers must be saved before the function starts
5015to use them for its own purposes.
5016@c i rewrote the first sentence above to fix an overfull hbox. --mew
5017@c 10feb93
5018@end defmac
5019
38f8b050 5020@defmac __builtin_next_arg (@var{lastarg})
c59a0a1d 5021This builtin returns the address of the first anonymous stack
38f8b050
JR
5022argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5023returns the address of the location above the first anonymous stack
5024argument. Use it in @code{va_start} to initialize the pointer for
5025fetching arguments from the stack. Also use it in @code{va_start} to
5026verify that the second parameter @var{lastarg} is the last named argument
5027of the current function.
5028@end defmac
5029
5030@defmac __builtin_classify_type (@var{object})
5031Since each machine has its own conventions for which data types are
5032passed in which kind of register, your implementation of @code{va_arg}
5033has to embody these conventions. The easiest way to categorize the
5034specified data type is to use @code{__builtin_classify_type} together
5035with @code{sizeof} and @code{__alignof__}.
5036
5037@code{__builtin_classify_type} ignores the value of @var{object},
5038considering only its data type. It returns an integer describing what
5039kind of type that is---integer, floating, pointer, structure, and so on.
5040
5041The file @file{typeclass.h} defines an enumeration that you can use to
5042interpret the values of @code{__builtin_classify_type}.
5043@end defmac
5044
5045These machine description macros help implement varargs:
5046
5047@hook TARGET_EXPAND_BUILTIN_SAVEREGS
5048If defined, this hook produces the machine-specific code for a call to
5049@code{__builtin_saveregs}. This code will be moved to the very
5050beginning of the function, before any parameter access are made. The
5051return value of this function should be an RTX that contains the value
5052to use as the return of @code{__builtin_saveregs}.
5053@end deftypefn
5054
5055@hook TARGET_SETUP_INCOMING_VARARGS
5056This target hook offers an alternative to using
5057@code{__builtin_saveregs} and defining the hook
5058@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5059register arguments into the stack so that all the arguments appear to
5060have been passed consecutively on the stack. Once this is done, you can
5061use the standard implementation of varargs that works for machines that
5062pass all their arguments on the stack.
5063
5064The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5065structure, containing the values that are obtained after processing the
5066named arguments. The arguments @var{mode} and @var{type} describe the
5067last named argument---its machine mode and its data type as a tree node.
5068
5069The target hook should do two things: first, push onto the stack all the
5070argument registers @emph{not} used for the named arguments, and second,
5071store the size of the data thus pushed into the @code{int}-valued
5072variable pointed to by @var{pretend_args_size}. The value that you
5073store here will serve as additional offset for setting up the stack
5074frame.
5075
5076Because you must generate code to push the anonymous arguments at
5077compile time without knowing their data types,
5078@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5079have just a single category of argument register and use it uniformly
5080for all data types.
5081
5082If the argument @var{second_time} is nonzero, it means that the
5083arguments of the function are being analyzed for the second time. This
5084happens for an inline function, which is not actually compiled until the
5085end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5086not generate any instructions in this case.
5087@end deftypefn
5088
5089@hook TARGET_STRICT_ARGUMENT_NAMING
5090Define this hook to return @code{true} if the location where a function
5091argument is passed depends on whether or not it is a named argument.
5092
5093This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
5094is set for varargs and stdarg functions. If this hook returns
5095@code{true}, the @var{named} argument is always true for named
5096arguments, and false for unnamed arguments. If it returns @code{false},
5097but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5098then all arguments are treated as named. Otherwise, all named arguments
5099except the last are treated as named.
5100
5101You need not define this hook if it always returns @code{false}.
5102@end deftypefn
5103
5104@hook TARGET_PRETEND_OUTGOING_VARARGS_NAMED
5105If you need to conditionally change ABIs so that one works with
5106@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5107@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5108defined, then define this hook to return @code{true} if
5109@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5110Otherwise, you should not define this hook.
5111@end deftypefn
5112
5113@node Trampolines
5114@section Trampolines for Nested Functions
5115@cindex trampolines for nested functions
5116@cindex nested functions, trampolines for
5117
5118A @dfn{trampoline} is a small piece of code that is created at run time
5119when the address of a nested function is taken. It normally resides on
5120the stack, in the stack frame of the containing function. These macros
5121tell GCC how to generate code to allocate and initialize a
5122trampoline.
5123
5124The instructions in the trampoline must do two things: load a constant
5125address into the static chain register, and jump to the real address of
5126the nested function. On CISC machines such as the m68k, this requires
5127two instructions, a move immediate and a jump. Then the two addresses
5128exist in the trampoline as word-long immediate operands. On RISC
5129machines, it is often necessary to load each address into a register in
5130two parts. Then pieces of each address form separate immediate
5131operands.
5132
5133The code generated to initialize the trampoline must store the variable
5134parts---the static chain value and the function address---into the
5135immediate operands of the instructions. On a CISC machine, this is
5136simply a matter of copying each address to a memory reference at the
5137proper offset from the start of the trampoline. On a RISC machine, it
5138may be necessary to take out pieces of the address and store them
5139separately.
5140
5141@hook TARGET_ASM_TRAMPOLINE_TEMPLATE
5142This hook is called by @code{assemble_trampoline_template} to output,
5143on the stream @var{f}, assembler code for a block of data that contains
5144the constant parts of a trampoline. This code should not include a
5145label---the label is taken care of automatically.
5146
5147If you do not define this hook, it means no template is needed
5148for the target. Do not define this hook on systems where the block move
5149code to copy the trampoline into place would be larger than the code
5150to generate it on the spot.
5151@end deftypefn
5152
5153@defmac TRAMPOLINE_SECTION
5154Return the section into which the trampoline template is to be placed
5155(@pxref{Sections}). The default value is @code{readonly_data_section}.
5156@end defmac
5157
5158@defmac TRAMPOLINE_SIZE
5159A C expression for the size in bytes of the trampoline, as an integer.
5160@end defmac
5161
5162@defmac TRAMPOLINE_ALIGNMENT
5163Alignment required for trampolines, in bits.
5164
5165If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5166is used for aligning trampolines.
5167@end defmac
5168
5169@hook TARGET_TRAMPOLINE_INIT
5170This hook is called to initialize a trampoline.
5171@var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5172is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5173RTX for the static chain value that should be passed to the function
5174when it is called.
5175
5176If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5177first thing this hook should do is emit a block move into @var{m_tramp}
5178from the memory block returned by @code{assemble_trampoline_template}.
5179Note that the block move need only cover the constant parts of the
5180trampoline. If the target isolates the variable parts of the trampoline
5181to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5182
5183If the target requires any other actions, such as flushing caches or
5184enabling stack execution, these actions should be performed after
5185initializing the trampoline proper.
5186@end deftypefn
5187
5188@hook TARGET_TRAMPOLINE_ADJUST_ADDRESS
5189This hook should perform any machine-specific adjustment in
5190the address of the trampoline. Its argument contains the address of the
5191memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5192the address to be used for a function call should be different from the
5193address at which the template was stored, the different address should
5194be returned; otherwise @var{addr} should be returned unchanged.
5195If this hook is not defined, @var{addr} will be used for function calls.
5196@end deftypefn
5197
5198Implementing trampolines is difficult on many machines because they have
5199separate instruction and data caches. Writing into a stack location
5200fails to clear the memory in the instruction cache, so when the program
5201jumps to that location, it executes the old contents.
5202
5203Here are two possible solutions. One is to clear the relevant parts of
5204the instruction cache whenever a trampoline is set up. The other is to
5205make all trampolines identical, by having them jump to a standard
5206subroutine. The former technique makes trampoline execution faster; the
5207latter makes initialization faster.
5208
5209To clear the instruction cache when a trampoline is initialized, define
5210the following macro.
5211
5212@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5213If defined, expands to a C expression clearing the @emph{instruction
5214cache} in the specified interval. The definition of this macro would
5215typically be a series of @code{asm} statements. Both @var{beg} and
5216@var{end} are both pointer expressions.
5217@end defmac
5218
5219The operating system may also require the stack to be made executable
5220before calling the trampoline. To implement this requirement, define
5221the following macro.
5222
5223@defmac ENABLE_EXECUTE_STACK
5224Define this macro if certain operations must be performed before executing
5225code located on the stack. The macro should expand to a series of C
5226file-scope constructs (e.g.@: functions) and provide a unique entry point
5227named @code{__enable_execute_stack}. The target is responsible for
5228emitting calls to the entry point in the code, for example from the
5229@code{TARGET_TRAMPOLINE_INIT} hook.
5230@end defmac
5231
5232To use a standard subroutine, define the following macro. In addition,
5233you must make sure that the instructions in a trampoline fill an entire
5234cache line with identical instructions, or else ensure that the
5235beginning of the trampoline code is always aligned at the same point in
5236its cache line. Look in @file{m68k.h} as a guide.
5237
5238@defmac TRANSFER_FROM_TRAMPOLINE
5239Define this macro if trampolines need a special subroutine to do their
5240work. The macro should expand to a series of @code{asm} statements
5241which will be compiled with GCC@. They go in a library function named
5242@code{__transfer_from_trampoline}.
5243
5244If you need to avoid executing the ordinary prologue code of a compiled
5245C function when you jump to the subroutine, you can do so by placing a
5246special label of your own in the assembler code. Use one @code{asm}
5247statement to generate an assembler label, and another to make the label
5248global. Then trampolines can use that label to jump directly to your
5249special assembler code.
5250@end defmac
5251
5252@node Library Calls
5253@section Implicit Calls to Library Routines
5254@cindex library subroutine names
5255@cindex @file{libgcc.a}
5256
5257@c prevent bad page break with this line
5258Here is an explanation of implicit calls to library routines.
5259
5260@defmac DECLARE_LIBRARY_RENAMES
5261This macro, if defined, should expand to a piece of C code that will get
5262expanded when compiling functions for libgcc.a. It can be used to
5263provide alternate names for GCC's internal library functions if there
5264are ABI-mandated names that the compiler should provide.
5265@end defmac
5266
5267@findex set_optab_libfunc
5268@findex init_one_libfunc
5269@hook TARGET_INIT_LIBFUNCS
5270This hook should declare additional library routines or rename
5271existing ones, using the functions @code{set_optab_libfunc} and
5272@code{init_one_libfunc} defined in @file{optabs.c}.
5273@code{init_optabs} calls this macro after initializing all the normal
5274library routines.
5275
5276The default is to do nothing. Most ports don't need to define this hook.
5277@end deftypefn
5278
5279@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5280This macro should return @code{true} if the library routine that
5281implements the floating point comparison operator @var{comparison} in
5282mode @var{mode} will return a boolean, and @var{false} if it will
5283return a tristate.
5284
5285GCC's own floating point libraries return tristates from the
5286comparison operators, so the default returns false always. Most ports
5287don't need to define this macro.
5288@end defmac
5289
5290@defmac TARGET_LIB_INT_CMP_BIASED
5291This macro should evaluate to @code{true} if the integer comparison
5292functions (like @code{__cmpdi2}) return 0 to indicate that the first
5293operand is smaller than the second, 1 to indicate that they are equal,
5294and 2 to indicate that the first operand is greater than the second.
5295If this macro evaluates to @code{false} the comparison functions return
5296@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5297in @file{libgcc.a}, you do not need to define this macro.
5298@end defmac
5299
5300@cindex US Software GOFAST, floating point emulation library
5301@cindex floating point emulation library, US Software GOFAST
5302@cindex GOFAST, floating point emulation library
5303@findex gofast_maybe_init_libfuncs
5304@defmac US_SOFTWARE_GOFAST
5305Define this macro if your system C library uses the US Software GOFAST
5306library to provide floating point emulation.
5307
5308In addition to defining this macro, your architecture must set
5309@code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
5310else call that function from its version of that hook. It is defined
5311in @file{config/gofast.h}, which must be included by your
5312architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
5313an example.
5314
5315If this macro is defined, the
5316@code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
5317false for @code{SFmode} and @code{DFmode} comparisons.
5318@end defmac
5319
5320@cindex @code{EDOM}, implicit usage
5321@findex matherr
5322@defmac TARGET_EDOM
5323The value of @code{EDOM} on the target machine, as a C integer constant
5324expression. If you don't define this macro, GCC does not attempt to
5325deposit the value of @code{EDOM} into @code{errno} directly. Look in
5326@file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5327system.
5328
5329If you do not define @code{TARGET_EDOM}, then compiled code reports
5330domain errors by calling the library function and letting it report the
5331error. If mathematical functions on your system use @code{matherr} when
5332there is an error, then you should leave @code{TARGET_EDOM} undefined so
5333that @code{matherr} is used normally.
5334@end defmac
5335
5336@cindex @code{errno}, implicit usage
5337@defmac GEN_ERRNO_RTX
5338Define this macro as a C expression to create an rtl expression that
5339refers to the global ``variable'' @code{errno}. (On certain systems,
5340@code{errno} may not actually be a variable.) If you don't define this
5341macro, a reasonable default is used.
5342@end defmac
5343
5344@cindex C99 math functions, implicit usage
5345@defmac TARGET_C99_FUNCTIONS
5346When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5347@code{sinf} and similarly for other functions defined by C99 standard. The
5348default is zero because a number of existing systems lack support for these
5349functions in their runtime so this macro needs to be redefined to one on
5350systems that do support the C99 runtime.
5351@end defmac
5352
5353@cindex sincos math function, implicit usage
5354@defmac TARGET_HAS_SINCOS
5355When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5356and @code{cos} with the same argument to a call to @code{sincos}. The
5357default is zero. The target has to provide the following functions:
5358@smallexample
5359void sincos(double x, double *sin, double *cos);
5360void sincosf(float x, float *sin, float *cos);
5361void sincosl(long double x, long double *sin, long double *cos);
5362@end smallexample
5363@end defmac
5364
5365@defmac NEXT_OBJC_RUNTIME
5366Define this macro to generate code for Objective-C message sending using
5367the calling convention of the NeXT system. This calling convention
5368involves passing the object, the selector and the method arguments all
5369at once to the method-lookup library function.
5370
5371The default calling convention passes just the object and the selector
5372to the lookup function, which returns a pointer to the method.
5373@end defmac
5374
5375@node Addressing Modes
5376@section Addressing Modes
5377@cindex addressing modes
5378
5379@c prevent bad page break with this line
5380This is about addressing modes.
5381
5382@defmac HAVE_PRE_INCREMENT
5383@defmacx HAVE_PRE_DECREMENT
5384@defmacx HAVE_POST_INCREMENT
5385@defmacx HAVE_POST_DECREMENT
5386A C expression that is nonzero if the machine supports pre-increment,
5387pre-decrement, post-increment, or post-decrement addressing respectively.
5388@end defmac
5389
5390@defmac HAVE_PRE_MODIFY_DISP
5391@defmacx HAVE_POST_MODIFY_DISP
5392A C expression that is nonzero if the machine supports pre- or
5393post-address side-effect generation involving constants other than
5394the size of the memory operand.
5395@end defmac
5396
5397@defmac HAVE_PRE_MODIFY_REG
5398@defmacx HAVE_POST_MODIFY_REG
5399A C expression that is nonzero if the machine supports pre- or
5400post-address side-effect generation involving a register displacement.
5401@end defmac
5402
5403@defmac CONSTANT_ADDRESS_P (@var{x})
5404A C expression that is 1 if the RTX @var{x} is a constant which
5405is a valid address. On most machines the default definition of
5406@code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5407is acceptable, but a few machines are more restrictive as to which
5408constant addresses are supported.
5409@end defmac
5410
5411@defmac CONSTANT_P (@var{x})
5412@code{CONSTANT_P}, which is defined by target-independent code,
5413accepts integer-values expressions whose values are not explicitly
5414known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5415expressions and @code{const} arithmetic expressions, in addition to
5416@code{const_int} and @code{const_double} expressions.
5417@end defmac
5418
5419@defmac MAX_REGS_PER_ADDRESS
5420A number, the maximum number of registers that can appear in a valid
5421memory address. Note that it is up to you to specify a value equal to
5422the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5423accept.
5424@end defmac
5425
5426@hook TARGET_LEGITIMATE_ADDRESS_P
5427A function that returns whether @var{x} (an RTX) is a legitimate memory
5428address on the target machine for a memory operand of mode @var{mode}.
5429
5430Legitimate addresses are defined in two variants: a strict variant and a
5431non-strict one. The @var{strict} parameter chooses which variant is
5432desired by the caller.
5433
5434The strict variant is used in the reload pass. It must be defined so
5435that any pseudo-register that has not been allocated a hard register is
5436considered a memory reference. This is because in contexts where some
5437kind of register is required, a pseudo-register with no hard register
5438must be rejected. For non-hard registers, the strict variant should look
5439up the @code{reg_renumber} array; it should then proceed using the hard
5440register number in the array, or treat the pseudo as a memory reference
5441if the array holds @code{-1}.
5442
5443The non-strict variant is used in other passes. It must be defined to
5444accept all pseudo-registers in every context where some kind of
5445register is required.
5446
5447Normally, constant addresses which are the sum of a @code{symbol_ref}
5448and an integer are stored inside a @code{const} RTX to mark them as
5449constant. Therefore, there is no need to recognize such sums
5450specifically as legitimate addresses. Normally you would simply
5451recognize any @code{const} as legitimate.
5452
5453Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5454sums that are not marked with @code{const}. It assumes that a naked
5455@code{plus} indicates indexing. If so, then you @emph{must} reject such
5456naked constant sums as illegitimate addresses, so that none of them will
5457be given to @code{PRINT_OPERAND_ADDRESS}.
5458
5459@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5460On some machines, whether a symbolic address is legitimate depends on
5461the section that the address refers to. On these machines, define the
5462target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5463into the @code{symbol_ref}, and then check for it here. When you see a
5464@code{const}, you will have to look inside it to find the
5465@code{symbol_ref} in order to determine the section. @xref{Assembler
5466Format}.
5467
5468@cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5469Some ports are still using a deprecated legacy substitute for
5470this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5471has this syntax:
5472
5473@example
5474#define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5475@end example
5476
5477@noindent
5478and should @code{goto @var{label}} if the address @var{x} is a valid
5479address on the target machine for a memory operand of mode @var{mode}.
5480Whether the strict or non-strict variants are desired is defined by
5481the @code{REG_OK_STRICT} macro introduced earlier in this section.
5482Using the hook is usually simpler because it limits the number of
5483files that are recompiled when changes are made.
5484@end deftypefn
5485
5486@defmac TARGET_MEM_CONSTRAINT
5487A single character to be used instead of the default @code{'m'}
5488character for general memory addresses. This defines the constraint
5489letter which matches the memory addresses accepted by
5490@code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5491support new address formats in your back end without changing the
5492semantics of the @code{'m'} constraint. This is necessary in order to
5493preserve functionality of inline assembly constructs using the
5494@code{'m'} constraint.
5495@end defmac
5496
5497@defmac FIND_BASE_TERM (@var{x})
5498A C expression to determine the base term of address @var{x},
5499or to provide a simplified version of @var{x} from which @file{alias.c}
5500can easily find the base term. This macro is used in only two places:
5501@code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5502
5503It is always safe for this macro to not be defined. It exists so
5504that alias analysis can understand machine-dependent addresses.
5505
5506The typical use of this macro is to handle addresses containing
5507a label_ref or symbol_ref within an UNSPEC@.
5508@end defmac
5509
5510@hook TARGET_LEGITIMIZE_ADDRESS
5511This hook is given an invalid memory address @var{x} for an
5512operand of mode @var{mode} and should try to return a valid memory
5513address.
5514
5515@findex break_out_memory_refs
5516@var{x} will always be the result of a call to @code{break_out_memory_refs},
5517and @var{oldx} will be the operand that was given to that function to produce
5518@var{x}.
5519
5520The code of the hook should not alter the substructure of
5521@var{x}. If it transforms @var{x} into a more legitimate form, it
5522should return the new @var{x}.
5523
5524It is not necessary for this hook to come up with a legitimate address.
5525The compiler has standard ways of doing so in all cases. In fact, it
5526is safe to omit this hook or make it return @var{x} if it cannot find
5527a valid way to legitimize the address. But often a machine-dependent
5528strategy can generate better code.
5529@end deftypefn
5530
5531@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5532A C compound statement that attempts to replace @var{x}, which is an address
5533that needs reloading, with a valid memory address for an operand of mode
5534@var{mode}. @var{win} will be a C statement label elsewhere in the code.
5535It is not necessary to define this macro, but it might be useful for
5536performance reasons.
5537
5538For example, on the i386, it is sometimes possible to use a single
5539reload register instead of two by reloading a sum of two pseudo
5540registers into a register. On the other hand, for number of RISC
5541processors offsets are limited so that often an intermediate address
5542needs to be generated in order to address a stack slot. By defining
5543@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5544generated for adjacent some stack slots can be made identical, and thus
5545be shared.
5546
5547@emph{Note}: This macro should be used with caution. It is necessary
5548to know something of how reload works in order to effectively use this,
5549and it is quite easy to produce macros that build in too much knowledge
5550of reload internals.
5551
5552@emph{Note}: This macro must be able to reload an address created by a
5553previous invocation of this macro. If it fails to handle such addresses
5554then the compiler may generate incorrect code or abort.
5555
5556@findex push_reload
5557The macro definition should use @code{push_reload} to indicate parts that
5558need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5559suitable to be passed unaltered to @code{push_reload}.
5560
5561The code generated by this macro must not alter the substructure of
5562@var{x}. If it transforms @var{x} into a more legitimate form, it
5563should assign @var{x} (which will always be a C variable) a new value.
5564This also applies to parts that you change indirectly by calling
5565@code{push_reload}.
5566
5567@findex strict_memory_address_p
5568The macro definition may use @code{strict_memory_address_p} to test if
5569the address has become legitimate.
5570
5571@findex copy_rtx
5572If you want to change only a part of @var{x}, one standard way of doing
5573this is to use @code{copy_rtx}. Note, however, that it unshares only a
5574single level of rtl. Thus, if the part to be changed is not at the
5575top level, you'll need to replace first the top level.
5576It is not necessary for this macro to come up with a legitimate
5577address; but often a machine-dependent strategy can generate better code.
5578@end defmac
5579
5580@hook TARGET_MODE_DEPENDENT_ADDRESS_P
5581This hook returns @code{true} if memory address @var{addr} can have
5582different meanings depending on the machine mode of the memory
5583reference it is used for or if the address is valid for some modes
5584but not others.
5585
5586Autoincrement and autodecrement addresses typically have mode-dependent
5587effects because the amount of the increment or decrement is the size
5588of the operand being addressed. Some machines have other mode-dependent
5589addresses. Many RISC machines have no mode-dependent addresses.
5590
5591You may assume that @var{addr} is a valid address for the machine.
5592
5593The default version of this hook returns @code{false}.
5594@end deftypefn
5595
5596@defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5597A C statement or compound statement with a conditional @code{goto
5598@var{label};} executed if memory address @var{x} (an RTX) can have
5599different meanings depending on the machine mode of the memory
5600reference it is used for or if the address is valid for some modes
5601but not others.
5602
5603Autoincrement and autodecrement addresses typically have mode-dependent
5604effects because the amount of the increment or decrement is the size
5605of the operand being addressed. Some machines have other mode-dependent
5606addresses. Many RISC machines have no mode-dependent addresses.
5607
5608You may assume that @var{addr} is a valid address for the machine.
5609
5610These are obsolete macros, replaced by the
5611@code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook.
5612@end defmac
5613
5614@defmac LEGITIMATE_CONSTANT_P (@var{x})
5615A C expression that is nonzero if @var{x} is a legitimate constant for
5616an immediate operand on the target machine. You can assume that
5617@var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5618@samp{1} is a suitable definition for this macro on machines where
5619anything @code{CONSTANT_P} is valid.
5620@end defmac
5621
5622@hook TARGET_DELEGITIMIZE_ADDRESS
5623This hook is used to undo the possibly obfuscating effects of the
5624@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5625macros. Some backend implementations of these macros wrap symbol
5626references inside an @code{UNSPEC} rtx to represent PIC or similar
5627addressing modes. This target hook allows GCC's optimizers to understand
5628the semantics of these opaque @code{UNSPEC}s by converting them back
5629into their original form.
5630@end deftypefn
5631
5632@hook TARGET_CANNOT_FORCE_CONST_MEM
5633This hook should return true if @var{x} is of a form that cannot (or
5634should not) be spilled to the constant pool. The default version of
5635this hook returns false.
5636
5637The primary reason to define this hook is to prevent reload from
5638deciding that a non-legitimate constant would be better reloaded
5639from the constant pool instead of spilling and reloading a register
5640holding the constant. This restriction is often true of addresses
5641of TLS symbols for various targets.
5642@end deftypefn
5643
5644@hook TARGET_USE_BLOCKS_FOR_CONSTANT_P
5645This hook should return true if pool entries for constant @var{x} can
5646be placed in an @code{object_block} structure. @var{mode} is the mode
5647of @var{x}.
5648
5649The default version returns false for all constants.
5650@end deftypefn
5651
89356d17 5652@hook TARGET_BUILTIN_RECIPROCAL
38f8b050
JR
5653This hook should return the DECL of a function that implements reciprocal of
5654the builtin function with builtin function code @var{fn}, or
5655@code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5656when @var{fn} is a code of a machine-dependent builtin function. When
5657@var{sqrt} is true, additional optimizations that apply only to the reciprocal
5658of a square root function are performed, and only reciprocals of @code{sqrt}
5659function are valid.
5660@end deftypefn
5661
5662@hook TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD
5663This hook should return the DECL of a function @var{f} that given an
5664address @var{addr} as an argument returns a mask @var{m} that can be
5665used to extract from two vectors the relevant data that resides in
5666@var{addr} in case @var{addr} is not properly aligned.
5667
5668The autovectorizer, when vectorizing a load operation from an address
5669@var{addr} that may be unaligned, will generate two vector loads from
5670the two aligned addresses around @var{addr}. It then generates a
5671@code{REALIGN_LOAD} operation to extract the relevant data from the
5672two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5673@var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5674the third argument, @var{OFF}, defines how the data will be extracted
5675from these two vectors: if @var{OFF} is 0, then the returned vector is
5676@var{v2}; otherwise, the returned vector is composed from the last
5677@var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5678@var{OFF} elements of @var{v2}.
5679
5680If this hook is defined, the autovectorizer will generate a call
5681to @var{f} (using the DECL tree that this hook returns) and will
5682use the return value of @var{f} as the argument @var{OFF} to
5683@code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5684should comply with the semantics expected by @code{REALIGN_LOAD}
5685described above.
5686If this hook is not defined, then @var{addr} will be used as
5687the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5688log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5689@end deftypefn
5690
5691@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN
5692This hook should return the DECL of a function @var{f} that implements
5693widening multiplication of the even elements of two input vectors of type @var{x}.
5694
5695If this hook is defined, the autovectorizer will use it along with the
5696@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5697widening multiplication in cases that the order of the results does not have to be
5698preserved (e.g.@: used only by a reduction computation). Otherwise, the
5699@code{widen_mult_hi/lo} idioms will be used.
5700@end deftypefn
5701
5702@hook TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD
5703This hook should return the DECL of a function @var{f} that implements
5704widening multiplication of the odd elements of two input vectors of type @var{x}.
5705
5706If this hook is defined, the autovectorizer will use it along with the
5707@code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5708widening multiplication in cases that the order of the results does not have to be
5709preserved (e.g.@: used only by a reduction computation). Otherwise, the
5710@code{widen_mult_hi/lo} idioms will be used.
5711@end deftypefn
5712
5713@hook TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
5714Returns cost of different scalar or vector statements for vectorization cost model.
720f5239
IR
5715For vector memory operations the cost may depend on type (@var{vectype}) and
5716misalignment value (@var{misalign}).
38f8b050
JR
5717@end deftypefn
5718
5719@hook TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE
5720Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5721@end deftypefn
5722
5723@hook TARGET_VECTORIZE_BUILTIN_VEC_PERM
5724Target builtin that implements vector permute.
5725@end deftypefn
5726
5727@hook TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK
5728Return true if a vector created for @code{builtin_vec_perm} is valid.
5729@end deftypefn
5730
5731@hook TARGET_VECTORIZE_BUILTIN_CONVERSION
5732This hook should return the DECL of a function that implements conversion of the
5733input vector of type @var{src_type} to type @var{dest_type}.
5734The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5735specifies how the conversion is to be applied
5736(truncation, rounding, etc.).
5737
5738If this hook is defined, the autovectorizer will use the
5739@code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5740conversion. Otherwise, it will return @code{NULL_TREE}.
5741@end deftypefn
5742
5743@hook TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION
5744This hook should return the decl of a function that implements the
5745vectorized variant of the builtin function with builtin function code
5746@var{code} or @code{NULL_TREE} if such a function is not available.
5747The value of @var{fndecl} is the builtin function declaration. The
5748return type of the vectorized function shall be of vector type
5749@var{vec_type_out} and the argument types should be @var{vec_type_in}.
5750@end deftypefn
5751
5752@hook TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT
5753This hook should return true if the target supports misaligned vector
5754store/load of a specific factor denoted in the @var{misalignment}
5755parameter. The vector store/load should be of machine mode @var{mode} and
5756the elements in the vectors should be of type @var{type}. @var{is_packed}
5757parameter is true if the memory access is defined in a packed struct.
5758@end deftypefn
5759
cc4b5170
RG
5760@hook TARGET_VECTORIZE_PREFERRED_SIMD_MODE
5761This hook should return the preferred mode for vectorizing scalar
5762mode @var{mode}. The default is
5763equal to @code{word_mode}, because the vectorizer can do some
26983c22
L
5764transformations even in absence of specialized @acronym{SIMD} hardware.
5765@end deftypefn
5766
767f865f
RG
5767@hook TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES
5768This hook should return a mask of sizes that should be iterated over
5769after trying to autovectorize using the vector size derived from the
5770mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5771The default is zero which means to not iterate over other vector sizes.
5772@end deftypefn
5773
38f8b050
JR
5774@node Anchored Addresses
5775@section Anchored Addresses
5776@cindex anchored addresses
5777@cindex @option{-fsection-anchors}
5778
5779GCC usually addresses every static object as a separate entity.
5780For example, if we have:
5781
5782@smallexample
5783static int a, b, c;
5784int foo (void) @{ return a + b + c; @}
5785@end smallexample
5786
5787the code for @code{foo} will usually calculate three separate symbolic
5788addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5789it would be better to calculate just one symbolic address and access
5790the three variables relative to it. The equivalent pseudocode would
5791be something like:
5792
5793@smallexample
5794int foo (void)
5795@{
5796 register int *xr = &x;
5797 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5798@}
5799@end smallexample
5800
5801(which isn't valid C). We refer to shared addresses like @code{x} as
5802``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5803
5804The hooks below describe the target properties that GCC needs to know
5805in order to make effective use of section anchors. It won't use
5806section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5807or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5808
5809@hook TARGET_MIN_ANCHOR_OFFSET
5810The minimum offset that should be applied to a section anchor.
5811On most targets, it should be the smallest offset that can be
5812applied to a base register while still giving a legitimate address
5813for every mode. The default value is 0.
5814@end deftypevr
5815
5816@hook TARGET_MAX_ANCHOR_OFFSET
5817Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5818offset that should be applied to section anchors. The default
5819value is 0.
5820@end deftypevr
5821
5822@hook TARGET_ASM_OUTPUT_ANCHOR
5823Write the assembly code to define section anchor @var{x}, which is a
5824@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5825The hook is called with the assembly output position set to the beginning
5826of @code{SYMBOL_REF_BLOCK (@var{x})}.
5827
5828If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5829it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5830If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5831is @code{NULL}, which disables the use of section anchors altogether.
5832@end deftypefn
5833
5834@hook TARGET_USE_ANCHORS_FOR_SYMBOL_P
5835Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5836@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5837@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5838
5839The default version is correct for most targets, but you might need to
5840intercept this hook to handle things like target-specific attributes
5841or target-specific sections.
5842@end deftypefn
5843
5844@node Condition Code
5845@section Condition Code Status
5846@cindex condition code status
5847
5848The macros in this section can be split in two families, according to the
5849two ways of representing condition codes in GCC.
5850
5851The first representation is the so called @code{(cc0)} representation
5852(@pxref{Jump Patterns}), where all instructions can have an implicit
5853clobber of the condition codes. The second is the condition code
5854register representation, which provides better schedulability for
5855architectures that do have a condition code register, but on which
5856most instructions do not affect it. The latter category includes
5857most RISC machines.
5858
5859The implicit clobbering poses a strong restriction on the placement of
5860the definition and use of the condition code, which need to be in adjacent
5861insns for machines using @code{(cc0)}. This can prevent important
5862optimizations on some machines. For example, on the IBM RS/6000, there
5863is a delay for taken branches unless the condition code register is set
5864three instructions earlier than the conditional branch. The instruction
5865scheduler cannot perform this optimization if it is not permitted to
5866separate the definition and use of the condition code register.
5867
5868For this reason, it is possible and suggested to use a register to
5869represent the condition code for new ports. If there is a specific
5870condition code register in the machine, use a hard register. If the
5871condition code or comparison result can be placed in any general register,
5872or if there are multiple condition registers, use a pseudo register.
5873Registers used to store the condition code value will usually have a mode
5874that is in class @code{MODE_CC}.
5875
5876Alternatively, you can use @code{BImode} if the comparison operator is
5877specified already in the compare instruction. In this case, you are not
5878interested in most macros in this section.
5879
5880@menu
5881* CC0 Condition Codes:: Old style representation of condition codes.
5882* MODE_CC Condition Codes:: Modern representation of condition codes.
5883* Cond. Exec. Macros:: Macros to control conditional execution.
5884@end menu
5885
5886@node CC0 Condition Codes
5887@subsection Representation of condition codes using @code{(cc0)}
5888@findex cc0
5889
5890@findex cc_status
5891The file @file{conditions.h} defines a variable @code{cc_status} to
5892describe how the condition code was computed (in case the interpretation of
5893the condition code depends on the instruction that it was set by). This
5894variable contains the RTL expressions on which the condition code is
5895currently based, and several standard flags.
5896
5897Sometimes additional machine-specific flags must be defined in the machine
5898description header file. It can also add additional machine-specific
5899information by defining @code{CC_STATUS_MDEP}.
5900
5901@defmac CC_STATUS_MDEP
5902C code for a data type which is used for declaring the @code{mdep}
5903component of @code{cc_status}. It defaults to @code{int}.
5904
5905This macro is not used on machines that do not use @code{cc0}.
5906@end defmac
5907
5908@defmac CC_STATUS_MDEP_INIT
5909A C expression to initialize the @code{mdep} field to ``empty''.
5910The default definition does nothing, since most machines don't use
5911the field anyway. If you want to use the field, you should probably
5912define this macro to initialize it.
5913
5914This macro is not used on machines that do not use @code{cc0}.
5915@end defmac
5916
5917@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5918A C compound statement to set the components of @code{cc_status}
5919appropriately for an insn @var{insn} whose body is @var{exp}. It is
5920this macro's responsibility to recognize insns that set the condition
5921code as a byproduct of other activity as well as those that explicitly
5922set @code{(cc0)}.
5923
5924This macro is not used on machines that do not use @code{cc0}.
5925
5926If there are insns that do not set the condition code but do alter
5927other machine registers, this macro must check to see whether they
5928invalidate the expressions that the condition code is recorded as
5929reflecting. For example, on the 68000, insns that store in address
5930registers do not set the condition code, which means that usually
5931@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5932insns. But suppose that the previous insn set the condition code
5933based on location @samp{a4@@(102)} and the current insn stores a new
5934value in @samp{a4}. Although the condition code is not changed by
5935this, it will no longer be true that it reflects the contents of
5936@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5937@code{cc_status} in this case to say that nothing is known about the
5938condition code value.
5939
5940The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5941with the results of peephole optimization: insns whose patterns are
5942@code{parallel} RTXs containing various @code{reg}, @code{mem} or
5943constants which are just the operands. The RTL structure of these
5944insns is not sufficient to indicate what the insns actually do. What
5945@code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5946@code{CC_STATUS_INIT}.
5947
5948A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5949that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5950@samp{cc}. This avoids having detailed information about patterns in
5951two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5952@end defmac
5953
5954@node MODE_CC Condition Codes
5955@subsection Representation of condition codes using registers
5956@findex CCmode
5957@findex MODE_CC
5958
5959@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5960On many machines, the condition code may be produced by other instructions
5961than compares, for example the branch can use directly the condition
5962code set by a subtract instruction. However, on some machines
5963when the condition code is set this way some bits (such as the overflow
5964bit) are not set in the same way as a test instruction, so that a different
5965branch instruction must be used for some conditional branches. When
5966this happens, use the machine mode of the condition code register to
5967record different formats of the condition code register. Modes can
5968also be used to record which compare instruction (e.g. a signed or an
5969unsigned comparison) produced the condition codes.
5970
5971If other modes than @code{CCmode} are required, add them to
5972@file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5973a mode given an operand of a compare. This is needed because the modes
5974have to be chosen not only during RTL generation but also, for example,
5975by instruction combination. The result of @code{SELECT_CC_MODE} should
5976be consistent with the mode used in the patterns; for example to support
5977the case of the add on the SPARC discussed above, we have the pattern
5978
5979@smallexample
5980(define_insn ""
5981 [(set (reg:CC_NOOV 0)
5982 (compare:CC_NOOV
5983 (plus:SI (match_operand:SI 0 "register_operand" "%r")
5984 (match_operand:SI 1 "arith_operand" "rI"))
5985 (const_int 0)))]
5986 ""
5987 "@dots{}")
5988@end smallexample
5989
5990@noindent
5991together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
5992for comparisons whose argument is a @code{plus}:
5993
5994@smallexample
5995#define SELECT_CC_MODE(OP,X,Y) \
5996 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5997 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5998 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5999 || GET_CODE (X) == NEG) \
6000 ? CC_NOOVmode : CCmode))
6001@end smallexample
6002
6003Another reason to use modes is to retain information on which operands
6004were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6005this section.
6006
6007You should define this macro if and only if you define extra CC modes
6008in @file{@var{machine}-modes.def}.
6009@end defmac
6010
6011@defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
6012On some machines not all possible comparisons are defined, but you can
6013convert an invalid comparison into a valid one. For example, the Alpha
6014does not have a @code{GT} comparison, but you can use an @code{LT}
6015comparison instead and swap the order of the operands.
6016
6017On such machines, define this macro to be a C statement to do any
6018required conversions. @var{code} is the initial comparison code
6019and @var{op0} and @var{op1} are the left and right operands of the
6020comparison, respectively. You should modify @var{code}, @var{op0}, and
6021@var{op1} as required.
6022
6023GCC will not assume that the comparison resulting from this macro is
6024valid but will see if the resulting insn matches a pattern in the
6025@file{md} file.
6026
6027You need not define this macro if it would never change the comparison
6028code or operands.
6029@end defmac
6030
6031@defmac REVERSIBLE_CC_MODE (@var{mode})
6032A C expression whose value is one if it is always safe to reverse a
6033comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6034can ever return @var{mode} for a floating-point inequality comparison,
6035then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6036
6037You need not define this macro if it would always returns zero or if the
6038floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6039For example, here is the definition used on the SPARC, where floating-point
6040inequality comparisons are always given @code{CCFPEmode}:
6041
6042@smallexample
6043#define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
6044@end smallexample
6045@end defmac
6046
6047@defmac REVERSE_CONDITION (@var{code}, @var{mode})
6048A C expression whose value is reversed condition code of the @var{code} for
6049comparison done in CC_MODE @var{mode}. The macro is used only in case
6050@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6051machine has some non-standard way how to reverse certain conditionals. For
6052instance in case all floating point conditions are non-trapping, compiler may
6053freely convert unordered compares to ordered one. Then definition may look
6054like:
6055
6056@smallexample
6057#define REVERSE_CONDITION(CODE, MODE) \
6058 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6059 : reverse_condition_maybe_unordered (CODE))
6060@end smallexample
6061@end defmac
6062
6063@hook TARGET_FIXED_CONDITION_CODE_REGS
6064On targets which do not use @code{(cc0)}, and which use a hard
6065register rather than a pseudo-register to hold condition codes, the
6066regular CSE passes are often not able to identify cases in which the
6067hard register is set to a common value. Use this hook to enable a
6068small pass which optimizes such cases. This hook should return true
6069to enable this pass, and it should set the integers to which its
6070arguments point to the hard register numbers used for condition codes.
6071When there is only one such register, as is true on most systems, the
6072integer pointed to by @var{p2} should be set to
6073@code{INVALID_REGNUM}.
6074
6075The default version of this hook returns false.
6076@end deftypefn
6077
6078@hook TARGET_CC_MODES_COMPATIBLE
6079On targets which use multiple condition code modes in class
6080@code{MODE_CC}, it is sometimes the case that a comparison can be
6081validly done in more than one mode. On such a system, define this
6082target hook to take two mode arguments and to return a mode in which
6083both comparisons may be validly done. If there is no such mode,
6084return @code{VOIDmode}.
6085
6086The default version of this hook checks whether the modes are the
6087same. If they are, it returns that mode. If they are different, it
6088returns @code{VOIDmode}.
6089@end deftypefn
6090
6091@node Cond. Exec. Macros
6092@subsection Macros to control conditional execution
6093@findex conditional execution
6094@findex predication
6095
6096There is one macro that may need to be defined for targets
6097supporting conditional execution, independent of how they
6098represent conditional branches.
6099
6100@defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
6101A C expression that returns true if the conditional execution predicate
6102@var{op1}, a comparison operation, is the inverse of @var{op2} and vice
6103versa. Define this to return 0 if the target has conditional execution
6104predicates that cannot be reversed safely. There is no need to validate
6105that the arguments of op1 and op2 are the same, this is done separately.
6106If no expansion is specified, this macro is defined as follows:
6107
6108@smallexample
6109#define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
6110 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
6111@end smallexample
6112@end defmac
6113
6114@node Costs
6115@section Describing Relative Costs of Operations
6116@cindex costs of instructions
6117@cindex relative costs
6118@cindex speed of instructions
6119
6120These macros let you describe the relative speed of various operations
6121on the target machine.
6122
6123@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6124A C expression for the cost of moving data of mode @var{mode} from a
6125register in class @var{from} to one in class @var{to}. The classes are
6126expressed using the enumeration values such as @code{GENERAL_REGS}. A
6127value of 2 is the default; other values are interpreted relative to
6128that.
6129
6130It is not required that the cost always equal 2 when @var{from} is the
6131same as @var{to}; on some machines it is expensive to move between
6132registers if they are not general registers.
6133
6134If reload sees an insn consisting of a single @code{set} between two
6135hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6136classes returns a value of 2, reload does not check to ensure that the
6137constraints of the insn are met. Setting a cost of other than 2 will
6138allow reload to verify that the constraints are met. You should do this
6139if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6140
6141These macros are obsolete, new ports should use the target hook
6142@code{TARGET_REGISTER_MOVE_COST} instead.
6143@end defmac
6144
6145@hook TARGET_REGISTER_MOVE_COST
6146This target hook should return the cost of moving data of mode @var{mode}
6147from a register in class @var{from} to one in class @var{to}. The classes
6148are expressed using the enumeration values such as @code{GENERAL_REGS}.
6149A value of 2 is the default; other values are interpreted relative to
6150that.
6151
6152It is not required that the cost always equal 2 when @var{from} is the
6153same as @var{to}; on some machines it is expensive to move between
6154registers if they are not general registers.
6155
6156If reload sees an insn consisting of a single @code{set} between two
6157hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6158classes returns a value of 2, reload does not check to ensure that the
6159constraints of the insn are met. Setting a cost of other than 2 will
6160allow reload to verify that the constraints are met. You should do this
6161if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6162
6163The default version of this function returns 2.
6164@end deftypefn
6165
6166@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6167A C expression for the cost of moving data of mode @var{mode} between a
6168register of class @var{class} and memory; @var{in} is zero if the value
6169is to be written to memory, nonzero if it is to be read in. This cost
6170is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6171registers and memory is more expensive than between two registers, you
6172should define this macro to express the relative cost.
6173
6174If you do not define this macro, GCC uses a default cost of 4 plus
6175the cost of copying via a secondary reload register, if one is
6176needed. If your machine requires a secondary reload register to copy
6177between memory and a register of @var{class} but the reload mechanism is
6178more complex than copying via an intermediate, define this macro to
6179reflect the actual cost of the move.
6180
6181GCC defines the function @code{memory_move_secondary_cost} if
6182secondary reloads are needed. It computes the costs due to copying via
6183a secondary register. If your machine copies from memory using a
6184secondary register in the conventional way but the default base value of
61854 is not correct for your machine, define this macro to add some other
6186value to the result of that function. The arguments to that function
6187are the same as to this macro.
6188
6189These macros are obsolete, new ports should use the target hook
6190@code{TARGET_MEMORY_MOVE_COST} instead.
6191@end defmac
6192
911852ff 6193@hook TARGET_MEMORY_MOVE_COST
38f8b050 6194This target hook should return the cost of moving data of mode @var{mode}
911852ff 6195between a register of class @var{rclass} and memory; @var{in} is @code{false}
38f8b050
JR
6196if the value is to be written to memory, @code{true} if it is to be read in.
6197This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6198If moving between registers and memory is more expensive than between two
6199registers, you should add this target hook to express the relative cost.
6200
6201If you do not add this target hook, GCC uses a default cost of 4 plus
6202the cost of copying via a secondary reload register, if one is
6203needed. If your machine requires a secondary reload register to copy
911852ff 6204between memory and a register of @var{rclass} but the reload mechanism is
38f8b050
JR
6205more complex than copying via an intermediate, use this target hook to
6206reflect the actual cost of the move.
6207
6208GCC defines the function @code{memory_move_secondary_cost} if
6209secondary reloads are needed. It computes the costs due to copying via
6210a secondary register. If your machine copies from memory using a
6211secondary register in the conventional way but the default base value of
62124 is not correct for your machine, use this target hook to add some other
6213value to the result of that function. The arguments to that function
6214are the same as to this target hook.
6215@end deftypefn
6216
6217@defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6218A C expression for the cost of a branch instruction. A value of 1 is the
6219default; other values are interpreted relative to that. Parameter @var{speed_p}
6220is true when the branch in question should be optimized for speed. When
6221it is false, @code{BRANCH_COST} should be returning value optimal for code size
6222rather then performance considerations. @var{predictable_p} is true for well
6223predictable branches. On many architectures the @code{BRANCH_COST} can be
6224reduced then.
6225@end defmac
6226
6227Here are additional macros which do not specify precise relative costs,
6228but only that certain actions are more expensive than GCC would
6229ordinarily expect.
6230
6231@defmac SLOW_BYTE_ACCESS
6232Define this macro as a C expression which is nonzero if accessing less
6233than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6234faster than accessing a word of memory, i.e., if such access
6235require more than one instruction or if there is no difference in cost
6236between byte and (aligned) word loads.
6237
6238When this macro is not defined, the compiler will access a field by
6239finding the smallest containing object; when it is defined, a fullword
6240load will be used if alignment permits. Unless bytes accesses are
6241faster than word accesses, using word accesses is preferable since it
6242may eliminate subsequent memory access if subsequent accesses occur to
6243other fields in the same word of the structure, but to different bytes.
6244@end defmac
6245
6246@defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6247Define this macro to be the value 1 if memory accesses described by the
6248@var{mode} and @var{alignment} parameters have a cost many times greater
6249than aligned accesses, for example if they are emulated in a trap
6250handler.
6251
6252When this macro is nonzero, the compiler will act as if
6253@code{STRICT_ALIGNMENT} were nonzero when generating code for block
6254moves. This can cause significantly more instructions to be produced.
6255Therefore, do not set this macro nonzero if unaligned accesses only add a
6256cycle or two to the time for a memory access.
6257
6258If the value of this macro is always zero, it need not be defined. If
6259this macro is defined, it should produce a nonzero value when
6260@code{STRICT_ALIGNMENT} is nonzero.
6261@end defmac
6262
6263@defmac MOVE_RATIO (@var{speed})
6264The threshold of number of scalar memory-to-memory move insns, @emph{below}
6265which a sequence of insns should be generated instead of a
6266string move insn or a library call. Increasing the value will always
6267make code faster, but eventually incurs high cost in increased code size.
6268
6269Note that on machines where the corresponding move insn is a
6270@code{define_expand} that emits a sequence of insns, this macro counts
6271the number of such sequences.
6272
6273The parameter @var{speed} is true if the code is currently being
6274optimized for speed rather than size.
6275
6276If you don't define this, a reasonable default is used.
6277@end defmac
6278
6279@defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6280A C expression used to determine whether @code{move_by_pieces} will be used to
6281copy a chunk of memory, or whether some other block move mechanism
6282will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6283than @code{MOVE_RATIO}.
6284@end defmac
6285
6286@defmac MOVE_MAX_PIECES
6287A C expression used by @code{move_by_pieces} to determine the largest unit
6288a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6289@end defmac
6290
6291@defmac CLEAR_RATIO (@var{speed})
6292The threshold of number of scalar move insns, @emph{below} which a sequence
6293of insns should be generated to clear memory instead of a string clear insn
6294or a library call. Increasing the value will always make code faster, but
6295eventually incurs high cost in increased code size.
6296
6297The parameter @var{speed} is true if the code is currently being
6298optimized for speed rather than size.
6299
6300If you don't define this, a reasonable default is used.
6301@end defmac
6302
6303@defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6304A C expression used to determine whether @code{clear_by_pieces} will be used
6305to clear a chunk of memory, or whether some other block clear mechanism
6306will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6307than @code{CLEAR_RATIO}.
6308@end defmac
6309
6310@defmac SET_RATIO (@var{speed})
6311The threshold of number of scalar move insns, @emph{below} which a sequence
6312of insns should be generated to set memory to a constant value, instead of
6313a block set insn or a library call.
6314Increasing the value will always make code faster, but
6315eventually incurs high cost in increased code size.
6316
6317The parameter @var{speed} is true if the code is currently being
6318optimized for speed rather than size.
6319
6320If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6321@end defmac
6322
6323@defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6324A C expression used to determine whether @code{store_by_pieces} will be
6325used to set a chunk of memory to a constant value, or whether some
6326other mechanism will be used. Used by @code{__builtin_memset} when
6327storing values other than constant zero.
6328Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6329than @code{SET_RATIO}.
6330@end defmac
6331
6332@defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6333A C expression used to determine whether @code{store_by_pieces} will be
6334used to set a chunk of memory to a constant string value, or whether some
6335other mechanism will be used. Used by @code{__builtin_strcpy} when
6336called with a constant source string.
6337Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6338than @code{MOVE_RATIO}.
6339@end defmac
6340
6341@defmac USE_LOAD_POST_INCREMENT (@var{mode})
6342A C expression used to determine whether a load postincrement is a good
6343thing to use for a given mode. Defaults to the value of
6344@code{HAVE_POST_INCREMENT}.
6345@end defmac
6346
6347@defmac USE_LOAD_POST_DECREMENT (@var{mode})
6348A C expression used to determine whether a load postdecrement is a good
6349thing to use for a given mode. Defaults to the value of
6350@code{HAVE_POST_DECREMENT}.
6351@end defmac
6352
6353@defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6354A C expression used to determine whether a load preincrement is a good
6355thing to use for a given mode. Defaults to the value of
6356@code{HAVE_PRE_INCREMENT}.
6357@end defmac
6358
6359@defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6360A C expression used to determine whether a load predecrement is a good
6361thing to use for a given mode. Defaults to the value of
6362@code{HAVE_PRE_DECREMENT}.
6363@end defmac
6364
6365@defmac USE_STORE_POST_INCREMENT (@var{mode})
6366A C expression used to determine whether a store postincrement is a good
6367thing to use for a given mode. Defaults to the value of
6368@code{HAVE_POST_INCREMENT}.
6369@end defmac
6370
6371@defmac USE_STORE_POST_DECREMENT (@var{mode})
6372A C expression used to determine whether a store postdecrement is a good
6373thing to use for a given mode. Defaults to the value of
6374@code{HAVE_POST_DECREMENT}.
6375@end defmac
6376
6377@defmac USE_STORE_PRE_INCREMENT (@var{mode})
6378This macro is used to determine whether a store preincrement is a good
6379thing to use for a given mode. Defaults to the value of
6380@code{HAVE_PRE_INCREMENT}.
6381@end defmac
6382
6383@defmac USE_STORE_PRE_DECREMENT (@var{mode})
6384This macro is used to determine whether a store predecrement is a good
6385thing to use for a given mode. Defaults to the value of
6386@code{HAVE_PRE_DECREMENT}.
6387@end defmac
6388
6389@defmac NO_FUNCTION_CSE
6390Define this macro if it is as good or better to call a constant
6391function address than to call an address kept in a register.
6392@end defmac
6393
6394@defmac RANGE_TEST_NON_SHORT_CIRCUIT
6395Define this macro if a non-short-circuit operation produced by
6396@samp{fold_range_test ()} is optimal. This macro defaults to true if
6397@code{BRANCH_COST} is greater than or equal to the value 2.
6398@end defmac
6399
6400@hook TARGET_RTX_COSTS
6401This target hook describes the relative costs of RTL expressions.
6402
6403The cost may depend on the precise form of the expression, which is
6404available for examination in @var{x}, and the rtx code of the expression
6405in which it is contained, found in @var{outer_code}. @var{code} is the
6406expression code---redundant, since it can be obtained with
6407@code{GET_CODE (@var{x})}.
6408
6409In implementing this hook, you can use the construct
6410@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6411instructions.
6412
6413On entry to the hook, @code{*@var{total}} contains a default estimate
6414for the cost of the expression. The hook should modify this value as
6415necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6416for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6417operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6418
6419When optimizing for code size, i.e.@: when @code{speed} is
6420false, this target hook should be used to estimate the relative
6421size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6422
6423The hook returns true when all subexpressions of @var{x} have been
6424processed, and false when @code{rtx_cost} should recurse.
6425@end deftypefn
6426
6427@hook TARGET_ADDRESS_COST
6428This hook computes the cost of an addressing mode that contains
6429@var{address}. If not defined, the cost is computed from
6430the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6431
6432For most CISC machines, the default cost is a good approximation of the
6433true cost of the addressing mode. However, on RISC machines, all
6434instructions normally have the same length and execution time. Hence
6435all addresses will have equal costs.
6436
6437In cases where more than one form of an address is known, the form with
6438the lowest cost will be used. If multiple forms have the same, lowest,
6439cost, the one that is the most complex will be used.
6440
6441For example, suppose an address that is equal to the sum of a register
6442and a constant is used twice in the same basic block. When this macro
6443is not defined, the address will be computed in a register and memory
6444references will be indirect through that register. On machines where
6445the cost of the addressing mode containing the sum is no higher than
6446that of a simple indirect reference, this will produce an additional
6447instruction and possibly require an additional register. Proper
6448specification of this macro eliminates this overhead for such machines.
6449
6450This hook is never called with an invalid address.
6451
6452On machines where an address involving more than one register is as
6453cheap as an address computation involving only one register, defining
6454@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6455be live over a region of code where only one would have been if
6456@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6457should be considered in the definition of this macro. Equivalent costs
6458should probably only be given to addresses with different numbers of
6459registers on machines with lots of registers.
6460@end deftypefn
6461
6462@node Scheduling
6463@section Adjusting the Instruction Scheduler
6464
6465The instruction scheduler may need a fair amount of machine-specific
6466adjustment in order to produce good code. GCC provides several target
6467hooks for this purpose. It is usually enough to define just a few of
6468them: try the first ones in this list first.
6469
6470@hook TARGET_SCHED_ISSUE_RATE
6471This hook returns the maximum number of instructions that can ever
6472issue at the same time on the target machine. The default is one.
6473Although the insn scheduler can define itself the possibility of issue
6474an insn on the same cycle, the value can serve as an additional
6475constraint to issue insns on the same simulated processor cycle (see
6476hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6477This value must be constant over the entire compilation. If you need
6478it to vary depending on what the instructions are, you must use
6479@samp{TARGET_SCHED_VARIABLE_ISSUE}.
6480@end deftypefn
6481
6482@hook TARGET_SCHED_VARIABLE_ISSUE
6483This hook is executed by the scheduler after it has scheduled an insn
6484from the ready list. It should return the number of insns which can
6485still be issued in the current cycle. The default is
6486@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6487@code{USE}, which normally are not counted against the issue rate.
6488You should define this hook if some insns take more machine resources
6489than others, so that fewer insns can follow them in the same cycle.
6490@var{file} is either a null pointer, or a stdio stream to write any
6491debug output to. @var{verbose} is the verbose level provided by
6492@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6493was scheduled.
6494@end deftypefn
6495
6496@hook TARGET_SCHED_ADJUST_COST
6497This function corrects the value of @var{cost} based on the
6498relationship between @var{insn} and @var{dep_insn} through the
6499dependence @var{link}. It should return the new value. The default
6500is to make no adjustment to @var{cost}. This can be used for example
6501to specify to the scheduler using the traditional pipeline description
6502that an output- or anti-dependence does not incur the same cost as a
6503data-dependence. If the scheduler using the automaton based pipeline
6504description, the cost of anti-dependence is zero and the cost of
6505output-dependence is maximum of one and the difference of latency
6506times of the first and the second insns. If these values are not
6507acceptable, you could use the hook to modify them too. See also
6508@pxref{Processor pipeline description}.
6509@end deftypefn
6510
6511@hook TARGET_SCHED_ADJUST_PRIORITY
6512This hook adjusts the integer scheduling priority @var{priority} of
6513@var{insn}. It should return the new priority. Increase the priority to
6514execute @var{insn} earlier, reduce the priority to execute @var{insn}
6515later. Do not define this hook if you do not need to adjust the
6516scheduling priorities of insns.
6517@end deftypefn
6518
6519@hook TARGET_SCHED_REORDER
6520This hook is executed by the scheduler after it has scheduled the ready
6521list, to allow the machine description to reorder it (for example to
6522combine two small instructions together on @samp{VLIW} machines).
6523@var{file} is either a null pointer, or a stdio stream to write any
6524debug output to. @var{verbose} is the verbose level provided by
6525@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6526list of instructions that are ready to be scheduled. @var{n_readyp} is
6527a pointer to the number of elements in the ready list. The scheduler
6528reads the ready list in reverse order, starting with
6529@var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6530is the timer tick of the scheduler. You may modify the ready list and
6531the number of ready insns. The return value is the number of insns that
6532can issue this cycle; normally this is just @code{issue_rate}. See also
6533@samp{TARGET_SCHED_REORDER2}.
6534@end deftypefn
6535
6536@hook TARGET_SCHED_REORDER2
6537Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6538function is called whenever the scheduler starts a new cycle. This one
6539is called once per iteration over a cycle, immediately after
6540@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6541return the number of insns to be scheduled in the same cycle. Defining
6542this hook can be useful if there are frequent situations where
6543scheduling one insn causes other insns to become ready in the same
6544cycle. These other insns can then be taken into account properly.
6545@end deftypefn
6546
6547@hook TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK
6548This hook is called after evaluation forward dependencies of insns in
6549chain given by two parameter values (@var{head} and @var{tail}
6550correspondingly) but before insns scheduling of the insn chain. For
6551example, it can be used for better insn classification if it requires
6552analysis of dependencies. This hook can use backward and forward
6553dependencies of the insn scheduler because they are already
6554calculated.
6555@end deftypefn
6556
6557@hook TARGET_SCHED_INIT
6558This hook is executed by the scheduler at the beginning of each block of
6559instructions that are to be scheduled. @var{file} is either a null
6560pointer, or a stdio stream to write any debug output to. @var{verbose}
6561is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6562@var{max_ready} is the maximum number of insns in the current scheduling
6563region that can be live at the same time. This can be used to allocate
6564scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6565@end deftypefn
6566
6567@hook TARGET_SCHED_FINISH
6568This hook is executed by the scheduler at the end of each block of
6569instructions that are to be scheduled. It can be used to perform
6570cleanup of any actions done by the other scheduling hooks. @var{file}
6571is either a null pointer, or a stdio stream to write any debug output
6572to. @var{verbose} is the verbose level provided by
6573@option{-fsched-verbose-@var{n}}.
6574@end deftypefn
6575
6576@hook TARGET_SCHED_INIT_GLOBAL
6577This hook is executed by the scheduler after function level initializations.
6578@var{file} is either a null pointer, or a stdio stream to write any debug output to.
6579@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6580@var{old_max_uid} is the maximum insn uid when scheduling begins.
6581@end deftypefn
6582
6583@hook TARGET_SCHED_FINISH_GLOBAL
6584This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6585@var{file} is either a null pointer, or a stdio stream to write any debug output to.
6586@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6587@end deftypefn
6588
6589@hook TARGET_SCHED_DFA_PRE_CYCLE_INSN
6590The hook returns an RTL insn. The automaton state used in the
6591pipeline hazard recognizer is changed as if the insn were scheduled
6592when the new simulated processor cycle starts. Usage of the hook may
6593simplify the automaton pipeline description for some @acronym{VLIW}
6594processors. If the hook is defined, it is used only for the automaton
6595based pipeline description. The default is not to change the state
6596when the new simulated processor cycle starts.
6597@end deftypefn
6598
6599@hook TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN
6600The hook can be used to initialize data used by the previous hook.
6601@end deftypefn
6602
6603@hook TARGET_SCHED_DFA_POST_CYCLE_INSN
6604The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6605to changed the state as if the insn were scheduled when the new
6606simulated processor cycle finishes.
6607@end deftypefn
6608
6609@hook TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN
6610The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6611used to initialize data used by the previous hook.
6612@end deftypefn
6613
6614@hook TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE
6615The hook to notify target that the current simulated cycle is about to finish.
6616The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6617to change the state in more complicated situations - e.g., when advancing
6618state on a single insn is not enough.
6619@end deftypefn
6620
6621@hook TARGET_SCHED_DFA_POST_ADVANCE_CYCLE
6622The hook to notify target that new simulated cycle has just started.
6623The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6624to change the state in more complicated situations - e.g., when advancing
6625state on a single insn is not enough.
6626@end deftypefn
6627
6628@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
6629This hook controls better choosing an insn from the ready insn queue
6630for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6631chooses the first insn from the queue. If the hook returns a positive
6632value, an additional scheduler code tries all permutations of
6633@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6634subsequent ready insns to choose an insn whose issue will result in
6635maximal number of issued insns on the same cycle. For the
6636@acronym{VLIW} processor, the code could actually solve the problem of
6637packing simple insns into the @acronym{VLIW} insn. Of course, if the
6638rules of @acronym{VLIW} packing are described in the automaton.
6639
6640This code also could be used for superscalar @acronym{RISC}
6641processors. Let us consider a superscalar @acronym{RISC} processor
6642with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6643@var{B}, some insns can be executed only in pipelines @var{B} or
6644@var{C}, and one insn can be executed in pipeline @var{B}. The
6645processor may issue the 1st insn into @var{A} and the 2nd one into
6646@var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6647until the next cycle. If the scheduler issues the 3rd insn the first,
6648the processor could issue all 3 insns per cycle.
6649
6650Actually this code demonstrates advantages of the automaton based
6651pipeline hazard recognizer. We try quickly and easy many insn
6652schedules to choose the best one.
6653
6654The default is no multipass scheduling.
6655@end deftypefn
6656
6657@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD
6658
6659This hook controls what insns from the ready insn queue will be
6660considered for the multipass insn scheduling. If the hook returns
6661zero for @var{insn}, the insn will be not chosen to
6662be issued.
6663
6664The default is that any ready insns can be chosen to be issued.
6665@end deftypefn
6666
c06bbdf7 6667@hook TARGET_SCHED_DFA_NEW_CYCLE
38f8b050
JR
6668This hook is called by the insn scheduler before issuing @var{insn}
6669on cycle @var{clock}. If the hook returns nonzero,
6670@var{insn} is not issued on this processor cycle. Instead,
6671the processor cycle is advanced. If *@var{sort_p}
6672is zero, the insn ready queue is not sorted on the new cycle
6673start as usually. @var{dump} and @var{verbose} specify the file and
6674verbosity level to use for debugging output.
6675@var{last_clock} and @var{clock} are, respectively, the
6676processor cycle on which the previous insn has been issued,
6677and the current processor cycle.
6678@end deftypefn
6679
6680@hook TARGET_SCHED_IS_COSTLY_DEPENDENCE
6681This hook is used to define which dependences are considered costly by
6682the target, so costly that it is not advisable to schedule the insns that
6683are involved in the dependence too close to one another. The parameters
6684to this hook are as follows: The first parameter @var{_dep} is the dependence
6685being evaluated. The second parameter @var{cost} is the cost of the
6686dependence as estimated by the scheduler, and the third
6687parameter @var{distance} is the distance in cycles between the two insns.
6688The hook returns @code{true} if considering the distance between the two
6689insns the dependence between them is considered costly by the target,
6690and @code{false} otherwise.
6691
6692Defining this hook can be useful in multiple-issue out-of-order machines,
6693where (a) it's practically hopeless to predict the actual data/resource
6694delays, however: (b) there's a better chance to predict the actual grouping
6695that will be formed, and (c) correctly emulating the grouping can be very
6696important. In such targets one may want to allow issuing dependent insns
6697closer to one another---i.e., closer than the dependence distance; however,
6698not in cases of ``costly dependences'', which this hooks allows to define.
6699@end deftypefn
6700
6701@hook TARGET_SCHED_H_I_D_EXTENDED
6702This hook is called by the insn scheduler after emitting a new instruction to
6703the instruction stream. The hook notifies a target backend to extend its
6704per instruction data structures.
6705@end deftypefn
6706
6707@hook TARGET_SCHED_ALLOC_SCHED_CONTEXT
6708Return a pointer to a store large enough to hold target scheduling context.
6709@end deftypefn
6710
6711@hook TARGET_SCHED_INIT_SCHED_CONTEXT
6712Initialize store pointed to by @var{tc} to hold target scheduling context.
6713It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6714beginning of the block. Otherwise, copy the current context into @var{tc}.
6715@end deftypefn
6716
6717@hook TARGET_SCHED_SET_SCHED_CONTEXT
6718Copy target scheduling context pointed to by @var{tc} to the current context.
6719@end deftypefn
6720
6721@hook TARGET_SCHED_CLEAR_SCHED_CONTEXT
6722Deallocate internal data in target scheduling context pointed to by @var{tc}.
6723@end deftypefn
6724
6725@hook TARGET_SCHED_FREE_SCHED_CONTEXT
6726Deallocate a store for target scheduling context pointed to by @var{tc}.
6727@end deftypefn
6728
6729@hook TARGET_SCHED_SPECULATE_INSN
6730This hook is called by the insn scheduler when @var{insn} has only
6731speculative dependencies and therefore can be scheduled speculatively.
6732The hook is used to check if the pattern of @var{insn} has a speculative
6733version and, in case of successful check, to generate that speculative
6734pattern. The hook should return 1, if the instruction has a speculative form,
6735or @minus{}1, if it doesn't. @var{request} describes the type of requested
6736speculation. If the return value equals 1 then @var{new_pat} is assigned
6737the generated speculative pattern.
6738@end deftypefn
6739
6740@hook TARGET_SCHED_NEEDS_BLOCK_P
6741This hook is called by the insn scheduler during generation of recovery code
6742for @var{insn}. It should return @code{true}, if the corresponding check
6743instruction should branch to recovery code, or @code{false} otherwise.
6744@end deftypefn
6745
6746@hook TARGET_SCHED_GEN_SPEC_CHECK
6747This hook is called by the insn scheduler to generate a pattern for recovery
6748check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6749speculative instruction for which the check should be generated.
6750@var{label} is either a label of a basic block, where recovery code should
6751be emitted, or a null pointer, when requested check doesn't branch to
6752recovery code (a simple check). If @var{mutate_p} is nonzero, then
6753a pattern for a branchy check corresponding to a simple check denoted by
6754@var{insn} should be generated. In this case @var{label} can't be null.
6755@end deftypefn
6756
6757@hook TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC
6758This hook is used as a workaround for
6759@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6760called on the first instruction of the ready list. The hook is used to
6761discard speculative instructions that stand first in the ready list from
6762being scheduled on the current cycle. If the hook returns @code{false},
6763@var{insn} will not be chosen to be issued.
6764For non-speculative instructions,
6765the hook should always return @code{true}. For example, in the ia64 backend
6766the hook is used to cancel data speculative insns when the ALAT table
6767is nearly full.
6768@end deftypefn
6769
6770@hook TARGET_SCHED_SET_SCHED_FLAGS
6771This hook is used by the insn scheduler to find out what features should be
6772enabled/used.
6773The structure *@var{spec_info} should be filled in by the target.
6774The structure describes speculation types that can be used in the scheduler.
6775@end deftypefn
6776
6777@hook TARGET_SCHED_SMS_RES_MII
6778This hook is called by the swing modulo scheduler to calculate a
6779resource-based lower bound which is based on the resources available in
6780the machine and the resources required by each instruction. The target
6781backend can use @var{g} to calculate such bound. A very simple lower
6782bound will be used in case this hook is not implemented: the total number
6783of instructions divided by the issue rate.
6784@end deftypefn
6785
7942e47e
RY
6786@hook TARGET_SCHED_DISPATCH
6787This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6788is supported in hardware and the condition specified in the parameter is true.
6789@end deftypefn
6790
6791@hook TARGET_SCHED_DISPATCH_DO
6792This hook is called by Haifa Scheduler. It performs the operation specified
6793in its second parameter.
6794@end deftypefn
6795
38f8b050
JR
6796@node Sections
6797@section Dividing the Output into Sections (Texts, Data, @dots{})
6798@c the above section title is WAY too long. maybe cut the part between
6799@c the (...)? --mew 10feb93
6800
6801An object file is divided into sections containing different types of
6802data. In the most common case, there are three sections: the @dfn{text
6803section}, which holds instructions and read-only data; the @dfn{data
6804section}, which holds initialized writable data; and the @dfn{bss
6805section}, which holds uninitialized data. Some systems have other kinds
6806of sections.
6807
6808@file{varasm.c} provides several well-known sections, such as
6809@code{text_section}, @code{data_section} and @code{bss_section}.
6810The normal way of controlling a @code{@var{foo}_section} variable
6811is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6812as described below. The macros are only read once, when @file{varasm.c}
6813initializes itself, so their values must be run-time constants.
6814They may however depend on command-line flags.
6815
6816@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6817use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6818to be string literals.
6819
6820Some assemblers require a different string to be written every time a
6821section is selected. If your assembler falls into this category, you
6822should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6823@code{get_unnamed_section} to set up the sections.
6824
6825You must always create a @code{text_section}, either by defining
6826@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6827in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6828@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6829create a distinct @code{readonly_data_section}, the default is to
6830reuse @code{text_section}.
6831
6832All the other @file{varasm.c} sections are optional, and are null
6833if the target does not provide them.
6834
6835@defmac TEXT_SECTION_ASM_OP
6836A C expression whose value is a string, including spacing, containing the
6837assembler operation that should precede instructions and read-only data.
6838Normally @code{"\t.text"} is right.
6839@end defmac
6840
6841@defmac HOT_TEXT_SECTION_NAME
6842If defined, a C string constant for the name of the section containing most
6843frequently executed functions of the program. If not defined, GCC will provide
6844a default definition if the target supports named sections.
6845@end defmac
6846
6847@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6848If defined, a C string constant for the name of the section containing unlikely
6849executed functions in the program.
6850@end defmac
6851
6852@defmac DATA_SECTION_ASM_OP
6853A C expression whose value is a string, including spacing, containing the
6854assembler operation to identify the following data as writable initialized
6855data. Normally @code{"\t.data"} is right.
6856@end defmac
6857
6858@defmac SDATA_SECTION_ASM_OP
6859If defined, a C expression whose value is a string, including spacing,
6860containing the assembler operation to identify the following data as
6861initialized, writable small data.
6862@end defmac
6863
6864@defmac READONLY_DATA_SECTION_ASM_OP
6865A C expression whose value is a string, including spacing, containing the
6866assembler operation to identify the following data as read-only initialized
6867data.
6868@end defmac
6869
6870@defmac BSS_SECTION_ASM_OP
6871If defined, a C expression whose value is a string, including spacing,
6872containing the assembler operation to identify the following data as
6873uninitialized global data. If not defined, and neither
6874@code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6875uninitialized global data will be output in the data section if
6876@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6877used.
6878@end defmac
6879
6880@defmac SBSS_SECTION_ASM_OP
6881If defined, a C expression whose value is a string, including spacing,
6882containing the assembler operation to identify the following data as
6883uninitialized, writable small data.
6884@end defmac
6885
6886@defmac TLS_COMMON_ASM_OP
6887If defined, a C expression whose value is a string containing the
6888assembler operation to identify the following data as thread-local
6889common data. The default is @code{".tls_common"}.
6890@end defmac
6891
6892@defmac TLS_SECTION_ASM_FLAG
6893If defined, a C expression whose value is a character constant
6894containing the flag used to mark a section as a TLS section. The
6895default is @code{'T'}.
6896@end defmac
6897
6898@defmac INIT_SECTION_ASM_OP
6899If defined, a C expression whose value is a string, including spacing,
6900containing the assembler operation to identify the following data as
6901initialization code. If not defined, GCC will assume such a section does
6902not exist. This section has no corresponding @code{init_section}
6903variable; it is used entirely in runtime code.
6904@end defmac
6905
6906@defmac FINI_SECTION_ASM_OP
6907If defined, a C expression whose value is a string, including spacing,
6908containing the assembler operation to identify the following data as
6909finalization code. If not defined, GCC will assume such a section does
6910not exist. This section has no corresponding @code{fini_section}
6911variable; it is used entirely in runtime code.
6912@end defmac
6913
6914@defmac INIT_ARRAY_SECTION_ASM_OP
6915If defined, a C expression whose value is a string, including spacing,
6916containing the assembler operation to identify the following data as
6917part of the @code{.init_array} (or equivalent) section. If not
6918defined, GCC will assume such a section does not exist. Do not define
6919both this macro and @code{INIT_SECTION_ASM_OP}.
6920@end defmac
6921
6922@defmac FINI_ARRAY_SECTION_ASM_OP
6923If defined, a C expression whose value is a string, including spacing,
6924containing the assembler operation to identify the following data as
6925part of the @code{.fini_array} (or equivalent) section. If not
6926defined, GCC will assume such a section does not exist. Do not define
6927both this macro and @code{FINI_SECTION_ASM_OP}.
6928@end defmac
6929
6930@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6931If defined, an ASM statement that switches to a different section
6932via @var{section_op}, calls @var{function}, and switches back to
6933the text section. This is used in @file{crtstuff.c} if
6934@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6935to initialization and finalization functions from the init and fini
6936sections. By default, this macro uses a simple function call. Some
6937ports need hand-crafted assembly code to avoid dependencies on
6938registers initialized in the function prologue or to ensure that
6939constant pools don't end up too far way in the text section.
6940@end defmac
6941
6942@defmac TARGET_LIBGCC_SDATA_SECTION
6943If defined, a string which names the section into which small
6944variables defined in crtstuff and libgcc should go. This is useful
6945when the target has options for optimizing access to small data, and
6946you want the crtstuff and libgcc routines to be conservative in what
6947they expect of your application yet liberal in what your application
6948expects. For example, for targets with a @code{.sdata} section (like
6949MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6950require small data support from your application, but use this macro
6951to put small data into @code{.sdata} so that your application can
6952access these variables whether it uses small data or not.
6953@end defmac
6954
6955@defmac FORCE_CODE_SECTION_ALIGN
6956If defined, an ASM statement that aligns a code section to some
6957arbitrary boundary. This is used to force all fragments of the
6958@code{.init} and @code{.fini} sections to have to same alignment
6959and thus prevent the linker from having to add any padding.
6960@end defmac
6961
6962@defmac JUMP_TABLES_IN_TEXT_SECTION
6963Define this macro to be an expression with a nonzero value if jump
6964tables (for @code{tablejump} insns) should be output in the text
6965section, along with the assembler instructions. Otherwise, the
6966readonly data section is used.
6967
6968This macro is irrelevant if there is no separate readonly data section.
6969@end defmac
6970
6971@hook TARGET_ASM_INIT_SECTIONS
6972Define this hook if you need to do something special to set up the
6973@file{varasm.c} sections, or if your target has some special sections
6974of its own that you need to create.
6975
6976GCC calls this hook after processing the command line, but before writing
6977any assembly code, and before calling any of the section-returning hooks
6978described below.
6979@end deftypefn
6980
6981@hook TARGET_ASM_RELOC_RW_MASK
6982Return a mask describing how relocations should be treated when
6983selecting sections. Bit 1 should be set if global relocations
6984should be placed in a read-write section; bit 0 should be set if
6985local relocations should be placed in a read-write section.
6986
6987The default version of this function returns 3 when @option{-fpic}
6988is in effect, and 0 otherwise. The hook is typically redefined
6989when the target cannot support (some kinds of) dynamic relocations
6990in read-only sections even in executables.
6991@end deftypefn
6992
6993@hook TARGET_ASM_SELECT_SECTION
6994Return the section into which @var{exp} should be placed. You can
6995assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6996some sort. @var{reloc} indicates whether the initial value of @var{exp}
6997requires link-time relocations. Bit 0 is set when variable contains
6998local relocations only, while bit 1 is set for global relocations.
6999@var{align} is the constant alignment in bits.
7000
7001The default version of this function takes care of putting read-only
7002variables in @code{readonly_data_section}.
7003
7004See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7005@end deftypefn
7006
7007@defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7008Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7009for @code{FUNCTION_DECL}s as well as for variables and constants.
7010
7011In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7012function has been determined to be likely to be called, and nonzero if
7013it is unlikely to be called.
7014@end defmac
7015
7016@hook TARGET_ASM_UNIQUE_SECTION
7017Build up a unique section name, expressed as a @code{STRING_CST} node,
7018and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7019As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7020the initial value of @var{exp} requires link-time relocations.
7021
7022The default version of this function appends the symbol name to the
7023ELF section name that would normally be used for the symbol. For
7024example, the function @code{foo} would be placed in @code{.text.foo}.
7025Whatever the actual target object format, this is often good enough.
7026@end deftypefn
7027
7028@hook TARGET_ASM_FUNCTION_RODATA_SECTION
7029Return the readonly data section associated with
7030@samp{DECL_SECTION_NAME (@var{decl})}.
7031The default version of this function selects @code{.gnu.linkonce.r.name} if
7032the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7033if function is in @code{.text.name}, and the normal readonly-data section
7034otherwise.
7035@end deftypefn
7036
7037@hook TARGET_ASM_SELECT_RTX_SECTION
7038Return the section into which a constant @var{x}, of mode @var{mode},
7039should be placed. You can assume that @var{x} is some kind of
7040constant in RTL@. The argument @var{mode} is redundant except in the
7041case of a @code{const_int} rtx. @var{align} is the constant alignment
7042in bits.
7043
7044The default version of this function takes care of putting symbolic
7045constants in @code{flag_pic} mode in @code{data_section} and everything
7046else in @code{readonly_data_section}.
7047@end deftypefn
7048
7049@hook TARGET_MANGLE_DECL_ASSEMBLER_NAME
7050Define this hook if you need to postprocess the assembler name generated
7051by target-independent code. The @var{id} provided to this hook will be
7052the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7053or the mangled name of the @var{decl} in C++). The return value of the
7054hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7055your target system. The default implementation of this hook just
7056returns the @var{id} provided.
7057@end deftypefn
7058
7059@hook TARGET_ENCODE_SECTION_INFO
7060Define this hook if references to a symbol or a constant must be
7061treated differently depending on something about the variable or
7062function named by the symbol (such as what section it is in).
7063
7064The hook is executed immediately after rtl has been created for
7065@var{decl}, which may be a variable or function declaration or
7066an entry in the constant pool. In either case, @var{rtl} is the
7067rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7068in this hook; that field may not have been initialized yet.
7069
7070In the case of a constant, it is safe to assume that the rtl is
7071a @code{mem} whose address is a @code{symbol_ref}. Most decls
7072will also have this form, but that is not guaranteed. Global
7073register variables, for instance, will have a @code{reg} for their
7074rtl. (Normally the right thing to do with such unusual rtl is
7075leave it alone.)
7076
7077The @var{new_decl_p} argument will be true if this is the first time
7078that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7079be false for subsequent invocations, which will happen for duplicate
7080declarations. Whether or not anything must be done for the duplicate
7081declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7082@var{new_decl_p} is always true when the hook is called for a constant.
7083
7084@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7085The usual thing for this hook to do is to record flags in the
7086@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7087Historically, the name string was modified if it was necessary to
7088encode more than one bit of information, but this practice is now
7089discouraged; use @code{SYMBOL_REF_FLAGS}.
7090
7091The default definition of this hook, @code{default_encode_section_info}
7092in @file{varasm.c}, sets a number of commonly-useful bits in
7093@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7094before overriding it.
7095@end deftypefn
7096
7097@hook TARGET_STRIP_NAME_ENCODING
7098Decode @var{name} and return the real name part, sans
7099the characters that @code{TARGET_ENCODE_SECTION_INFO}
7100may have added.
7101@end deftypefn
7102
7103@hook TARGET_IN_SMALL_DATA_P
7104Returns true if @var{exp} should be placed into a ``small data'' section.
7105The default version of this hook always returns false.
7106@end deftypefn
7107
7108@hook TARGET_HAVE_SRODATA_SECTION
7109Contains the value true if the target places read-only
7110``small data'' into a separate section. The default value is false.
7111@end deftypevr
7112
3c5273a9
KT
7113@hook TARGET_PROFILE_BEFORE_PROLOGUE
7114
38f8b050
JR
7115@hook TARGET_BINDS_LOCAL_P
7116Returns true if @var{exp} names an object for which name resolution
7117rules must resolve to the current ``module'' (dynamic shared library
7118or executable image).
7119
7120The default version of this hook implements the name resolution rules
7121for ELF, which has a looser model of global name binding than other
7122currently supported object file formats.
7123@end deftypefn
7124
7125@hook TARGET_HAVE_TLS
7126Contains the value true if the target supports thread-local storage.
7127The default value is false.
7128@end deftypevr
7129
7130
7131@node PIC
7132@section Position Independent Code
7133@cindex position independent code
7134@cindex PIC
7135
7136This section describes macros that help implement generation of position
7137independent code. Simply defining these macros is not enough to
7138generate valid PIC; you must also add support to the hook
7139@code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7140@code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7141must modify the definition of @samp{movsi} to do something appropriate
7142when the source operand contains a symbolic address. You may also
7143need to alter the handling of switch statements so that they use
7144relative addresses.
7145@c i rearranged the order of the macros above to try to force one of
7146@c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7147
7148@defmac PIC_OFFSET_TABLE_REGNUM
7149The register number of the register used to address a table of static
7150data addresses in memory. In some cases this register is defined by a
7151processor's ``application binary interface'' (ABI)@. When this macro
7152is defined, RTL is generated for this register once, as with the stack
7153pointer and frame pointer registers. If this macro is not defined, it
7154is up to the machine-dependent files to allocate such a register (if
7155necessary). Note that this register must be fixed when in use (e.g.@:
7156when @code{flag_pic} is true).
7157@end defmac
7158
7159@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
f8fe0a4a
JM
7160A C expression that is nonzero if the register defined by
7161@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7162the default is zero. Do not define
38f8b050
JR
7163this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7164@end defmac
7165
7166@defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7167A C expression that is nonzero if @var{x} is a legitimate immediate
7168operand on the target machine when generating position independent code.
7169You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7170check this. You can also assume @var{flag_pic} is true, so you need not
7171check it either. You need not define this macro if all constants
7172(including @code{SYMBOL_REF}) can be immediate operands when generating
7173position independent code.
7174@end defmac
7175
7176@node Assembler Format
7177@section Defining the Output Assembler Language
7178
7179This section describes macros whose principal purpose is to describe how
7180to write instructions in assembler language---rather than what the
7181instructions do.
7182
7183@menu
7184* File Framework:: Structural information for the assembler file.
7185* Data Output:: Output of constants (numbers, strings, addresses).
7186* Uninitialized Data:: Output of uninitialized variables.
7187* Label Output:: Output and generation of labels.
7188* Initialization:: General principles of initialization
7189 and termination routines.
7190* Macros for Initialization::
7191 Specific macros that control the handling of
7192 initialization and termination routines.
7193* Instruction Output:: Output of actual instructions.
7194* Dispatch Tables:: Output of jump tables.
7195* Exception Region Output:: Output of exception region code.
7196* Alignment Output:: Pseudo ops for alignment and skipping data.
7197@end menu
7198
7199@node File Framework
7200@subsection The Overall Framework of an Assembler File
7201@cindex assembler format
7202@cindex output of assembler code
7203
7204@c prevent bad page break with this line
7205This describes the overall framework of an assembly file.
7206
7207@findex default_file_start
7208@hook TARGET_ASM_FILE_START
7209Output to @code{asm_out_file} any text which the assembler expects to
7210find at the beginning of a file. The default behavior is controlled
7211by two flags, documented below. Unless your target's assembler is
7212quite unusual, if you override the default, you should call
7213@code{default_file_start} at some point in your target hook. This
7214lets other target files rely on these variables.
7215@end deftypefn
7216
7217@hook TARGET_ASM_FILE_START_APP_OFF
7218If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7219printed as the very first line in the assembly file, unless
7220@option{-fverbose-asm} is in effect. (If that macro has been defined
7221to the empty string, this variable has no effect.) With the normal
7222definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7223assembler that it need not bother stripping comments or extra
7224whitespace from its input. This allows it to work a bit faster.
7225
7226The default is false. You should not set it to true unless you have
7227verified that your port does not generate any extra whitespace or
7228comments that will cause GAS to issue errors in NO_APP mode.
7229@end deftypevr
7230
7231@hook TARGET_ASM_FILE_START_FILE_DIRECTIVE
7232If this flag is true, @code{output_file_directive} will be called
7233for the primary source file, immediately after printing
7234@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7235this to be done. The default is false.
7236@end deftypevr
7237
7238@hook TARGET_ASM_FILE_END
7239Output to @code{asm_out_file} any text which the assembler expects
7240to find at the end of a file. The default is to output nothing.
7241@end deftypefn
7242
7243@deftypefun void file_end_indicate_exec_stack ()
7244Some systems use a common convention, the @samp{.note.GNU-stack}
7245special section, to indicate whether or not an object file relies on
7246the stack being executable. If your system uses this convention, you
7247should define @code{TARGET_ASM_FILE_END} to this function. If you
7248need to do other things in that hook, have your hook function call
7249this function.
7250@end deftypefun
7251
7252@hook TARGET_ASM_LTO_START
7253Output to @code{asm_out_file} any text which the assembler expects
7254to find at the start of an LTO section. The default is to output
7255nothing.
7256@end deftypefn
7257
7258@hook TARGET_ASM_LTO_END
7259Output to @code{asm_out_file} any text which the assembler expects
7260to find at the end of an LTO section. The default is to output
7261nothing.
7262@end deftypefn
7263
7264@hook TARGET_ASM_CODE_END
7265Output to @code{asm_out_file} any text which is needed before emitting
7266unwind info and debug info at the end of a file. Some targets emit
7267here PIC setup thunks that cannot be emitted at the end of file,
7268because they couldn't have unwind info then. The default is to output
7269nothing.
7270@end deftypefn
7271
7272@defmac ASM_COMMENT_START
7273A C string constant describing how to begin a comment in the target
7274assembler language. The compiler assumes that the comment will end at
7275the end of the line.
7276@end defmac
7277
7278@defmac ASM_APP_ON
7279A C string constant for text to be output before each @code{asm}
7280statement or group of consecutive ones. Normally this is
7281@code{"#APP"}, which is a comment that has no effect on most
7282assemblers but tells the GNU assembler that it must check the lines
7283that follow for all valid assembler constructs.
7284@end defmac
7285
7286@defmac ASM_APP_OFF
7287A C string constant for text to be output after each @code{asm}
7288statement or group of consecutive ones. Normally this is
7289@code{"#NO_APP"}, which tells the GNU assembler to resume making the
7290time-saving assumptions that are valid for ordinary compiler output.
7291@end defmac
7292
7293@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7294A C statement to output COFF information or DWARF debugging information
7295which indicates that filename @var{name} is the current source file to
7296the stdio stream @var{stream}.
7297
7298This macro need not be defined if the standard form of output
7299for the file format in use is appropriate.
7300@end defmac
7301
b5f5d41d
AS
7302@hook TARGET_ASM_OUTPUT_SOURCE_FILENAME
7303
38f8b050
JR
7304@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7305A C statement to output the string @var{string} to the stdio stream
7306@var{stream}. If you do not call the function @code{output_quoted_string}
7307in your config files, GCC will only call it to output filenames to
7308the assembler source. So you can use it to canonicalize the format
7309of the filename using this macro.
7310@end defmac
7311
7312@defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7313A C statement to output something to the assembler file to handle a
7314@samp{#ident} directive containing the text @var{string}. If this
7315macro is not defined, nothing is output for a @samp{#ident} directive.
7316@end defmac
7317
7318@hook TARGET_ASM_NAMED_SECTION
7319Output assembly directives to switch to section @var{name}. The section
7320should have attributes as specified by @var{flags}, which is a bit mask
7321of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7322is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7323this section is associated.
7324@end deftypefn
7325
7326@hook TARGET_HAVE_NAMED_SECTIONS
7327This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7328@end deftypevr
7329
7330@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7331@hook TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7332This flag is true if we can create zeroed data by switching to a BSS
7333section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7334This is true on most ELF targets.
7335@end deftypevr
7336
7337@hook TARGET_SECTION_TYPE_FLAGS
7338Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7339based on a variable or function decl, a section name, and whether or not the
7340declaration's initializer may contain runtime relocations. @var{decl} may be
7341null, in which case read-write data should be assumed.
7342
7343The default version of this function handles choosing code vs data,
7344read-only vs read-write data, and @code{flag_pic}. You should only
7345need to override this if your target has special flags that might be
7346set via @code{__attribute__}.
7347@end deftypefn
7348
7349@hook TARGET_ASM_RECORD_GCC_SWITCHES
7350Provides the target with the ability to record the gcc command line
7351switches that have been passed to the compiler, and options that are
7352enabled. The @var{type} argument specifies what is being recorded.
7353It can take the following values:
7354
7355@table @gcctabopt
7356@item SWITCH_TYPE_PASSED
7357@var{text} is a command line switch that has been set by the user.
7358
7359@item SWITCH_TYPE_ENABLED
7360@var{text} is an option which has been enabled. This might be as a
7361direct result of a command line switch, or because it is enabled by
7362default or because it has been enabled as a side effect of a different
7363command line switch. For example, the @option{-O2} switch enables
7364various different individual optimization passes.
7365
7366@item SWITCH_TYPE_DESCRIPTIVE
7367@var{text} is either NULL or some descriptive text which should be
7368ignored. If @var{text} is NULL then it is being used to warn the
7369target hook that either recording is starting or ending. The first
7370time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7371warning is for start up and the second time the warning is for
7372wind down. This feature is to allow the target hook to make any
7373necessary preparations before it starts to record switches and to
7374perform any necessary tidying up after it has finished recording
7375switches.
7376
7377@item SWITCH_TYPE_LINE_START
7378This option can be ignored by this target hook.
7379
7380@item SWITCH_TYPE_LINE_END
7381This option can be ignored by this target hook.
7382@end table
7383
7384The hook's return value must be zero. Other return values may be
7385supported in the future.
7386
7387By default this hook is set to NULL, but an example implementation is
7388provided for ELF based targets. Called @var{elf_record_gcc_switches},
7389it records the switches as ASCII text inside a new, string mergeable
7390section in the assembler output file. The name of the new section is
7391provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7392hook.
7393@end deftypefn
7394
7395@hook TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7396This is the name of the section that will be created by the example
7397ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7398hook.
7399@end deftypevr
7400
7401@need 2000
7402@node Data Output
7403@subsection Output of Data
7404
7405
7406@hook TARGET_ASM_BYTE_OP
7407@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7408@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7409@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7410@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7411@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7412@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7413@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7414@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7415These hooks specify assembly directives for creating certain kinds
7416of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7417byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7418aligned two-byte object, and so on. Any of the hooks may be
7419@code{NULL}, indicating that no suitable directive is available.
7420
7421The compiler will print these strings at the start of a new line,
7422followed immediately by the object's initial value. In most cases,
7423the string should contain a tab, a pseudo-op, and then another tab.
7424@end deftypevr
7425
7426@hook TARGET_ASM_INTEGER
7427The @code{assemble_integer} function uses this hook to output an
7428integer object. @var{x} is the object's value, @var{size} is its size
7429in bytes and @var{aligned_p} indicates whether it is aligned. The
7430function should return @code{true} if it was able to output the
7431object. If it returns false, @code{assemble_integer} will try to
7432split the object into smaller parts.
7433
7434The default implementation of this hook will use the
7435@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7436when the relevant string is @code{NULL}.
7437@end deftypefn
7438
6cbd8875
AS
7439@hook TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA
7440A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7441can't deal with, and output assembly code to @var{file} corresponding to
7442the pattern @var{x}. This may be used to allow machine-dependent
7443@code{UNSPEC}s to appear within constants.
7444
7445If target hook fails to recognize a pattern, it must return @code{false},
7446so that a standard error message is printed. If it prints an error message
7447itself, by calling, for example, @code{output_operand_lossage}, it may just
7448return @code{true}.
7449@end deftypefn
7450
38f8b050
JR
7451@defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
7452A C statement to recognize @var{rtx} patterns that
7453@code{output_addr_const} can't deal with, and output assembly code to
7454@var{stream} corresponding to the pattern @var{x}. This may be used to
7455allow machine-dependent @code{UNSPEC}s to appear within constants.
7456
7457If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
7458@code{goto fail}, so that a standard error message is printed. If it
7459prints an error message itself, by calling, for example,
7460@code{output_operand_lossage}, it may just complete normally.
7461@end defmac
7462
7463@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7464A C statement to output to the stdio stream @var{stream} an assembler
7465instruction to assemble a string constant containing the @var{len}
7466bytes at @var{ptr}. @var{ptr} will be a C expression of type
7467@code{char *} and @var{len} a C expression of type @code{int}.
7468
7469If the assembler has a @code{.ascii} pseudo-op as found in the
7470Berkeley Unix assembler, do not define the macro
7471@code{ASM_OUTPUT_ASCII}.
7472@end defmac
7473
7474@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7475A C statement to output word @var{n} of a function descriptor for
7476@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7477is defined, and is otherwise unused.
7478@end defmac
7479
7480@defmac CONSTANT_POOL_BEFORE_FUNCTION
7481You may define this macro as a C expression. You should define the
7482expression to have a nonzero value if GCC should output the constant
7483pool for a function before the code for the function, or a zero value if
7484GCC should output the constant pool after the function. If you do
7485not define this macro, the usual case, GCC will output the constant
7486pool before the function.
7487@end defmac
7488
7489@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7490A C statement to output assembler commands to define the start of the
7491constant pool for a function. @var{funname} is a string giving
7492the name of the function. Should the return type of the function
7493be required, it can be obtained via @var{fundecl}. @var{size}
7494is the size, in bytes, of the constant pool that will be written
7495immediately after this call.
7496
7497If no constant-pool prefix is required, the usual case, this macro need
7498not be defined.
7499@end defmac
7500
7501@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7502A C statement (with or without semicolon) to output a constant in the
7503constant pool, if it needs special treatment. (This macro need not do
7504anything for RTL expressions that can be output normally.)
7505
7506The argument @var{file} is the standard I/O stream to output the
7507assembler code on. @var{x} is the RTL expression for the constant to
7508output, and @var{mode} is the machine mode (in case @var{x} is a
7509@samp{const_int}). @var{align} is the required alignment for the value
7510@var{x}; you should output an assembler directive to force this much
7511alignment.
7512
7513The argument @var{labelno} is a number to use in an internal label for
7514the address of this pool entry. The definition of this macro is
7515responsible for outputting the label definition at the proper place.
7516Here is how to do this:
7517
7518@smallexample
7519@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7520@end smallexample
7521
7522When you output a pool entry specially, you should end with a
7523@code{goto} to the label @var{jumpto}. This will prevent the same pool
7524entry from being output a second time in the usual manner.
7525
7526You need not define this macro if it would do nothing.
7527@end defmac
7528
7529@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7530A C statement to output assembler commands to at the end of the constant
7531pool for a function. @var{funname} is a string giving the name of the
7532function. Should the return type of the function be required, you can
7533obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7534constant pool that GCC wrote immediately before this call.
7535
7536If no constant-pool epilogue is required, the usual case, you need not
7537define this macro.
7538@end defmac
7539
7540@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7541Define this macro as a C expression which is nonzero if @var{C} is
7542used as a logical line separator by the assembler. @var{STR} points
7543to the position in the string where @var{C} was found; this can be used if
7544a line separator uses multiple characters.
7545
7546If you do not define this macro, the default is that only
7547the character @samp{;} is treated as a logical line separator.
7548@end defmac
7549
7550@hook TARGET_ASM_OPEN_PAREN
7551These target hooks are C string constants, describing the syntax in the
7552assembler for grouping arithmetic expressions. If not overridden, they
7553default to normal parentheses, which is correct for most assemblers.
7554@end deftypevr
7555
7556These macros are provided by @file{real.h} for writing the definitions
7557of @code{ASM_OUTPUT_DOUBLE} and the like:
7558
7559@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7560@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7561@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7562@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7563@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7564@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7565These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7566target's floating point representation, and store its bit pattern in
7567the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7568@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7569simple @code{long int}. For the others, it should be an array of
7570@code{long int}. The number of elements in this array is determined
7571by the size of the desired target floating point data type: 32 bits of
7572it go in each @code{long int} array element. Each array element holds
757332 bits of the result, even if @code{long int} is wider than 32 bits
7574on the host machine.
7575
7576The array element values are designed so that you can print them out
7577using @code{fprintf} in the order they should appear in the target
7578machine's memory.
7579@end defmac
7580
7581@node Uninitialized Data
7582@subsection Output of Uninitialized Variables
7583
7584Each of the macros in this section is used to do the whole job of
7585outputting a single uninitialized variable.
7586
7587@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7588A C statement (sans semicolon) to output to the stdio stream
7589@var{stream} the assembler definition of a common-label named
7590@var{name} whose size is @var{size} bytes. The variable @var{rounded}
7591is the size rounded up to whatever alignment the caller wants. It is
7592possible that @var{size} may be zero, for instance if a struct with no
7593other member than a zero-length array is defined. In this case, the
7594backend must output a symbol definition that allocates at least one
7595byte, both so that the address of the resulting object does not compare
7596equal to any other, and because some object formats cannot even express
7597the concept of a zero-sized common symbol, as that is how they represent
7598an ordinary undefined external.
7599
7600Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7601output the name itself; before and after that, output the additional
7602assembler syntax for defining the name, and a newline.
7603
7604This macro controls how the assembler definitions of uninitialized
7605common global variables are output.
7606@end defmac
7607
7608@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7609Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7610separate, explicit argument. If you define this macro, it is used in
7611place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7612handling the required alignment of the variable. The alignment is specified
7613as the number of bits.
7614@end defmac
7615
7616@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7617Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7618variable to be output, if there is one, or @code{NULL_TREE} if there
7619is no corresponding variable. If you define this macro, GCC will use it
7620in place of both @code{ASM_OUTPUT_COMMON} and
7621@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7622the variable's decl in order to chose what to output.
7623@end defmac
7624
7625@defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
7626A C statement (sans semicolon) to output to the stdio stream
7627@var{stream} the assembler definition of uninitialized global @var{decl} named
7628@var{name} whose size is @var{size} bytes. The variable @var{rounded}
7629is the size rounded up to whatever alignment the caller wants.
7630
7631Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
7632defining this macro. If unable, use the expression
7633@code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7634before and after that, output the additional assembler syntax for defining
7635the name, and a newline.
7636
7637There are two ways of handling global BSS@. One is to define either
7638this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
7639The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7640switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7641You do not need to do both.
7642
7643Some languages do not have @code{common} data, and require a
7644non-common form of global BSS in order to handle uninitialized globals
7645efficiently. C++ is one example of this. However, if the target does
7646not support global BSS, the front end may choose to make globals
7647common in order to save space in the object file.
7648@end defmac
7649
7650@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7651Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
7652separate, explicit argument. If you define this macro, it is used in
7653place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
7654handling the required alignment of the variable. The alignment is specified
7655as the number of bits.
7656
7657Try to use function @code{asm_output_aligned_bss} defined in file
7658@file{varasm.c} when defining this macro.
7659@end defmac
7660
7661@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7662A C statement (sans semicolon) to output to the stdio stream
7663@var{stream} the assembler definition of a local-common-label named
7664@var{name} whose size is @var{size} bytes. The variable @var{rounded}
7665is the size rounded up to whatever alignment the caller wants.
7666
7667Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7668output the name itself; before and after that, output the additional
7669assembler syntax for defining the name, and a newline.
7670
7671This macro controls how the assembler definitions of uninitialized
7672static variables are output.
7673@end defmac
7674
7675@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7676Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7677separate, explicit argument. If you define this macro, it is used in
7678place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7679handling the required alignment of the variable. The alignment is specified
7680as the number of bits.
7681@end defmac
7682
7683@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7684Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7685variable to be output, if there is one, or @code{NULL_TREE} if there
7686is no corresponding variable. If you define this macro, GCC will use it
7687in place of both @code{ASM_OUTPUT_DECL} and
7688@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7689the variable's decl in order to chose what to output.
7690@end defmac
7691
7692@node Label Output
7693@subsection Output and Generation of Labels
7694
7695@c prevent bad page break with this line
7696This is about outputting labels.
7697
7698@findex assemble_name
7699@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7700A C statement (sans semicolon) to output to the stdio stream
7701@var{stream} the assembler definition of a label named @var{name}.
7702Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7703output the name itself; before and after that, output the additional
7704assembler syntax for defining the name, and a newline. A default
7705definition of this macro is provided which is correct for most systems.
7706@end defmac
7707
135a687e
KT
7708@defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7709A C statement (sans semicolon) to output to the stdio stream
7710@var{stream} the assembler definition of a label named @var{name} of
7711a function.
7712Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7713output the name itself; before and after that, output the additional
7714assembler syntax for defining the name, and a newline. A default
7715definition of this macro is provided which is correct for most systems.
7716
7717If this macro is not defined, then the function name is defined in the
7718usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7719@end defmac
7720
38f8b050
JR
7721@findex assemble_name_raw
7722@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7723Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7724to refer to a compiler-generated label. The default definition uses
7725@code{assemble_name_raw}, which is like @code{assemble_name} except
7726that it is more efficient.
7727@end defmac
7728
7729@defmac SIZE_ASM_OP
7730A C string containing the appropriate assembler directive to specify the
7731size of a symbol, without any arguments. On systems that use ELF, the
7732default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7733systems, the default is not to define this macro.
7734
7735Define this macro only if it is correct to use the default definitions
7736of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7737for your system. If you need your own custom definitions of those
7738macros, or if you do not need explicit symbol sizes at all, do not
7739define this macro.
7740@end defmac
7741
7742@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7743A C statement (sans semicolon) to output to the stdio stream
7744@var{stream} a directive telling the assembler that the size of the
7745symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7746If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7747provided.
7748@end defmac
7749
7750@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7751A C statement (sans semicolon) to output to the stdio stream
7752@var{stream} a directive telling the assembler to calculate the size of
7753the symbol @var{name} by subtracting its address from the current
7754address.
7755
7756If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7757provided. The default assumes that the assembler recognizes a special
7758@samp{.} symbol as referring to the current address, and can calculate
7759the difference between this and another symbol. If your assembler does
7760not recognize @samp{.} or cannot do calculations with it, you will need
7761to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7762@end defmac
7763
7764@defmac TYPE_ASM_OP
7765A C string containing the appropriate assembler directive to specify the
7766type of a symbol, without any arguments. On systems that use ELF, the
7767default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7768systems, the default is not to define this macro.
7769
7770Define this macro only if it is correct to use the default definition of
7771@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7772custom definition of this macro, or if you do not need explicit symbol
7773types at all, do not define this macro.
7774@end defmac
7775
7776@defmac TYPE_OPERAND_FMT
7777A C string which specifies (using @code{printf} syntax) the format of
7778the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7779default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7780the default is not to define this macro.
7781
7782Define this macro only if it is correct to use the default definition of
7783@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7784custom definition of this macro, or if you do not need explicit symbol
7785types at all, do not define this macro.
7786@end defmac
7787
7788@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7789A C statement (sans semicolon) to output to the stdio stream
7790@var{stream} a directive telling the assembler that the type of the
7791symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7792that string is always either @samp{"function"} or @samp{"object"}, but
7793you should not count on this.
7794
7795If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7796definition of this macro is provided.
7797@end defmac
7798
7799@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7800A C statement (sans semicolon) to output to the stdio stream
7801@var{stream} any text necessary for declaring the name @var{name} of a
7802function which is being defined. This macro is responsible for
7803outputting the label definition (perhaps using
135a687e 7804@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
38f8b050
JR
7805@code{FUNCTION_DECL} tree node representing the function.
7806
7807If this macro is not defined, then the function name is defined in the
135a687e 7808usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
38f8b050
JR
7809
7810You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7811of this macro.
7812@end defmac
7813
7814@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7815A C statement (sans semicolon) to output to the stdio stream
7816@var{stream} any text necessary for declaring the size of a function
7817which is being defined. The argument @var{name} is the name of the
7818function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7819representing the function.
7820
7821If this macro is not defined, then the function size is not defined.
7822
7823You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7824of this macro.
7825@end defmac
7826
7827@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7828A C statement (sans semicolon) to output to the stdio stream
7829@var{stream} any text necessary for declaring the name @var{name} of an
7830initialized variable which is being defined. This macro must output the
7831label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7832@var{decl} is the @code{VAR_DECL} tree node representing the variable.
7833
7834If this macro is not defined, then the variable name is defined in the
7835usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7836
7837You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7838@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7839@end defmac
7840
ad78130c 7841@hook TARGET_ASM_DECLARE_CONSTANT_NAME
38f8b050
JR
7842A target hook to output to the stdio stream @var{file} any text necessary
7843for declaring the name @var{name} of a constant which is being defined. This
7844target hook is responsible for outputting the label definition (perhaps using
7845@code{assemble_label}). The argument @var{exp} is the value of the constant,
7846and @var{size} is the size of the constant in bytes. The @var{name}
7847will be an internal label.
7848
7849The default version of this target hook, define the @var{name} in the
7850usual manner as a label (by means of @code{assemble_label}).
7851
7852You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7853@end deftypefn
7854
7855@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7856A C statement (sans semicolon) to output to the stdio stream
7857@var{stream} any text necessary for claiming a register @var{regno}
7858for a global variable @var{decl} with name @var{name}.
7859
7860If you don't define this macro, that is equivalent to defining it to do
7861nothing.
7862@end defmac
7863
7864@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7865A C statement (sans semicolon) to finish up declaring a variable name
7866once the compiler has processed its initializer fully and thus has had a
7867chance to determine the size of an array when controlled by an
7868initializer. This is used on systems where it's necessary to declare
7869something about the size of the object.
7870
7871If you don't define this macro, that is equivalent to defining it to do
7872nothing.
7873
7874You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7875@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7876@end defmac
7877
7878@hook TARGET_ASM_GLOBALIZE_LABEL
7879This target hook is a function to output to the stdio stream
7880@var{stream} some commands that will make the label @var{name} global;
7881that is, available for reference from other files.
7882
7883The default implementation relies on a proper definition of
7884@code{GLOBAL_ASM_OP}.
7885@end deftypefn
7886
7887@hook TARGET_ASM_GLOBALIZE_DECL_NAME
7888This target hook is a function to output to the stdio stream
7889@var{stream} some commands that will make the name associated with @var{decl}
7890global; that is, available for reference from other files.
7891
7892The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7893@end deftypefn
7894
7895@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7896A C statement (sans semicolon) to output to the stdio stream
7897@var{stream} some commands that will make the label @var{name} weak;
7898that is, available for reference from other files but only used if
7899no other definition is available. Use the expression
7900@code{assemble_name (@var{stream}, @var{name})} to output the name
7901itself; before and after that, output the additional assembler syntax
7902for making that name weak, and a newline.
7903
7904If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7905support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7906macro.
7907@end defmac
7908
7909@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7910Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7911@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7912or variable decl. If @var{value} is not @code{NULL}, this C statement
7913should output to the stdio stream @var{stream} assembler code which
7914defines (equates) the weak symbol @var{name} to have the value
7915@var{value}. If @var{value} is @code{NULL}, it should output commands
7916to make @var{name} weak.
7917@end defmac
7918
7919@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7920Outputs a directive that enables @var{name} to be used to refer to
7921symbol @var{value} with weak-symbol semantics. @code{decl} is the
7922declaration of @code{name}.
7923@end defmac
7924
7925@defmac SUPPORTS_WEAK
74b90fe2
JDA
7926A preprocessor constant expression which evaluates to true if the target
7927supports weak symbols.
38f8b050
JR
7928
7929If you don't define this macro, @file{defaults.h} provides a default
7930definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
74b90fe2
JDA
7931is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
7932@end defmac
7933
7934@defmac TARGET_SUPPORTS_WEAK
7935A C expression which evaluates to true if the target supports weak symbols.
7936
7937If you don't define this macro, @file{defaults.h} provides a default
7938definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
7939this macro if you want to control weak symbol support with a compiler
7940flag such as @option{-melf}.
38f8b050
JR
7941@end defmac
7942
7943@defmac MAKE_DECL_ONE_ONLY (@var{decl})
7944A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7945public symbol such that extra copies in multiple translation units will
7946be discarded by the linker. Define this macro if your object file
7947format provides support for this concept, such as the @samp{COMDAT}
7948section flags in the Microsoft Windows PE/COFF format, and this support
7949requires changes to @var{decl}, such as putting it in a separate section.
7950@end defmac
7951
7952@defmac SUPPORTS_ONE_ONLY
7953A C expression which evaluates to true if the target supports one-only
7954semantics.
7955
7956If you don't define this macro, @file{varasm.c} provides a default
7957definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7958definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7959you want to control one-only symbol support with a compiler flag, or if
7960setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7961be emitted as one-only.
7962@end defmac
7963
7964@hook TARGET_ASM_ASSEMBLE_VISIBILITY
7965This target hook is a function to output to @var{asm_out_file} some
7966commands that will make the symbol(s) associated with @var{decl} have
7967hidden, protected or internal visibility as specified by @var{visibility}.
7968@end deftypefn
7969
7970@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7971A C expression that evaluates to true if the target's linker expects
7972that weak symbols do not appear in a static archive's table of contents.
7973The default is @code{0}.
7974
7975Leaving weak symbols out of an archive's table of contents means that,
7976if a symbol will only have a definition in one translation unit and
7977will have undefined references from other translation units, that
7978symbol should not be weak. Defining this macro to be nonzero will
7979thus have the effect that certain symbols that would normally be weak
7980(explicit template instantiations, and vtables for polymorphic classes
7981with noninline key methods) will instead be nonweak.
7982
7983The C++ ABI requires this macro to be zero. Define this macro for
7984targets where full C++ ABI compliance is impossible and where linker
7985restrictions require weak symbols to be left out of a static archive's
7986table of contents.
7987@end defmac
7988
7989@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7990A C statement (sans semicolon) to output to the stdio stream
7991@var{stream} any text necessary for declaring the name of an external
7992symbol named @var{name} which is referenced in this compilation but
7993not defined. The value of @var{decl} is the tree node for the
7994declaration.
7995
7996This macro need not be defined if it does not need to output anything.
7997The GNU assembler and most Unix assemblers don't require anything.
7998@end defmac
7999
8000@hook TARGET_ASM_EXTERNAL_LIBCALL
8001This target hook is a function to output to @var{asm_out_file} an assembler
8002pseudo-op to declare a library function name external. The name of the
8003library function is given by @var{symref}, which is a @code{symbol_ref}.
8004@end deftypefn
8005
8006@hook TARGET_ASM_MARK_DECL_PRESERVED
8007This target hook is a function to output to @var{asm_out_file} an assembler
8008directive to annotate @var{symbol} as used. The Darwin target uses the
8009.no_dead_code_strip directive.
8010@end deftypefn
8011
8012@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8013A C statement (sans semicolon) to output to the stdio stream
8014@var{stream} a reference in assembler syntax to a label named
8015@var{name}. This should add @samp{_} to the front of the name, if that
8016is customary on your operating system, as it is in most Berkeley Unix
8017systems. This macro is used in @code{assemble_name}.
8018@end defmac
8019
8020@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8021A C statement (sans semicolon) to output a reference to
8022@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8023will be used to output the name of the symbol. This macro may be used
8024to modify the way a symbol is referenced depending on information
8025encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8026@end defmac
8027
8028@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8029A C statement (sans semicolon) to output a reference to @var{buf}, the
8030result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8031@code{assemble_name} will be used to output the name of the symbol.
8032This macro is not used by @code{output_asm_label}, or the @code{%l}
8033specifier that calls it; the intention is that this macro should be set
8034when it is necessary to output a label differently when its address is
8035being taken.
8036@end defmac
8037
8038@hook TARGET_ASM_INTERNAL_LABEL
8039A function to output to the stdio stream @var{stream} a label whose
8040name is made from the string @var{prefix} and the number @var{labelno}.
8041
8042It is absolutely essential that these labels be distinct from the labels
8043used for user-level functions and variables. Otherwise, certain programs
8044will have name conflicts with internal labels.
8045
8046It is desirable to exclude internal labels from the symbol table of the
8047object file. Most assemblers have a naming convention for labels that
8048should be excluded; on many systems, the letter @samp{L} at the
8049beginning of a label has this effect. You should find out what
8050convention your system uses, and follow it.
8051
8052The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8053@end deftypefn
8054
8055@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8056A C statement to output to the stdio stream @var{stream} a debug info
8057label whose name is made from the string @var{prefix} and the number
8058@var{num}. This is useful for VLIW targets, where debug info labels
8059may need to be treated differently than branch target labels. On some
8060systems, branch target labels must be at the beginning of instruction
8061bundles, but debug info labels can occur in the middle of instruction
8062bundles.
8063
8064If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8065used.
8066@end defmac
8067
8068@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8069A C statement to store into the string @var{string} a label whose name
8070is made from the string @var{prefix} and the number @var{num}.
8071
8072This string, when output subsequently by @code{assemble_name}, should
8073produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8074with the same @var{prefix} and @var{num}.
8075
8076If the string begins with @samp{*}, then @code{assemble_name} will
8077output the rest of the string unchanged. It is often convenient for
8078@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8079string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8080to output the string, and may change it. (Of course,
8081@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8082you should know what it does on your machine.)
8083@end defmac
8084
8085@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8086A C expression to assign to @var{outvar} (which is a variable of type
8087@code{char *}) a newly allocated string made from the string
8088@var{name} and the number @var{number}, with some suitable punctuation
8089added. Use @code{alloca} to get space for the string.
8090
8091The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8092produce an assembler label for an internal static variable whose name is
8093@var{name}. Therefore, the string must be such as to result in valid
8094assembler code. The argument @var{number} is different each time this
8095macro is executed; it prevents conflicts between similarly-named
8096internal static variables in different scopes.
8097
8098Ideally this string should not be a valid C identifier, to prevent any
8099conflict with the user's own symbols. Most assemblers allow periods
8100or percent signs in assembler symbols; putting at least one of these
8101between the name and the number will suffice.
8102
8103If this macro is not defined, a default definition will be provided
8104which is correct for most systems.
8105@end defmac
8106
8107@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8108A C statement to output to the stdio stream @var{stream} assembler code
8109which defines (equates) the symbol @var{name} to have the value @var{value}.
8110
8111@findex SET_ASM_OP
8112If @code{SET_ASM_OP} is defined, a default definition is provided which is
8113correct for most systems.
8114@end defmac
8115
8116@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8117A C statement to output to the stdio stream @var{stream} assembler code
8118which defines (equates) the symbol whose tree node is @var{decl_of_name}
8119to have the value of the tree node @var{decl_of_value}. This macro will
8120be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8121the tree nodes are available.
8122
8123@findex SET_ASM_OP
8124If @code{SET_ASM_OP} is defined, a default definition is provided which is
8125correct for most systems.
8126@end defmac
8127
8128@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8129A C statement that evaluates to true if the assembler code which defines
8130(equates) the symbol whose tree node is @var{decl_of_name} to have the value
8131of the tree node @var{decl_of_value} should be emitted near the end of the
8132current compilation unit. The default is to not defer output of defines.
8133This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8134@samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8135@end defmac
8136
8137@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8138A C statement to output to the stdio stream @var{stream} assembler code
8139which defines (equates) the weak symbol @var{name} to have the value
8140@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8141an undefined weak symbol.
8142
8143Define this macro if the target only supports weak aliases; define
8144@code{ASM_OUTPUT_DEF} instead if possible.
8145@end defmac
8146
8147@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8148Define this macro to override the default assembler names used for
8149Objective-C methods.
8150
8151The default name is a unique method number followed by the name of the
8152class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8153the category is also included in the assembler name (e.g.@:
8154@samp{_1_Foo_Bar}).
8155
8156These names are safe on most systems, but make debugging difficult since
8157the method's selector is not present in the name. Therefore, particular
8158systems define other ways of computing names.
8159
8160@var{buf} is an expression of type @code{char *} which gives you a
8161buffer in which to store the name; its length is as long as
8162@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
816350 characters extra.
8164
8165The argument @var{is_inst} specifies whether the method is an instance
8166method or a class method; @var{class_name} is the name of the class;
8167@var{cat_name} is the name of the category (or @code{NULL} if the method is not
8168in a category); and @var{sel_name} is the name of the selector.
8169
8170On systems where the assembler can handle quoted names, you can use this
8171macro to provide more human-readable names.
8172@end defmac
8173
8174@defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
8175A C statement (sans semicolon) to output to the stdio stream
8176@var{stream} commands to declare that the label @var{name} is an
8177Objective-C class reference. This is only needed for targets whose
8178linkers have special support for NeXT-style runtimes.
8179@end defmac
8180
8181@defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
8182A C statement (sans semicolon) to output to the stdio stream
8183@var{stream} commands to declare that the label @var{name} is an
8184unresolved Objective-C class reference. This is only needed for targets
8185whose linkers have special support for NeXT-style runtimes.
8186@end defmac
8187
8188@node Initialization
8189@subsection How Initialization Functions Are Handled
8190@cindex initialization routines
8191@cindex termination routines
8192@cindex constructors, output of
8193@cindex destructors, output of
8194
8195The compiled code for certain languages includes @dfn{constructors}
8196(also called @dfn{initialization routines})---functions to initialize
8197data in the program when the program is started. These functions need
8198to be called before the program is ``started''---that is to say, before
8199@code{main} is called.
8200
8201Compiling some languages generates @dfn{destructors} (also called
8202@dfn{termination routines}) that should be called when the program
8203terminates.
8204
8205To make the initialization and termination functions work, the compiler
8206must output something in the assembler code to cause those functions to
8207be called at the appropriate time. When you port the compiler to a new
8208system, you need to specify how to do this.
8209
8210There are two major ways that GCC currently supports the execution of
8211initialization and termination functions. Each way has two variants.
8212Much of the structure is common to all four variations.
8213
8214@findex __CTOR_LIST__
8215@findex __DTOR_LIST__
8216The linker must build two lists of these functions---a list of
8217initialization functions, called @code{__CTOR_LIST__}, and a list of
8218termination functions, called @code{__DTOR_LIST__}.
8219
8220Each list always begins with an ignored function pointer (which may hold
82210, @minus{}1, or a count of the function pointers after it, depending on
8222the environment). This is followed by a series of zero or more function
8223pointers to constructors (or destructors), followed by a function
8224pointer containing zero.
8225
8226Depending on the operating system and its executable file format, either
8227@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8228time and exit time. Constructors are called in reverse order of the
8229list; destructors in forward order.
8230
8231The best way to handle static constructors works only for object file
8232formats which provide arbitrarily-named sections. A section is set
8233aside for a list of constructors, and another for a list of destructors.
8234Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8235object file that defines an initialization function also puts a word in
8236the constructor section to point to that function. The linker
8237accumulates all these words into one contiguous @samp{.ctors} section.
8238Termination functions are handled similarly.
8239
8240This method will be chosen as the default by @file{target-def.h} if
8241@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8242support arbitrary sections, but does support special designated
8243constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8244and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8245
8246When arbitrary sections are available, there are two variants, depending
8247upon how the code in @file{crtstuff.c} is called. On systems that
8248support a @dfn{.init} section which is executed at program startup,
8249parts of @file{crtstuff.c} are compiled into that section. The
8250program is linked by the @command{gcc} driver like this:
8251
8252@smallexample
8253ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8254@end smallexample
8255
8256The prologue of a function (@code{__init}) appears in the @code{.init}
8257section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8258for the function @code{__fini} in the @dfn{.fini} section. Normally these
8259files are provided by the operating system or by the GNU C library, but
8260are provided by GCC for a few targets.
8261
8262The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8263compiled from @file{crtstuff.c}. They contain, among other things, code
8264fragments within the @code{.init} and @code{.fini} sections that branch
8265to routines in the @code{.text} section. The linker will pull all parts
8266of a section together, which results in a complete @code{__init} function
8267that invokes the routines we need at startup.
8268
8269To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8270macro properly.
8271
8272If no init section is available, when GCC compiles any function called
8273@code{main} (or more accurately, any function designated as a program
8274entry point by the language front end calling @code{expand_main_function}),
8275it inserts a procedure call to @code{__main} as the first executable code
8276after the function prologue. The @code{__main} function is defined
8277in @file{libgcc2.c} and runs the global constructors.
8278
8279In file formats that don't support arbitrary sections, there are again
8280two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8281and an `a.out' format must be used. In this case,
8282@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8283entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8284and with the address of the void function containing the initialization
8285code as its value. The GNU linker recognizes this as a request to add
8286the value to a @dfn{set}; the values are accumulated, and are eventually
8287placed in the executable as a vector in the format described above, with
8288a leading (ignored) count and a trailing zero element.
8289@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8290section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8291the compilation of @code{main} to call @code{__main} as above, starting
8292the initialization process.
8293
8294The last variant uses neither arbitrary sections nor the GNU linker.
8295This is preferable when you want to do dynamic linking and when using
8296file formats which the GNU linker does not support, such as `ECOFF'@. In
8297this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8298termination functions are recognized simply by their names. This requires
8299an extra program in the linkage step, called @command{collect2}. This program
8300pretends to be the linker, for use with GCC; it does its job by running
8301the ordinary linker, but also arranges to include the vectors of
8302initialization and termination functions. These functions are called
8303via @code{__main} as described above. In order to use this method,
8304@code{use_collect2} must be defined in the target in @file{config.gcc}.
8305
8306@ifinfo
8307The following section describes the specific macros that control and
8308customize the handling of initialization and termination functions.
8309@end ifinfo
8310
8311@node Macros for Initialization
8312@subsection Macros Controlling Initialization Routines
8313
8314Here are the macros that control how the compiler handles initialization
8315and termination functions:
8316
8317@defmac INIT_SECTION_ASM_OP
8318If defined, a C string constant, including spacing, for the assembler
8319operation to identify the following data as initialization code. If not
8320defined, GCC will assume such a section does not exist. When you are
8321using special sections for initialization and termination functions, this
8322macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8323run the initialization functions.
8324@end defmac
8325
8326@defmac HAS_INIT_SECTION
8327If defined, @code{main} will not call @code{__main} as described above.
8328This macro should be defined for systems that control start-up code
8329on a symbol-by-symbol basis, such as OSF/1, and should not
8330be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8331@end defmac
8332
8333@defmac LD_INIT_SWITCH
8334If defined, a C string constant for a switch that tells the linker that
8335the following symbol is an initialization routine.
8336@end defmac
8337
8338@defmac LD_FINI_SWITCH
8339If defined, a C string constant for a switch that tells the linker that
8340the following symbol is a finalization routine.
8341@end defmac
8342
8343@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8344If defined, a C statement that will write a function that can be
8345automatically called when a shared library is loaded. The function
8346should call @var{func}, which takes no arguments. If not defined, and
8347the object format requires an explicit initialization function, then a
8348function called @code{_GLOBAL__DI} will be generated.
8349
8350This function and the following one are used by collect2 when linking a
8351shared library that needs constructors or destructors, or has DWARF2
8352exception tables embedded in the code.
8353@end defmac
8354
8355@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8356If defined, a C statement that will write a function that can be
8357automatically called when a shared library is unloaded. The function
8358should call @var{func}, which takes no arguments. If not defined, and
8359the object format requires an explicit finalization function, then a
8360function called @code{_GLOBAL__DD} will be generated.
8361@end defmac
8362
8363@defmac INVOKE__main
8364If defined, @code{main} will call @code{__main} despite the presence of
8365@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8366where the init section is not actually run automatically, but is still
8367useful for collecting the lists of constructors and destructors.
8368@end defmac
8369
8370@defmac SUPPORTS_INIT_PRIORITY
8371If nonzero, the C++ @code{init_priority} attribute is supported and the
8372compiler should emit instructions to control the order of initialization
8373of objects. If zero, the compiler will issue an error message upon
8374encountering an @code{init_priority} attribute.
8375@end defmac
8376
8377@hook TARGET_HAVE_CTORS_DTORS
8378This value is true if the target supports some ``native'' method of
8379collecting constructors and destructors to be run at startup and exit.
8380It is false if we must use @command{collect2}.
8381@end deftypevr
8382
8383@hook TARGET_ASM_CONSTRUCTOR
8384If defined, a function that outputs assembler code to arrange to call
8385the function referenced by @var{symbol} at initialization time.
8386
8387Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8388no arguments and with no return value. If the target supports initialization
8389priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8390otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8391
8392If this macro is not defined by the target, a suitable default will
8393be chosen if (1) the target supports arbitrary section names, (2) the
8394target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8395is not defined.
8396@end deftypefn
8397
8398@hook TARGET_ASM_DESTRUCTOR
8399This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8400functions rather than initialization functions.
8401@end deftypefn
8402
8403If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8404generated for the generated object file will have static linkage.
8405
8406If your system uses @command{collect2} as the means of processing
8407constructors, then that program normally uses @command{nm} to scan
8408an object file for constructor functions to be called.
8409
8410On certain kinds of systems, you can define this macro to make
8411@command{collect2} work faster (and, in some cases, make it work at all):
8412
8413@defmac OBJECT_FORMAT_COFF
8414Define this macro if the system uses COFF (Common Object File Format)
8415object files, so that @command{collect2} can assume this format and scan
8416object files directly for dynamic constructor/destructor functions.
8417
8418This macro is effective only in a native compiler; @command{collect2} as
8419part of a cross compiler always uses @command{nm} for the target machine.
8420@end defmac
8421
8422@defmac REAL_NM_FILE_NAME
8423Define this macro as a C string constant containing the file name to use
8424to execute @command{nm}. The default is to search the path normally for
8425@command{nm}.
8426
8427If your system supports shared libraries and has a program to list the
8428dynamic dependencies of a given library or executable, you can define
8429these macros to enable support for running initialization and
8430termination functions in shared libraries:
8431@end defmac
8432
8433@defmac LDD_SUFFIX
8434Define this macro to a C string constant containing the name of the program
8435which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
8436@end defmac
8437
8438@defmac PARSE_LDD_OUTPUT (@var{ptr})
8439Define this macro to be C code that extracts filenames from the output
8440of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8441of type @code{char *} that points to the beginning of a line of output
8442from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8443code must advance @var{ptr} to the beginning of the filename on that
8444line. Otherwise, it must set @var{ptr} to @code{NULL}.
8445@end defmac
8446
8447@defmac SHLIB_SUFFIX
8448Define this macro to a C string constant containing the default shared
8449library extension of the target (e.g., @samp{".so"}). @command{collect2}
8450strips version information after this suffix when generating global
8451constructor and destructor names. This define is only needed on targets
8452that use @command{collect2} to process constructors and destructors.
8453@end defmac
8454
8455@node Instruction Output
8456@subsection Output of Assembler Instructions
8457
8458@c prevent bad page break with this line
8459This describes assembler instruction output.
8460
8461@defmac REGISTER_NAMES
8462A C initializer containing the assembler's names for the machine
8463registers, each one as a C string constant. This is what translates
8464register numbers in the compiler into assembler language.
8465@end defmac
8466
8467@defmac ADDITIONAL_REGISTER_NAMES
8468If defined, a C initializer for an array of structures containing a name
8469and a register number. This macro defines additional names for hard
8470registers, thus allowing the @code{asm} option in declarations to refer
8471to registers using alternate names.
8472@end defmac
8473
8474@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8475Define this macro if you are using an unusual assembler that
8476requires different names for the machine instructions.
8477
8478The definition is a C statement or statements which output an
8479assembler instruction opcode to the stdio stream @var{stream}. The
8480macro-operand @var{ptr} is a variable of type @code{char *} which
8481points to the opcode name in its ``internal'' form---the form that is
8482written in the machine description. The definition should output the
8483opcode name to @var{stream}, performing any translation you desire, and
8484increment the variable @var{ptr} to point at the end of the opcode
8485so that it will not be output twice.
8486
8487In fact, your macro definition may process less than the entire opcode
8488name, or more than the opcode name; but if you want to process text
8489that includes @samp{%}-sequences to substitute operands, you must take
8490care of the substitution yourself. Just be sure to increment
8491@var{ptr} over whatever text should not be output normally.
8492
8493@findex recog_data.operand
8494If you need to look at the operand values, they can be found as the
8495elements of @code{recog_data.operand}.
8496
8497If the macro definition does nothing, the instruction is output
8498in the usual way.
8499@end defmac
8500
8501@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8502If defined, a C statement to be executed just prior to the output of
8503assembler code for @var{insn}, to modify the extracted operands so
8504they will be output differently.
8505
8506Here the argument @var{opvec} is the vector containing the operands
8507extracted from @var{insn}, and @var{noperands} is the number of
8508elements of the vector which contain meaningful data for this insn.
8509The contents of this vector are what will be used to convert the insn
8510template into assembler code, so you can change the assembler output
8511by changing the contents of the vector.
8512
8513This macro is useful when various assembler syntaxes share a single
8514file of instruction patterns; by defining this macro differently, you
8515can cause a large class of instructions to be output differently (such
8516as with rearranged operands). Naturally, variations in assembler
8517syntax affecting individual insn patterns ought to be handled by
8518writing conditional output routines in those patterns.
8519
8520If this macro is not defined, it is equivalent to a null statement.
8521@end defmac
8522
8523@hook TARGET_ASM_FINAL_POSTSCAN_INSN
8524If defined, this target hook is a function which is executed just after the
8525output of assembler code for @var{insn}, to change the mode of the assembler
8526if necessary.
8527
8528Here the argument @var{opvec} is the vector containing the operands
8529extracted from @var{insn}, and @var{noperands} is the number of
8530elements of the vector which contain meaningful data for this insn.
8531The contents of this vector are what was used to convert the insn
8532template into assembler code, so you can change the assembler mode
8533by checking the contents of the vector.
8534@end deftypefn
8535
8536@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8537A C compound statement to output to stdio stream @var{stream} the
8538assembler syntax for an instruction operand @var{x}. @var{x} is an
8539RTL expression.
8540
8541@var{code} is a value that can be used to specify one of several ways
8542of printing the operand. It is used when identical operands must be
8543printed differently depending on the context. @var{code} comes from
8544the @samp{%} specification that was used to request printing of the
8545operand. If the specification was just @samp{%@var{digit}} then
8546@var{code} is 0; if the specification was @samp{%@var{ltr}
8547@var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8548
8549@findex reg_names
8550If @var{x} is a register, this macro should print the register's name.
8551The names can be found in an array @code{reg_names} whose type is
8552@code{char *[]}. @code{reg_names} is initialized from
8553@code{REGISTER_NAMES}.
8554
8555When the machine description has a specification @samp{%@var{punct}}
8556(a @samp{%} followed by a punctuation character), this macro is called
8557with a null pointer for @var{x} and the punctuation character for
8558@var{code}.
8559@end defmac
8560
8561@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8562A C expression which evaluates to true if @var{code} is a valid
8563punctuation character for use in the @code{PRINT_OPERAND} macro. If
8564@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8565punctuation characters (except for the standard one, @samp{%}) are used
8566in this way.
8567@end defmac
8568
8569@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8570A C compound statement to output to stdio stream @var{stream} the
8571assembler syntax for an instruction operand that is a memory reference
8572whose address is @var{x}. @var{x} is an RTL expression.
8573
8574@cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8575On some machines, the syntax for a symbolic address depends on the
8576section that the address refers to. On these machines, define the hook
8577@code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8578@code{symbol_ref}, and then check for it here. @xref{Assembler
8579Format}.
8580@end defmac
8581
8582@findex dbr_sequence_length
8583@defmac DBR_OUTPUT_SEQEND (@var{file})
8584A C statement, to be executed after all slot-filler instructions have
8585been output. If necessary, call @code{dbr_sequence_length} to
8586determine the number of slots filled in a sequence (zero if not
8587currently outputting a sequence), to decide how many no-ops to output,
8588or whatever.
8589
8590Don't define this macro if it has nothing to do, but it is helpful in
8591reading assembly output if the extent of the delay sequence is made
8592explicit (e.g.@: with white space).
8593@end defmac
8594
8595@findex final_sequence
8596Note that output routines for instructions with delay slots must be
8597prepared to deal with not being output as part of a sequence
8598(i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8599found.) The variable @code{final_sequence} is null when not
8600processing a sequence, otherwise it contains the @code{sequence} rtx
8601being output.
8602
8603@findex asm_fprintf
8604@defmac REGISTER_PREFIX
8605@defmacx LOCAL_LABEL_PREFIX
8606@defmacx USER_LABEL_PREFIX
8607@defmacx IMMEDIATE_PREFIX
8608If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8609@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8610@file{final.c}). These are useful when a single @file{md} file must
8611support multiple assembler formats. In that case, the various @file{tm.h}
8612files can define these macros differently.
8613@end defmac
8614
8615@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8616If defined this macro should expand to a series of @code{case}
8617statements which will be parsed inside the @code{switch} statement of
8618the @code{asm_fprintf} function. This allows targets to define extra
8619printf formats which may useful when generating their assembler
8620statements. Note that uppercase letters are reserved for future
8621generic extensions to asm_fprintf, and so are not available to target
8622specific code. The output file is given by the parameter @var{file}.
8623The varargs input pointer is @var{argptr} and the rest of the format
8624string, starting the character after the one that is being switched
8625upon, is pointed to by @var{format}.
8626@end defmac
8627
8628@defmac ASSEMBLER_DIALECT
8629If your target supports multiple dialects of assembler language (such as
8630different opcodes), define this macro as a C expression that gives the
8631numeric index of the assembler language dialect to use, with zero as the
8632first variant.
8633
8634If this macro is defined, you may use constructs of the form
8635@smallexample
8636@samp{@{option0|option1|option2@dots{}@}}
8637@end smallexample
8638@noindent
8639in the output templates of patterns (@pxref{Output Template}) or in the
8640first argument of @code{asm_fprintf}. This construct outputs
8641@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8642@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8643within these strings retain their usual meaning. If there are fewer
8644alternatives within the braces than the value of
8645@code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8646
8647If you do not define this macro, the characters @samp{@{}, @samp{|} and
8648@samp{@}} do not have any special meaning when used in templates or
8649operands to @code{asm_fprintf}.
8650
8651Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8652@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8653the variations in assembler language syntax with that mechanism. Define
8654@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8655if the syntax variant are larger and involve such things as different
8656opcodes or operand order.
8657@end defmac
8658
8659@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8660A C expression to output to @var{stream} some assembler code
8661which will push hard register number @var{regno} onto the stack.
8662The code need not be optimal, since this macro is used only when
8663profiling.
8664@end defmac
8665
8666@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8667A C expression to output to @var{stream} some assembler code
8668which will pop hard register number @var{regno} off of the stack.
8669The code need not be optimal, since this macro is used only when
8670profiling.
8671@end defmac
8672
8673@node Dispatch Tables
8674@subsection Output of Dispatch Tables
8675
8676@c prevent bad page break with this line
8677This concerns dispatch tables.
8678
8679@cindex dispatch table
8680@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8681A C statement to output to the stdio stream @var{stream} an assembler
8682pseudo-instruction to generate a difference between two labels.
8683@var{value} and @var{rel} are the numbers of two internal labels. The
8684definitions of these labels are output using
8685@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8686way here. For example,
8687
8688@smallexample
8689fprintf (@var{stream}, "\t.word L%d-L%d\n",
8690 @var{value}, @var{rel})
8691@end smallexample
8692
8693You must provide this macro on machines where the addresses in a
8694dispatch table are relative to the table's own address. If defined, GCC
8695will also use this macro on all machines when producing PIC@.
8696@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8697mode and flags can be read.
8698@end defmac
8699
8700@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8701This macro should be provided on machines where the addresses
8702in a dispatch table are absolute.
8703
8704The definition should be a C statement to output to the stdio stream
8705@var{stream} an assembler pseudo-instruction to generate a reference to
8706a label. @var{value} is the number of an internal label whose
8707definition is output using @code{(*targetm.asm_out.internal_label)}.
8708For example,
8709
8710@smallexample
8711fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8712@end smallexample
8713@end defmac
8714
8715@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8716Define this if the label before a jump-table needs to be output
8717specially. The first three arguments are the same as for
8718@code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8719jump-table which follows (a @code{jump_insn} containing an
8720@code{addr_vec} or @code{addr_diff_vec}).
8721
8722This feature is used on system V to output a @code{swbeg} statement
8723for the table.
8724
8725If this macro is not defined, these labels are output with
8726@code{(*targetm.asm_out.internal_label)}.
8727@end defmac
8728
8729@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8730Define this if something special must be output at the end of a
8731jump-table. The definition should be a C statement to be executed
8732after the assembler code for the table is written. It should write
8733the appropriate code to stdio stream @var{stream}. The argument
8734@var{table} is the jump-table insn, and @var{num} is the label-number
8735of the preceding label.
8736
8737If this macro is not defined, nothing special is output at the end of
8738the jump-table.
8739@end defmac
8740
8741@hook TARGET_ASM_EMIT_UNWIND_LABEL
8742This target hook emits a label at the beginning of each FDE@. It
8743should be defined on targets where FDEs need special labels, and it
8744should write the appropriate label, for the FDE associated with the
8745function declaration @var{decl}, to the stdio stream @var{stream}.
8746The third argument, @var{for_eh}, is a boolean: true if this is for an
8747exception table. The fourth argument, @var{empty}, is a boolean:
8748true if this is a placeholder label for an omitted FDE@.
8749
8750The default is that FDEs are not given nonlocal labels.
8751@end deftypefn
8752
8753@hook TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL
8754This target hook emits a label at the beginning of the exception table.
8755It should be defined on targets where it is desirable for the table
8756to be broken up according to function.
8757
8758The default is that no label is emitted.
8759@end deftypefn
8760
a68b5e52
RH
8761@hook TARGET_ASM_EMIT_EXCEPT_PERSONALITY
8762
38f8b050
JR
8763@hook TARGET_ASM_UNWIND_EMIT
8764This target hook emits assembly directives required to unwind the
f0a0390e
RH
8765given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8766returns @code{UI_TARGET}.
38f8b050
JR
8767@end deftypefn
8768
3bc6b3e6
RH
8769@hook TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8770
38f8b050
JR
8771@node Exception Region Output
8772@subsection Assembler Commands for Exception Regions
8773
8774@c prevent bad page break with this line
8775
8776This describes commands marking the start and the end of an exception
8777region.
8778
8779@defmac EH_FRAME_SECTION_NAME
8780If defined, a C string constant for the name of the section containing
8781exception handling frame unwind information. If not defined, GCC will
8782provide a default definition if the target supports named sections.
8783@file{crtstuff.c} uses this macro to switch to the appropriate section.
8784
8785You should define this symbol if your target supports DWARF 2 frame
8786unwind information and the default definition does not work.
8787@end defmac
8788
8789@defmac EH_FRAME_IN_DATA_SECTION
8790If defined, DWARF 2 frame unwind information will be placed in the
8791data section even though the target supports named sections. This
8792might be necessary, for instance, if the system linker does garbage
8793collection and sections cannot be marked as not to be collected.
8794
8795Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8796also defined.
8797@end defmac
8798
8799@defmac EH_TABLES_CAN_BE_READ_ONLY
8800Define this macro to 1 if your target is such that no frame unwind
8801information encoding used with non-PIC code will ever require a
8802runtime relocation, but the linker may not support merging read-only
8803and read-write sections into a single read-write section.
8804@end defmac
8805
8806@defmac MASK_RETURN_ADDR
8807An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8808that it does not contain any extraneous set bits in it.
8809@end defmac
8810
8811@defmac DWARF2_UNWIND_INFO
8812Define this macro to 0 if your target supports DWARF 2 frame unwind
8813information, but it does not yet work with exception handling.
8814Otherwise, if your target supports this information (if it defines
f0a0390e
RH
8815@code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP}
8816or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8817@end defmac
38f8b050 8818
f0a0390e
RH
8819@hook TARGET_EXCEPT_UNWIND_INFO
8820This hook defines the mechanism that will be used for exception handling
8821by the target. If the target has ABI specified unwind tables, the hook
8822should return @code{UI_TARGET}. If the target is to use the
8823@code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8824should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8825information, the hook should return @code{UI_DWARF2}.
38f8b050 8826
f0a0390e
RH
8827A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8828This may end up simplifying other parts of target-specific code. The
8829default implementation of this hook never returns @code{UI_NONE}.
38f8b050 8830
f0a0390e
RH
8831Note that the value returned by this hook should be constant. It should
8832not depend on anything except command-line switches. In particular, the
8833setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8834macros and builtin functions related to exception handling are set up
8835depending on this setting.
8836
8837The default implementation of the hook first honors the
8838@option{--enable-sjlj-exceptions} configure option, then
8839@code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}.
8840@end deftypefn
38f8b050
JR
8841
8842@hook TARGET_UNWIND_TABLES_DEFAULT
8843This variable should be set to @code{true} if the target ABI requires unwinding
8844tables even when exceptions are not used.
8845@end deftypevr
8846
8847@defmac MUST_USE_SJLJ_EXCEPTIONS
8848This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
8849runtime-variable. In that case, @file{except.h} cannot correctly
8850determine the corresponding definition of @code{MUST_USE_SJLJ_EXCEPTIONS},
8851so the target must provide it directly.
8852@end defmac
8853
8854@defmac DONT_USE_BUILTIN_SETJMP
8855Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8856should use the @code{setjmp}/@code{longjmp} functions from the C library
8857instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8858@end defmac
8859
8860@defmac DWARF_CIE_DATA_ALIGNMENT
8861This macro need only be defined if the target might save registers in the
8862function prologue at an offset to the stack pointer that is not aligned to
8863@code{UNITS_PER_WORD}. The definition should be the negative minimum
8864alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8865minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8866the target supports DWARF 2 frame unwind information.
8867@end defmac
8868
8869@hook TARGET_TERMINATE_DW2_EH_FRAME_INFO
8870Contains the value true if the target should add a zero word onto the
8871end of a Dwarf-2 frame info section when used for exception handling.
8872Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8873true otherwise.
8874@end deftypevr
8875
8876@hook TARGET_DWARF_REGISTER_SPAN
8877Given a register, this hook should return a parallel of registers to
8878represent where to find the register pieces. Define this hook if the
8879register and its mode are represented in Dwarf in non-contiguous
8880locations, or if the register should be represented in more than one
8881register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8882If not defined, the default is to return @code{NULL_RTX}.
8883@end deftypefn
8884
8885@hook TARGET_INIT_DWARF_REG_SIZES_EXTRA
8886If some registers are represented in Dwarf-2 unwind information in
8887multiple pieces, define this hook to fill in information about the
8888sizes of those pieces in the table used by the unwinder at runtime.
8889It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8890filling in a single size corresponding to each hard register;
8891@var{address} is the address of the table.
8892@end deftypefn
8893
8894@hook TARGET_ASM_TTYPE
8895This hook is used to output a reference from a frame unwinding table to
8896the type_info object identified by @var{sym}. It should return @code{true}
8897if the reference was output. Returning @code{false} will cause the
8898reference to be output using the normal Dwarf2 routines.
8899@end deftypefn
8900
8901@hook TARGET_ARM_EABI_UNWINDER
8902This flag should be set to @code{true} on targets that use an ARM EABI
8903based unwinding library, and @code{false} on other targets. This effects
8904the format of unwinding tables, and how the unwinder in entered after
8905running a cleanup. The default is @code{false}.
8906@end deftypevr
8907
8908@node Alignment Output
8909@subsection Assembler Commands for Alignment
8910
8911@c prevent bad page break with this line
8912This describes commands for alignment.
8913
8914@defmac JUMP_ALIGN (@var{label})
8915The alignment (log base 2) to put in front of @var{label}, which is
8916a common destination of jumps and has no fallthru incoming edge.
8917
8918This macro need not be defined if you don't want any special alignment
8919to be done at such a time. Most machine descriptions do not currently
8920define the macro.
8921
8922Unless it's necessary to inspect the @var{label} parameter, it is better
8923to set the variable @var{align_jumps} in the target's
8924@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8925selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8926@end defmac
8927
ad0c4c36
DD
8928@hook TARGET_ASM_JUMP_ALIGN_MAX_SKIP
8929The maximum number of bytes to skip before @var{label} when applying
8930@code{JUMP_ALIGN}. This works only if
8931@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8932@end deftypefn
8933
38f8b050
JR
8934@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8935The alignment (log base 2) to put in front of @var{label}, which follows
8936a @code{BARRIER}.
8937
8938This macro need not be defined if you don't want any special alignment
8939to be done at such a time. Most machine descriptions do not currently
8940define the macro.
8941@end defmac
8942
ad0c4c36
DD
8943@hook TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8944The maximum number of bytes to skip before @var{label} when applying
38f8b050
JR
8945@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8946@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
ad0c4c36 8947@end deftypefn
38f8b050
JR
8948
8949@defmac LOOP_ALIGN (@var{label})
8950The alignment (log base 2) to put in front of @var{label}, which follows
8951a @code{NOTE_INSN_LOOP_BEG} note.
8952
8953This macro need not be defined if you don't want any special alignment
8954to be done at such a time. Most machine descriptions do not currently
8955define the macro.
8956
8957Unless it's necessary to inspect the @var{label} parameter, it is better
8958to set the variable @code{align_loops} in the target's
8959@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8960selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8961@end defmac
8962
ad0c4c36
DD
8963@hook TARGET_ASM_LOOP_ALIGN_MAX_SKIP
8964The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to
8965@var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is
8966defined.
8967@end deftypefn
38f8b050
JR
8968
8969@defmac LABEL_ALIGN (@var{label})
8970The alignment (log base 2) to put in front of @var{label}.
8971If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8972the maximum of the specified values is used.
8973
8974Unless it's necessary to inspect the @var{label} parameter, it is better
8975to set the variable @code{align_labels} in the target's
8976@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
8977selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8978@end defmac
8979
ad0c4c36
DD
8980@hook TARGET_ASM_LABEL_ALIGN_MAX_SKIP
8981The maximum number of bytes to skip when applying @code{LABEL_ALIGN}
8982to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN}
8983is defined.
8984@end deftypefn
38f8b050
JR
8985
8986@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8987A C statement to output to the stdio stream @var{stream} an assembler
8988instruction to advance the location counter by @var{nbytes} bytes.
8989Those bytes should be zero when loaded. @var{nbytes} will be a C
8990expression of type @code{unsigned HOST_WIDE_INT}.
8991@end defmac
8992
8993@defmac ASM_NO_SKIP_IN_TEXT
8994Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
8995text section because it fails to put zeros in the bytes that are skipped.
8996This is true on many Unix systems, where the pseudo--op to skip bytes
8997produces no-op instructions rather than zeros when used in the text
8998section.
8999@end defmac
9000
9001@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9002A C statement to output to the stdio stream @var{stream} an assembler
9003command to advance the location counter to a multiple of 2 to the
9004@var{power} bytes. @var{power} will be a C expression of type @code{int}.
9005@end defmac
9006
9007@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9008Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9009for padding, if necessary.
9010@end defmac
9011
9012@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9013A C statement to output to the stdio stream @var{stream} an assembler
9014command to advance the location counter to a multiple of 2 to the
9015@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9016satisfy the alignment request. @var{power} and @var{max_skip} will be
9017a C expression of type @code{int}.
9018@end defmac
9019
9020@need 3000
9021@node Debugging Info
9022@section Controlling Debugging Information Format
9023
9024@c prevent bad page break with this line
9025This describes how to specify debugging information.
9026
9027@menu
9028* All Debuggers:: Macros that affect all debugging formats uniformly.
9029* DBX Options:: Macros enabling specific options in DBX format.
9030* DBX Hooks:: Hook macros for varying DBX format.
9031* File Names and DBX:: Macros controlling output of file names in DBX format.
9032* SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
9033* VMS Debug:: Macros for VMS debug format.
9034@end menu
9035
9036@node All Debuggers
9037@subsection Macros Affecting All Debugging Formats
9038
9039@c prevent bad page break with this line
9040These macros affect all debugging formats.
9041
9042@defmac DBX_REGISTER_NUMBER (@var{regno})
9043A C expression that returns the DBX register number for the compiler
9044register number @var{regno}. In the default macro provided, the value
9045of this expression will be @var{regno} itself. But sometimes there are
9046some registers that the compiler knows about and DBX does not, or vice
9047versa. In such cases, some register may need to have one number in the
9048compiler and another for DBX@.
9049
9050If two registers have consecutive numbers inside GCC, and they can be
9051used as a pair to hold a multiword value, then they @emph{must} have
9052consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9053Otherwise, debuggers will be unable to access such a pair, because they
9054expect register pairs to be consecutive in their own numbering scheme.
9055
9056If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9057does not preserve register pairs, then what you must do instead is
9058redefine the actual register numbering scheme.
9059@end defmac
9060
9061@defmac DEBUGGER_AUTO_OFFSET (@var{x})
9062A C expression that returns the integer offset value for an automatic
9063variable having address @var{x} (an RTL expression). The default
9064computation assumes that @var{x} is based on the frame-pointer and
9065gives the offset from the frame-pointer. This is required for targets
9066that produce debugging output for DBX or COFF-style debugging output
9067for SDB and allow the frame-pointer to be eliminated when the
9068@option{-g} options is used.
9069@end defmac
9070
9071@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9072A C expression that returns the integer offset value for an argument
9073having address @var{x} (an RTL expression). The nominal offset is
9074@var{offset}.
9075@end defmac
9076
9077@defmac PREFERRED_DEBUGGING_TYPE
9078A C expression that returns the type of debugging output GCC should
9079produce when the user specifies just @option{-g}. Define
9080this if you have arranged for GCC to support more than one format of
9081debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9082@code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9083@code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9084
9085When the user specifies @option{-ggdb}, GCC normally also uses the
9086value of this macro to select the debugging output format, but with two
9087exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9088value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9089defined, GCC uses @code{DBX_DEBUG}.
9090
9091The value of this macro only affects the default debugging output; the
9092user can always get a specific type of output by using @option{-gstabs},
9093@option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9094@end defmac
9095
9096@node DBX Options
9097@subsection Specific Options for DBX Output
9098
9099@c prevent bad page break with this line
9100These are specific options for DBX output.
9101
9102@defmac DBX_DEBUGGING_INFO
9103Define this macro if GCC should produce debugging output for DBX
9104in response to the @option{-g} option.
9105@end defmac
9106
9107@defmac XCOFF_DEBUGGING_INFO
9108Define this macro if GCC should produce XCOFF format debugging output
9109in response to the @option{-g} option. This is a variant of DBX format.
9110@end defmac
9111
9112@defmac DEFAULT_GDB_EXTENSIONS
9113Define this macro to control whether GCC should by default generate
9114GDB's extended version of DBX debugging information (assuming DBX-format
9115debugging information is enabled at all). If you don't define the
9116macro, the default is 1: always generate the extended information
9117if there is any occasion to.
9118@end defmac
9119
9120@defmac DEBUG_SYMS_TEXT
9121Define this macro if all @code{.stabs} commands should be output while
9122in the text section.
9123@end defmac
9124
9125@defmac ASM_STABS_OP
9126A C string constant, including spacing, naming the assembler pseudo op to
9127use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9128If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9129applies only to DBX debugging information format.
9130@end defmac
9131
9132@defmac ASM_STABD_OP
9133A C string constant, including spacing, naming the assembler pseudo op to
9134use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9135value is the current location. If you don't define this macro,
9136@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9137information format.
9138@end defmac
9139
9140@defmac ASM_STABN_OP
9141A C string constant, including spacing, naming the assembler pseudo op to
9142use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9143name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9144macro applies only to DBX debugging information format.
9145@end defmac
9146
9147@defmac DBX_NO_XREFS
9148Define this macro if DBX on your system does not support the construct
9149@samp{xs@var{tagname}}. On some systems, this construct is used to
9150describe a forward reference to a structure named @var{tagname}.
9151On other systems, this construct is not supported at all.
9152@end defmac
9153
9154@defmac DBX_CONTIN_LENGTH
9155A symbol name in DBX-format debugging information is normally
9156continued (split into two separate @code{.stabs} directives) when it
9157exceeds a certain length (by default, 80 characters). On some
9158operating systems, DBX requires this splitting; on others, splitting
9159must not be done. You can inhibit splitting by defining this macro
9160with the value zero. You can override the default splitting-length by
9161defining this macro as an expression for the length you desire.
9162@end defmac
9163
9164@defmac DBX_CONTIN_CHAR
9165Normally continuation is indicated by adding a @samp{\} character to
9166the end of a @code{.stabs} string when a continuation follows. To use
9167a different character instead, define this macro as a character
9168constant for the character you want to use. Do not define this macro
9169if backslash is correct for your system.
9170@end defmac
9171
9172@defmac DBX_STATIC_STAB_DATA_SECTION
9173Define this macro if it is necessary to go to the data section before
9174outputting the @samp{.stabs} pseudo-op for a non-global static
9175variable.
9176@end defmac
9177
9178@defmac DBX_TYPE_DECL_STABS_CODE
9179The value to use in the ``code'' field of the @code{.stabs} directive
9180for a typedef. The default is @code{N_LSYM}.
9181@end defmac
9182
9183@defmac DBX_STATIC_CONST_VAR_CODE
9184The value to use in the ``code'' field of the @code{.stabs} directive
9185for a static variable located in the text section. DBX format does not
9186provide any ``right'' way to do this. The default is @code{N_FUN}.
9187@end defmac
9188
9189@defmac DBX_REGPARM_STABS_CODE
9190The value to use in the ``code'' field of the @code{.stabs} directive
9191for a parameter passed in registers. DBX format does not provide any
9192``right'' way to do this. The default is @code{N_RSYM}.
9193@end defmac
9194
9195@defmac DBX_REGPARM_STABS_LETTER
9196The letter to use in DBX symbol data to identify a symbol as a parameter
9197passed in registers. DBX format does not customarily provide any way to
9198do this. The default is @code{'P'}.
9199@end defmac
9200
9201@defmac DBX_FUNCTION_FIRST
9202Define this macro if the DBX information for a function and its
9203arguments should precede the assembler code for the function. Normally,
9204in DBX format, the debugging information entirely follows the assembler
9205code.
9206@end defmac
9207
9208@defmac DBX_BLOCKS_FUNCTION_RELATIVE
9209Define this macro, with value 1, if the value of a symbol describing
9210the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9211relative to the start of the enclosing function. Normally, GCC uses
9212an absolute address.
9213@end defmac
9214
9215@defmac DBX_LINES_FUNCTION_RELATIVE
9216Define this macro, with value 1, if the value of a symbol indicating
9217the current line number (@code{N_SLINE}) should be relative to the
9218start of the enclosing function. Normally, GCC uses an absolute address.
9219@end defmac
9220
9221@defmac DBX_USE_BINCL
9222Define this macro if GCC should generate @code{N_BINCL} and
9223@code{N_EINCL} stabs for included header files, as on Sun systems. This
9224macro also directs GCC to output a type number as a pair of a file
9225number and a type number within the file. Normally, GCC does not
9226generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9227number for a type number.
9228@end defmac
9229
9230@node DBX Hooks
9231@subsection Open-Ended Hooks for DBX Format
9232
9233@c prevent bad page break with this line
9234These are hooks for DBX format.
9235
9236@defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
9237Define this macro to say how to output to @var{stream} the debugging
9238information for the start of a scope level for variable names. The
9239argument @var{name} is the name of an assembler symbol (for use with
9240@code{assemble_name}) whose value is the address where the scope begins.
9241@end defmac
9242
9243@defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
9244Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
9245@end defmac
9246
9247@defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
9248Define this macro if the target machine requires special handling to
9249output an @code{N_FUN} entry for the function @var{decl}.
9250@end defmac
9251
9252@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9253A C statement to output DBX debugging information before code for line
9254number @var{line} of the current source file to the stdio stream
9255@var{stream}. @var{counter} is the number of time the macro was
9256invoked, including the current invocation; it is intended to generate
9257unique labels in the assembly output.
9258
9259This macro should not be defined if the default output is correct, or
9260if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9261@end defmac
9262
9263@defmac NO_DBX_FUNCTION_END
9264Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9265@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9266On those machines, define this macro to turn this feature off without
9267disturbing the rest of the gdb extensions.
9268@end defmac
9269
9270@defmac NO_DBX_BNSYM_ENSYM
9271Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9272extension construct. On those machines, define this macro to turn this
9273feature off without disturbing the rest of the gdb extensions.
9274@end defmac
9275
9276@node File Names and DBX
9277@subsection File Names in DBX Format
9278
9279@c prevent bad page break with this line
9280This describes file names in DBX format.
9281
9282@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9283A C statement to output DBX debugging information to the stdio stream
9284@var{stream}, which indicates that file @var{name} is the main source
9285file---the file specified as the input file for compilation.
9286This macro is called only once, at the beginning of compilation.
9287
9288This macro need not be defined if the standard form of output
9289for DBX debugging information is appropriate.
9290
9291It may be necessary to refer to a label equal to the beginning of the
9292text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9293to do so. If you do this, you must also set the variable
9294@var{used_ltext_label_name} to @code{true}.
9295@end defmac
9296
9297@defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9298Define this macro, with value 1, if GCC should not emit an indication
9299of the current directory for compilation and current source language at
9300the beginning of the file.
9301@end defmac
9302
9303@defmac NO_DBX_GCC_MARKER
9304Define this macro, with value 1, if GCC should not emit an indication
9305that this object file was compiled by GCC@. The default is to emit
9306an @code{N_OPT} stab at the beginning of every source file, with
9307@samp{gcc2_compiled.} for the string and value 0.
9308@end defmac
9309
9310@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9311A C statement to output DBX debugging information at the end of
9312compilation of the main source file @var{name}. Output should be
9313written to the stdio stream @var{stream}.
9314
9315If you don't define this macro, nothing special is output at the end
9316of compilation, which is correct for most machines.
9317@end defmac
9318
9319@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9320Define this macro @emph{instead of} defining
9321@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9322the end of compilation is an @code{N_SO} stab with an empty string,
9323whose value is the highest absolute text address in the file.
9324@end defmac
9325
9326@need 2000
9327@node SDB and DWARF
9328@subsection Macros for SDB and DWARF Output
9329
9330@c prevent bad page break with this line
9331Here are macros for SDB and DWARF output.
9332
9333@defmac SDB_DEBUGGING_INFO
9334Define this macro if GCC should produce COFF-style debugging output
9335for SDB in response to the @option{-g} option.
9336@end defmac
9337
9338@defmac DWARF2_DEBUGGING_INFO
9339Define this macro if GCC should produce dwarf version 2 format
9340debugging output in response to the @option{-g} option.
9341
9342@hook TARGET_DWARF_CALLING_CONVENTION
9343Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9344be emitted for each function. Instead of an integer return the enum
9345value for the @code{DW_CC_} tag.
9346@end deftypefn
9347
9348To support optional call frame debugging information, you must also
9349define @code{INCOMING_RETURN_ADDR_RTX} and either set
9350@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9351prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9352as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9353@end defmac
9354
9355@defmac DWARF2_FRAME_INFO
9356Define this macro to a nonzero value if GCC should always output
f0a0390e
RH
9357Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9358(@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9359exceptions are enabled, GCC will output this information not matter
9360how you define @code{DWARF2_FRAME_INFO}.
38f8b050
JR
9361@end defmac
9362
f0a0390e
RH
9363@hook TARGET_DEBUG_UNWIND_INFO
9364This hook defines the mechanism that will be used for describing frame
9365unwind information to the debugger. Normally the hook will return
9366@code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9367return @code{UI_NONE} otherwise.
9368
9369A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9370is disabled in order to always output DWARF 2 frame information.
9371
9372A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9373This will suppress generation of the normal debug frame unwind information.
9374@end deftypefn
9375
38f8b050
JR
9376@defmac DWARF2_ASM_LINE_DEBUG_INFO
9377Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9378line debug info sections. This will result in much more compact line number
9379tables, and hence is desirable if it works.
9380@end defmac
9381
9730bc27
TT
9382@hook TARGET_WANT_DEBUG_PUB_SECTIONS
9383
38f8b050
JR
9384@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9385A C statement to issue assembly directives that create a difference
9386@var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9387@end defmac
9388
9389@defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9390A C statement to issue assembly directives that create a difference
9391between the two given labels in system defined units, e.g. instruction
9392slots on IA64 VMS, using an integer of the given size.
9393@end defmac
9394
9395@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9396A C statement to issue assembly directives that create a
9397section-relative reference to the given @var{label}, using an integer of the
9398given @var{size}. The label is known to be defined in the given @var{section}.
9399@end defmac
9400
9401@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9402A C statement to issue assembly directives that create a self-relative
9403reference to the given @var{label}, using an integer of the given @var{size}.
9404@end defmac
9405
9406@defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9407A C statement to issue assembly directives that create a reference to
9408the DWARF table identifier @var{label} from the current section. This
9409is used on some systems to avoid garbage collecting a DWARF table which
9410is referenced by a function.
9411@end defmac
9412
9413@hook TARGET_ASM_OUTPUT_DWARF_DTPREL
9414If defined, this target hook is a function which outputs a DTP-relative
9415reference to the given TLS symbol of the specified size.
9416@end deftypefn
9417
9418@defmac PUT_SDB_@dots{}
9419Define these macros to override the assembler syntax for the special
9420SDB assembler directives. See @file{sdbout.c} for a list of these
9421macros and their arguments. If the standard syntax is used, you need
9422not define them yourself.
9423@end defmac
9424
9425@defmac SDB_DELIM
9426Some assemblers do not support a semicolon as a delimiter, even between
9427SDB assembler directives. In that case, define this macro to be the
9428delimiter to use (usually @samp{\n}). It is not necessary to define
9429a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9430required.
9431@end defmac
9432
9433@defmac SDB_ALLOW_UNKNOWN_REFERENCES
9434Define this macro to allow references to unknown structure,
9435union, or enumeration tags to be emitted. Standard COFF does not
9436allow handling of unknown references, MIPS ECOFF has support for
9437it.
9438@end defmac
9439
9440@defmac SDB_ALLOW_FORWARD_REFERENCES
9441Define this macro to allow references to structure, union, or
9442enumeration tags that have not yet been seen to be handled. Some
9443assemblers choke if forward tags are used, while some require it.
9444@end defmac
9445
9446@defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9447A C statement to output SDB debugging information before code for line
9448number @var{line} of the current source file to the stdio stream
9449@var{stream}. The default is to emit an @code{.ln} directive.
9450@end defmac
9451
9452@need 2000
9453@node VMS Debug
9454@subsection Macros for VMS Debug Format
9455
9456@c prevent bad page break with this line
9457Here are macros for VMS debug format.
9458
9459@defmac VMS_DEBUGGING_INFO
9460Define this macro if GCC should produce debugging output for VMS
9461in response to the @option{-g} option. The default behavior for VMS
9462is to generate minimal debug info for a traceback in the absence of
9463@option{-g} unless explicitly overridden with @option{-g0}. This
fac0f722 9464behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
38f8b050
JR
9465@code{TARGET_OPTION_OVERRIDE}.
9466@end defmac
9467
9468@node Floating Point
9469@section Cross Compilation and Floating Point
9470@cindex cross compilation and floating point
9471@cindex floating point and cross compilation
9472
9473While all modern machines use twos-complement representation for integers,
9474there are a variety of representations for floating point numbers. This
9475means that in a cross-compiler the representation of floating point numbers
9476in the compiled program may be different from that used in the machine
9477doing the compilation.
9478
9479Because different representation systems may offer different amounts of
9480range and precision, all floating point constants must be represented in
9481the target machine's format. Therefore, the cross compiler cannot
9482safely use the host machine's floating point arithmetic; it must emulate
9483the target's arithmetic. To ensure consistency, GCC always uses
9484emulation to work with floating point values, even when the host and
9485target floating point formats are identical.
9486
9487The following macros are provided by @file{real.h} for the compiler to
9488use. All parts of the compiler which generate or optimize
9489floating-point calculations must use these macros. They may evaluate
9490their operands more than once, so operands must not have side effects.
9491
9492@defmac REAL_VALUE_TYPE
9493The C data type to be used to hold a floating point value in the target
9494machine's format. Typically this is a @code{struct} containing an
9495array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9496quantity.
9497@end defmac
9498
9499@deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9500Compares for equality the two values, @var{x} and @var{y}. If the target
9501floating point format supports negative zeroes and/or NaNs,
9502@samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9503@samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9504@end deftypefn
9505
9506@deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9507Tests whether @var{x} is less than @var{y}.
9508@end deftypefn
9509
9510@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9511Truncates @var{x} to a signed integer, rounding toward zero.
9512@end deftypefn
9513
9514@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9515Truncates @var{x} to an unsigned integer, rounding toward zero. If
9516@var{x} is negative, returns zero.
9517@end deftypefn
9518
9519@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9520Converts @var{string} into a floating point number in the target machine's
9521representation for mode @var{mode}. This routine can handle both
9522decimal and hexadecimal floating point constants, using the syntax
9523defined by the C language for both.
9524@end deftypefn
9525
9526@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9527Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9528@end deftypefn
9529
9530@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9531Determines whether @var{x} represents infinity (positive or negative).
9532@end deftypefn
9533
9534@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9535Determines whether @var{x} represents a ``NaN'' (not-a-number).
9536@end deftypefn
9537
9538@deftypefn Macro void REAL_ARITHMETIC (REAL_VALUE_TYPE @var{output}, enum tree_code @var{code}, REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9539Calculates an arithmetic operation on the two floating point values
9540@var{x} and @var{y}, storing the result in @var{output} (which must be a
9541variable).
9542
9543The operation to be performed is specified by @var{code}. Only the
9544following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9545@code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9546
9547If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9548target's floating point format cannot represent infinity, it will call
9549@code{abort}. Callers should check for this situation first, using
9550@code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9551@end deftypefn
9552
9553@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9554Returns the negative of the floating point value @var{x}.
9555@end deftypefn
9556
9557@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9558Returns the absolute value of @var{x}.
9559@end deftypefn
9560
9561@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
9562Truncates the floating point value @var{x} to fit in @var{mode}. The
9563return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9564appropriate bit pattern to be output as a floating constant whose
9565precision accords with mode @var{mode}.
9566@end deftypefn
9567
9568@deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9569Converts a floating point value @var{x} into a double-precision integer
9570which is then stored into @var{low} and @var{high}. If the value is not
9571integral, it is truncated.
9572@end deftypefn
9573
9574@deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode})
9575Converts a double-precision integer found in @var{low} and @var{high},
9576into a floating point value which is then stored into @var{x}. The
9577value is truncated to fit in mode @var{mode}.
9578@end deftypefn
9579
9580@node Mode Switching
9581@section Mode Switching Instructions
9582@cindex mode switching
9583The following macros control mode switching optimizations:
9584
9585@defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9586Define this macro if the port needs extra instructions inserted for mode
9587switching in an optimizing compilation.
9588
9589For an example, the SH4 can perform both single and double precision
9590floating point operations, but to perform a single precision operation,
9591the FPSCR PR bit has to be cleared, while for a double precision
9592operation, this bit has to be set. Changing the PR bit requires a general
9593purpose register as a scratch register, hence these FPSCR sets have to
9594be inserted before reload, i.e.@: you can't put this into instruction emitting
9595or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9596
9597You can have multiple entities that are mode-switched, and select at run time
9598which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9599return nonzero for any @var{entity} that needs mode-switching.
9600If you define this macro, you also have to define
9601@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9602@code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9603@code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9604are optional.
9605@end defmac
9606
9607@defmac NUM_MODES_FOR_MODE_SWITCHING
9608If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9609initializer for an array of integers. Each initializer element
9610N refers to an entity that needs mode switching, and specifies the number
9611of different modes that might need to be set for this entity.
9612The position of the initializer in the initializer---starting counting at
9613zero---determines the integer that is used to refer to the mode-switched
9614entity in question.
9615In macros that take mode arguments / yield a mode result, modes are
9616represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9617switch is needed / supplied.
9618@end defmac
9619
9620@defmac MODE_NEEDED (@var{entity}, @var{insn})
9621@var{entity} is an integer specifying a mode-switched entity. If
9622@code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9623return an integer value not larger than the corresponding element in
9624@code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9625be switched into prior to the execution of @var{insn}.
9626@end defmac
9627
9628@defmac MODE_AFTER (@var{mode}, @var{insn})
9629If this macro is defined, it is evaluated for every @var{insn} during
9630mode switching. It determines the mode that an insn results in (if
9631different from the incoming mode).
9632@end defmac
9633
9634@defmac MODE_ENTRY (@var{entity})
9635If this macro is defined, it is evaluated for every @var{entity} that needs
9636mode switching. It should evaluate to an integer, which is a mode that
9637@var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9638is defined then @code{MODE_EXIT} must be defined.
9639@end defmac
9640
9641@defmac MODE_EXIT (@var{entity})
9642If this macro is defined, it is evaluated for every @var{entity} that needs
9643mode switching. It should evaluate to an integer, which is a mode that
9644@var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9645is defined then @code{MODE_ENTRY} must be defined.
9646@end defmac
9647
9648@defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9649This macro specifies the order in which modes for @var{entity} are processed.
96500 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9651lowest. The value of the macro should be an integer designating a mode
9652for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9653(@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9654@code{num_modes_for_mode_switching[@var{entity}] - 1}.
9655@end defmac
9656
9657@defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9658Generate one or more insns to set @var{entity} to @var{mode}.
9659@var{hard_reg_live} is the set of hard registers live at the point where
9660the insn(s) are to be inserted.
9661@end defmac
9662
9663@node Target Attributes
9664@section Defining target-specific uses of @code{__attribute__}
9665@cindex target attributes
9666@cindex machine attributes
9667@cindex attributes, target-specific
9668
9669Target-specific attributes may be defined for functions, data and types.
9670These are described using the following target hooks; they also need to
9671be documented in @file{extend.texi}.
9672
9673@hook TARGET_ATTRIBUTE_TABLE
9674If defined, this target hook points to an array of @samp{struct
9675attribute_spec} (defined in @file{tree.h}) specifying the machine
9676specific attributes for this target and some of the restrictions on the
9677entities to which these attributes are applied and the arguments they
9678take.
9679@end deftypevr
9680
9681@hook TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P
9682If defined, this target hook is a function which returns true if the
9683machine-specific attribute named @var{name} expects an identifier
9684given as its first argument to be passed on as a plain identifier, not
9685subjected to name lookup. If this is not defined, the default is
9686false for all machine-specific attributes.
9687@end deftypefn
9688
9689@hook TARGET_COMP_TYPE_ATTRIBUTES
9690If defined, this target hook is a function which returns zero if the attributes on
9691@var{type1} and @var{type2} are incompatible, one if they are compatible,
9692and two if they are nearly compatible (which causes a warning to be
9693generated). If this is not defined, machine-specific attributes are
9694supposed always to be compatible.
9695@end deftypefn
9696
9697@hook TARGET_SET_DEFAULT_TYPE_ATTRIBUTES
9698If defined, this target hook is a function which assigns default attributes to
9699the newly defined @var{type}.
9700@end deftypefn
9701
9702@hook TARGET_MERGE_TYPE_ATTRIBUTES
9703Define this target hook if the merging of type attributes needs special
9704handling. If defined, the result is a list of the combined
9705@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9706that @code{comptypes} has already been called and returned 1. This
9707function may call @code{merge_attributes} to handle machine-independent
9708merging.
9709@end deftypefn
9710
9711@hook TARGET_MERGE_DECL_ATTRIBUTES
9712Define this target hook if the merging of decl attributes needs special
9713handling. If defined, the result is a list of the combined
9714@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9715@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9716when this is needed are when one attribute overrides another, or when an
9717attribute is nullified by a subsequent definition. This function may
9718call @code{merge_attributes} to handle machine-independent merging.
9719
9720@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9721If the only target-specific handling you require is @samp{dllimport}
9722for Microsoft Windows targets, you should define the macro
9723@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9724will then define a function called
9725@code{merge_dllimport_decl_attributes} which can then be defined as
9726the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9727add @code{handle_dll_attribute} in the attribute table for your port
9728to perform initial processing of the @samp{dllimport} and
9729@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9730@file{i386/i386.c}, for example.
9731@end deftypefn
9732
9733@hook TARGET_VALID_DLLIMPORT_ATTRIBUTE_P
9734
9735@defmac TARGET_DECLSPEC
9736Define this macro to a nonzero value if you want to treat
9737@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9738default, this behavior is enabled only for targets that define
9739@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9740of @code{__declspec} is via a built-in macro, but you should not rely
9741on this implementation detail.
9742@end defmac
9743
9744@hook TARGET_INSERT_ATTRIBUTES
9745Define this target hook if you want to be able to add attributes to a decl
9746when it is being created. This is normally useful for back ends which
9747wish to implement a pragma by using the attributes which correspond to
9748the pragma's effect. The @var{node} argument is the decl which is being
9749created. The @var{attr_ptr} argument is a pointer to the attribute list
9750for this decl. The list itself should not be modified, since it may be
9751shared with other decls, but attributes may be chained on the head of
9752the list and @code{*@var{attr_ptr}} modified to point to the new
9753attributes, or a copy of the list may be made if further changes are
9754needed.
9755@end deftypefn
9756
9757@hook TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P
9758@cindex inlining
9759This target hook returns @code{true} if it is ok to inline @var{fndecl}
9760into the current function, despite its having target-specific
9761attributes, @code{false} otherwise. By default, if a function has a
9762target specific attribute attached to it, it will not be inlined.
9763@end deftypefn
9764
9765@hook TARGET_OPTION_VALID_ATTRIBUTE_P
9766This hook is called to parse the @code{attribute(option("..."))}, and
9767it allows the function to set different target machine compile time
9768options for the current function that might be different than the
9769options specified on the command line. The hook should return
9770@code{true} if the options are valid.
9771
9772The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9773the function declaration to hold a pointer to a target specific
9774@var{struct cl_target_option} structure.
9775@end deftypefn
9776
9777@hook TARGET_OPTION_SAVE
9778This hook is called to save any additional target specific information
9779in the @var{struct cl_target_option} structure for function specific
9780options.
9781@xref{Option file format}.
9782@end deftypefn
9783
9784@hook TARGET_OPTION_RESTORE
9785This hook is called to restore any additional target specific
9786information in the @var{struct cl_target_option} structure for
9787function specific options.
9788@end deftypefn
9789
9790@hook TARGET_OPTION_PRINT
9791This hook is called to print any additional target specific
9792information in the @var{struct cl_target_option} structure for
9793function specific options.
9794@end deftypefn
9795
56cb42ea 9796@hook TARGET_OPTION_PRAGMA_PARSE
38f8b050
JR
9797This target hook parses the options for @code{#pragma GCC option} to
9798set the machine specific options for functions that occur later in the
9799input stream. The options should be the same as handled by the
56cb42ea 9800@code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
38f8b050
JR
9801@end deftypefn
9802
9803@hook TARGET_OPTION_OVERRIDE
9804Sometimes certain combinations of command options do not make sense on
9805a particular target machine. You can override the hook
9806@code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9807once just after all the command options have been parsed.
9808
9809Don't use this hook to turn on various extra optimizations for
fac0f722 9810@option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
38f8b050
JR
9811
9812If you need to do something whenever the optimization level is
9813changed via the optimize attribute or pragma, see
9814@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9815@end deftypefn
9816
9817@hook TARGET_CAN_INLINE_P
9818This target hook returns @code{false} if the @var{caller} function
9819cannot inline @var{callee}, based on target specific information. By
9820default, inlining is not allowed if the callee function has function
9821specific target options and the caller does not use the same options.
9822@end deftypefn
9823
9824@node Emulated TLS
9825@section Emulating TLS
9826@cindex Emulated TLS
9827
9828For targets whose psABI does not provide Thread Local Storage via
9829specific relocations and instruction sequences, an emulation layer is
9830used. A set of target hooks allows this emulation layer to be
9831configured for the requirements of a particular target. For instance
9832the psABI may in fact specify TLS support in terms of an emulation
9833layer.
9834
9835The emulation layer works by creating a control object for every TLS
9836object. To access the TLS object, a lookup function is provided
9837which, when given the address of the control object, will return the
9838address of the current thread's instance of the TLS object.
9839
9840@hook TARGET_EMUTLS_GET_ADDRESS
9841Contains the name of the helper function that uses a TLS control
9842object to locate a TLS instance. The default causes libgcc's
9843emulated TLS helper function to be used.
9844@end deftypevr
9845
9846@hook TARGET_EMUTLS_REGISTER_COMMON
9847Contains the name of the helper function that should be used at
9848program startup to register TLS objects that are implicitly
9849initialized to zero. If this is @code{NULL}, all TLS objects will
9850have explicit initializers. The default causes libgcc's emulated TLS
9851registration function to be used.
9852@end deftypevr
9853
9854@hook TARGET_EMUTLS_VAR_SECTION
9855Contains the name of the section in which TLS control variables should
9856be placed. The default of @code{NULL} allows these to be placed in
9857any section.
9858@end deftypevr
9859
9860@hook TARGET_EMUTLS_TMPL_SECTION
9861Contains the name of the section in which TLS initializers should be
9862placed. The default of @code{NULL} allows these to be placed in any
9863section.
9864@end deftypevr
9865
9866@hook TARGET_EMUTLS_VAR_PREFIX
9867Contains the prefix to be prepended to TLS control variable names.
9868The default of @code{NULL} uses a target-specific prefix.
9869@end deftypevr
9870
9871@hook TARGET_EMUTLS_TMPL_PREFIX
9872Contains the prefix to be prepended to TLS initializer objects. The
9873default of @code{NULL} uses a target-specific prefix.
9874@end deftypevr
9875
9876@hook TARGET_EMUTLS_VAR_FIELDS
9877Specifies a function that generates the FIELD_DECLs for a TLS control
9878object type. @var{type} is the RECORD_TYPE the fields are for and
9879@var{name} should be filled with the structure tag, if the default of
9880@code{__emutls_object} is unsuitable. The default creates a type suitable
9881for libgcc's emulated TLS function.
9882@end deftypefn
9883
9884@hook TARGET_EMUTLS_VAR_INIT
9885Specifies a function that generates the CONSTRUCTOR to initialize a
9886TLS control object. @var{var} is the TLS control object, @var{decl}
9887is the TLS object and @var{tmpl_addr} is the address of the
9888initializer. The default initializes libgcc's emulated TLS control object.
9889@end deftypefn
9890
9891@hook TARGET_EMUTLS_VAR_ALIGN_FIXED
9892Specifies whether the alignment of TLS control variable objects is
9893fixed and should not be increased as some backends may do to optimize
9894single objects. The default is false.
9895@end deftypevr
9896
9897@hook TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9898Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9899may be used to describe emulated TLS control objects.
9900@end deftypevr
9901
9902@node MIPS Coprocessors
9903@section Defining coprocessor specifics for MIPS targets.
9904@cindex MIPS coprocessor-definition macros
9905
9906The MIPS specification allows MIPS implementations to have as many as 4
9907coprocessors, each with as many as 32 private registers. GCC supports
9908accessing these registers and transferring values between the registers
9909and memory using asm-ized variables. For example:
9910
9911@smallexample
9912 register unsigned int cp0count asm ("c0r1");
9913 unsigned int d;
9914
9915 d = cp0count + 3;
9916@end smallexample
9917
9918(``c0r1'' is the default name of register 1 in coprocessor 0; alternate
9919names may be added as described below, or the default names may be
9920overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
9921
9922Coprocessor registers are assumed to be epilogue-used; sets to them will
9923be preserved even if it does not appear that the register is used again
9924later in the function.
9925
9926Another note: according to the MIPS spec, coprocessor 1 (if present) is
9927the FPU@. One accesses COP1 registers through standard mips
9928floating-point support; they are not included in this mechanism.
9929
9930There is one macro used in defining the MIPS coprocessor interface which
9931you may want to override in subtargets; it is described below.
9932
9933@defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
9934A comma-separated list (with leading comma) of pairs describing the
9935alternate names of coprocessor registers. The format of each entry should be
9936@smallexample
9937@{ @var{alternatename}, @var{register_number}@}
9938@end smallexample
9939Default: empty.
9940@end defmac
9941
9942@node PCH Target
9943@section Parameters for Precompiled Header Validity Checking
9944@cindex parameters, precompiled headers
9945
9946@hook TARGET_GET_PCH_VALIDITY
9947This hook returns a pointer to the data needed by
9948@code{TARGET_PCH_VALID_P} and sets
9949@samp{*@var{sz}} to the size of the data in bytes.
9950@end deftypefn
9951
9952@hook TARGET_PCH_VALID_P
9953This hook checks whether the options used to create a PCH file are
9954compatible with the current settings. It returns @code{NULL}
9955if so and a suitable error message if not. Error messages will
9956be presented to the user and must be localized using @samp{_(@var{msg})}.
9957
9958@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
9959when the PCH file was created and @var{sz} is the size of that data in bytes.
9960It's safe to assume that the data was created by the same version of the
9961compiler, so no format checking is needed.
9962
9963The default definition of @code{default_pch_valid_p} should be
9964suitable for most targets.
9965@end deftypefn
9966
9967@hook TARGET_CHECK_PCH_TARGET_FLAGS
9968If this hook is nonnull, the default implementation of
9969@code{TARGET_PCH_VALID_P} will use it to check for compatible values
9970of @code{target_flags}. @var{pch_flags} specifies the value that
9971@code{target_flags} had when the PCH file was created. The return
9972value is the same as for @code{TARGET_PCH_VALID_P}.
9973@end deftypefn
9974
9975@node C++ ABI
9976@section C++ ABI parameters
9977@cindex parameters, c++ abi
9978
9979@hook TARGET_CXX_GUARD_TYPE
9980Define this hook to override the integer type used for guard variables.
9981These are used to implement one-time construction of static objects. The
9982default is long_long_integer_type_node.
9983@end deftypefn
9984
9985@hook TARGET_CXX_GUARD_MASK_BIT
9986This hook determines how guard variables are used. It should return
9987@code{false} (the default) if the first byte should be used. A return value of
9988@code{true} indicates that only the least significant bit should be used.
9989@end deftypefn
9990
9991@hook TARGET_CXX_GET_COOKIE_SIZE
9992This hook returns the size of the cookie to use when allocating an array
9993whose elements have the indicated @var{type}. Assumes that it is already
9994known that a cookie is needed. The default is
9995@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
9996IA64/Generic C++ ABI@.
9997@end deftypefn
9998
9999@hook TARGET_CXX_COOKIE_HAS_SIZE
10000This hook should return @code{true} if the element size should be stored in
10001array cookies. The default is to return @code{false}.
10002@end deftypefn
10003
10004@hook TARGET_CXX_IMPORT_EXPORT_CLASS
10005If defined by a backend this hook allows the decision made to export
10006class @var{type} to be overruled. Upon entry @var{import_export}
10007will contain 1 if the class is going to be exported, @minus{}1 if it is going
10008to be imported and 0 otherwise. This function should return the
10009modified value and perform any other actions necessary to support the
10010backend's targeted operating system.
10011@end deftypefn
10012
10013@hook TARGET_CXX_CDTOR_RETURNS_THIS
10014This hook should return @code{true} if constructors and destructors return
10015the address of the object created/destroyed. The default is to return
10016@code{false}.
10017@end deftypefn
10018
10019@hook TARGET_CXX_KEY_METHOD_MAY_BE_INLINE
10020This hook returns true if the key method for a class (i.e., the method
10021which, if defined in the current translation unit, causes the virtual
10022table to be emitted) may be an inline function. Under the standard
10023Itanium C++ ABI the key method may be an inline function so long as
10024the function is not declared inline in the class definition. Under
10025some variants of the ABI, an inline function can never be the key
10026method. The default is to return @code{true}.
10027@end deftypefn
10028
10029@hook TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY
10030
10031@hook TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT
10032This hook returns true (the default) if virtual tables and other
10033similar implicit class data objects are always COMDAT if they have
10034external linkage. If this hook returns false, then class data for
10035classes whose virtual table will be emitted in only one translation
10036unit will not be COMDAT.
10037@end deftypefn
10038
10039@hook TARGET_CXX_LIBRARY_RTTI_COMDAT
10040This hook returns true (the default) if the RTTI information for
10041the basic types which is defined in the C++ runtime should always
10042be COMDAT, false if it should not be COMDAT.
10043@end deftypefn
10044
10045@hook TARGET_CXX_USE_AEABI_ATEXIT
10046This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10047should be used to register static destructors when @option{-fuse-cxa-atexit}
10048is in effect. The default is to return false to use @code{__cxa_atexit}.
10049@end deftypefn
10050
10051@hook TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT
10052This hook returns true if the target @code{atexit} function can be used
10053in the same manner as @code{__cxa_atexit} to register C++ static
10054destructors. This requires that @code{atexit}-registered functions in
10055shared libraries are run in the correct order when the libraries are
10056unloaded. The default is to return false.
10057@end deftypefn
10058
10059@hook TARGET_CXX_ADJUST_CLASS_AT_DEFINITION
10060
10061@node Named Address Spaces
10062@section Adding support for named address spaces
10063@cindex named address spaces
10064
10065The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10066standards committee, @cite{Programming Languages - C - Extensions to
10067support embedded processors}, specifies a syntax for embedded
10068processors to specify alternate address spaces. You can configure a
10069GCC port to support section 5.1 of the draft report to add support for
10070address spaces other than the default address space. These address
10071spaces are new keywords that are similar to the @code{volatile} and
10072@code{const} type attributes.
10073
10074Pointers to named address spaces can have a different size than
10075pointers to the generic address space.
10076
10077For example, the SPU port uses the @code{__ea} address space to refer
10078to memory in the host processor, rather than memory local to the SPU
10079processor. Access to memory in the @code{__ea} address space involves
10080issuing DMA operations to move data between the host processor and the
10081local processor memory address space. Pointers in the @code{__ea}
10082address space are either 32 bits or 64 bits based on the
10083@option{-mea32} or @option{-mea64} switches (native SPU pointers are
10084always 32 bits).
10085
10086Internally, address spaces are represented as a small integer in the
10087range 0 to 15 with address space 0 being reserved for the generic
10088address space.
10089
10090To register a named address space qualifier keyword with the C front end,
10091the target may call the @code{c_register_addr_space} routine. For example,
10092the SPU port uses the following to declare @code{__ea} as the keyword for
10093named address space #1:
10094@smallexample
10095#define ADDR_SPACE_EA 1
10096c_register_addr_space ("__ea", ADDR_SPACE_EA);
10097@end smallexample
10098
10099@hook TARGET_ADDR_SPACE_POINTER_MODE
10100Define this to return the machine mode to use for pointers to
10101@var{address_space} if the target supports named address spaces.
10102The default version of this hook returns @code{ptr_mode} for the
10103generic address space only.
10104@end deftypefn
10105
10106@hook TARGET_ADDR_SPACE_ADDRESS_MODE
10107Define this to return the machine mode to use for addresses in
10108@var{address_space} if the target supports named address spaces.
10109The default version of this hook returns @code{Pmode} for the
10110generic address space only.
10111@end deftypefn
10112
10113@hook TARGET_ADDR_SPACE_VALID_POINTER_MODE
10114Define this to return nonzero if the port can handle pointers
10115with machine mode @var{mode} to address space @var{as}. This target
10116hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10117except that it includes explicit named address space support. The default
10118version of this hook returns true for the modes returned by either the
10119@code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10120target hooks for the given address space.
10121@end deftypefn
10122
10123@hook TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P
10124Define this to return true if @var{exp} is a valid address for mode
10125@var{mode} in the named address space @var{as}. The @var{strict}
10126parameter says whether strict addressing is in effect after reload has
10127finished. This target hook is the same as the
10128@code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10129explicit named address space support.
10130@end deftypefn
10131
10132@hook TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS
10133Define this to modify an invalid address @var{x} to be a valid address
10134with mode @var{mode} in the named address space @var{as}. This target
10135hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10136except that it includes explicit named address space support.
10137@end deftypefn
10138
10139@hook TARGET_ADDR_SPACE_SUBSET_P
10140Define this to return whether the @var{subset} named address space is
10141contained within the @var{superset} named address space. Pointers to
10142a named address space that is a subset of another named address space
10143will be converted automatically without a cast if used together in
10144arithmetic operations. Pointers to a superset address space can be
10145converted to pointers to a subset address space via explicit casts.
10146@end deftypefn
10147
10148@hook TARGET_ADDR_SPACE_CONVERT
10149Define this to convert the pointer expression represented by the RTL
10150@var{op} with type @var{from_type} that points to a named address
10151space to a new pointer expression with type @var{to_type} that points
10152to a different named address space. When this hook it called, it is
10153guaranteed that one of the two address spaces is a subset of the other,
10154as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10155@end deftypefn
10156
10157@node Misc
10158@section Miscellaneous Parameters
10159@cindex parameters, miscellaneous
10160
10161@c prevent bad page break with this line
10162Here are several miscellaneous parameters.
10163
10164@defmac HAS_LONG_COND_BRANCH
10165Define this boolean macro to indicate whether or not your architecture
10166has conditional branches that can span all of memory. It is used in
10167conjunction with an optimization that partitions hot and cold basic
10168blocks into separate sections of the executable. If this macro is
10169set to false, gcc will convert any conditional branches that attempt
10170to cross between sections into unconditional branches or indirect jumps.
10171@end defmac
10172
10173@defmac HAS_LONG_UNCOND_BRANCH
10174Define this boolean macro to indicate whether or not your architecture
10175has unconditional branches that can span all of memory. It is used in
10176conjunction with an optimization that partitions hot and cold basic
10177blocks into separate sections of the executable. If this macro is
10178set to false, gcc will convert any unconditional branches that attempt
10179to cross between sections into indirect jumps.
10180@end defmac
10181
10182@defmac CASE_VECTOR_MODE
10183An alias for a machine mode name. This is the machine mode that
10184elements of a jump-table should have.
10185@end defmac
10186
10187@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10188Optional: return the preferred mode for an @code{addr_diff_vec}
10189when the minimum and maximum offset are known. If you define this,
10190it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10191To make this work, you also have to define @code{INSN_ALIGN} and
10192make the alignment for @code{addr_diff_vec} explicit.
10193The @var{body} argument is provided so that the offset_unsigned and scale
10194flags can be updated.
10195@end defmac
10196
10197@defmac CASE_VECTOR_PC_RELATIVE
10198Define this macro to be a C expression to indicate when jump-tables
10199should contain relative addresses. You need not define this macro if
10200jump-tables never contain relative addresses, or jump-tables should
10201contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10202is in effect.
10203@end defmac
10204
10205@hook TARGET_CASE_VALUES_THRESHOLD
10206This function return the smallest number of different values for which it
10207is best to use a jump-table instead of a tree of conditional branches.
10208The default is four for machines with a @code{casesi} instruction and
10209five otherwise. This is best for most machines.
10210@end deftypefn
10211
10212@defmac CASE_USE_BIT_TESTS
10213Define this macro to be a C expression to indicate whether C switch
10214statements may be implemented by a sequence of bit tests. This is
10215advantageous on processors that can efficiently implement left shift
10216of 1 by the number of bits held in a register, but inappropriate on
10217targets that would require a loop. By default, this macro returns
10218@code{true} if the target defines an @code{ashlsi3} pattern, and
10219@code{false} otherwise.
10220@end defmac
10221
10222@defmac WORD_REGISTER_OPERATIONS
10223Define this macro if operations between registers with integral mode
10224smaller than a word are always performed on the entire register.
10225Most RISC machines have this property and most CISC machines do not.
10226@end defmac
10227
10228@defmac LOAD_EXTEND_OP (@var{mem_mode})
10229Define this macro to be a C expression indicating when insns that read
10230memory in @var{mem_mode}, an integral mode narrower than a word, set the
10231bits outside of @var{mem_mode} to be either the sign-extension or the
10232zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10233of @var{mem_mode} for which the
10234insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10235@code{UNKNOWN} for other modes.
10236
10237This macro is not called with @var{mem_mode} non-integral or with a width
10238greater than or equal to @code{BITS_PER_WORD}, so you may return any
10239value in this case. Do not define this macro if it would always return
10240@code{UNKNOWN}. On machines where this macro is defined, you will normally
10241define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10242
10243You may return a non-@code{UNKNOWN} value even if for some hard registers
10244the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10245of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10246when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10247integral mode larger than this but not larger than @code{word_mode}.
10248
10249You must return @code{UNKNOWN} if for some hard registers that allow this
10250mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10251@code{word_mode}, but that they can change to another integral mode that
10252is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10253@end defmac
10254
10255@defmac SHORT_IMMEDIATES_SIGN_EXTEND
10256Define this macro if loading short immediate values into registers sign
10257extends.
10258@end defmac
10259
10260@defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
10261Define this macro if the same instructions that convert a floating
10262point number to a signed fixed point number also convert validly to an
10263unsigned one.
10264@end defmac
10265
10266@hook TARGET_MIN_DIVISIONS_FOR_RECIP_MUL
10267When @option{-ffast-math} is in effect, GCC tries to optimize
10268divisions by the same divisor, by turning them into multiplications by
10269the reciprocal. This target hook specifies the minimum number of divisions
10270that should be there for GCC to perform the optimization for a variable
10271of mode @var{mode}. The default implementation returns 3 if the machine
10272has an instruction for the division, and 2 if it does not.
10273@end deftypefn
10274
10275@defmac MOVE_MAX
10276The maximum number of bytes that a single instruction can move quickly
10277between memory and registers or between two memory locations.
10278@end defmac
10279
10280@defmac MAX_MOVE_MAX
10281The maximum number of bytes that a single instruction can move quickly
10282between memory and registers or between two memory locations. If this
10283is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10284constant value that is the largest value that @code{MOVE_MAX} can have
10285at run-time.
10286@end defmac
10287
10288@defmac SHIFT_COUNT_TRUNCATED
10289A C expression that is nonzero if on this machine the number of bits
10290actually used for the count of a shift operation is equal to the number
10291of bits needed to represent the size of the object being shifted. When
10292this macro is nonzero, the compiler will assume that it is safe to omit
10293a sign-extend, zero-extend, and certain bitwise `and' instructions that
10294truncates the count of a shift operation. On machines that have
10295instructions that act on bit-fields at variable positions, which may
10296include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10297also enables deletion of truncations of the values that serve as
10298arguments to bit-field instructions.
10299
10300If both types of instructions truncate the count (for shifts) and
10301position (for bit-field operations), or if no variable-position bit-field
10302instructions exist, you should define this macro.
10303
10304However, on some machines, such as the 80386 and the 680x0, truncation
10305only applies to shift operations and not the (real or pretended)
10306bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10307such machines. Instead, add patterns to the @file{md} file that include
10308the implied truncation of the shift instructions.
10309
10310You need not define this macro if it would always have the value of zero.
10311@end defmac
10312
10313@anchor{TARGET_SHIFT_TRUNCATION_MASK}
10314@hook TARGET_SHIFT_TRUNCATION_MASK
10315This function describes how the standard shift patterns for @var{mode}
10316deal with shifts by negative amounts or by more than the width of the mode.
10317@xref{shift patterns}.
10318
10319On many machines, the shift patterns will apply a mask @var{m} to the
10320shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10321equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10322this is true for mode @var{mode}, the function should return @var{m},
10323otherwise it should return 0. A return value of 0 indicates that no
10324particular behavior is guaranteed.
10325
10326Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10327@emph{not} apply to general shift rtxes; it applies only to instructions
10328that are generated by the named shift patterns.
10329
10330The default implementation of this function returns
10331@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10332and 0 otherwise. This definition is always safe, but if
10333@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10334nevertheless truncate the shift count, you may get better code
10335by overriding it.
10336@end deftypefn
10337
10338@defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10339A C expression which is nonzero if on this machine it is safe to
10340``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10341bits (where @var{outprec} is smaller than @var{inprec}) by merely
10342operating on it as if it had only @var{outprec} bits.
10343
10344On many machines, this expression can be 1.
10345
10346@c rearranged this, removed the phrase "it is reported that". this was
10347@c to fix an overfull hbox. --mew 10feb93
10348When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10349modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10350If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10351such cases may improve things.
10352@end defmac
10353
10354@hook TARGET_MODE_REP_EXTENDED
10355The representation of an integral mode can be such that the values
10356are always extended to a wider integral mode. Return
10357@code{SIGN_EXTEND} if values of @var{mode} are represented in
10358sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10359otherwise. (Currently, none of the targets use zero-extended
10360representation this way so unlike @code{LOAD_EXTEND_OP},
10361@code{TARGET_MODE_REP_EXTENDED} is expected to return either
10362@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10363@var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10364widest integral mode and currently we take advantage of this fact.)
10365
10366Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10367value even if the extension is not performed on certain hard registers
10368as long as for the @code{REGNO_REG_CLASS} of these hard registers
10369@code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10370
10371Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10372describe two related properties. If you define
10373@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10374to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10375extension.
10376
10377In order to enforce the representation of @code{mode},
10378@code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10379@code{mode}.
10380@end deftypefn
10381
10382@defmac STORE_FLAG_VALUE
10383A C expression describing the value returned by a comparison operator
10384with an integral mode and stored by a store-flag instruction
10385(@samp{cstore@var{mode}4}) when the condition is true. This description must
10386apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10387comparison operators whose results have a @code{MODE_INT} mode.
10388
10389A value of 1 or @minus{}1 means that the instruction implementing the
10390comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10391and 0 when the comparison is false. Otherwise, the value indicates
10392which bits of the result are guaranteed to be 1 when the comparison is
10393true. This value is interpreted in the mode of the comparison
10394operation, which is given by the mode of the first operand in the
10395@samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10396@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10397the compiler.
10398
10399If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10400generate code that depends only on the specified bits. It can also
10401replace comparison operators with equivalent operations if they cause
10402the required bits to be set, even if the remaining bits are undefined.
10403For example, on a machine whose comparison operators return an
10404@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10405@samp{0x80000000}, saying that just the sign bit is relevant, the
10406expression
10407
10408@smallexample
10409(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10410@end smallexample
10411
10412@noindent
10413can be converted to
10414
10415@smallexample
10416(ashift:SI @var{x} (const_int @var{n}))
10417@end smallexample
10418
10419@noindent
10420where @var{n} is the appropriate shift count to move the bit being
10421tested into the sign bit.
10422
10423There is no way to describe a machine that always sets the low-order bit
10424for a true value, but does not guarantee the value of any other bits,
10425but we do not know of any machine that has such an instruction. If you
10426are trying to port GCC to such a machine, include an instruction to
10427perform a logical-and of the result with 1 in the pattern for the
10428comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10429
10430Often, a machine will have multiple instructions that obtain a value
10431from a comparison (or the condition codes). Here are rules to guide the
10432choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10433to be used:
10434
10435@itemize @bullet
10436@item
10437Use the shortest sequence that yields a valid definition for
10438@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10439``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10440comparison operators to do so because there may be opportunities to
10441combine the normalization with other operations.
10442
10443@item
10444For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10445slightly preferred on machines with expensive jumps and 1 preferred on
10446other machines.
10447
10448@item
10449As a second choice, choose a value of @samp{0x80000001} if instructions
10450exist that set both the sign and low-order bits but do not define the
10451others.
10452
10453@item
10454Otherwise, use a value of @samp{0x80000000}.
10455@end itemize
10456
10457Many machines can produce both the value chosen for
10458@code{STORE_FLAG_VALUE} and its negation in the same number of
10459instructions. On those machines, you should also define a pattern for
10460those cases, e.g., one matching
10461
10462@smallexample
10463(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10464@end smallexample
10465
10466Some machines can also perform @code{and} or @code{plus} operations on
10467condition code values with less instructions than the corresponding
10468@samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10469machines, define the appropriate patterns. Use the names @code{incscc}
10470and @code{decscc}, respectively, for the patterns which perform
10471@code{plus} or @code{minus} operations on condition code values. See
10472@file{rs6000.md} for some examples. The GNU Superoptizer can be used to
10473find such instruction sequences on other machines.
10474
10475If this macro is not defined, the default value, 1, is used. You need
10476not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10477instructions, or if the value generated by these instructions is 1.
10478@end defmac
10479
10480@defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10481A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10482returned when comparison operators with floating-point results are true.
10483Define this macro on machines that have comparison operations that return
10484floating-point values. If there are no such operations, do not define
10485this macro.
10486@end defmac
10487
10488@defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10489A C expression that gives a rtx representing the nonzero true element
10490for vector comparisons. The returned rtx should be valid for the inner
10491mode of @var{mode} which is guaranteed to be a vector mode. Define
10492this macro on machines that have vector comparison operations that
10493return a vector result. If there are no such operations, do not define
10494this macro. Typically, this macro is defined as @code{const1_rtx} or
10495@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10496the compiler optimizing such vector comparison operations for the
10497given mode.
10498@end defmac
10499
10500@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10501@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10502A C expression that indicates whether the architecture defines a value
10503for @code{clz} or @code{ctz} with a zero operand.
10504A result of @code{0} indicates the value is undefined.
10505If the value is defined for only the RTL expression, the macro should
10506evaluate to @code{1}; if the value applies also to the corresponding optab
10507entry (which is normally the case if it expands directly into
10508the corresponding RTL), then the macro should evaluate to @code{2}.
10509In the cases where the value is defined, @var{value} should be set to
10510this value.
10511
10512If this macro is not defined, the value of @code{clz} or
10513@code{ctz} at zero is assumed to be undefined.
10514
10515This macro must be defined if the target's expansion for @code{ffs}
10516relies on a particular value to get correct results. Otherwise it
10517is not necessary, though it may be used to optimize some corner cases, and
10518to provide a default expansion for the @code{ffs} optab.
10519
10520Note that regardless of this macro the ``definedness'' of @code{clz}
10521and @code{ctz} at zero do @emph{not} extend to the builtin functions
10522visible to the user. Thus one may be free to adjust the value at will
10523to match the target expansion of these operations without fear of
10524breaking the API@.
10525@end defmac
10526
10527@defmac Pmode
10528An alias for the machine mode for pointers. On most machines, define
10529this to be the integer mode corresponding to the width of a hardware
10530pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10531On some machines you must define this to be one of the partial integer
10532modes, such as @code{PSImode}.
10533
10534The width of @code{Pmode} must be at least as large as the value of
10535@code{POINTER_SIZE}. If it is not equal, you must define the macro
10536@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10537to @code{Pmode}.
10538@end defmac
10539
10540@defmac FUNCTION_MODE
10541An alias for the machine mode used for memory references to functions
10542being called, in @code{call} RTL expressions. On most CISC machines,
10543where an instruction can begin at any byte address, this should be
10544@code{QImode}. On most RISC machines, where all instructions have fixed
10545size and alignment, this should be a mode with the same size and alignment
10546as the machine instruction words - typically @code{SImode} or @code{HImode}.
10547@end defmac
10548
10549@defmac STDC_0_IN_SYSTEM_HEADERS
10550In normal operation, the preprocessor expands @code{__STDC__} to the
10551constant 1, to signify that GCC conforms to ISO Standard C@. On some
10552hosts, like Solaris, the system compiler uses a different convention,
10553where @code{__STDC__} is normally 0, but is 1 if the user specifies
10554strict conformance to the C Standard.
10555
10556Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10557convention when processing system header files, but when processing user
10558files @code{__STDC__} will always expand to 1.
10559@end defmac
10560
10561@defmac NO_IMPLICIT_EXTERN_C
10562Define this macro if the system header files support C++ as well as C@.
10563This macro inhibits the usual method of using system header files in
10564C++, which is to pretend that the file's contents are enclosed in
10565@samp{extern "C" @{@dots{}@}}.
10566@end defmac
10567
10568@findex #pragma
10569@findex pragma
10570@defmac REGISTER_TARGET_PRAGMAS ()
10571Define this macro if you want to implement any target-specific pragmas.
10572If defined, it is a C expression which makes a series of calls to
10573@code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10574for each pragma. The macro may also do any
10575setup required for the pragmas.
10576
10577The primary reason to define this macro is to provide compatibility with
10578other compilers for the same target. In general, we discourage
10579definition of target-specific pragmas for GCC@.
10580
10581If the pragma can be implemented by attributes then you should consider
10582defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10583
10584Preprocessor macros that appear on pragma lines are not expanded. All
10585@samp{#pragma} directives that do not match any registered pragma are
10586silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10587@end defmac
10588
10589@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10590@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10591
10592Each call to @code{c_register_pragma} or
10593@code{c_register_pragma_with_expansion} establishes one pragma. The
10594@var{callback} routine will be called when the preprocessor encounters a
10595pragma of the form
10596
10597@smallexample
10598#pragma [@var{space}] @var{name} @dots{}
10599@end smallexample
10600
10601@var{space} is the case-sensitive namespace of the pragma, or
10602@code{NULL} to put the pragma in the global namespace. The callback
10603routine receives @var{pfile} as its first argument, which can be passed
10604on to cpplib's functions if necessary. You can lex tokens after the
10605@var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10606callback will be silently ignored. The end of the line is indicated by
10607a token of type @code{CPP_EOF}. Macro expansion occurs on the
10608arguments of pragmas registered with
10609@code{c_register_pragma_with_expansion} but not on the arguments of
10610pragmas registered with @code{c_register_pragma}.
10611
10612Note that the use of @code{pragma_lex} is specific to the C and C++
10613compilers. It will not work in the Java or Fortran compilers, or any
10614other language compilers for that matter. Thus if @code{pragma_lex} is going
10615to be called from target-specific code, it must only be done so when
10616building the C and C++ compilers. This can be done by defining the
10617variables @code{c_target_objs} and @code{cxx_target_objs} in the
10618target entry in the @file{config.gcc} file. These variables should name
10619the target-specific, language-specific object file which contains the
10620code that uses @code{pragma_lex}. Note it will also be necessary to add a
10621rule to the makefile fragment pointed to by @code{tmake_file} that shows
10622how to build this object file.
10623@end deftypefun
10624
10625@findex #pragma
10626@findex pragma
10627@defmac HANDLE_SYSV_PRAGMA
10628Define this macro (to a value of 1) if you want the System V style
10629pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
10630[=<value>]} to be supported by gcc.
10631
10632The pack pragma specifies the maximum alignment (in bytes) of fields
10633within a structure, in much the same way as the @samp{__aligned__} and
10634@samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
10635the behavior to the default.
10636
10637A subtlety for Microsoft Visual C/C++ style bit-field packing
10638(e.g.@: -mms-bitfields) for targets that support it:
10639When a bit-field is inserted into a packed record, the whole size
10640of the underlying type is used by one or more same-size adjacent
10641bit-fields (that is, if its long:3, 32 bits is used in the record,
10642and any additional adjacent long bit-fields are packed into the same
10643chunk of 32 bits. However, if the size changes, a new field of that
10644size is allocated).
10645
10646If both MS bit-fields and @samp{__attribute__((packed))} are used,
10647the latter will take precedence. If @samp{__attribute__((packed))} is
10648used on a single field when MS bit-fields are in use, it will take
10649precedence for that field, but the alignment of the rest of the structure
10650may affect its placement.
10651
10652The weak pragma only works if @code{SUPPORTS_WEAK} and
10653@code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
10654of specifically named weak labels, optionally with a value.
10655@end defmac
10656
10657@findex #pragma
10658@findex pragma
10659@defmac HANDLE_PRAGMA_PACK_PUSH_POP
10660Define this macro (to a value of 1) if you want to support the Win32
10661style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
10662pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
10663alignment (in bytes) of fields within a structure, in much the same way as
10664the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
10665pack value of zero resets the behavior to the default. Successive
10666invocations of this pragma cause the previous values to be stacked, so
10667that invocations of @samp{#pragma pack(pop)} will return to the previous
10668value.
10669@end defmac
10670
10671@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10672Define this macro, as well as
10673@code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
10674arguments of @samp{#pragma pack}.
10675@end defmac
10676
10677@hook TARGET_HANDLE_PRAGMA_EXTERN_PREFIX
10678
10679@defmac TARGET_DEFAULT_PACK_STRUCT
10680If your target requires a structure packing default other than 0 (meaning
10681the machine default), define this macro to the necessary value (in bytes).
10682This must be a value that would also be valid to use with
10683@samp{#pragma pack()} (that is, a small power of two).
10684@end defmac
10685
10686@defmac DOLLARS_IN_IDENTIFIERS
10687Define this macro to control use of the character @samp{$} in
10688identifier names for the C family of languages. 0 means @samp{$} is
10689not allowed by default; 1 means it is allowed. 1 is the default;
10690there is no need to define this macro in that case.
10691@end defmac
10692
10693@defmac NO_DOLLAR_IN_LABEL
10694Define this macro if the assembler does not accept the character
10695@samp{$} in label names. By default constructors and destructors in
10696G++ have @samp{$} in the identifiers. If this macro is defined,
10697@samp{.} is used instead.
10698@end defmac
10699
10700@defmac NO_DOT_IN_LABEL
10701Define this macro if the assembler does not accept the character
10702@samp{.} in label names. By default constructors and destructors in G++
10703have names that use @samp{.}. If this macro is defined, these names
10704are rewritten to avoid @samp{.}.
10705@end defmac
10706
10707@defmac INSN_SETS_ARE_DELAYED (@var{insn})
10708Define this macro as a C expression that is nonzero if it is safe for the
10709delay slot scheduler to place instructions in the delay slot of @var{insn},
10710even if they appear to use a resource set or clobbered in @var{insn}.
10711@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10712every @code{call_insn} has this behavior. On machines where some @code{insn}
10713or @code{jump_insn} is really a function call and hence has this behavior,
10714you should define this macro.
10715
10716You need not define this macro if it would always return zero.
10717@end defmac
10718
10719@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10720Define this macro as a C expression that is nonzero if it is safe for the
10721delay slot scheduler to place instructions in the delay slot of @var{insn},
10722even if they appear to set or clobber a resource referenced in @var{insn}.
10723@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10724some @code{insn} or @code{jump_insn} is really a function call and its operands
10725are registers whose use is actually in the subroutine it calls, you should
10726define this macro. Doing so allows the delay slot scheduler to move
10727instructions which copy arguments into the argument registers into the delay
10728slot of @var{insn}.
10729
10730You need not define this macro if it would always return zero.
10731@end defmac
10732
10733@defmac MULTIPLE_SYMBOL_SPACES
10734Define this macro as a C expression that is nonzero if, in some cases,
10735global symbols from one translation unit may not be bound to undefined
10736symbols in another translation unit without user intervention. For
10737instance, under Microsoft Windows symbols must be explicitly imported
10738from shared libraries (DLLs).
10739
10740You need not define this macro if it would always evaluate to zero.
10741@end defmac
10742
10743@hook TARGET_MD_ASM_CLOBBERS
10744This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10745any hard regs the port wishes to automatically clobber for an asm.
10746It should return the result of the last @code{tree_cons} used to add a
10747clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10748corresponding parameters to the asm and may be inspected to avoid
10749clobbering a register that is an input or output of the asm. You can use
10750@code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10751for overlap with regards to asm-declared registers.
10752@end deftypefn
10753
10754@defmac MATH_LIBRARY
10755Define this macro as a C string constant for the linker argument to link
d9d16a19
JM
10756in the system math library, minus the initial @samp{"-l"}, or
10757@samp{""} if the target does not have a
38f8b050
JR
10758separate math library.
10759
d9d16a19 10760You need only define this macro if the default of @samp{"m"} is wrong.
38f8b050
JR
10761@end defmac
10762
10763@defmac LIBRARY_PATH_ENV
10764Define this macro as a C string constant for the environment variable that
10765specifies where the linker should look for libraries.
10766
10767You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10768is wrong.
10769@end defmac
10770
10771@defmac TARGET_POSIX_IO
10772Define this macro if the target supports the following POSIX@ file
10773functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10774Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10775to use file locking when exiting a program, which avoids race conditions
10776if the program has forked. It will also create directories at run-time
10777for cross-profiling.
10778@end defmac
10779
10780@defmac MAX_CONDITIONAL_EXECUTE
10781
10782A C expression for the maximum number of instructions to execute via
10783conditional execution instructions instead of a branch. A value of
10784@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
107851 if it does use cc0.
10786@end defmac
10787
10788@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10789Used if the target needs to perform machine-dependent modifications on the
10790conditionals used for turning basic blocks into conditionally executed code.
10791@var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10792contains information about the currently processed blocks. @var{true_expr}
10793and @var{false_expr} are the tests that are used for converting the
10794then-block and the else-block, respectively. Set either @var{true_expr} or
10795@var{false_expr} to a null pointer if the tests cannot be converted.
10796@end defmac
10797
10798@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10799Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10800if-statements into conditions combined by @code{and} and @code{or} operations.
10801@var{bb} contains the basic block that contains the test that is currently
10802being processed and about to be turned into a condition.
10803@end defmac
10804
10805@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10806A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10807be converted to conditional execution format. @var{ce_info} points to
10808a data structure, @code{struct ce_if_block}, which contains information
10809about the currently processed blocks.
10810@end defmac
10811
10812@defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10813A C expression to perform any final machine dependent modifications in
10814converting code to conditional execution. The involved basic blocks
10815can be found in the @code{struct ce_if_block} structure that is pointed
10816to by @var{ce_info}.
10817@end defmac
10818
10819@defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10820A C expression to cancel any machine dependent modifications in
10821converting code to conditional execution. The involved basic blocks
10822can be found in the @code{struct ce_if_block} structure that is pointed
10823to by @var{ce_info}.
10824@end defmac
10825
10826@defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10827A C expression to initialize any extra fields in a @code{struct ce_if_block}
10828structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10829@end defmac
10830
10831@defmac IFCVT_EXTRA_FIELDS
10832If defined, it should expand to a set of field declarations that will be
10833added to the @code{struct ce_if_block} structure. These should be initialized
10834by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10835@end defmac
10836
10837@hook TARGET_MACHINE_DEPENDENT_REORG
10838If non-null, this hook performs a target-specific pass over the
10839instruction stream. The compiler will run it at all optimization levels,
10840just before the point at which it normally does delayed-branch scheduling.
10841
10842The exact purpose of the hook varies from target to target. Some use
10843it to do transformations that are necessary for correctness, such as
10844laying out in-function constant pools or avoiding hardware hazards.
10845Others use it as an opportunity to do some machine-dependent optimizations.
10846
10847You need not implement the hook if it has nothing to do. The default
10848definition is null.
10849@end deftypefn
10850
10851@hook TARGET_INIT_BUILTINS
10852Define this hook if you have any machine-specific built-in functions
10853that need to be defined. It should be a function that performs the
10854necessary setup.
10855
10856Machine specific built-in functions can be useful to expand special machine
10857instructions that would otherwise not normally be generated because
10858they have no equivalent in the source language (for example, SIMD vector
10859instructions or prefetch instructions).
10860
10861To create a built-in function, call the function
10862@code{lang_hooks.builtin_function}
10863which is defined by the language front end. You can use any type nodes set
10864up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
10865only language front ends that use those two functions will call
10866@samp{TARGET_INIT_BUILTINS}.
10867@end deftypefn
10868
10869@hook TARGET_BUILTIN_DECL
10870Define this hook if you have any machine-specific built-in functions
10871that need to be defined. It should be a function that returns the
10872builtin function declaration for the builtin function code @var{code}.
10873If there is no such builtin and it cannot be initialized at this time
10874if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10875If @var{code} is out of range the function should return
10876@code{error_mark_node}.
10877@end deftypefn
10878
10879@hook TARGET_EXPAND_BUILTIN
10880
10881Expand a call to a machine specific built-in function that was set up by
10882@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10883function call; the result should go to @var{target} if that is
10884convenient, and have mode @var{mode} if that is convenient.
10885@var{subtarget} may be used as the target for computing one of
10886@var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10887ignored. This function should return the result of the call to the
10888built-in function.
10889@end deftypefn
10890
d66f5459 10891@hook TARGET_RESOLVE_OVERLOADED_BUILTIN
38f8b050
JR
10892Select a replacement for a machine specific built-in function that
10893was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10894@emph{before} regular type checking, and so allows the target to
10895implement a crude form of function overloading. @var{fndecl} is the
10896declaration of the built-in function. @var{arglist} is the list of
10897arguments passed to the built-in function. The result is a
10898complete expression that implements the operation, usually
10899another @code{CALL_EXPR}.
10900@var{arglist} really has type @samp{VEC(tree,gc)*}
10901@end deftypefn
10902
08914aaa 10903@hook TARGET_FOLD_BUILTIN
38f8b050
JR
10904Fold a call to a machine specific built-in function that was set up by
10905@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10906built-in function. @var{n_args} is the number of arguments passed to
10907the function; the arguments themselves are pointed to by @var{argp}.
10908The result is another tree containing a simplified expression for the
10909call's result. If @var{ignore} is true the value will be ignored.
10910@end deftypefn
10911
10912@hook TARGET_INVALID_WITHIN_DOLOOP
10913
10914Take an instruction in @var{insn} and return NULL if it is valid within a
10915low-overhead loop, otherwise return a string explaining why doloop
10916could not be applied.
10917
10918Many targets use special registers for low-overhead looping. For any
10919instruction that clobbers these this function should return a string indicating
10920the reason why the doloop could not be applied.
10921By default, the RTL loop optimizer does not use a present doloop pattern for
10922loops containing function calls or branch on table instructions.
10923@end deftypefn
10924
10925@defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10926
10927Take a branch insn in @var{branch1} and another in @var{branch2}.
10928Return true if redirecting @var{branch1} to the destination of
10929@var{branch2} is possible.
10930
10931On some targets, branches may have a limited range. Optimizing the
10932filling of delay slots can result in branches being redirected, and this
10933may in turn cause a branch offset to overflow.
10934@end defmac
10935
10936@hook TARGET_COMMUTATIVE_P
10937This target hook returns @code{true} if @var{x} is considered to be commutative.
10938Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10939PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10940of the enclosing rtl, if known, otherwise it is UNKNOWN.
10941@end deftypefn
10942
10943@hook TARGET_ALLOCATE_INITIAL_VALUE
10944
10945When the initial value of a hard register has been copied in a pseudo
10946register, it is often not necessary to actually allocate another register
10947to this pseudo register, because the original hard register or a stack slot
10948it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10949is called at the start of register allocation once for each hard register
10950that had its initial value copied by using
10951@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10952Possible values are @code{NULL_RTX}, if you don't want
10953to do any special allocation, a @code{REG} rtx---that would typically be
10954the hard register itself, if it is known not to be clobbered---or a
10955@code{MEM}.
10956If you are returning a @code{MEM}, this is only a hint for the allocator;
10957it might decide to use another register anyways.
10958You may use @code{current_function_leaf_function} in the hook, functions
10959that use @code{REG_N_SETS}, to determine if the hard
10960register in question will not be clobbered.
10961The default value of this hook is @code{NULL}, which disables any special
10962allocation.
10963@end deftypefn
10964
10965@hook TARGET_UNSPEC_MAY_TRAP_P
10966This target hook returns nonzero if @var{x}, an @code{unspec} or
10967@code{unspec_volatile} operation, might cause a trap. Targets can use
10968this hook to enhance precision of analysis for @code{unspec} and
10969@code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10970to analyze inner elements of @var{x} in which case @var{flags} should be
10971passed along.
10972@end deftypefn
10973
10974@hook TARGET_SET_CURRENT_FUNCTION
10975The compiler invokes this hook whenever it changes its current function
10976context (@code{cfun}). You can define this function if
10977the back end needs to perform any initialization or reset actions on a
10978per-function basis. For example, it may be used to implement function
10979attributes that affect register usage or code generation patterns.
10980The argument @var{decl} is the declaration for the new function context,
10981and may be null to indicate that the compiler has left a function context
10982and is returning to processing at the top level.
10983The default hook function does nothing.
10984
10985GCC sets @code{cfun} to a dummy function context during initialization of
10986some parts of the back end. The hook function is not invoked in this
10987situation; you need not worry about the hook being invoked recursively,
10988or when the back end is in a partially-initialized state.
10989@code{cfun} might be @code{NULL} to indicate processing at top level,
10990outside of any function scope.
10991@end deftypefn
10992
10993@defmac TARGET_OBJECT_SUFFIX
10994Define this macro to be a C string representing the suffix for object
10995files on your target machine. If you do not define this macro, GCC will
10996use @samp{.o} as the suffix for object files.
10997@end defmac
10998
10999@defmac TARGET_EXECUTABLE_SUFFIX
11000Define this macro to be a C string representing the suffix to be
11001automatically added to executable files on your target machine. If you
11002do not define this macro, GCC will use the null string as the suffix for
11003executable files.
11004@end defmac
11005
11006@defmac COLLECT_EXPORT_LIST
11007If defined, @code{collect2} will scan the individual object files
11008specified on its command line and create an export list for the linker.
11009Define this macro for systems like AIX, where the linker discards
11010object files that are not referenced from @code{main} and uses export
11011lists.
11012@end defmac
11013
11014@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
11015Define this macro to a C expression representing a variant of the
11016method call @var{mdecl}, if Java Native Interface (JNI) methods
11017must be invoked differently from other methods on your target.
11018For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
11019the @code{stdcall} calling convention and this macro is then
11020defined as this expression:
11021
11022@smallexample
11023build_type_attribute_variant (@var{mdecl},
11024 build_tree_list
11025 (get_identifier ("stdcall"),
11026 NULL))
11027@end smallexample
11028@end defmac
11029
11030@hook TARGET_CANNOT_MODIFY_JUMPS_P
11031This target hook returns @code{true} past the point in which new jump
11032instructions could be created. On machines that require a register for
11033every jump such as the SHmedia ISA of SH5, this point would typically be
11034reload, so this target hook should be defined to a function such as:
11035
11036@smallexample
11037static bool
11038cannot_modify_jumps_past_reload_p ()
11039@{
11040 return (reload_completed || reload_in_progress);
11041@}
11042@end smallexample
11043@end deftypefn
11044
11045@hook TARGET_BRANCH_TARGET_REGISTER_CLASS
11046This target hook returns a register class for which branch target register
11047optimizations should be applied. All registers in this class should be
11048usable interchangeably. After reload, registers in this class will be
11049re-allocated and loads will be hoisted out of loops and be subjected
11050to inter-block scheduling.
11051@end deftypefn
11052
11053@hook TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED
11054Branch target register optimization will by default exclude callee-saved
11055registers
11056that are not already live during the current function; if this target hook
11057returns true, they will be included. The target code must than make sure
11058that all target registers in the class returned by
11059@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
11060saved. @var{after_prologue_epilogue_gen} indicates if prologues and
11061epilogues have already been generated. Note, even if you only return
11062true when @var{after_prologue_epilogue_gen} is false, you still are likely
11063to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
11064to reserve space for caller-saved target registers.
11065@end deftypefn
11066
11067@hook TARGET_HAVE_CONDITIONAL_EXECUTION
11068This target hook returns true if the target supports conditional execution.
11069This target hook is required only when the target has several different
11070modes and they have different conditional execution capability, such as ARM.
11071@end deftypefn
11072
11073@hook TARGET_LOOP_UNROLL_ADJUST
11074This target hook returns a new value for the number of times @var{loop}
11075should be unrolled. The parameter @var{nunroll} is the number of times
11076the loop is to be unrolled. The parameter @var{loop} is a pointer to
11077the loop, which is going to be checked for unrolling. This target hook
11078is required only when the target has special constraints like maximum
11079number of memory accesses.
11080@end deftypefn
11081
11082@defmac POWI_MAX_MULTS
11083If defined, this macro is interpreted as a signed integer C expression
11084that specifies the maximum number of floating point multiplications
11085that should be emitted when expanding exponentiation by an integer
11086constant inline. When this value is defined, exponentiation requiring
11087more than this number of multiplications is implemented by calling the
11088system library's @code{pow}, @code{powf} or @code{powl} routines.
11089The default value places no upper bound on the multiplication count.
11090@end defmac
11091
11092@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11093This target hook should register any extra include files for the
11094target. The parameter @var{stdinc} indicates if normal include files
11095are present. The parameter @var{sysroot} is the system root directory.
11096The parameter @var{iprefix} is the prefix for the gcc directory.
11097@end deftypefn
11098
11099@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11100This target hook should register any extra include files for the
11101target before any standard headers. The parameter @var{stdinc}
11102indicates if normal include files are present. The parameter
11103@var{sysroot} is the system root directory. The parameter
11104@var{iprefix} is the prefix for the gcc directory.
11105@end deftypefn
11106
11107@deftypefn Macro void TARGET_OPTF (char *@var{path})
11108This target hook should register special include paths for the target.
11109The parameter @var{path} is the include to register. On Darwin
11110systems, this is used for Framework includes, which have semantics
11111that are different from @option{-I}.
11112@end deftypefn
11113
11114@defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11115This target macro returns @code{true} if it is safe to use a local alias
11116for a virtual function @var{fndecl} when constructing thunks,
11117@code{false} otherwise. By default, the macro returns @code{true} for all
11118functions, if a target supports aliases (i.e.@: defines
11119@code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11120@end defmac
11121
11122@defmac TARGET_FORMAT_TYPES
11123If defined, this macro is the name of a global variable containing
11124target-specific format checking information for the @option{-Wformat}
11125option. The default is to have no target-specific format checks.
11126@end defmac
11127
11128@defmac TARGET_N_FORMAT_TYPES
11129If defined, this macro is the number of entries in
11130@code{TARGET_FORMAT_TYPES}.
11131@end defmac
11132
11133@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11134If defined, this macro is the name of a global variable containing
11135target-specific format overrides for the @option{-Wformat} option. The
11136default is to have no target-specific format overrides. If defined,
11137@code{TARGET_FORMAT_TYPES} must be defined, too.
11138@end defmac
11139
11140@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11141If defined, this macro specifies the number of entries in
11142@code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11143@end defmac
11144
11145@defmac TARGET_OVERRIDES_FORMAT_INIT
11146If defined, this macro specifies the optional initialization
11147routine for target specific customizations of the system printf
11148and scanf formatter settings.
11149@end defmac
11150
11151@hook TARGET_RELAXED_ORDERING
11152If set to @code{true}, means that the target's memory model does not
11153guarantee that loads which do not depend on one another will access
11154main memory in the order of the instruction stream; if ordering is
11155important, an explicit memory barrier must be used. This is true of
11156many recent processors which implement a policy of ``relaxed,''
11157``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11158and ia64. The default is @code{false}.
11159@end deftypevr
11160
11161@hook TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN
11162If defined, this macro returns the diagnostic message when it is
11163illegal to pass argument @var{val} to function @var{funcdecl}
11164with prototype @var{typelist}.
11165@end deftypefn
11166
11167@hook TARGET_INVALID_CONVERSION
11168If defined, this macro returns the diagnostic message when it is
11169invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11170if validity should be determined by the front end.
11171@end deftypefn
11172
11173@hook TARGET_INVALID_UNARY_OP
11174If defined, this macro returns the diagnostic message when it is
11175invalid to apply operation @var{op} (where unary plus is denoted by
11176@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11177if validity should be determined by the front end.
11178@end deftypefn
11179
11180@hook TARGET_INVALID_BINARY_OP
11181If defined, this macro returns the diagnostic message when it is
11182invalid to apply operation @var{op} to operands of types @var{type1}
11183and @var{type2}, or @code{NULL} if validity should be determined by
11184the front end.
11185@end deftypefn
11186
11187@hook TARGET_INVALID_PARAMETER_TYPE
11188If defined, this macro returns the diagnostic message when it is
11189invalid for functions to include parameters of type @var{type},
11190or @code{NULL} if validity should be determined by
11191the front end. This is currently used only by the C and C++ front ends.
11192@end deftypefn
11193
11194@hook TARGET_INVALID_RETURN_TYPE
11195If defined, this macro returns the diagnostic message when it is
11196invalid for functions to have return type @var{type},
11197or @code{NULL} if validity should be determined by
11198the front end. This is currently used only by the C and C++ front ends.
11199@end deftypefn
11200
11201@hook TARGET_PROMOTED_TYPE
11202If defined, this target hook returns the type to which values of
11203@var{type} should be promoted when they appear in expressions,
11204analogous to the integer promotions, or @code{NULL_TREE} to use the
11205front end's normal promotion rules. This hook is useful when there are
11206target-specific types with special promotion rules.
11207This is currently used only by the C and C++ front ends.
11208@end deftypefn
11209
11210@hook TARGET_CONVERT_TO_TYPE
11211If defined, this hook returns the result of converting @var{expr} to
11212@var{type}. It should return the converted expression,
11213or @code{NULL_TREE} to apply the front end's normal conversion rules.
11214This hook is useful when there are target-specific types with special
11215conversion rules.
11216This is currently used only by the C and C++ front ends.
11217@end deftypefn
11218
11219@defmac TARGET_USE_JCR_SECTION
11220This macro determines whether to use the JCR section to register Java
11221classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11222SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11223@end defmac
11224
11225@defmac OBJC_JBLEN
11226This macro determines the size of the objective C jump buffer for the
11227NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11228@end defmac
11229
11230@defmac LIBGCC2_UNWIND_ATTRIBUTE
11231Define this macro if any target-specific attributes need to be attached
11232to the functions in @file{libgcc} that provide low-level support for
11233call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11234and the associated definitions of those functions.
11235@end defmac
11236
11237@hook TARGET_UPDATE_STACK_BOUNDARY
11238Define this macro to update the current function stack boundary if
11239necessary.
11240@end deftypefn
11241
11242@hook TARGET_GET_DRAP_RTX
11243This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11244different argument pointer register is needed to access the function's
11245argument list due to stack realignment. Return @code{NULL} if no DRAP
11246is needed.
11247@end deftypefn
11248
11249@hook TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS
11250When optimization is disabled, this hook indicates whether or not
11251arguments should be allocated to stack slots. Normally, GCC allocates
11252stacks slots for arguments when not optimizing in order to make
11253debugging easier. However, when a function is declared with
11254@code{__attribute__((naked))}, there is no stack frame, and the compiler
11255cannot safely move arguments from the registers in which they are passed
11256to the stack. Therefore, this hook should return true in general, but
11257false for naked functions. The default implementation always returns true.
11258@end deftypefn
11259
11260@hook TARGET_CONST_ANCHOR
11261On some architectures it can take multiple instructions to synthesize
11262a constant. If there is another constant already in a register that
11263is close enough in value then it is preferable that the new constant
11264is computed from this register using immediate addition or
11265subtraction. We accomplish this through CSE. Besides the value of
11266the constant we also add a lower and an upper constant anchor to the
11267available expressions. These are then queried when encountering new
11268constants. The anchors are computed by rounding the constant up and
11269down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11270@code{TARGET_CONST_ANCHOR} should be the maximum positive value
11271accepted by immediate-add plus one. We currently assume that the
11272value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11273MIPS, where add-immediate takes a 16-bit signed value,
11274@code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11275is zero, which disables this optimization. @end deftypevr
This page took 1.245151 seconds and 5 git commands to generate.