1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005, 2006 Free Software Foundation, Inc.
3 @c This is part of the GCC manual.
4 @c For copying conditions, see the file gcc.texi.
7 @chapter Target Description Macros and Functions
8 @cindex machine description macros
9 @cindex target description macros
10 @cindex macros, target description
11 @cindex @file{tm.h} macros
13 In addition to the file @file{@var{machine}.md}, a machine description
14 includes a C header file conventionally given the name
15 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
16 The header file defines numerous macros that convey the information
17 about the target machine that does not fit into the scheme of the
18 @file{.md} file. The file @file{tm.h} should be a link to
19 @file{@var{machine}.h}. The header file @file{config.h} includes
20 @file{tm.h} and most compiler source files include @file{config.h}. The
21 source file defines a variable @code{targetm}, which is a structure
22 containing pointers to functions and data relating to the target
23 machine. @file{@var{machine}.c} should also contain their definitions,
24 if they are not defined elsewhere in GCC, and other functions called
25 through the macros defined in the @file{.h} file.
28 * Target Structure:: The @code{targetm} variable.
29 * Driver:: Controlling how the driver runs the compilation passes.
30 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
31 * Per-Function Data:: Defining data structures for per-function information.
32 * Storage Layout:: Defining sizes and alignments of data.
33 * Type Layout:: Defining sizes and properties of basic user data types.
34 * Registers:: Naming and describing the hardware registers.
35 * Register Classes:: Defining the classes of hardware registers.
36 * Old Constraints:: The old way to define machine-specific constraints.
37 * Stack and Calling:: Defining which way the stack grows and by how much.
38 * Varargs:: Defining the varargs macros.
39 * Trampolines:: Code set up at run time to enter a nested function.
40 * Library Calls:: Controlling how library routines are implicitly called.
41 * Addressing Modes:: Defining addressing modes valid for memory operands.
42 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
43 * Condition Code:: Defining how insns update the condition code.
44 * Costs:: Defining relative costs of different operations.
45 * Scheduling:: Adjusting the behavior of the instruction scheduler.
46 * Sections:: Dividing storage into text, data, and other sections.
47 * PIC:: Macros for position independent code.
48 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
49 * Debugging Info:: Defining the format of debugging output.
50 * Floating Point:: Handling floating point for cross-compilers.
51 * Mode Switching:: Insertion of mode-switching instructions.
52 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
53 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
54 * PCH Target:: Validity checking for precompiled headers.
55 * C++ ABI:: Controlling C++ ABI changes.
56 * Misc:: Everything else.
59 @node Target Structure
60 @section The Global @code{targetm} Variable
62 @cindex target functions
64 @deftypevar {struct gcc_target} targetm
65 The target @file{.c} file must define the global @code{targetm} variable
66 which contains pointers to functions and data relating to the target
67 machine. The variable is declared in @file{target.h};
68 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
69 used to initialize the variable, and macros for the default initializers
70 for elements of the structure. The @file{.c} file should override those
71 macros for which the default definition is inappropriate. For example:
74 #include "target-def.h"
76 /* @r{Initialize the GCC target structure.} */
78 #undef TARGET_COMP_TYPE_ATTRIBUTES
79 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
81 struct gcc_target targetm = TARGET_INITIALIZER;
85 Where a macro should be defined in the @file{.c} file in this manner to
86 form part of the @code{targetm} structure, it is documented below as a
87 ``Target Hook'' with a prototype. Many macros will change in future
88 from being defined in the @file{.h} file to being part of the
89 @code{targetm} structure.
92 @section Controlling the Compilation Driver, @file{gcc}
94 @cindex controlling the compilation driver
96 @c prevent bad page break with this line
97 You can control the compilation driver.
99 @defmac SWITCH_TAKES_ARG (@var{char})
100 A C expression which determines whether the option @option{-@var{char}}
101 takes arguments. The value should be the number of arguments that
102 option takes--zero, for many options.
104 By default, this macro is defined as
105 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
106 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
107 wish to add additional options which take arguments. Any redefinition
108 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
112 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
113 A C expression which determines whether the option @option{-@var{name}}
114 takes arguments. The value should be the number of arguments that
115 option takes--zero, for many options. This macro rather than
116 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
118 By default, this macro is defined as
119 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
120 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
121 wish to add additional options which take arguments. Any redefinition
122 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
126 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
127 A C expression which determines whether the option @option{-@var{char}}
128 stops compilation before the generation of an executable. The value is
129 boolean, nonzero if the option does stop an executable from being
130 generated, zero otherwise.
132 By default, this macro is defined as
133 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
134 options properly. You need not define
135 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
136 options which affect the generation of an executable. Any redefinition
137 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
138 for additional options.
141 @defmac SWITCHES_NEED_SPACES
142 A string-valued C expression which enumerates the options for which
143 the linker needs a space between the option and its argument.
145 If this macro is not defined, the default value is @code{""}.
148 @defmac TARGET_OPTION_TRANSLATE_TABLE
149 If defined, a list of pairs of strings, the first of which is a
150 potential command line target to the @file{gcc} driver program, and the
151 second of which is a space-separated (tabs and other whitespace are not
152 supported) list of options with which to replace the first option. The
153 target defining this list is responsible for assuring that the results
154 are valid. Replacement options may not be the @code{--opt} style, they
155 must be the @code{-opt} style. It is the intention of this macro to
156 provide a mechanism for substitution that affects the multilibs chosen,
157 such as one option that enables many options, some of which select
158 multilibs. Example nonsensical definition, where @option{-malt-abi},
159 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
162 #define TARGET_OPTION_TRANSLATE_TABLE \
163 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
164 @{ "-compat", "-EB -malign=4 -mspoo" @}
168 @defmac DRIVER_SELF_SPECS
169 A list of specs for the driver itself. It should be a suitable
170 initializer for an array of strings, with no surrounding braces.
172 The driver applies these specs to its own command line between loading
173 default @file{specs} files (but not command-line specified ones) and
174 choosing the multilib directory or running any subcommands. It
175 applies them in the order given, so each spec can depend on the
176 options added by earlier ones. It is also possible to remove options
177 using @samp{%<@var{option}} in the usual way.
179 This macro can be useful when a port has several interdependent target
180 options. It provides a way of standardizing the command line so
181 that the other specs are easier to write.
183 Do not define this macro if it does not need to do anything.
186 @defmac OPTION_DEFAULT_SPECS
187 A list of specs used to support configure-time default options (i.e.@:
188 @option{--with} options) in the driver. It should be a suitable initializer
189 for an array of structures, each containing two strings, without the
190 outermost pair of surrounding braces.
192 The first item in the pair is the name of the default. This must match
193 the code in @file{config.gcc} for the target. The second item is a spec
194 to apply if a default with this name was specified. The string
195 @samp{%(VALUE)} in the spec will be replaced by the value of the default
196 everywhere it occurs.
198 The driver will apply these specs to its own command line between loading
199 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
200 the same mechanism as @code{DRIVER_SELF_SPECS}.
202 Do not define this macro if it does not need to do anything.
206 A C string constant that tells the GCC driver program options to
207 pass to CPP@. It can also specify how to translate options you
208 give to GCC into options for GCC to pass to the CPP@.
210 Do not define this macro if it does not need to do anything.
213 @defmac CPLUSPLUS_CPP_SPEC
214 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
215 than C@. If you do not define this macro, then the value of
216 @code{CPP_SPEC} (if any) will be used instead.
220 A C string constant that tells the GCC driver program options to
221 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
223 It can also specify how to translate options you give to GCC into options
224 for GCC to pass to front ends.
226 Do not define this macro if it does not need to do anything.
230 A C string constant that tells the GCC driver program options to
231 pass to @code{cc1plus}. It can also specify how to translate options you
232 give to GCC into options for GCC to pass to the @code{cc1plus}.
234 Do not define this macro if it does not need to do anything.
235 Note that everything defined in CC1_SPEC is already passed to
236 @code{cc1plus} so there is no need to duplicate the contents of
237 CC1_SPEC in CC1PLUS_SPEC@.
241 A C string constant that tells the GCC driver program options to
242 pass to the assembler. It can also specify how to translate options
243 you give to GCC into options for GCC to pass to the assembler.
244 See the file @file{sun3.h} for an example of this.
246 Do not define this macro if it does not need to do anything.
249 @defmac ASM_FINAL_SPEC
250 A C string constant that tells the GCC driver program how to
251 run any programs which cleanup after the normal assembler.
252 Normally, this is not needed. See the file @file{mips.h} for
255 Do not define this macro if it does not need to do anything.
258 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
259 Define this macro, with no value, if the driver should give the assembler
260 an argument consisting of a single dash, @option{-}, to instruct it to
261 read from its standard input (which will be a pipe connected to the
262 output of the compiler proper). This argument is given after any
263 @option{-o} option specifying the name of the output file.
265 If you do not define this macro, the assembler is assumed to read its
266 standard input if given no non-option arguments. If your assembler
267 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
268 see @file{mips.h} for instance.
272 A C string constant that tells the GCC driver program options to
273 pass to the linker. It can also specify how to translate options you
274 give to GCC into options for GCC to pass to the linker.
276 Do not define this macro if it does not need to do anything.
280 Another C string constant used much like @code{LINK_SPEC}. The difference
281 between the two is that @code{LIB_SPEC} is used at the end of the
282 command given to the linker.
284 If this macro is not defined, a default is provided that
285 loads the standard C library from the usual place. See @file{gcc.c}.
289 Another C string constant that tells the GCC driver program
290 how and when to place a reference to @file{libgcc.a} into the
291 linker command line. This constant is placed both before and after
292 the value of @code{LIB_SPEC}.
294 If this macro is not defined, the GCC driver provides a default that
295 passes the string @option{-lgcc} to the linker.
298 @defmac REAL_LIBGCC_SPEC
299 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
300 @code{LIBGCC_SPEC} is not directly used by the driver program but is
301 instead modified to refer to different versions of @file{libgcc.a}
302 depending on the values of the command line flags @option{-static},
303 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
304 targets where these modifications are inappropriate, define
305 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
306 driver how to place a reference to @file{libgcc} on the link command
307 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
310 @defmac USE_LD_AS_NEEDED
311 A macro that controls the modifications to @code{LIBGCC_SPEC}
312 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
313 generated that uses --as-needed and the shared libgcc in place of the
314 static exception handler library, when linking without any of
315 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
319 If defined, this C string constant is added to @code{LINK_SPEC}.
320 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
321 the modifications to @code{LIBGCC_SPEC} mentioned in
322 @code{REAL_LIBGCC_SPEC}.
325 @defmac STARTFILE_SPEC
326 Another C string constant used much like @code{LINK_SPEC}. The
327 difference between the two is that @code{STARTFILE_SPEC} is used at
328 the very beginning of the command given to the linker.
330 If this macro is not defined, a default is provided that loads the
331 standard C startup file from the usual place. See @file{gcc.c}.
335 Another C string constant used much like @code{LINK_SPEC}. The
336 difference between the two is that @code{ENDFILE_SPEC} is used at
337 the very end of the command given to the linker.
339 Do not define this macro if it does not need to do anything.
342 @defmac THREAD_MODEL_SPEC
343 GCC @code{-v} will print the thread model GCC was configured to use.
344 However, this doesn't work on platforms that are multilibbed on thread
345 models, such as AIX 4.3. On such platforms, define
346 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
347 blanks that names one of the recognized thread models. @code{%*}, the
348 default value of this macro, will expand to the value of
349 @code{thread_file} set in @file{config.gcc}.
352 @defmac SYSROOT_SUFFIX_SPEC
353 Define this macro to add a suffix to the target sysroot when GCC is
354 configured with a sysroot. This will cause GCC to search for usr/lib,
355 et al, within sysroot+suffix.
358 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
359 Define this macro to add a headers_suffix to the target sysroot when
360 GCC is configured with a sysroot. This will cause GCC to pass the
361 updated sysroot+headers_suffix to CPP, causing it to search for
362 usr/include, et al, within sysroot+headers_suffix.
366 Define this macro to provide additional specifications to put in the
367 @file{specs} file that can be used in various specifications like
370 The definition should be an initializer for an array of structures,
371 containing a string constant, that defines the specification name, and a
372 string constant that provides the specification.
374 Do not define this macro if it does not need to do anything.
376 @code{EXTRA_SPECS} is useful when an architecture contains several
377 related targets, which have various @code{@dots{}_SPECS} which are similar
378 to each other, and the maintainer would like one central place to keep
381 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
382 define either @code{_CALL_SYSV} when the System V calling sequence is
383 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
386 The @file{config/rs6000/rs6000.h} target file defines:
389 #define EXTRA_SPECS \
390 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
392 #define CPP_SYS_DEFAULT ""
395 The @file{config/rs6000/sysv.h} target file defines:
399 "%@{posix: -D_POSIX_SOURCE @} \
400 %@{mcall-sysv: -D_CALL_SYSV @} \
401 %@{!mcall-sysv: %(cpp_sysv_default) @} \
402 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
404 #undef CPP_SYSV_DEFAULT
405 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
408 while the @file{config/rs6000/eabiaix.h} target file defines
409 @code{CPP_SYSV_DEFAULT} as:
412 #undef CPP_SYSV_DEFAULT
413 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
417 @defmac LINK_LIBGCC_SPECIAL_1
418 Define this macro if the driver program should find the library
419 @file{libgcc.a}. If you do not define this macro, the driver program will pass
420 the argument @option{-lgcc} to tell the linker to do the search.
423 @defmac LINK_GCC_C_SEQUENCE_SPEC
424 The sequence in which libgcc and libc are specified to the linker.
425 By default this is @code{%G %L %G}.
428 @defmac LINK_COMMAND_SPEC
429 A C string constant giving the complete command line need to execute the
430 linker. When you do this, you will need to update your port each time a
431 change is made to the link command line within @file{gcc.c}. Therefore,
432 define this macro only if you need to completely redefine the command
433 line for invoking the linker and there is no other way to accomplish
434 the effect you need. Overriding this macro may be avoidable by overriding
435 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
438 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
439 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
440 directories from linking commands. Do not give it a nonzero value if
441 removing duplicate search directories changes the linker's semantics.
444 @defmac MULTILIB_DEFAULTS
445 Define this macro as a C expression for the initializer of an array of
446 string to tell the driver program which options are defaults for this
447 target and thus do not need to be handled specially when using
448 @code{MULTILIB_OPTIONS}.
450 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
451 the target makefile fragment or if none of the options listed in
452 @code{MULTILIB_OPTIONS} are set by default.
453 @xref{Target Fragment}.
456 @defmac RELATIVE_PREFIX_NOT_LINKDIR
457 Define this macro to tell @command{gcc} that it should only translate
458 a @option{-B} prefix into a @option{-L} linker option if the prefix
459 indicates an absolute file name.
462 @defmac MD_EXEC_PREFIX
463 If defined, this macro is an additional prefix to try after
464 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
465 when the @option{-b} option is used, or the compiler is built as a cross
466 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
467 to the list of directories used to find the assembler in @file{configure.in}.
470 @defmac STANDARD_STARTFILE_PREFIX
471 Define this macro as a C string constant if you wish to override the
472 standard choice of @code{libdir} as the default prefix to
473 try when searching for startup files such as @file{crt0.o}.
474 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
475 is built as a cross compiler.
478 @defmac STANDARD_STARTFILE_PREFIX_1
479 Define this macro as a C string constant if you wish to override the
480 standard choice of @code{/lib} as a prefix to try after the default prefix
481 when searching for startup files such as @file{crt0.o}.
482 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
483 is built as a cross compiler.
486 @defmac STANDARD_STARTFILE_PREFIX_2
487 Define this macro as a C string constant if you wish to override the
488 standard choice of @code{/lib} as yet another prefix to try after the
489 default prefix when searching for startup files such as @file{crt0.o}.
490 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
491 is built as a cross compiler.
494 @defmac MD_STARTFILE_PREFIX
495 If defined, this macro supplies an additional prefix to try after the
496 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
497 @option{-b} option is used, or when the compiler is built as a cross
501 @defmac MD_STARTFILE_PREFIX_1
502 If defined, this macro supplies yet another prefix to try after the
503 standard prefixes. It is not searched when the @option{-b} option is
504 used, or when the compiler is built as a cross compiler.
507 @defmac INIT_ENVIRONMENT
508 Define this macro as a C string constant if you wish to set environment
509 variables for programs called by the driver, such as the assembler and
510 loader. The driver passes the value of this macro to @code{putenv} to
511 initialize the necessary environment variables.
514 @defmac LOCAL_INCLUDE_DIR
515 Define this macro as a C string constant if you wish to override the
516 standard choice of @file{/usr/local/include} as the default prefix to
517 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
518 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
520 Cross compilers do not search either @file{/usr/local/include} or its
524 @defmac MODIFY_TARGET_NAME
525 Define this macro if you wish to define command-line switches that
526 modify the default target name.
528 For each switch, you can include a string to be appended to the first
529 part of the configuration name or a string to be deleted from the
530 configuration name, if present. The definition should be an initializer
531 for an array of structures. Each array element should have three
532 elements: the switch name (a string constant, including the initial
533 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
534 indicate whether the string should be inserted or deleted, and the string
535 to be inserted or deleted (a string constant).
537 For example, on a machine where @samp{64} at the end of the
538 configuration name denotes a 64-bit target and you want the @option{-32}
539 and @option{-64} switches to select between 32- and 64-bit targets, you would
543 #define MODIFY_TARGET_NAME \
544 @{ @{ "-32", DELETE, "64"@}, \
545 @{"-64", ADD, "64"@}@}
549 @defmac SYSTEM_INCLUDE_DIR
550 Define this macro as a C string constant if you wish to specify a
551 system-specific directory to search for header files before the standard
552 directory. @code{SYSTEM_INCLUDE_DIR} comes before
553 @code{STANDARD_INCLUDE_DIR} in the search order.
555 Cross compilers do not use this macro and do not search the directory
559 @defmac STANDARD_INCLUDE_DIR
560 Define this macro as a C string constant if you wish to override the
561 standard choice of @file{/usr/include} as the default prefix to
562 try when searching for header files.
564 Cross compilers ignore this macro and do not search either
565 @file{/usr/include} or its replacement.
568 @defmac STANDARD_INCLUDE_COMPONENT
569 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
570 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
571 If you do not define this macro, no component is used.
574 @defmac INCLUDE_DEFAULTS
575 Define this macro if you wish to override the entire default search path
576 for include files. For a native compiler, the default search path
577 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
578 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
579 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
580 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
581 and specify private search areas for GCC@. The directory
582 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
584 The definition should be an initializer for an array of structures.
585 Each array element should have four elements: the directory name (a
586 string constant), the component name (also a string constant), a flag
587 for C++-only directories,
588 and a flag showing that the includes in the directory don't need to be
589 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
590 the array with a null element.
592 The component name denotes what GNU package the include file is part of,
593 if any, in all uppercase letters. For example, it might be @samp{GCC}
594 or @samp{BINUTILS}. If the package is part of a vendor-supplied
595 operating system, code the component name as @samp{0}.
597 For example, here is the definition used for VAX/VMS:
600 #define INCLUDE_DEFAULTS \
602 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
603 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
604 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
611 Here is the order of prefixes tried for exec files:
615 Any prefixes specified by the user with @option{-B}.
618 The environment variable @code{GCC_EXEC_PREFIX}, if any.
621 The directories specified by the environment variable @code{COMPILER_PATH}.
624 The macro @code{STANDARD_EXEC_PREFIX}.
627 @file{/usr/lib/gcc/}.
630 The macro @code{MD_EXEC_PREFIX}, if any.
633 Here is the order of prefixes tried for startfiles:
637 Any prefixes specified by the user with @option{-B}.
640 The environment variable @code{GCC_EXEC_PREFIX}, if any.
643 The directories specified by the environment variable @code{LIBRARY_PATH}
644 (or port-specific name; native only, cross compilers do not use this).
647 The macro @code{STANDARD_EXEC_PREFIX}.
650 @file{/usr/lib/gcc/}.
653 The macro @code{MD_EXEC_PREFIX}, if any.
656 The macro @code{MD_STARTFILE_PREFIX}, if any.
659 The macro @code{STANDARD_STARTFILE_PREFIX}.
668 @node Run-time Target
669 @section Run-time Target Specification
670 @cindex run-time target specification
671 @cindex predefined macros
672 @cindex target specifications
674 @c prevent bad page break with this line
675 Here are run-time target specifications.
677 @defmac TARGET_CPU_CPP_BUILTINS ()
678 This function-like macro expands to a block of code that defines
679 built-in preprocessor macros and assertions for the target cpu, using
680 the functions @code{builtin_define}, @code{builtin_define_std} and
681 @code{builtin_assert}. When the front end
682 calls this macro it provides a trailing semicolon, and since it has
683 finished command line option processing your code can use those
686 @code{builtin_assert} takes a string in the form you pass to the
687 command-line option @option{-A}, such as @code{cpu=mips}, and creates
688 the assertion. @code{builtin_define} takes a string in the form
689 accepted by option @option{-D} and unconditionally defines the macro.
691 @code{builtin_define_std} takes a string representing the name of an
692 object-like macro. If it doesn't lie in the user's namespace,
693 @code{builtin_define_std} defines it unconditionally. Otherwise, it
694 defines a version with two leading underscores, and another version
695 with two leading and trailing underscores, and defines the original
696 only if an ISO standard was not requested on the command line. For
697 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
698 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
699 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
700 defines only @code{_ABI64}.
702 You can also test for the C dialect being compiled. The variable
703 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
704 or @code{clk_objective_c}. Note that if we are preprocessing
705 assembler, this variable will be @code{clk_c} but the function-like
706 macro @code{preprocessing_asm_p()} will return true, so you might want
707 to check for that first. If you need to check for strict ANSI, the
708 variable @code{flag_iso} can be used. The function-like macro
709 @code{preprocessing_trad_p()} can be used to check for traditional
713 @defmac TARGET_OS_CPP_BUILTINS ()
714 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
715 and is used for the target operating system instead.
718 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
719 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
720 and is used for the target object format. @file{elfos.h} uses this
721 macro to define @code{__ELF__}, so you probably do not need to define
725 @deftypevar {extern int} target_flags
726 This variable is declared in @file{options.h}, which is included before
727 any target-specific headers.
730 @deftypevar {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
731 This variable specifies the initial value of @code{target_flags}.
732 Its default setting is 0.
735 @cindex optional hardware or system features
736 @cindex features, optional, in system conventions
738 @deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
739 This hook is called whenever the user specifies one of the
740 target-specific options described by the @file{.opt} definition files
741 (@pxref{Options}). It has the opportunity to do some option-specific
742 processing and should return true if the option is valid. The default
743 definition does nothing but return true.
745 @var{code} specifies the @code{OPT_@var{name}} enumeration value
746 associated with the selected option; @var{name} is just a rendering of
747 the option name in which non-alphanumeric characters are replaced by
748 underscores. @var{arg} specifies the string argument and is null if
749 no argument was given. If the option is flagged as a @code{UInteger}
750 (@pxref{Option properties}), @var{value} is the numeric value of the
751 argument. Otherwise @var{value} is 1 if the positive form of the
752 option was used and 0 if the ``no-'' form was.
755 @defmac TARGET_VERSION
756 This macro is a C statement to print on @code{stderr} a string
757 describing the particular machine description choice. Every machine
758 description should define @code{TARGET_VERSION}. For example:
762 #define TARGET_VERSION \
763 fprintf (stderr, " (68k, Motorola syntax)");
765 #define TARGET_VERSION \
766 fprintf (stderr, " (68k, MIT syntax)");
771 @defmac OVERRIDE_OPTIONS
772 Sometimes certain combinations of command options do not make sense on
773 a particular target machine. You can define a macro
774 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
775 defined, is executed once just after all the command options have been
778 Don't use this macro to turn on various extra optimizations for
779 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
782 @defmac C_COMMON_OVERRIDE_OPTIONS
783 This is similar to @code{OVERRIDE_OPTIONS} but is only used in the C
784 language frontends (C, Objective-C, C++, Objective-C++) and so can be
785 used to alter option flag variables which only exist in those
789 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
790 Some machines may desire to change what optimizations are performed for
791 various optimization levels. This macro, if defined, is executed once
792 just after the optimization level is determined and before the remainder
793 of the command options have been parsed. Values set in this macro are
794 used as the default values for the other command line options.
796 @var{level} is the optimization level specified; 2 if @option{-O2} is
797 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
799 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
801 You should not use this macro to change options that are not
802 machine-specific. These should uniformly selected by the same
803 optimization level on all supported machines. Use this macro to enable
804 machine-specific optimizations.
806 @strong{Do not examine @code{write_symbols} in
807 this macro!} The debugging options are not supposed to alter the
811 @defmac CAN_DEBUG_WITHOUT_FP
812 Define this macro if debugging can be performed even without a frame
813 pointer. If this macro is defined, GCC will turn on the
814 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
817 @node Per-Function Data
818 @section Defining data structures for per-function information.
819 @cindex per-function data
820 @cindex data structures
822 If the target needs to store information on a per-function basis, GCC
823 provides a macro and a couple of variables to allow this. Note, just
824 using statics to store the information is a bad idea, since GCC supports
825 nested functions, so you can be halfway through encoding one function
826 when another one comes along.
828 GCC defines a data structure called @code{struct function} which
829 contains all of the data specific to an individual function. This
830 structure contains a field called @code{machine} whose type is
831 @code{struct machine_function *}, which can be used by targets to point
832 to their own specific data.
834 If a target needs per-function specific data it should define the type
835 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
836 This macro should be used to initialize the function pointer
837 @code{init_machine_status}. This pointer is explained below.
839 One typical use of per-function, target specific data is to create an
840 RTX to hold the register containing the function's return address. This
841 RTX can then be used to implement the @code{__builtin_return_address}
842 function, for level 0.
844 Note---earlier implementations of GCC used a single data area to hold
845 all of the per-function information. Thus when processing of a nested
846 function began the old per-function data had to be pushed onto a
847 stack, and when the processing was finished, it had to be popped off the
848 stack. GCC used to provide function pointers called
849 @code{save_machine_status} and @code{restore_machine_status} to handle
850 the saving and restoring of the target specific information. Since the
851 single data area approach is no longer used, these pointers are no
854 @defmac INIT_EXPANDERS
855 Macro called to initialize any target specific information. This macro
856 is called once per function, before generation of any RTL has begun.
857 The intention of this macro is to allow the initialization of the
858 function pointer @code{init_machine_status}.
861 @deftypevar {void (*)(struct function *)} init_machine_status
862 If this function pointer is non-@code{NULL} it will be called once per
863 function, before function compilation starts, in order to allow the
864 target to perform any target specific initialization of the
865 @code{struct function} structure. It is intended that this would be
866 used to initialize the @code{machine} of that structure.
868 @code{struct machine_function} structures are expected to be freed by GC@.
869 Generally, any memory that they reference must be allocated by using
870 @code{ggc_alloc}, including the structure itself.
874 @section Storage Layout
875 @cindex storage layout
877 Note that the definitions of the macros in this table which are sizes or
878 alignments measured in bits do not need to be constant. They can be C
879 expressions that refer to static variables, such as the @code{target_flags}.
880 @xref{Run-time Target}.
882 @defmac BITS_BIG_ENDIAN
883 Define this macro to have the value 1 if the most significant bit in a
884 byte has the lowest number; otherwise define it to have the value zero.
885 This means that bit-field instructions count from the most significant
886 bit. If the machine has no bit-field instructions, then this must still
887 be defined, but it doesn't matter which value it is defined to. This
888 macro need not be a constant.
890 This macro does not affect the way structure fields are packed into
891 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
894 @defmac BYTES_BIG_ENDIAN
895 Define this macro to have the value 1 if the most significant byte in a
896 word has the lowest number. This macro need not be a constant.
899 @defmac WORDS_BIG_ENDIAN
900 Define this macro to have the value 1 if, in a multiword object, the
901 most significant word has the lowest number. This applies to both
902 memory locations and registers; GCC fundamentally assumes that the
903 order of words in memory is the same as the order in registers. This
904 macro need not be a constant.
907 @defmac LIBGCC2_WORDS_BIG_ENDIAN
908 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
909 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
910 used only when compiling @file{libgcc2.c}. Typically the value will be set
911 based on preprocessor defines.
914 @defmac FLOAT_WORDS_BIG_ENDIAN
915 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
916 @code{TFmode} floating point numbers are stored in memory with the word
917 containing the sign bit at the lowest address; otherwise define it to
918 have the value 0. This macro need not be a constant.
920 You need not define this macro if the ordering is the same as for
924 @defmac BITS_PER_UNIT
925 Define this macro to be the number of bits in an addressable storage
926 unit (byte). If you do not define this macro the default is 8.
929 @defmac BITS_PER_WORD
930 Number of bits in a word. If you do not define this macro, the default
931 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
934 @defmac MAX_BITS_PER_WORD
935 Maximum number of bits in a word. If this is undefined, the default is
936 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
937 largest value that @code{BITS_PER_WORD} can have at run-time.
940 @defmac UNITS_PER_WORD
941 Number of storage units in a word; normally the size of a general-purpose
942 register, a power of two from 1 or 8.
945 @defmac MIN_UNITS_PER_WORD
946 Minimum number of units in a word. If this is undefined, the default is
947 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
948 smallest value that @code{UNITS_PER_WORD} can have at run-time.
951 @defmac UNITS_PER_SIMD_WORD
952 Number of units in the vectors that the vectorizer can produce.
953 The default is equal to @code{UNITS_PER_WORD}, because the vectorizer
954 can do some transformations even in absence of specialized @acronym{SIMD}
959 Width of a pointer, in bits. You must specify a value no wider than the
960 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
961 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
962 a value the default is @code{BITS_PER_WORD}.
965 @defmac POINTERS_EXTEND_UNSIGNED
966 A C expression whose value is greater than zero if pointers that need to be
967 extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to
968 be zero-extended and zero if they are to be sign-extended. If the value
969 is less then zero then there must be an "ptr_extend" instruction that
970 extends a pointer from @code{POINTER_SIZE} to @code{Pmode}.
972 You need not define this macro if the @code{POINTER_SIZE} is equal
973 to the width of @code{Pmode}.
976 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
977 A macro to update @var{m} and @var{unsignedp} when an object whose type
978 is @var{type} and which has the specified mode and signedness is to be
979 stored in a register. This macro is only called when @var{type} is a
982 On most RISC machines, which only have operations that operate on a full
983 register, define this macro to set @var{m} to @code{word_mode} if
984 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
985 cases, only integer modes should be widened because wider-precision
986 floating-point operations are usually more expensive than their narrower
989 For most machines, the macro definition does not change @var{unsignedp}.
990 However, some machines, have instructions that preferentially handle
991 either signed or unsigned quantities of certain modes. For example, on
992 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
993 sign-extend the result to 64 bits. On such machines, set
994 @var{unsignedp} according to which kind of extension is more efficient.
996 Do not define this macro if it would never modify @var{m}.
999 @defmac PROMOTE_FUNCTION_MODE
1000 Like @code{PROMOTE_MODE}, but is applied to outgoing function arguments or
1001 function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS}
1002 and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively.
1004 The default is @code{PROMOTE_MODE}.
1007 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype})
1008 This target hook should return @code{true} if the promotion described by
1009 @code{PROMOTE_FUNCTION_MODE} should be done for outgoing function
1013 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype})
1014 This target hook should return @code{true} if the promotion described by
1015 @code{PROMOTE_FUNCTION_MODE} should be done for the return value of
1018 If this target hook returns @code{true}, @code{TARGET_FUNCTION_VALUE}
1019 must perform the same promotions done by @code{PROMOTE_FUNCTION_MODE}.
1022 @defmac PARM_BOUNDARY
1023 Normal alignment required for function parameters on the stack, in
1024 bits. All stack parameters receive at least this much alignment
1025 regardless of data type. On most machines, this is the same as the
1029 @defmac STACK_BOUNDARY
1030 Define this macro to the minimum alignment enforced by hardware for the
1031 stack pointer on this machine. The definition is a C expression for the
1032 desired alignment (measured in bits). This value is used as a default
1033 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1034 this should be the same as @code{PARM_BOUNDARY}.
1037 @defmac PREFERRED_STACK_BOUNDARY
1038 Define this macro if you wish to preserve a certain alignment for the
1039 stack pointer, greater than what the hardware enforces. The definition
1040 is a C expression for the desired alignment (measured in bits). This
1041 macro must evaluate to a value equal to or larger than
1042 @code{STACK_BOUNDARY}.
1045 @defmac FUNCTION_BOUNDARY
1046 Alignment required for a function entry point, in bits.
1049 @defmac BIGGEST_ALIGNMENT
1050 Biggest alignment that any data type can require on this machine, in bits.
1053 @defmac MINIMUM_ATOMIC_ALIGNMENT
1054 If defined, the smallest alignment, in bits, that can be given to an
1055 object that can be referenced in one operation, without disturbing any
1056 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1057 on machines that don't have byte or half-word store operations.
1060 @defmac BIGGEST_FIELD_ALIGNMENT
1061 Biggest alignment that any structure or union field can require on this
1062 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1063 structure and union fields only, unless the field alignment has been set
1064 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1067 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1068 An expression for the alignment of a structure field @var{field} if the
1069 alignment computed in the usual way (including applying of
1070 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1071 alignment) is @var{computed}. It overrides alignment only if the
1072 field alignment has not been set by the
1073 @code{__attribute__ ((aligned (@var{n})))} construct.
1076 @defmac MAX_OFILE_ALIGNMENT
1077 Biggest alignment supported by the object file format of this machine.
1078 Use this macro to limit the alignment which can be specified using the
1079 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1080 the default value is @code{BIGGEST_ALIGNMENT}.
1083 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1084 If defined, a C expression to compute the alignment for a variable in
1085 the static store. @var{type} is the data type, and @var{basic-align} is
1086 the alignment that the object would ordinarily have. The value of this
1087 macro is used instead of that alignment to align the object.
1089 If this macro is not defined, then @var{basic-align} is used.
1092 One use of this macro is to increase alignment of medium-size data to
1093 make it all fit in fewer cache lines. Another is to cause character
1094 arrays to be word-aligned so that @code{strcpy} calls that copy
1095 constants to character arrays can be done inline.
1098 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1099 If defined, a C expression to compute the alignment given to a constant
1100 that is being placed in memory. @var{constant} is the constant and
1101 @var{basic-align} is the alignment that the object would ordinarily
1102 have. The value of this macro is used instead of that alignment to
1105 If this macro is not defined, then @var{basic-align} is used.
1107 The typical use of this macro is to increase alignment for string
1108 constants to be word aligned so that @code{strcpy} calls that copy
1109 constants can be done inline.
1112 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1113 If defined, a C expression to compute the alignment for a variable in
1114 the local store. @var{type} is the data type, and @var{basic-align} is
1115 the alignment that the object would ordinarily have. The value of this
1116 macro is used instead of that alignment to align the object.
1118 If this macro is not defined, then @var{basic-align} is used.
1120 One use of this macro is to increase alignment of medium-size data to
1121 make it all fit in fewer cache lines.
1124 @defmac EMPTY_FIELD_BOUNDARY
1125 Alignment in bits to be given to a structure bit-field that follows an
1126 empty field such as @code{int : 0;}.
1128 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1131 @defmac STRUCTURE_SIZE_BOUNDARY
1132 Number of bits which any structure or union's size must be a multiple of.
1133 Each structure or union's size is rounded up to a multiple of this.
1135 If you do not define this macro, the default is the same as
1136 @code{BITS_PER_UNIT}.
1139 @defmac STRICT_ALIGNMENT
1140 Define this macro to be the value 1 if instructions will fail to work
1141 if given data not on the nominal alignment. If instructions will merely
1142 go slower in that case, define this macro as 0.
1145 @defmac PCC_BITFIELD_TYPE_MATTERS
1146 Define this if you wish to imitate the way many other C compilers handle
1147 alignment of bit-fields and the structures that contain them.
1149 The behavior is that the type written for a named bit-field (@code{int},
1150 @code{short}, or other integer type) imposes an alignment for the entire
1151 structure, as if the structure really did contain an ordinary field of
1152 that type. In addition, the bit-field is placed within the structure so
1153 that it would fit within such a field, not crossing a boundary for it.
1155 Thus, on most machines, a named bit-field whose type is written as
1156 @code{int} would not cross a four-byte boundary, and would force
1157 four-byte alignment for the whole structure. (The alignment used may
1158 not be four bytes; it is controlled by the other alignment parameters.)
1160 An unnamed bit-field will not affect the alignment of the containing
1163 If the macro is defined, its definition should be a C expression;
1164 a nonzero value for the expression enables this behavior.
1166 Note that if this macro is not defined, or its value is zero, some
1167 bit-fields may cross more than one alignment boundary. The compiler can
1168 support such references if there are @samp{insv}, @samp{extv}, and
1169 @samp{extzv} insns that can directly reference memory.
1171 The other known way of making bit-fields work is to define
1172 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1173 Then every structure can be accessed with fullwords.
1175 Unless the machine has bit-field instructions or you define
1176 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1177 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1179 If your aim is to make GCC use the same conventions for laying out
1180 bit-fields as are used by another compiler, here is how to investigate
1181 what the other compiler does. Compile and run this program:
1200 printf ("Size of foo1 is %d\n",
1201 sizeof (struct foo1));
1202 printf ("Size of foo2 is %d\n",
1203 sizeof (struct foo2));
1208 If this prints 2 and 5, then the compiler's behavior is what you would
1209 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1212 @defmac BITFIELD_NBYTES_LIMITED
1213 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1214 to aligning a bit-field within the structure.
1217 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELDS (void)
1218 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1219 whether unnamed bitfields affect the alignment of the containing
1220 structure. The hook should return true if the structure should inherit
1221 the alignment requirements of an unnamed bitfield's type.
1224 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELDS (void)
1225 This target hook should return @code{true} if accesses to volatile bitfields
1226 should use the narrowest mode possible. It should return @code{false} if
1227 these accesses should use the bitfield container type.
1229 The default is @code{!TARGET_STRICT_ALIGN}.
1232 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1233 Return 1 if a structure or array containing @var{field} should be accessed using
1236 If @var{field} is the only field in the structure, @var{mode} is its
1237 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1238 case where structures of one field would require the structure's mode to
1239 retain the field's mode.
1241 Normally, this is not needed. See the file @file{c4x.h} for an example
1242 of how to use this macro to prevent a structure having a floating point
1243 field from being accessed in an integer mode.
1246 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1247 Define this macro as an expression for the alignment of a type (given
1248 by @var{type} as a tree node) if the alignment computed in the usual
1249 way is @var{computed} and the alignment explicitly specified was
1252 The default is to use @var{specified} if it is larger; otherwise, use
1253 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1256 @defmac MAX_FIXED_MODE_SIZE
1257 An integer expression for the size in bits of the largest integer
1258 machine mode that should actually be used. All integer machine modes of
1259 this size or smaller can be used for structures and unions with the
1260 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1261 (DImode)} is assumed.
1264 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1265 If defined, an expression of type @code{enum machine_mode} that
1266 specifies the mode of the save area operand of a
1267 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1268 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1269 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1270 having its mode specified.
1272 You need not define this macro if it always returns @code{Pmode}. You
1273 would most commonly define this macro if the
1274 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1278 @defmac STACK_SIZE_MODE
1279 If defined, an expression of type @code{enum machine_mode} that
1280 specifies the mode of the size increment operand of an
1281 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1283 You need not define this macro if it always returns @code{word_mode}.
1284 You would most commonly define this macro if the @code{allocate_stack}
1285 pattern needs to support both a 32- and a 64-bit mode.
1288 @defmac TARGET_FLOAT_FORMAT
1289 A code distinguishing the floating point format of the target machine.
1290 There are four defined values:
1293 @item IEEE_FLOAT_FORMAT
1294 This code indicates IEEE floating point. It is the default; there is no
1295 need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@.
1297 @item VAX_FLOAT_FORMAT
1298 This code indicates the ``F float'' (for @code{float}) and ``D float''
1299 or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@.
1301 @item IBM_FLOAT_FORMAT
1302 This code indicates the format used on the IBM System/370.
1304 @item C4X_FLOAT_FORMAT
1305 This code indicates the format used on the TMS320C3x/C4x.
1308 If your target uses a floating point format other than these, you must
1309 define a new @var{name}_FLOAT_FORMAT code for it, and add support for
1310 it to @file{real.c}.
1312 The ordering of the component words of floating point values stored in
1313 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}.
1316 @defmac MODE_HAS_NANS (@var{mode})
1317 When defined, this macro should be true if @var{mode} has a NaN
1318 representation. The compiler assumes that NaNs are not equal to
1319 anything (including themselves) and that addition, subtraction,
1320 multiplication and division all return NaNs when one operand is
1323 By default, this macro is true if @var{mode} is a floating-point
1324 mode and the target floating-point format is IEEE@.
1327 @defmac MODE_HAS_INFINITIES (@var{mode})
1328 This macro should be true if @var{mode} can represent infinity. At
1329 present, the compiler uses this macro to decide whether @samp{x - x}
1330 is always defined. By default, the macro is true when @var{mode}
1331 is a floating-point mode and the target format is IEEE@.
1334 @defmac MODE_HAS_SIGNED_ZEROS (@var{mode})
1335 True if @var{mode} distinguishes between positive and negative zero.
1336 The rules are expected to follow the IEEE standard:
1340 @samp{x + x} has the same sign as @samp{x}.
1343 If the sum of two values with opposite sign is zero, the result is
1344 positive for all rounding modes expect towards @minus{}infinity, for
1345 which it is negative.
1348 The sign of a product or quotient is negative when exactly one
1349 of the operands is negative.
1352 The default definition is true if @var{mode} is a floating-point
1353 mode and the target format is IEEE@.
1356 @defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode})
1357 If defined, this macro should be true for @var{mode} if it has at
1358 least one rounding mode in which @samp{x} and @samp{-x} can be
1359 rounded to numbers of different magnitude. Two such modes are
1360 towards @minus{}infinity and towards +infinity.
1362 The default definition of this macro is true if @var{mode} is
1363 a floating-point mode and the target format is IEEE@.
1366 @defmac ROUND_TOWARDS_ZERO
1367 If defined, this macro should be true if the prevailing rounding
1368 mode is towards zero. A true value has the following effects:
1372 @code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes.
1375 @file{libgcc.a}'s floating-point emulator will round towards zero
1376 rather than towards nearest.
1379 The compiler's floating-point emulator will round towards zero after
1380 doing arithmetic, and when converting from the internal float format to
1384 The macro does not affect the parsing of string literals. When the
1385 primary rounding mode is towards zero, library functions like
1386 @code{strtod} might still round towards nearest, and the compiler's
1387 parser should behave like the target's @code{strtod} where possible.
1389 Not defining this macro is equivalent to returning zero.
1392 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1393 This macro should return true if floats with @var{size}
1394 bits do not have a NaN or infinity representation, but use the largest
1395 exponent for normal numbers instead.
1397 Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS}
1398 and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes.
1399 It also affects the way @file{libgcc.a} and @file{real.c} emulate
1400 floating-point arithmetic.
1402 The default definition of this macro returns false for all sizes.
1405 @deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type})
1406 This target hook should return @code{true} a vector is opaque. That
1407 is, if no cast is needed when copying a vector value of type
1408 @var{type} into another vector lvalue of the same size. Vector opaque
1409 types cannot be initialized. The default is that there are no such
1413 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type})
1414 This target hook returns @code{true} if bit-fields in the given
1415 @var{record_type} are to be laid out following the rules of Microsoft
1416 Visual C/C++, namely: (i) a bit-field won't share the same storage
1417 unit with the previous bit-field if their underlying types have
1418 different sizes, and the bit-field will be aligned to the highest
1419 alignment of the underlying types of itself and of the previous
1420 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1421 the whole enclosing structure, even if it is unnamed; except that
1422 (iii) a zero-sized bit-field will be disregarded unless it follows
1423 another bit-field of nonzero size. If this hook returns @code{true},
1424 other macros that control bit-field layout are ignored.
1426 When a bit-field is inserted into a packed record, the whole size
1427 of the underlying type is used by one or more same-size adjacent
1428 bit-fields (that is, if its long:3, 32 bits is used in the record,
1429 and any additional adjacent long bit-fields are packed into the same
1430 chunk of 32 bits. However, if the size changes, a new field of that
1431 size is allocated). In an unpacked record, this is the same as using
1432 alignment, but not equivalent when packing.
1434 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1435 the latter will take precedence. If @samp{__attribute__((packed))} is
1436 used on a single field when MS bit-fields are in use, it will take
1437 precedence for that field, but the alignment of the rest of the structure
1438 may affect its placement.
1441 @deftypefn {Target Hook} {bool} TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1442 Returns true if the target supports decimal floating point.
1445 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_FUNDAMENTAL_TYPE (tree @var{type})
1446 If your target defines any fundamental types, define this hook to
1447 return the appropriate encoding for these types as part of a C++
1448 mangled name. The @var{type} argument is the tree structure
1449 representing the type to be mangled. The hook may be applied to trees
1450 which are not target-specific fundamental types; it should return
1451 @code{NULL} for all such types, as well as arguments it does not
1452 recognize. If the return value is not @code{NULL}, it must point to
1453 a statically-allocated string constant.
1455 Target-specific fundamental types might be new fundamental types or
1456 qualified versions of ordinary fundamental types. Encode new
1457 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1458 is the name used for the type in source code, and @var{n} is the
1459 length of @var{name} in decimal. Encode qualified versions of
1460 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1461 @var{name} is the name used for the type qualifier in source code,
1462 @var{n} is the length of @var{name} as above, and @var{code} is the
1463 code used to represent the unqualified version of this type. (See
1464 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1465 codes.) In both cases the spaces are for clarity; do not include any
1466 spaces in your string.
1468 The default version of this hook always returns @code{NULL}, which is
1469 appropriate for a target that does not define any new fundamental
1474 @section Layout of Source Language Data Types
1476 These macros define the sizes and other characteristics of the standard
1477 basic data types used in programs being compiled. Unlike the macros in
1478 the previous section, these apply to specific features of C and related
1479 languages, rather than to fundamental aspects of storage layout.
1481 @defmac INT_TYPE_SIZE
1482 A C expression for the size in bits of the type @code{int} on the
1483 target machine. If you don't define this, the default is one word.
1486 @defmac SHORT_TYPE_SIZE
1487 A C expression for the size in bits of the type @code{short} on the
1488 target 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
1493 @defmac LONG_TYPE_SIZE
1494 A C expression for the size in bits of the type @code{long} on the
1495 target machine. If you don't define this, the default is one word.
1498 @defmac ADA_LONG_TYPE_SIZE
1499 On 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
1501 that situation, define this macro to be a C expression to be used for
1502 the size of that type. If you don't define this, the default is the
1503 value of @code{LONG_TYPE_SIZE}.
1506 @defmac LONG_LONG_TYPE_SIZE
1507 A C expression for the size in bits of the type @code{long long} on the
1508 target machine. If you don't define this, the default is two
1509 words. If you want to support GNU Ada on your machine, the value of this
1510 macro must be at least 64.
1513 @defmac CHAR_TYPE_SIZE
1514 A C expression for the size in bits of the type @code{char} on the
1515 target machine. If you don't define this, the default is
1516 @code{BITS_PER_UNIT}.
1519 @defmac BOOL_TYPE_SIZE
1520 A C expression for the size in bits of the C++ type @code{bool} and
1521 C99 type @code{_Bool} on the target machine. If you don't define
1522 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1525 @defmac FLOAT_TYPE_SIZE
1526 A C expression for the size in bits of the type @code{float} on the
1527 target machine. If you don't define this, the default is one word.
1530 @defmac DOUBLE_TYPE_SIZE
1531 A C expression for the size in bits of the type @code{double} on the
1532 target machine. If you don't define this, the default is two
1536 @defmac LONG_DOUBLE_TYPE_SIZE
1537 A C expression for the size in bits of the type @code{long double} on
1538 the target machine. If you don't define this, the default is two
1542 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1543 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1544 if you want routines in @file{libgcc2.a} for a size other than
1545 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1546 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1549 @defmac LIBGCC2_HAS_DF_MODE
1550 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1551 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1552 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1553 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1554 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1558 @defmac LIBGCC2_HAS_XF_MODE
1559 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1560 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1561 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1562 is 80 then the default is 1, otherwise it is 0.
1565 @defmac LIBGCC2_HAS_TF_MODE
1566 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1567 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1568 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1569 is 128 then the default is 1, otherwise it is 0.
1576 Define these macros to be the size in bits of the mantissa of
1577 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1578 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1579 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1580 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1581 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1582 @code{LIBGCC2_DOUBLE_TYPE_SIZE} or
1583 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1586 @defmac TARGET_FLT_EVAL_METHOD
1587 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1588 assuming, if applicable, that the floating-point control word is in its
1589 default state. If you do not define this macro the value of
1590 @code{FLT_EVAL_METHOD} will be zero.
1593 @defmac WIDEST_HARDWARE_FP_SIZE
1594 A C expression for the size in bits of the widest floating-point format
1595 supported by the hardware. If you define this macro, you must specify a
1596 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1597 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1601 @defmac DEFAULT_SIGNED_CHAR
1602 An expression whose value is 1 or 0, according to whether the type
1603 @code{char} should be signed or unsigned by default. The user can
1604 always override this default with the options @option{-fsigned-char}
1605 and @option{-funsigned-char}.
1608 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1609 This target hook should return true if the compiler should give an
1610 @code{enum} type only as many bytes as it takes to represent the range
1611 of possible values of that type. It should return false if all
1612 @code{enum} types should be allocated like @code{int}.
1614 The default is to return false.
1618 A C expression for a string describing the name of the data type to use
1619 for size values. The typedef name @code{size_t} is defined using the
1620 contents of the string.
1622 The string can contain more than one keyword. If so, separate them with
1623 spaces, and write first any length keyword, then @code{unsigned} if
1624 appropriate, and finally @code{int}. The string must exactly match one
1625 of the data type names defined in the function
1626 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1627 omit @code{int} or change the order---that would cause the compiler to
1630 If you don't define this macro, the default is @code{"long unsigned
1634 @defmac PTRDIFF_TYPE
1635 A C expression for a string describing the name of the data type to use
1636 for the result of subtracting two pointers. The typedef name
1637 @code{ptrdiff_t} is defined using the contents of the string. See
1638 @code{SIZE_TYPE} above for more information.
1640 If you don't define this macro, the default is @code{"long int"}.
1644 A C expression for a string describing the name of the data type to use
1645 for wide characters. The typedef name @code{wchar_t} is defined using
1646 the contents of the string. See @code{SIZE_TYPE} above for more
1649 If you don't define this macro, the default is @code{"int"}.
1652 @defmac WCHAR_TYPE_SIZE
1653 A C expression for the size in bits of the data type for wide
1654 characters. This is used in @code{cpp}, which cannot make use of
1659 A C expression for a string describing the name of the data type to
1660 use for wide characters passed to @code{printf} and returned from
1661 @code{getwc}. The typedef name @code{wint_t} is defined using the
1662 contents of the string. See @code{SIZE_TYPE} above for more
1665 If you don't define this macro, the default is @code{"unsigned int"}.
1669 A C expression for a string describing the name of the data type that
1670 can represent any value of any standard or extended signed integer type.
1671 The typedef name @code{intmax_t} is defined using the contents of the
1672 string. See @code{SIZE_TYPE} above for more information.
1674 If you don't define this macro, the default is the first of
1675 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1676 much precision as @code{long long int}.
1679 @defmac UINTMAX_TYPE
1680 A C expression for a string describing the name of the data type that
1681 can represent any value of any standard or extended unsigned integer
1682 type. The typedef name @code{uintmax_t} is defined using the contents
1683 of the string. See @code{SIZE_TYPE} above for more information.
1685 If you don't define this macro, the default is the first of
1686 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1687 unsigned int"} that has as much precision as @code{long long unsigned
1691 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1692 The C++ compiler represents a pointer-to-member-function with a struct
1699 ptrdiff_t vtable_index;
1706 The C++ compiler must use one bit to indicate whether the function that
1707 will be called through a pointer-to-member-function is virtual.
1708 Normally, we assume that the low-order bit of a function pointer must
1709 always be zero. Then, by ensuring that the vtable_index is odd, we can
1710 distinguish which variant of the union is in use. But, on some
1711 platforms function pointers can be odd, and so this doesn't work. In
1712 that case, we use the low-order bit of the @code{delta} field, and shift
1713 the remainder of the @code{delta} field to the left.
1715 GCC will automatically make the right selection about where to store
1716 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1717 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1718 set such that functions always start at even addresses, but the lowest
1719 bit of pointers to functions indicate whether the function at that
1720 address is in ARM or Thumb mode. If this is the case of your
1721 architecture, you should define this macro to
1722 @code{ptrmemfunc_vbit_in_delta}.
1724 In general, you should not have to define this macro. On architectures
1725 in which function addresses are always even, according to
1726 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1727 @code{ptrmemfunc_vbit_in_pfn}.
1730 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1731 Normally, the C++ compiler uses function pointers in vtables. This
1732 macro allows the target to change to use ``function descriptors''
1733 instead. Function descriptors are found on targets for whom a
1734 function pointer is actually a small data structure. Normally the
1735 data structure consists of the actual code address plus a data
1736 pointer to which the function's data is relative.
1738 If vtables are used, the value of this macro should be the number
1739 of words that the function descriptor occupies.
1742 @defmac TARGET_VTABLE_ENTRY_ALIGN
1743 By default, the vtable entries are void pointers, the so the alignment
1744 is the same as pointer alignment. The value of this macro specifies
1745 the alignment of the vtable entry in bits. It should be defined only
1746 when special alignment is necessary. */
1749 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1750 There are a few non-descriptor entries in the vtable at offsets below
1751 zero. If these entries must be padded (say, to preserve the alignment
1752 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1753 of words in each data entry.
1757 @section Register Usage
1758 @cindex register usage
1760 This section explains how to describe what registers the target machine
1761 has, and how (in general) they can be used.
1763 The description of which registers a specific instruction can use is
1764 done with register classes; see @ref{Register Classes}. For information
1765 on using registers to access a stack frame, see @ref{Frame Registers}.
1766 For passing values in registers, see @ref{Register Arguments}.
1767 For returning values in registers, see @ref{Scalar Return}.
1770 * Register Basics:: Number and kinds of registers.
1771 * Allocation Order:: Order in which registers are allocated.
1772 * Values in Registers:: What kinds of values each reg can hold.
1773 * Leaf Functions:: Renumbering registers for leaf functions.
1774 * Stack Registers:: Handling a register stack such as 80387.
1777 @node Register Basics
1778 @subsection Basic Characteristics of Registers
1780 @c prevent bad page break with this line
1781 Registers have various characteristics.
1783 @defmac FIRST_PSEUDO_REGISTER
1784 Number of hardware registers known to the compiler. They receive
1785 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1786 pseudo register's number really is assigned the number
1787 @code{FIRST_PSEUDO_REGISTER}.
1790 @defmac FIXED_REGISTERS
1791 @cindex fixed register
1792 An initializer that says which registers are used for fixed purposes
1793 all throughout the compiled code and are therefore not available for
1794 general allocation. These would include the stack pointer, the frame
1795 pointer (except on machines where that can be used as a general
1796 register when no frame pointer is needed), the program counter on
1797 machines where that is considered one of the addressable registers,
1798 and any other numbered register with a standard use.
1800 This information is expressed as a sequence of numbers, separated by
1801 commas and surrounded by braces. The @var{n}th number is 1 if
1802 register @var{n} is fixed, 0 otherwise.
1804 The table initialized from this macro, and the table initialized by
1805 the following one, may be overridden at run time either automatically,
1806 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1807 the user with the command options @option{-ffixed-@var{reg}},
1808 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1811 @defmac CALL_USED_REGISTERS
1812 @cindex call-used register
1813 @cindex call-clobbered register
1814 @cindex call-saved register
1815 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1816 clobbered (in general) by function calls as well as for fixed
1817 registers. This macro therefore identifies the registers that are not
1818 available for general allocation of values that must live across
1821 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1822 automatically saves it on function entry and restores it on function
1823 exit, if the register is used within the function.
1826 @defmac CALL_REALLY_USED_REGISTERS
1827 @cindex call-used register
1828 @cindex call-clobbered register
1829 @cindex call-saved register
1830 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1831 that the entire set of @code{FIXED_REGISTERS} be included.
1832 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1833 This macro is optional. If not specified, it defaults to the value
1834 of @code{CALL_USED_REGISTERS}.
1837 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1838 @cindex call-used register
1839 @cindex call-clobbered register
1840 @cindex call-saved register
1841 A C expression that is nonzero if it is not permissible to store a
1842 value of mode @var{mode} in hard register number @var{regno} across a
1843 call without some part of it being clobbered. For most machines this
1844 macro need not be defined. It is only required for machines that do not
1845 preserve the entire contents of a register across a call.
1849 @findex call_used_regs
1852 @findex reg_class_contents
1853 @defmac CONDITIONAL_REGISTER_USAGE
1854 Zero or more C statements that may conditionally modify five variables
1855 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1856 @code{reg_names}, and @code{reg_class_contents}, to take into account
1857 any dependence of these register sets on target flags. The first three
1858 of these are of type @code{char []} (interpreted as Boolean vectors).
1859 @code{global_regs} is a @code{const char *[]}, and
1860 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1861 called, @code{fixed_regs}, @code{call_used_regs},
1862 @code{reg_class_contents}, and @code{reg_names} have been initialized
1863 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1864 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1865 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1866 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1867 command options have been applied.
1869 You need not define this macro if it has no work to do.
1871 @cindex disabling certain registers
1872 @cindex controlling register usage
1873 If the usage of an entire class of registers depends on the target
1874 flags, you may indicate this to GCC by using this macro to modify
1875 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1876 registers in the classes which should not be used by GCC@. Also define
1877 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1878 to return @code{NO_REGS} if it
1879 is called with a letter for a class that shouldn't be used.
1881 (However, if this class is not included in @code{GENERAL_REGS} and all
1882 of the insn patterns whose constraints permit this class are
1883 controlled by target switches, then GCC will automatically avoid using
1884 these registers when the target switches are opposed to them.)
1887 @defmac INCOMING_REGNO (@var{out})
1888 Define this macro if the target machine has register windows. This C
1889 expression returns the register number as seen by the called function
1890 corresponding to the register number @var{out} as seen by the calling
1891 function. Return @var{out} if register number @var{out} is not an
1895 @defmac OUTGOING_REGNO (@var{in})
1896 Define this macro if the target machine has register windows. This C
1897 expression returns the register number as seen by the calling function
1898 corresponding to the register number @var{in} as seen by the called
1899 function. Return @var{in} if register number @var{in} is not an inbound
1903 @defmac LOCAL_REGNO (@var{regno})
1904 Define this macro if the target machine has register windows. This C
1905 expression returns true if the register is call-saved but is in the
1906 register window. Unlike most call-saved registers, such registers
1907 need not be explicitly restored on function exit or during non-local
1912 If the program counter has a register number, define this as that
1913 register number. Otherwise, do not define it.
1916 @node Allocation Order
1917 @subsection Order of Allocation of Registers
1918 @cindex order of register allocation
1919 @cindex register allocation order
1921 @c prevent bad page break with this line
1922 Registers are allocated in order.
1924 @defmac REG_ALLOC_ORDER
1925 If defined, an initializer for a vector of integers, containing the
1926 numbers of hard registers in the order in which GCC should prefer
1927 to use them (from most preferred to least).
1929 If this macro is not defined, registers are used lowest numbered first
1930 (all else being equal).
1932 One use of this macro is on machines where the highest numbered
1933 registers must always be saved and the save-multiple-registers
1934 instruction supports only sequences of consecutive registers. On such
1935 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1936 the highest numbered allocable register first.
1939 @defmac ORDER_REGS_FOR_LOCAL_ALLOC
1940 A C statement (sans semicolon) to choose the order in which to allocate
1941 hard registers for pseudo-registers local to a basic block.
1943 Store the desired register order in the array @code{reg_alloc_order}.
1944 Element 0 should be the register to allocate first; element 1, the next
1945 register; and so on.
1947 The macro body should not assume anything about the contents of
1948 @code{reg_alloc_order} before execution of the macro.
1950 On most machines, it is not necessary to define this macro.
1953 @node Values in Registers
1954 @subsection How Values Fit in Registers
1956 This section discusses the macros that describe which kinds of values
1957 (specifically, which machine modes) each register can hold, and how many
1958 consecutive registers are needed for a given mode.
1960 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
1961 A C expression for the number of consecutive hard registers, starting
1962 at register number @var{regno}, required to hold a value of mode
1965 On a machine where all registers are exactly one word, a suitable
1966 definition of this macro is
1969 #define HARD_REGNO_NREGS(REGNO, MODE) \
1970 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
1975 @defmac REGMODE_NATURAL_SIZE (@var{mode})
1976 Define this macro if the natural size of registers that hold values
1977 of mode @var{mode} is not the word size. It is a C expression that
1978 should give the natural size in bytes for the specified mode. It is
1979 used by the register allocator to try to optimize its results. This
1980 happens for example on SPARC 64-bit where the natural size of
1981 floating-point registers is still 32-bit.
1984 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
1985 A C expression that is nonzero if it is permissible to store a value
1986 of mode @var{mode} in hard register number @var{regno} (or in several
1987 registers starting with that one). For a machine where all registers
1988 are equivalent, a suitable definition is
1991 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
1994 You need not include code to check for the numbers of fixed registers,
1995 because the allocation mechanism considers them to be always occupied.
1997 @cindex register pairs
1998 On some machines, double-precision values must be kept in even/odd
1999 register pairs. You can implement that by defining this macro to reject
2000 odd register numbers for such modes.
2002 The minimum requirement for a mode to be OK in a register is that the
2003 @samp{mov@var{mode}} instruction pattern support moves between the
2004 register and other hard register in the same class and that moving a
2005 value into the register and back out not alter it.
2007 Since the same instruction used to move @code{word_mode} will work for
2008 all narrower integer modes, it is not necessary on any machine for
2009 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2010 you define patterns @samp{movhi}, etc., to take advantage of this. This
2011 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2012 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2015 Many machines have special registers for floating point arithmetic.
2016 Often people assume that floating point machine modes are allowed only
2017 in floating point registers. This is not true. Any registers that
2018 can hold integers can safely @emph{hold} a floating point machine
2019 mode, whether or not floating arithmetic can be done on it in those
2020 registers. Integer move instructions can be used to move the values.
2022 On some machines, though, the converse is true: fixed-point machine
2023 modes may not go in floating registers. This is true if the floating
2024 registers normalize any value stored in them, because storing a
2025 non-floating value there would garble it. In this case,
2026 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2027 floating registers. But if the floating registers do not automatically
2028 normalize, if you can store any bit pattern in one and retrieve it
2029 unchanged without a trap, then any machine mode may go in a floating
2030 register, so you can define this macro to say so.
2032 The primary significance of special floating registers is rather that
2033 they are the registers acceptable in floating point arithmetic
2034 instructions. However, this is of no concern to
2035 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2036 constraints for those instructions.
2038 On some machines, the floating registers are especially slow to access,
2039 so that it is better to store a value in a stack frame than in such a
2040 register if floating point arithmetic is not being done. As long as the
2041 floating registers are not in class @code{GENERAL_REGS}, they will not
2042 be used unless some pattern's constraint asks for one.
2045 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2046 A C expression that is nonzero if it is OK to rename a hard register
2047 @var{from} to another hard register @var{to}.
2049 One common use of this macro is to prevent renaming of a register to
2050 another register that is not saved by a prologue in an interrupt
2053 The default is always nonzero.
2056 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2057 A C expression that is nonzero if a value of mode
2058 @var{mode1} is accessible in mode @var{mode2} without copying.
2060 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2061 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2062 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2063 should be nonzero. If they differ for any @var{r}, you should define
2064 this macro to return zero unless some other mechanism ensures the
2065 accessibility of the value in a narrower mode.
2067 You should define this macro to return nonzero in as many cases as
2068 possible since doing so will allow GCC to perform better register
2072 @defmac AVOID_CCMODE_COPIES
2073 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2074 registers. You should only define this macro if support for copying to/from
2075 @code{CCmode} is incomplete.
2078 @node Leaf Functions
2079 @subsection Handling Leaf Functions
2081 @cindex leaf functions
2082 @cindex functions, leaf
2083 On some machines, a leaf function (i.e., one which makes no calls) can run
2084 more efficiently if it does not make its own register window. Often this
2085 means it is required to receive its arguments in the registers where they
2086 are passed by the caller, instead of the registers where they would
2089 The special treatment for leaf functions generally applies only when
2090 other conditions are met; for example, often they may use only those
2091 registers for its own variables and temporaries. We use the term ``leaf
2092 function'' to mean a function that is suitable for this special
2093 handling, so that functions with no calls are not necessarily ``leaf
2096 GCC assigns register numbers before it knows whether the function is
2097 suitable for leaf function treatment. So it needs to renumber the
2098 registers in order to output a leaf function. The following macros
2101 @defmac LEAF_REGISTERS
2102 Name of a char vector, indexed by hard register number, which
2103 contains 1 for a register that is allowable in a candidate for leaf
2106 If leaf function treatment involves renumbering the registers, then the
2107 registers marked here should be the ones before renumbering---those that
2108 GCC would ordinarily allocate. The registers which will actually be
2109 used in the assembler code, after renumbering, should not be marked with 1
2112 Define this macro only if the target machine offers a way to optimize
2113 the treatment of leaf functions.
2116 @defmac LEAF_REG_REMAP (@var{regno})
2117 A C expression whose value is the register number to which @var{regno}
2118 should be renumbered, when a function is treated as a leaf function.
2120 If @var{regno} is a register number which should not appear in a leaf
2121 function before renumbering, then the expression should yield @minus{}1, which
2122 will cause the compiler to abort.
2124 Define this macro only if the target machine offers a way to optimize the
2125 treatment of leaf functions, and registers need to be renumbered to do
2129 @findex current_function_is_leaf
2130 @findex current_function_uses_only_leaf_regs
2131 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2132 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2133 specially. They can test the C variable @code{current_function_is_leaf}
2134 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2135 set prior to local register allocation and is valid for the remaining
2136 compiler passes. They can also test the C variable
2137 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2138 functions which only use leaf registers.
2139 @code{current_function_uses_only_leaf_regs} is valid after all passes
2140 that modify the instructions have been run and is only useful if
2141 @code{LEAF_REGISTERS} is defined.
2142 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2143 @c of the next paragraph?! --mew 2feb93
2145 @node Stack Registers
2146 @subsection Registers That Form a Stack
2148 There are special features to handle computers where some of the
2149 ``registers'' form a stack. Stack registers are normally written by
2150 pushing onto the stack, and are numbered relative to the top of the
2153 Currently, GCC can only handle one group of stack-like registers, and
2154 they must be consecutively numbered. Furthermore, the existing
2155 support for stack-like registers is specific to the 80387 floating
2156 point coprocessor. If you have a new architecture that uses
2157 stack-like registers, you will need to do substantial work on
2158 @file{reg-stack.c} and write your machine description to cooperate
2159 with it, as well as defining these macros.
2162 Define this if the machine has any stack-like registers.
2165 @defmac FIRST_STACK_REG
2166 The number of the first stack-like register. This one is the top
2170 @defmac LAST_STACK_REG
2171 The number of the last stack-like register. This one is the bottom of
2175 @node Register Classes
2176 @section Register Classes
2177 @cindex register class definitions
2178 @cindex class definitions, register
2180 On many machines, the numbered registers are not all equivalent.
2181 For example, certain registers may not be allowed for indexed addressing;
2182 certain registers may not be allowed in some instructions. These machine
2183 restrictions are described to the compiler using @dfn{register classes}.
2185 You define a number of register classes, giving each one a name and saying
2186 which of the registers belong to it. Then you can specify register classes
2187 that are allowed as operands to particular instruction patterns.
2191 In general, each register will belong to several classes. In fact, one
2192 class must be named @code{ALL_REGS} and contain all the registers. Another
2193 class must be named @code{NO_REGS} and contain no registers. Often the
2194 union of two classes will be another class; however, this is not required.
2196 @findex GENERAL_REGS
2197 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2198 terribly special about the name, but the operand constraint letters
2199 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2200 the same as @code{ALL_REGS}, just define it as a macro which expands
2203 Order the classes so that if class @var{x} is contained in class @var{y}
2204 then @var{x} has a lower class number than @var{y}.
2206 The way classes other than @code{GENERAL_REGS} are specified in operand
2207 constraints is through machine-dependent operand constraint letters.
2208 You can define such letters to correspond to various classes, then use
2209 them in operand constraints.
2211 You should define a class for the union of two classes whenever some
2212 instruction allows both classes. For example, if an instruction allows
2213 either a floating point (coprocessor) register or a general register for a
2214 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2215 which includes both of them. Otherwise you will get suboptimal code.
2217 You must also specify certain redundant information about the register
2218 classes: for each class, which classes contain it and which ones are
2219 contained in it; for each pair of classes, the largest class contained
2222 When a value occupying several consecutive registers is expected in a
2223 certain class, all the registers used must belong to that class.
2224 Therefore, register classes cannot be used to enforce a requirement for
2225 a register pair to start with an even-numbered register. The way to
2226 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2228 Register classes used for input-operands of bitwise-and or shift
2229 instructions have a special requirement: each such class must have, for
2230 each fixed-point machine mode, a subclass whose registers can transfer that
2231 mode to or from memory. For example, on some machines, the operations for
2232 single-byte values (@code{QImode}) are limited to certain registers. When
2233 this is so, each register class that is used in a bitwise-and or shift
2234 instruction must have a subclass consisting of registers from which
2235 single-byte values can be loaded or stored. This is so that
2236 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2238 @deftp {Data type} {enum reg_class}
2239 An enumerated type that must be defined with all the register class names
2240 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2241 must be the last register class, followed by one more enumerated value,
2242 @code{LIM_REG_CLASSES}, which is not a register class but rather
2243 tells how many classes there are.
2245 Each register class has a number, which is the value of casting
2246 the class name to type @code{int}. The number serves as an index
2247 in many of the tables described below.
2250 @defmac N_REG_CLASSES
2251 The number of distinct register classes, defined as follows:
2254 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2258 @defmac REG_CLASS_NAMES
2259 An initializer containing the names of the register classes as C string
2260 constants. These names are used in writing some of the debugging dumps.
2263 @defmac REG_CLASS_CONTENTS
2264 An initializer containing the contents of the register classes, as integers
2265 which are bit masks. The @var{n}th integer specifies the contents of class
2266 @var{n}. The way the integer @var{mask} is interpreted is that
2267 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2269 When the machine has more than 32 registers, an integer does not suffice.
2270 Then the integers are replaced by sub-initializers, braced groupings containing
2271 several integers. Each sub-initializer must be suitable as an initializer
2272 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2273 In this situation, the first integer in each sub-initializer corresponds to
2274 registers 0 through 31, the second integer to registers 32 through 63, and
2278 @defmac REGNO_REG_CLASS (@var{regno})
2279 A C expression whose value is a register class containing hard register
2280 @var{regno}. In general there is more than one such class; choose a class
2281 which is @dfn{minimal}, meaning that no smaller class also contains the
2285 @defmac BASE_REG_CLASS
2286 A macro whose definition is the name of the class to which a valid
2287 base register must belong. A base register is one used in an address
2288 which is the register value plus a displacement.
2291 @defmac MODE_BASE_REG_CLASS (@var{mode})
2292 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2293 the selection of a base register in a mode dependent manner. If
2294 @var{mode} is VOIDmode then it should return the same value as
2295 @code{BASE_REG_CLASS}.
2298 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2299 A C expression whose value is the register class to which a valid
2300 base register must belong in order to be used in a base plus index
2301 register address. You should define this macro if base plus index
2302 addresses have different requirements than other base register uses.
2305 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2306 A C expression whose value is the register class to which a valid
2307 base register must belong. @var{outer_code} and @var{index_code} define the
2308 context in which the base register occurs. @var{outer_code} is the code of
2309 the immediately enclosing expression (@code{MEM} for the top level of an
2310 address, @code{ADDRESS} for something that occurs in an
2311 @code{address_operand}). @var{index_code} is the code of the corresponding
2312 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2315 @defmac INDEX_REG_CLASS
2316 A macro whose definition is the name of the class to which a valid
2317 index register must belong. An index register is one used in an
2318 address where its value is either multiplied by a scale factor or
2319 added to another register (as well as added to a displacement).
2322 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2323 A C expression which is nonzero if register number @var{num} is
2324 suitable for use as a base register in operand addresses. It may be
2325 either a suitable hard register or a pseudo register that has been
2326 allocated such a hard register.
2329 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2330 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2331 that expression may examine the mode of the memory reference in
2332 @var{mode}. You should define this macro if the mode of the memory
2333 reference affects whether a register may be used as a base register. If
2334 you define this macro, the compiler will use it instead of
2335 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for addresses
2336 that appear outside a @code{MEM}, i.e. as an @code{address_operand}.
2340 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2341 A C expression which is nonzero if register number @var{num} is suitable for
2342 use as a base register in base plus index operand addresses, accessing
2343 memory in mode @var{mode}. It may be either a suitable hard register or a
2344 pseudo register that has been allocated such a hard register. You should
2345 define this macro if base plus index addresses have different requirements
2346 than other base register uses.
2348 Use of this macro is deprecated; please use the more general
2349 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2352 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2353 A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except that
2354 that expression may examine the context in which the register appears in the
2355 memory reference. @var{outer_code} is the code of the immediately enclosing
2356 expression (@code{MEM} if at the top level of the address, @code{ADDRESS} for
2357 something that occurs in an @code{address_operand}). @var{index_code} is the
2358 code of the corresponding index expression if @var{outer_code} is @code{PLUS};
2359 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2360 that appear outside a @code{MEM}, i.e. as an @code{address_operand}.
2363 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2364 A C expression which is nonzero if register number @var{num} is
2365 suitable for use as an index register in operand addresses. It may be
2366 either a suitable hard register or a pseudo register that has been
2367 allocated such a hard register.
2369 The difference between an index register and a base register is that
2370 the index register may be scaled. If an address involves the sum of
2371 two registers, neither one of them scaled, then either one may be
2372 labeled the ``base'' and the other the ``index''; but whichever
2373 labeling is used must fit the machine's constraints of which registers
2374 may serve in each capacity. The compiler will try both labelings,
2375 looking for one that is valid, and will reload one or both registers
2376 only if neither labeling works.
2379 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2380 A C expression that places additional restrictions on the register class
2381 to use when it is necessary to copy value @var{x} into a register in class
2382 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2383 another, smaller class. On many machines, the following definition is
2387 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2390 Sometimes returning a more restrictive class makes better code. For
2391 example, on the 68000, when @var{x} is an integer constant that is in range
2392 for a @samp{moveq} instruction, the value of this macro is always
2393 @code{DATA_REGS} as long as @var{class} includes the data registers.
2394 Requiring a data register guarantees that a @samp{moveq} will be used.
2396 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2397 @var{class} is if @var{x} is a legitimate constant which cannot be
2398 loaded into some register class. By returning @code{NO_REGS} you can
2399 force @var{x} into a memory location. For example, rs6000 can load
2400 immediate values into general-purpose registers, but does not have an
2401 instruction for loading an immediate value into a floating-point
2402 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2403 @var{x} is a floating-point constant. If the constant can't be loaded
2404 into any kind of register, code generation will be better if
2405 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2406 of using @code{PREFERRED_RELOAD_CLASS}.
2408 If an insn has pseudos in it after register allocation, reload will go
2409 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2410 to find the best one. Returning @code{NO_REGS}, in this case, makes
2411 reload add a @code{!} in front of the constraint: the x86 back-end uses
2412 this feature to discourage usage of 387 registers when math is done in
2413 the SSE registers (and vice versa).
2416 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2417 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2418 input reloads. If you don't define this macro, the default is to use
2419 @var{class}, unchanged.
2421 You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2422 reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2425 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2426 A C expression that places additional restrictions on the register class
2427 to use when it is necessary to be able to hold a value of mode
2428 @var{mode} in a reload register for which class @var{class} would
2431 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2432 there are certain modes that simply can't go in certain reload classes.
2434 The value is a register class; perhaps @var{class}, or perhaps another,
2437 Don't define this macro unless the target machine has limitations which
2438 require the macro to do something nontrivial.
2441 @deftypefn {Target Hook} enum reg_class TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, enum reg_class @var{reload_class}, enum machine_mode @var{reload_mode}, secondary_reload_info *@var{sri})
2442 Many machines have some registers that cannot be copied directly to or
2443 from memory or even from other types of registers. An example is the
2444 @samp{MQ} register, which on most machines, can only be copied to or
2445 from general registers, but not memory. Below, we shall be using the
2446 term 'intermediate register' when a move operation cannot be performed
2447 directly, but has to be done by copying the source into the intermediate
2448 register first, and then copying the intermediate register to the
2449 destination. An intermediate register always has the same mode as
2450 source and destination. Since it holds the actual value being copied,
2451 reload might apply optimizations to re-use an intermediate register
2452 and eliding the copy from the source when it can determine that the
2453 intermediate register still holds the required value.
2455 Another kind of secondary reload is required on some machines which
2456 allow copying all registers to and from memory, but require a scratch
2457 register for stores to some memory locations (e.g., those with symbolic
2458 address on the RT, and those with certain symbolic address on the SPARC
2459 when compiling PIC)@. Scratch registers need not have the same mode
2460 as the value being copied, and usually hold a different value that
2461 that being copied. Special patterns in the md file are needed to
2462 describe how the copy is performed with the help of the scratch register;
2463 these patterns also describe the number, register class(es) and mode(s)
2464 of the scratch register(s).
2466 In some cases, both an intermediate and a scratch register are required.
2468 For input reloads, this target hook is called with nonzero @var{in_p},
2469 and @var{x} is an rtx that needs to be copied to a register in of class
2470 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2471 hook is called with zero @var{in_p}, and a register of class @var{reload_mode}
2472 needs to be copied to rtx @var{x} in @var{reload_mode}.
2474 If copying a register of @var{reload_class} from/to @var{x} requires
2475 an intermediate register, the hook @code{secondary_reload} should
2476 return the register class required for this intermediate register.
2477 If no intermediate register is required, it should return NO_REGS.
2478 If more than one intermediate register is required, describe the one
2479 that is closest in the copy chain to the reload register.
2481 If scratch registers are needed, you also have to describe how to
2482 perform the copy from/to the reload register to/from this
2483 closest intermediate register. Or if no intermediate register is
2484 required, but still a scratch register is needed, describe the
2485 copy from/to the reload register to/from the reload operand @var{x}.
2487 You do this by setting @code{sri->icode} to the instruction code of a pattern
2488 in the md file which performs the move. Operands 0 and 1 are the output
2489 and input of this copy, respectively. Operands from operand 2 onward are
2490 for scratch operands. These scratch operands must have a mode, and a
2491 single-register-class
2492 @c [later: or memory]
2495 When an intermediate register is used, the @code{secondary_reload}
2496 hook will be called again to determine how to copy the intermediate
2497 register to/from the reload operand @var{x}, so your hook must also
2498 have code to handle the register class of the intermediate operand.
2500 @c [For later: maybe we'll allow multi-alternative reload patterns -
2501 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2502 @c and match the constraints of input and output to determine the required
2503 @c alternative. A restriction would be that constraints used to match
2504 @c against reloads registers would have to be written as register class
2505 @c constraints, or we need a new target macro / hook that tells us if an
2506 @c arbitrary constraint can match an unknown register of a given class.
2507 @c Such a macro / hook would also be useful in other places.]
2510 @var{x} might be a pseudo-register or a @code{subreg} of a
2511 pseudo-register, which could either be in a hard register or in memory.
2512 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2513 in memory and the hard register number if it is in a register.
2515 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2516 currently not supported. For the time being, you will have to continue
2517 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2519 @code{copy_cost} also uses this target hook to find out how values are
2520 copied. If you want it to include some extra cost for the need to allocate
2521 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2522 Or if two dependent moves are supposed to have a lower cost than the sum
2523 of the individual moves due to expected fortuitous scheduling and/or special
2524 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2527 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2528 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2529 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2530 These macros are obsolete, new ports should use the target hook
2531 @code{TARGET_SECONDARY_RELOAD} instead.
2533 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2534 target hook. Older ports still define these macros to indicate to the
2535 reload phase that it may
2536 need to allocate at least one register for a reload in addition to the
2537 register to contain the data. Specifically, if copying @var{x} to a
2538 register @var{class} in @var{mode} requires an intermediate register,
2539 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2540 largest register class all of whose registers can be used as
2541 intermediate registers or scratch registers.
2543 If copying a register @var{class} in @var{mode} to @var{x} requires an
2544 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2545 was supposed to be defined be defined to return the largest register
2546 class required. If the
2547 requirements for input and output reloads were the same, the macro
2548 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2551 The values returned by these macros are often @code{GENERAL_REGS}.
2552 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2553 can be directly copied to or from a register of @var{class} in
2554 @var{mode} without requiring a scratch register. Do not define this
2555 macro if it would always return @code{NO_REGS}.
2557 If a scratch register is required (either with or without an
2558 intermediate register), you were supposed to define patterns for
2559 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2560 (@pxref{Standard Names}. These patterns, which were normally
2561 implemented with a @code{define_expand}, should be similar to the
2562 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2565 These patterns need constraints for the reload register and scratch
2567 contain a single register class. If the original reload register (whose
2568 class is @var{class}) can meet the constraint given in the pattern, the
2569 value returned by these macros is used for the class of the scratch
2570 register. Otherwise, two additional reload registers are required.
2571 Their classes are obtained from the constraints in the insn pattern.
2573 @var{x} might be a pseudo-register or a @code{subreg} of a
2574 pseudo-register, which could either be in a hard register or in memory.
2575 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2576 in memory and the hard register number if it is in a register.
2578 These macros should not be used in the case where a particular class of
2579 registers can only be copied to memory and not to another class of
2580 registers. In that case, secondary reload registers are not needed and
2581 would not be helpful. Instead, a stack location must be used to perform
2582 the copy and the @code{mov@var{m}} pattern should use memory as an
2583 intermediate storage. This case often occurs between floating-point and
2587 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2588 Certain machines have the property that some registers cannot be copied
2589 to some other registers without using memory. Define this macro on
2590 those machines to be a C expression that is nonzero if objects of mode
2591 @var{m} in registers of @var{class1} can only be copied to registers of
2592 class @var{class2} by storing a register of @var{class1} into memory
2593 and loading that memory location into a register of @var{class2}.
2595 Do not define this macro if its value would always be zero.
2598 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2599 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2600 allocates a stack slot for a memory location needed for register copies.
2601 If this macro is defined, the compiler instead uses the memory location
2602 defined by this macro.
2604 Do not define this macro if you do not define
2605 @code{SECONDARY_MEMORY_NEEDED}.
2608 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2609 When the compiler needs a secondary memory location to copy between two
2610 registers of mode @var{mode}, it normally allocates sufficient memory to
2611 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2612 load operations in a mode that many bits wide and whose class is the
2613 same as that of @var{mode}.
2615 This is right thing to do on most machines because it ensures that all
2616 bits of the register are copied and prevents accesses to the registers
2617 in a narrower mode, which some machines prohibit for floating-point
2620 However, this default behavior is not correct on some machines, such as
2621 the DEC Alpha, that store short integers in floating-point registers
2622 differently than in integer registers. On those machines, the default
2623 widening will not work correctly and you must define this macro to
2624 suppress that widening in some cases. See the file @file{alpha.h} for
2627 Do not define this macro if you do not define
2628 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2629 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2632 @defmac SMALL_REGISTER_CLASSES
2633 On some machines, it is risky to let hard registers live across arbitrary
2634 insns. Typically, these machines have instructions that require values
2635 to be in specific registers (like an accumulator), and reload will fail
2636 if the required hard register is used for another purpose across such an
2639 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero
2640 value on these machines. When this macro has a nonzero value, the
2641 compiler will try to minimize the lifetime of hard registers.
2643 It is always safe to define this macro with a nonzero value, but if you
2644 unnecessarily define it, you will reduce the amount of optimizations
2645 that can be performed in some cases. If you do not define this macro
2646 with a nonzero value when it is required, the compiler will run out of
2647 spill registers and print a fatal error message. For most machines, you
2648 should not define this macro at all.
2651 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2652 A C expression whose value is nonzero if pseudos that have been assigned
2653 to registers of class @var{class} would likely be spilled because
2654 registers of @var{class} are needed for spill registers.
2656 The default value of this macro returns 1 if @var{class} has exactly one
2657 register and zero otherwise. On most machines, this default should be
2658 used. Only define this macro to some other expression if pseudos
2659 allocated by @file{local-alloc.c} end up in memory because their hard
2660 registers were needed for spill registers. If this macro returns nonzero
2661 for those classes, those pseudos will only be allocated by
2662 @file{global.c}, which knows how to reallocate the pseudo to another
2663 register. If there would not be another register available for
2664 reallocation, you should not change the definition of this macro since
2665 the only effect of such a definition would be to slow down register
2669 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2670 A C expression for the maximum number of consecutive registers
2671 of class @var{class} needed to hold a value of mode @var{mode}.
2673 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2674 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2675 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2676 @var{mode})} for all @var{regno} values in the class @var{class}.
2678 This macro helps control the handling of multiple-word values
2682 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2683 If defined, a C expression that returns nonzero for a @var{class} for which
2684 a change from mode @var{from} to mode @var{to} is invalid.
2686 For the example, loading 32-bit integer or floating-point objects into
2687 floating-point registers on the Alpha extends them to 64 bits.
2688 Therefore loading a 64-bit object and then storing it as a 32-bit object
2689 does not store the low-order 32 bits, as would be the case for a normal
2690 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2694 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2695 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2696 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2700 @node Old Constraints
2701 @section Obsolete Macros for Defining Constraints
2702 @cindex defining constraints, obsolete method
2703 @cindex constraints, defining, obsolete method
2705 Machine-specific constraints can be defined with these macros instead
2706 of the machine description constructs described in @ref{Define
2707 Constraints}. This mechanism is obsolete. New ports should not use
2708 it; old ports should convert to the new mechanism.
2710 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2711 For the constraint at the start of @var{str}, which starts with the letter
2712 @var{c}, return the length. This allows you to have register class /
2713 constant / extra constraints that are longer than a single letter;
2714 you don't need to define this macro if you can do with single-letter
2715 constraints only. The definition of this macro should use
2716 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2717 to handle specially.
2718 There are some sanity checks in genoutput.c that check the constraint lengths
2719 for the md file, so you can also use this macro to help you while you are
2720 transitioning from a byzantine single-letter-constraint scheme: when you
2721 return a negative length for a constraint you want to re-use, genoutput
2722 will complain about every instance where it is used in the md file.
2725 @defmac REG_CLASS_FROM_LETTER (@var{char})
2726 A C expression which defines the machine-dependent operand constraint
2727 letters for register classes. If @var{char} is such a letter, the
2728 value should be the register class corresponding to it. Otherwise,
2729 the value should be @code{NO_REGS}. The register letter @samp{r},
2730 corresponding to class @code{GENERAL_REGS}, will not be passed
2731 to this macro; you do not need to handle it.
2734 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2735 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2736 passed in @var{str}, so that you can use suffixes to distinguish between
2740 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2741 A C expression that defines the machine-dependent operand constraint
2742 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2743 particular ranges of integer values. If @var{c} is one of those
2744 letters, the expression should check that @var{value}, an integer, is in
2745 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2746 not one of those letters, the value should be 0 regardless of
2750 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2751 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2752 string passed in @var{str}, so that you can use suffixes to distinguish
2753 between different variants.
2756 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2757 A C expression that defines the machine-dependent operand constraint
2758 letters that specify particular ranges of @code{const_double} values
2759 (@samp{G} or @samp{H}).
2761 If @var{c} is one of those letters, the expression should check that
2762 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2763 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2764 letters, the value should be 0 regardless of @var{value}.
2766 @code{const_double} is used for all floating-point constants and for
2767 @code{DImode} fixed-point constants. A given letter can accept either
2768 or both kinds of values. It can use @code{GET_MODE} to distinguish
2769 between these kinds.
2772 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2773 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2774 string passed in @var{str}, so that you can use suffixes to distinguish
2775 between different variants.
2778 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2779 A C expression that defines the optional machine-dependent constraint
2780 letters that can be used to segregate specific types of operands, usually
2781 memory references, for the target machine. Any letter that is not
2782 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2783 @code{REG_CLASS_FROM_CONSTRAINT}
2784 may be used. Normally this macro will not be defined.
2786 If it is required for a particular target machine, it should return 1
2787 if @var{value} corresponds to the operand type represented by the
2788 constraint letter @var{c}. If @var{c} is not defined as an extra
2789 constraint, the value returned should be 0 regardless of @var{value}.
2791 For example, on the ROMP, load instructions cannot have their output
2792 in r0 if the memory reference contains a symbolic address. Constraint
2793 letter @samp{Q} is defined as representing a memory address that does
2794 @emph{not} contain a symbolic address. An alternative is specified with
2795 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2796 alternative specifies @samp{m} on the input and a register class that
2797 does not include r0 on the output.
2800 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2801 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2802 in @var{str}, so that you can use suffixes to distinguish between different
2806 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2807 A C expression that defines the optional machine-dependent constraint
2808 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2809 be treated like memory constraints by the reload pass.
2811 It should return 1 if the operand type represented by the constraint
2812 at the start of @var{str}, the first letter of which is the letter @var{c},
2813 comprises a subset of all memory references including
2814 all those whose address is simply a base register. This allows the reload
2815 pass to reload an operand, if it does not directly correspond to the operand
2816 type of @var{c}, by copying its address into a base register.
2818 For example, on the S/390, some instructions do not accept arbitrary
2819 memory references, but only those that do not make use of an index
2820 register. The constraint letter @samp{Q} is defined via
2821 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2822 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2823 a @samp{Q} constraint can handle any memory operand, because the
2824 reload pass knows it can be reloaded by copying the memory address
2825 into a base register if required. This is analogous to the way
2826 a @samp{o} constraint can handle any memory operand.
2829 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2830 A C expression that defines the optional machine-dependent constraint
2831 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2832 @code{EXTRA_CONSTRAINT_STR}, that should
2833 be treated like address constraints by the reload pass.
2835 It should return 1 if the operand type represented by the constraint
2836 at the start of @var{str}, which starts with the letter @var{c}, comprises
2837 a subset of all memory addresses including
2838 all those that consist of just a base register. This allows the reload
2839 pass to reload an operand, if it does not directly correspond to the operand
2840 type of @var{str}, by copying it into a base register.
2842 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
2843 be used with the @code{address_operand} predicate. It is treated
2844 analogously to the @samp{p} constraint.
2847 @node Stack and Calling
2848 @section Stack Layout and Calling Conventions
2849 @cindex calling conventions
2851 @c prevent bad page break with this line
2852 This describes the stack layout and calling conventions.
2856 * Exception Handling::
2861 * Register Arguments::
2863 * Aggregate Return::
2868 * Stack Smashing Protection::
2872 @subsection Basic Stack Layout
2873 @cindex stack frame layout
2874 @cindex frame layout
2876 @c prevent bad page break with this line
2877 Here is the basic stack layout.
2879 @defmac STACK_GROWS_DOWNWARD
2880 Define this macro if pushing a word onto the stack moves the stack
2881 pointer to a smaller address.
2883 When we say, ``define this macro if @dots{}'', it means that the
2884 compiler checks this macro only with @code{#ifdef} so the precise
2885 definition used does not matter.
2888 @defmac STACK_PUSH_CODE
2889 This macro defines the operation used when something is pushed
2890 on the stack. In RTL, a push operation will be
2891 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
2893 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
2894 and @code{POST_INC}. Which of these is correct depends on
2895 the stack direction and on whether the stack pointer points
2896 to the last item on the stack or whether it points to the
2897 space for the next item on the stack.
2899 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
2900 defined, which is almost always right, and @code{PRE_INC} otherwise,
2901 which is often wrong.
2904 @defmac FRAME_GROWS_DOWNWARD
2905 Define this macro to nonzero value if the addresses of local variable slots
2906 are at negative offsets from the frame pointer.
2909 @defmac ARGS_GROW_DOWNWARD
2910 Define this macro if successive arguments to a function occupy decreasing
2911 addresses on the stack.
2914 @defmac STARTING_FRAME_OFFSET
2915 Offset from the frame pointer to the first local variable slot to be allocated.
2917 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2918 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2919 Otherwise, it is found by adding the length of the first slot to the
2920 value @code{STARTING_FRAME_OFFSET}.
2921 @c i'm not sure if the above is still correct.. had to change it to get
2922 @c rid of an overfull. --mew 2feb93
2925 @defmac STACK_ALIGNMENT_NEEDED
2926 Define to zero to disable final alignment of the stack during reload.
2927 The nonzero default for this macro is suitable for most ports.
2929 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
2930 is a register save block following the local block that doesn't require
2931 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
2932 stack alignment and do it in the backend.
2935 @defmac STACK_POINTER_OFFSET
2936 Offset from the stack pointer register to the first location at which
2937 outgoing arguments are placed. If not specified, the default value of
2938 zero is used. This is the proper value for most machines.
2940 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2941 the first location at which outgoing arguments are placed.
2944 @defmac FIRST_PARM_OFFSET (@var{fundecl})
2945 Offset from the argument pointer register to the first argument's
2946 address. On some machines it may depend on the data type of the
2949 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2950 the first argument's address.
2953 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
2954 Offset from the stack pointer register to an item dynamically allocated
2955 on the stack, e.g., by @code{alloca}.
2957 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
2958 length of the outgoing arguments. The default is correct for most
2959 machines. See @file{function.c} for details.
2962 @defmac INITIAL_FRAME_ADDRESS_RTX
2963 A C expression whose value is RTL representing the address of the initial
2964 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
2965 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
2966 default value will be used. Define this macro in order to make frame pointer
2967 elimination work in the presence of @code{__builtin_frame_address (count)} and
2968 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
2971 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
2972 A C expression whose value is RTL representing the address in a stack
2973 frame where the pointer to the caller's frame is stored. Assume that
2974 @var{frameaddr} is an RTL expression for the address of the stack frame
2977 If you don't define this macro, the default is to return the value
2978 of @var{frameaddr}---that is, the stack frame address is also the
2979 address of the stack word that points to the previous frame.
2982 @defmac SETUP_FRAME_ADDRESSES
2983 If defined, a C expression that produces the machine-specific code to
2984 setup the stack so that arbitrary frames can be accessed. For example,
2985 on the SPARC, we must flush all of the register windows to the stack
2986 before we can access arbitrary stack frames. You will seldom need to
2990 @deftypefn {Target Hook} bool TARGET_BUILTIN_SETJMP_FRAME_VALUE ()
2991 This target hook should return an rtx that is used to store
2992 the address of the current frame into the built in @code{setjmp} buffer.
2993 The default value, @code{virtual_stack_vars_rtx}, is correct for most
2994 machines. One reason you may need to define this target hook is if
2995 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
2998 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
2999 A C expression whose value is RTL representing the value of the return
3000 address for the frame @var{count} steps up from the current frame, after
3001 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3002 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3003 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3005 The value of the expression must always be the correct address when
3006 @var{count} is zero, but may be @code{NULL_RTX} if there is not way to
3007 determine the return address of other frames.
3010 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3011 Define this if the return address of a particular stack frame is accessed
3012 from the frame pointer of the previous stack frame.
3015 @defmac INCOMING_RETURN_ADDR_RTX
3016 A C expression whose value is RTL representing the location of the
3017 incoming return address at the beginning of any function, before the
3018 prologue. This RTL is either a @code{REG}, indicating that the return
3019 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3022 You only need to define this macro if you want to support call frame
3023 debugging information like that provided by DWARF 2.
3025 If this RTL is a @code{REG}, you should also define
3026 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3029 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3030 A C expression whose value is an integer giving a DWARF 2 column
3031 number that may be used as an alternate return column. This should
3032 be defined only if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3033 general register, but an alternate column needs to be used for
3037 @defmac DWARF_ZERO_REG
3038 A C expression whose value is an integer giving a DWARF 2 register
3039 number that is considered to always have the value zero. This should
3040 only be defined if the target has an architected zero register, and
3041 someone decided it was a good idea to use that register number to
3042 terminate the stack backtrace. New ports should avoid this.
3045 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3046 This target hook allows the backend to emit frame-related insns that
3047 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3048 info engine will invoke it on insns of the form
3050 (set (reg) (unspec [...] UNSPEC_INDEX))
3054 (set (reg) (unspec_volatile [...] UNSPECV_INDEX)).
3056 to let the backend emit the call frame instructions. @var{label} is
3057 the CFI label attached to the insn, @var{pattern} is the pattern of
3058 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3061 @defmac INCOMING_FRAME_SP_OFFSET
3062 A C expression whose value is an integer giving the offset, in bytes,
3063 from the value of the stack pointer register to the top of the stack
3064 frame at the beginning of any function, before the prologue. The top of
3065 the frame is defined to be the value of the stack pointer in the
3066 previous frame, just before the call instruction.
3068 You only need to define this macro if you want to support call frame
3069 debugging information like that provided by DWARF 2.
3072 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3073 A C expression whose value is an integer giving the offset, in bytes,
3074 from the argument pointer to the canonical frame address (cfa). The
3075 final value should coincide with that calculated by
3076 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3077 during virtual register instantiation.
3079 The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)},
3080 which is correct for most machines; in general, the arguments are found
3081 immediately before the stack frame. Note that this is not the case on
3082 some targets that save registers into the caller's frame, such as SPARC
3083 and rs6000, and so such targets need to define this macro.
3085 You only need to define this macro if the default is incorrect, and you
3086 want to support call frame debugging information like that provided by
3090 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3091 If defined, a C expression whose value is an integer giving the offset
3092 in bytes from the frame pointer to the canonical frame address (cfa).
3093 The final value should coincide with that calculated by
3094 @code{INCOMING_FRAME_SP_OFFSET}.
3096 Normally the CFA is calculated as an offset from the argument pointer,
3097 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3098 variable due to the ABI, this may not be possible. If this macro is
3099 defined, it implies that the virtual register instantiation should be
3100 based on the frame pointer instead of the argument pointer. Only one
3101 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3105 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3106 If defined, a C expression whose value is an integer giving the offset
3107 in bytes from the canonical frame address (cfa) to the frame base used
3108 in DWARF 2 debug information. The default is zero. A different value
3109 may reduce the size of debug information on some ports.
3112 @node Exception Handling
3113 @subsection Exception Handling Support
3114 @cindex exception handling
3116 @defmac EH_RETURN_DATA_REGNO (@var{N})
3117 A C expression whose value is the @var{N}th register number used for
3118 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3119 @var{N} registers are usable.
3121 The exception handling library routines communicate with the exception
3122 handlers via a set of agreed upon registers. Ideally these registers
3123 should be call-clobbered; it is possible to use call-saved registers,
3124 but may negatively impact code size. The target must support at least
3125 2 data registers, but should define 4 if there are enough free registers.
3127 You must define this macro if you want to support call frame exception
3128 handling like that provided by DWARF 2.
3131 @defmac EH_RETURN_STACKADJ_RTX
3132 A C expression whose value is RTL representing a location in which
3133 to store a stack adjustment to be applied before function return.
3134 This is used to unwind the stack to an exception handler's call frame.
3135 It will be assigned zero on code paths that return normally.
3137 Typically this is a call-clobbered hard register that is otherwise
3138 untouched by the epilogue, but could also be a stack slot.
3140 Do not define this macro if the stack pointer is saved and restored
3141 by the regular prolog and epilog code in the call frame itself; in
3142 this case, the exception handling library routines will update the
3143 stack location to be restored in place. Otherwise, you must define
3144 this macro if you want to support call frame exception handling like
3145 that provided by DWARF 2.
3148 @defmac EH_RETURN_HANDLER_RTX
3149 A C expression whose value is RTL representing a location in which
3150 to store the address of an exception handler to which we should
3151 return. It will not be assigned on code paths that return normally.
3153 Typically this is the location in the call frame at which the normal
3154 return address is stored. For targets that return by popping an
3155 address off the stack, this might be a memory address just below
3156 the @emph{target} call frame rather than inside the current call
3157 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3158 been assigned, so it may be used to calculate the location of the
3161 Some targets have more complex requirements than storing to an
3162 address calculable during initial code generation. In that case
3163 the @code{eh_return} instruction pattern should be used instead.
3165 If you want to support call frame exception handling, you must
3166 define either this macro or the @code{eh_return} instruction pattern.
3169 @defmac RETURN_ADDR_OFFSET
3170 If defined, an integer-valued C expression for which rtl will be generated
3171 to add it to the exception handler address before it is searched in the
3172 exception handling tables, and to subtract it again from the address before
3173 using it to return to the exception handler.
3176 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3177 This macro chooses the encoding of pointers embedded in the exception
3178 handling sections. If at all possible, this should be defined such
3179 that the exception handling section will not require dynamic relocations,
3180 and so may be read-only.
3182 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3183 @var{global} is true if the symbol may be affected by dynamic relocations.
3184 The macro should return a combination of the @code{DW_EH_PE_*} defines
3185 as found in @file{dwarf2.h}.
3187 If this macro is not defined, pointers will not be encoded but
3188 represented directly.
3191 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3192 This macro allows the target to emit whatever special magic is required
3193 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3194 Generic code takes care of pc-relative and indirect encodings; this must
3195 be defined if the target uses text-relative or data-relative encodings.
3197 This is a C statement that branches to @var{done} if the format was
3198 handled. @var{encoding} is the format chosen, @var{size} is the number
3199 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3203 @defmac MD_UNWIND_SUPPORT
3204 A string specifying a file to be #include'd in unwind-dw2.c. The file
3205 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3208 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3209 This macro allows the target to add cpu and operating system specific
3210 code to the call-frame unwinder for use when there is no unwind data
3211 available. The most common reason to implement this macro is to unwind
3212 through signal frames.
3214 This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c}
3215 and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3216 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3217 for the address of the code being executed and @code{context->cfa} for
3218 the stack pointer value. If the frame can be decoded, the register save
3219 addresses should be updated in @var{fs} and the macro should evaluate to
3220 @code{_URC_NO_REASON}. If the frame cannot be decoded, the macro should
3221 evaluate to @code{_URC_END_OF_STACK}.
3223 For proper signal handling in Java this macro is accompanied by
3224 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3227 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3228 This macro allows the target to add operating system specific code to the
3229 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3230 usually used for signal or interrupt frames.
3232 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3233 @var{context} is an @code{_Unwind_Context};
3234 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3235 for the abi and context in the @code{.unwabi} directive. If the
3236 @code{.unwabi} directive can be handled, the register save addresses should
3237 be updated in @var{fs}.
3240 @defmac TARGET_USES_WEAK_UNWIND_INFO
3241 A C expression that evaluates to true if the target requires unwind
3242 info to be given comdat linkage. Define it to be @code{1} if comdat
3243 linkage is necessary. The default is @code{0}.
3246 @node Stack Checking
3247 @subsection Specifying How Stack Checking is Done
3249 GCC will check that stack references are within the boundaries of
3250 the stack, if the @option{-fstack-check} is specified, in one of three ways:
3254 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3255 will assume that you have arranged for stack checking to be done at
3256 appropriate places in the configuration files, e.g., in
3257 @code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special
3261 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
3262 called @code{check_stack} in your @file{md} file, GCC will call that
3263 pattern with one argument which is the address to compare the stack
3264 value against. You must arrange for this pattern to report an error if
3265 the stack pointer is out of range.
3268 If neither of the above are true, GCC will generate code to periodically
3269 ``probe'' the stack pointer using the values of the macros defined below.
3272 Normally, you will use the default values of these macros, so GCC
3273 will use the third approach.
3275 @defmac STACK_CHECK_BUILTIN
3276 A nonzero value if stack checking is done by the configuration files in a
3277 machine-dependent manner. You should define this macro if stack checking
3278 is require by the ABI of your machine or if you would like to have to stack
3279 checking in some more efficient way than GCC's portable approach.
3280 The default value of this macro is zero.
3283 @defmac STACK_CHECK_PROBE_INTERVAL
3284 An integer representing the interval at which GCC must generate stack
3285 probe instructions. You will normally define this macro to be no larger
3286 than the size of the ``guard pages'' at the end of a stack area. The
3287 default value of 4096 is suitable for most systems.
3290 @defmac STACK_CHECK_PROBE_LOAD
3291 A integer which is nonzero if GCC should perform the stack probe
3292 as a load instruction and zero if GCC should use a store instruction.
3293 The default is zero, which is the most efficient choice on most systems.
3296 @defmac STACK_CHECK_PROTECT
3297 The number of bytes of stack needed to recover from a stack overflow,
3298 for languages where such a recovery is supported. The default value of
3299 75 words should be adequate for most machines.
3302 @defmac STACK_CHECK_MAX_FRAME_SIZE
3303 The maximum size of a stack frame, in bytes. GCC will generate probe
3304 instructions in non-leaf functions to ensure at least this many bytes of
3305 stack are available. If a stack frame is larger than this size, stack
3306 checking will not be reliable and GCC will issue a warning. The
3307 default is chosen so that GCC only generates one instruction on most
3308 systems. You should normally not change the default value of this macro.
3311 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3312 GCC uses this value to generate the above warning message. It
3313 represents the amount of fixed frame used by a function, not including
3314 space for any callee-saved registers, temporaries and user variables.
3315 You need only specify an upper bound for this amount and will normally
3316 use the default of four words.
3319 @defmac STACK_CHECK_MAX_VAR_SIZE
3320 The maximum size, in bytes, of an object that GCC will place in the
3321 fixed area of the stack frame when the user specifies
3322 @option{-fstack-check}.
3323 GCC computed the default from the values of the above macros and you will
3324 normally not need to override that default.
3328 @node Frame Registers
3329 @subsection Registers That Address the Stack Frame
3331 @c prevent bad page break with this line
3332 This discusses registers that address the stack frame.
3334 @defmac STACK_POINTER_REGNUM
3335 The register number of the stack pointer register, which must also be a
3336 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3337 the hardware determines which register this is.
3340 @defmac FRAME_POINTER_REGNUM
3341 The register number of the frame pointer register, which is used to
3342 access automatic variables in the stack frame. On some machines, the
3343 hardware determines which register this is. On other machines, you can
3344 choose any register you wish for this purpose.
3347 @defmac HARD_FRAME_POINTER_REGNUM
3348 On some machines the offset between the frame pointer and starting
3349 offset of the automatic variables is not known until after register
3350 allocation has been done (for example, because the saved registers are
3351 between these two locations). On those machines, define
3352 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3353 be used internally until the offset is known, and define
3354 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3355 used for the frame pointer.
3357 You should define this macro only in the very rare circumstances when it
3358 is not possible to calculate the offset between the frame pointer and
3359 the automatic variables until after register allocation has been
3360 completed. When this macro is defined, you must also indicate in your
3361 definition of @code{ELIMINABLE_REGS} how to eliminate
3362 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3363 or @code{STACK_POINTER_REGNUM}.
3365 Do not define this macro if it would be the same as
3366 @code{FRAME_POINTER_REGNUM}.
3369 @defmac ARG_POINTER_REGNUM
3370 The register number of the arg pointer register, which is used to access
3371 the function's argument list. On some machines, this is the same as the
3372 frame pointer register. On some machines, the hardware determines which
3373 register this is. On other machines, you can choose any register you
3374 wish for this purpose. If this is not the same register as the frame
3375 pointer register, then you must mark it as a fixed register according to
3376 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3377 (@pxref{Elimination}).
3380 @defmac RETURN_ADDRESS_POINTER_REGNUM
3381 The register number of the return address pointer register, which is used to
3382 access the current function's return address from the stack. On some
3383 machines, the return address is not at a fixed offset from the frame
3384 pointer or stack pointer or argument pointer. This register can be defined
3385 to point to the return address on the stack, and then be converted by
3386 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3388 Do not define this macro unless there is no other way to get the return
3389 address from the stack.
3392 @defmac STATIC_CHAIN_REGNUM
3393 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3394 Register numbers used for passing a function's static chain pointer. If
3395 register windows are used, the register number as seen by the called
3396 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3397 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3398 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3401 The static chain register need not be a fixed register.
3403 If the static chain is passed in memory, these macros should not be
3404 defined; instead, the next two macros should be defined.
3407 @defmac STATIC_CHAIN
3408 @defmacx STATIC_CHAIN_INCOMING
3409 If the static chain is passed in memory, these macros provide rtx giving
3410 @code{mem} expressions that denote where they are stored.
3411 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
3412 as seen by the calling and called functions, respectively. Often the former
3413 will be at an offset from the stack pointer and the latter at an offset from
3416 @findex stack_pointer_rtx
3417 @findex frame_pointer_rtx
3418 @findex arg_pointer_rtx
3419 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3420 @code{arg_pointer_rtx} will have been initialized prior to the use of these
3421 macros and should be used to refer to those items.
3423 If the static chain is passed in a register, the two previous macros should
3427 @defmac DWARF_FRAME_REGISTERS
3428 This macro specifies the maximum number of hard registers that can be
3429 saved in a call frame. This is used to size data structures used in
3430 DWARF2 exception handling.
3432 Prior to GCC 3.0, this macro was needed in order to establish a stable
3433 exception handling ABI in the face of adding new hard registers for ISA
3434 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3435 in the number of hard registers. Nevertheless, this macro can still be
3436 used to reduce the runtime memory requirements of the exception handling
3437 routines, which can be substantial if the ISA contains a lot of
3438 registers that are not call-saved.
3440 If this macro is not defined, it defaults to
3441 @code{FIRST_PSEUDO_REGISTER}.
3444 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3446 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3447 for backward compatibility in pre GCC 3.0 compiled code.
3449 If this macro is not defined, it defaults to
3450 @code{DWARF_FRAME_REGISTERS}.
3453 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3455 Define this macro if the target's representation for dwarf registers
3456 is different than the internal representation for unwind column.
3457 Given a dwarf register, this macro should return the internal unwind
3458 column number to use instead.
3460 See the PowerPC's SPE target for an example.
3463 @defmac DWARF_FRAME_REGNUM (@var{regno})
3465 Define this macro if the target's representation for dwarf registers
3466 used in .eh_frame or .debug_frame is different from that used in other
3467 debug info sections. Given a GCC hard register number, this macro
3468 should return the .eh_frame register number. The default is
3469 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3473 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3475 Define this macro to map register numbers held in the call frame info
3476 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3477 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3478 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3479 return @code{@var{regno}}.
3484 @subsection Eliminating Frame Pointer and Arg Pointer
3486 @c prevent bad page break with this line
3487 This is about eliminating the frame pointer and arg pointer.
3489 @defmac FRAME_POINTER_REQUIRED
3490 A C expression which is nonzero if a function must have and use a frame
3491 pointer. This expression is evaluated in the reload pass. If its value is
3492 nonzero the function will have a frame pointer.
3494 The expression can in principle examine the current function and decide
3495 according to the facts, but on most machines the constant 0 or the
3496 constant 1 suffices. Use 0 when the machine allows code to be generated
3497 with no frame pointer, and doing so saves some time or space. Use 1
3498 when there is no possible advantage to avoiding a frame pointer.
3500 In certain cases, the compiler does not know how to produce valid code
3501 without a frame pointer. The compiler recognizes those cases and
3502 automatically gives the function a frame pointer regardless of what
3503 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
3506 In a function that does not require a frame pointer, the frame pointer
3507 register can be allocated for ordinary usage, unless you mark it as a
3508 fixed register. See @code{FIXED_REGISTERS} for more information.
3511 @findex get_frame_size
3512 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3513 A C statement to store in the variable @var{depth-var} the difference
3514 between the frame pointer and the stack pointer values immediately after
3515 the function prologue. The value would be computed from information
3516 such as the result of @code{get_frame_size ()} and the tables of
3517 registers @code{regs_ever_live} and @code{call_used_regs}.
3519 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3520 need not be defined. Otherwise, it must be defined even if
3521 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
3522 case, you may set @var{depth-var} to anything.
3525 @defmac ELIMINABLE_REGS
3526 If defined, this macro specifies a table of register pairs used to
3527 eliminate unneeded registers that point into the stack frame. If it is not
3528 defined, the only elimination attempted by the compiler is to replace
3529 references to the frame pointer with references to the stack pointer.
3531 The definition of this macro is a list of structure initializations, each
3532 of which specifies an original and replacement register.
3534 On some machines, the position of the argument pointer is not known until
3535 the compilation is completed. In such a case, a separate hard register
3536 must be used for the argument pointer. This register can be eliminated by
3537 replacing it with either the frame pointer or the argument pointer,
3538 depending on whether or not the frame pointer has been eliminated.
3540 In this case, you might specify:
3542 #define ELIMINABLE_REGS \
3543 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3544 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3545 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3548 Note that the elimination of the argument pointer with the stack pointer is
3549 specified first since that is the preferred elimination.
3552 @defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
3553 A C expression that returns nonzero if the compiler is allowed to try
3554 to replace register number @var{from-reg} with register number
3555 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
3556 is defined, and will usually be the constant 1, since most of the cases
3557 preventing register elimination are things that the compiler already
3561 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3562 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3563 specifies the initial difference between the specified pair of
3564 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3568 @node Stack Arguments
3569 @subsection Passing Function Arguments on the Stack
3570 @cindex arguments on stack
3571 @cindex stack arguments
3573 The macros in this section control how arguments are passed
3574 on the stack. See the following section for other macros that
3575 control passing certain arguments in registers.
3577 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype})
3578 This target hook returns @code{true} if an argument declared in a
3579 prototype as an integral type smaller than @code{int} should actually be
3580 passed as an @code{int}. In addition to avoiding errors in certain
3581 cases of mismatch, it also makes for better code on certain machines.
3582 The default is to not promote prototypes.
3586 A C expression. If nonzero, push insns will be used to pass
3588 If the target machine does not have a push instruction, set it to zero.
3589 That directs GCC to use an alternate strategy: to
3590 allocate the entire argument block and then store the arguments into
3591 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3594 @defmac PUSH_ARGS_REVERSED
3595 A C expression. If nonzero, function arguments will be evaluated from
3596 last to first, rather than from first to last. If this macro is not
3597 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3598 and args grow in opposite directions, and 0 otherwise.
3601 @defmac PUSH_ROUNDING (@var{npushed})
3602 A C expression that is the number of bytes actually pushed onto the
3603 stack when an instruction attempts to push @var{npushed} bytes.
3605 On some machines, the definition
3608 #define PUSH_ROUNDING(BYTES) (BYTES)
3612 will suffice. But on other machines, instructions that appear
3613 to push one byte actually push two bytes in an attempt to maintain
3614 alignment. Then the definition should be
3617 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3621 @findex current_function_outgoing_args_size
3622 @defmac ACCUMULATE_OUTGOING_ARGS
3623 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3624 will be computed and placed into the variable
3625 @code{current_function_outgoing_args_size}. No space will be pushed
3626 onto the stack for each call; instead, the function prologue should
3627 increase the stack frame size by this amount.
3629 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3633 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3634 Define this macro if functions should assume that stack space has been
3635 allocated for arguments even when their values are passed in
3638 The value of this macro is the size, in bytes, of the area reserved for
3639 arguments passed in registers for the function represented by @var{fndecl},
3640 which can be zero if GCC is calling a library function.
3642 This space can be allocated by the caller, or be a part of the
3643 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3646 @c above is overfull. not sure what to do. --mew 5feb93 did
3647 @c something, not sure if it looks good. --mew 10feb93
3649 @defmac OUTGOING_REG_PARM_STACK_SPACE
3650 Define this if it is the responsibility of the caller to allocate the area
3651 reserved for arguments passed in registers.
3653 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3654 whether the space for these arguments counts in the value of
3655 @code{current_function_outgoing_args_size}.
3658 @defmac STACK_PARMS_IN_REG_PARM_AREA
3659 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3660 stack parameters don't skip the area specified by it.
3661 @c i changed this, makes more sens and it should have taken care of the
3662 @c overfull.. not as specific, tho. --mew 5feb93
3664 Normally, when a parameter is not passed in registers, it is placed on the
3665 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3666 suppresses this behavior and causes the parameter to be passed on the
3667 stack in its natural location.
3670 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3671 A C expression that should indicate the number of bytes of its own
3672 arguments that a function pops on returning, or 0 if the
3673 function pops no arguments and the caller must therefore pop them all
3674 after the function returns.
3676 @var{fundecl} is a C variable whose value is a tree node that describes
3677 the function in question. Normally it is a node of type
3678 @code{FUNCTION_DECL} that describes the declaration of the function.
3679 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3681 @var{funtype} is a C variable whose value is a tree node that
3682 describes the function in question. Normally it is a node of type
3683 @code{FUNCTION_TYPE} that describes the data type of the function.
3684 From this it is possible to obtain the data types of the value and
3685 arguments (if known).
3687 When a call to a library function is being considered, @var{fundecl}
3688 will contain an identifier node for the library function. Thus, if
3689 you need to distinguish among various library functions, you can do so
3690 by their names. Note that ``library function'' in this context means
3691 a function used to perform arithmetic, whose name is known specially
3692 in the compiler and was not mentioned in the C code being compiled.
3694 @var{stack-size} is the number of bytes of arguments passed on the
3695 stack. If a variable number of bytes is passed, it is zero, and
3696 argument popping will always be the responsibility of the calling function.
3698 On the VAX, all functions always pop their arguments, so the definition
3699 of this macro is @var{stack-size}. On the 68000, using the standard
3700 calling convention, no functions pop their arguments, so the value of
3701 the macro is always 0 in this case. But an alternative calling
3702 convention is available in which functions that take a fixed number of
3703 arguments pop them but other functions (such as @code{printf}) pop
3704 nothing (the caller pops all). When this convention is in use,
3705 @var{funtype} is examined to determine whether a function takes a fixed
3706 number of arguments.
3709 @defmac CALL_POPS_ARGS (@var{cum})
3710 A C expression that should indicate the number of bytes a call sequence
3711 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3712 when compiling a function call.
3714 @var{cum} is the variable in which all arguments to the called function
3715 have been accumulated.
3717 On certain architectures, such as the SH5, a call trampoline is used
3718 that pops certain registers off the stack, depending on the arguments
3719 that have been passed to the function. Since this is a property of the
3720 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3724 @node Register Arguments
3725 @subsection Passing Arguments in Registers
3726 @cindex arguments in registers
3727 @cindex registers arguments
3729 This section describes the macros which let you control how various
3730 types of arguments are passed in registers or how they are arranged in
3733 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3734 A C expression that controls whether a function argument is passed
3735 in a register, and which register.
3737 The arguments are @var{cum}, which summarizes all the previous
3738 arguments; @var{mode}, the machine mode of the argument; @var{type},
3739 the data type of the argument as a tree node or 0 if that is not known
3740 (which happens for C support library functions); and @var{named},
3741 which is 1 for an ordinary argument and 0 for nameless arguments that
3742 correspond to @samp{@dots{}} in the called function's prototype.
3743 @var{type} can be an incomplete type if a syntax error has previously
3746 The value of the expression is usually either a @code{reg} RTX for the
3747 hard register in which to pass the argument, or zero to pass the
3748 argument on the stack.
3750 For machines like the VAX and 68000, where normally all arguments are
3751 pushed, zero suffices as a definition.
3753 The value of the expression can also be a @code{parallel} RTX@. This is
3754 used when an argument is passed in multiple locations. The mode of the
3755 @code{parallel} should be the mode of the entire argument. The
3756 @code{parallel} holds any number of @code{expr_list} pairs; each one
3757 describes where part of the argument is passed. In each
3758 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3759 register in which to pass this part of the argument, and the mode of the
3760 register RTX indicates how large this part of the argument is. The
3761 second operand of the @code{expr_list} is a @code{const_int} which gives
3762 the offset in bytes into the entire argument of where this part starts.
3763 As a special exception the first @code{expr_list} in the @code{parallel}
3764 RTX may have a first operand of zero. This indicates that the entire
3765 argument is also stored on the stack.
3767 The last time this macro is called, it is called with @code{MODE ==
3768 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3769 pattern as operands 2 and 3 respectively.
3771 @cindex @file{stdarg.h} and register arguments
3772 The usual way to make the ISO library @file{stdarg.h} work on a machine
3773 where some arguments are usually passed in registers, is to cause
3774 nameless arguments to be passed on the stack instead. This is done
3775 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
3777 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
3778 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
3779 You may use the hook @code{targetm.calls.must_pass_in_stack}
3780 in the definition of this macro to determine if this argument is of a
3781 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3782 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
3783 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3784 defined, the argument will be computed in the stack and then loaded into
3788 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, tree @var{type})
3789 This target hook should return @code{true} if we should not pass @var{type}
3790 solely in registers. The file @file{expr.h} defines a
3791 definition that is usually appropriate, refer to @file{expr.h} for additional
3795 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3796 Define this macro if the target machine has ``register windows'', so
3797 that the register in which a function sees an arguments is not
3798 necessarily the same as the one in which the caller passed the
3801 For such machines, @code{FUNCTION_ARG} computes the register in which
3802 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
3803 be defined in a similar fashion to tell the function being called
3804 where the arguments will arrive.
3806 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
3807 serves both purposes.
3810 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3811 This target hook returns the number of bytes at the beginning of an
3812 argument that must be put in registers. The value must be zero for
3813 arguments that are passed entirely in registers or that are entirely
3814 pushed on the stack.
3816 On some machines, certain arguments must be passed partially in
3817 registers and partially in memory. On these machines, typically the
3818 first few words of arguments are passed in registers, and the rest
3819 on the stack. If a multi-word argument (a @code{double} or a
3820 structure) crosses that boundary, its first few words must be passed
3821 in registers and the rest must be pushed. This macro tells the
3822 compiler when this occurs, and how many bytes should go in registers.
3824 @code{FUNCTION_ARG} for these arguments should return the first
3825 register to be used by the caller for this argument; likewise
3826 @code{FUNCTION_INCOMING_ARG}, for the called function.
3829 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3830 This target hook should return @code{true} if an argument at the
3831 position indicated by @var{cum} should be passed by reference. This
3832 predicate is queried after target independent reasons for being
3833 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
3835 If the hook returns true, a copy of that argument is made in memory and a
3836 pointer to the argument is passed instead of the argument itself.
3837 The pointer is passed in whatever way is appropriate for passing a pointer
3841 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3842 The function argument described by the parameters to this hook is
3843 known to be passed by reference. The hook should return true if the
3844 function argument should be copied by the callee instead of copied
3847 For any argument for which the hook returns true, if it can be
3848 determined that the argument is not modified, then a copy need
3851 The default version of this hook always returns false.
3854 @defmac CUMULATIVE_ARGS
3855 A C type for declaring a variable that is used as the first argument of
3856 @code{FUNCTION_ARG} and other related values. For some target machines,
3857 the type @code{int} suffices and can hold the number of bytes of
3860 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
3861 arguments that have been passed on the stack. The compiler has other
3862 variables to keep track of that. For target machines on which all
3863 arguments are passed on the stack, there is no need to store anything in
3864 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
3865 should not be empty, so use @code{int}.
3868 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
3869 A C statement (sans semicolon) for initializing the variable
3870 @var{cum} for the state at the beginning of the argument list. The
3871 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
3872 is the tree node for the data type of the function which will receive
3873 the args, or 0 if the args are to a compiler support library function.
3874 For direct calls that are not libcalls, @var{fndecl} contain the
3875 declaration node of the function. @var{fndecl} is also set when
3876 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
3877 being compiled. @var{n_named_args} is set to the number of named
3878 arguments, including a structure return address if it is passed as a
3879 parameter, when making a call. When processing incoming arguments,
3880 @var{n_named_args} is set to @minus{}1.
3882 When processing a call to a compiler support library function,
3883 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
3884 contains the name of the function, as a string. @var{libname} is 0 when
3885 an ordinary C function call is being processed. Thus, each time this
3886 macro is called, either @var{libname} or @var{fntype} is nonzero, but
3887 never both of them at once.
3890 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
3891 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
3892 it gets a @code{MODE} argument instead of @var{fntype}, that would be
3893 @code{NULL}. @var{indirect} would always be zero, too. If this macro
3894 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
3895 0)} is used instead.
3898 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
3899 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
3900 finding the arguments for the function being compiled. If this macro is
3901 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
3903 The value passed for @var{libname} is always 0, since library routines
3904 with special calling conventions are never compiled with GCC@. The
3905 argument @var{libname} exists for symmetry with
3906 @code{INIT_CUMULATIVE_ARGS}.
3907 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
3908 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
3911 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
3912 A C statement (sans semicolon) to update the summarizer variable
3913 @var{cum} to advance past an argument in the argument list. The
3914 values @var{mode}, @var{type} and @var{named} describe that argument.
3915 Once this is done, the variable @var{cum} is suitable for analyzing
3916 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
3918 This macro need not do anything if the argument in question was passed
3919 on the stack. The compiler knows how to track the amount of stack space
3920 used for arguments without any special help.
3923 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
3924 If defined, a C expression which determines whether, and in which direction,
3925 to pad out an argument with extra space. The value should be of type
3926 @code{enum direction}: either @code{upward} to pad above the argument,
3927 @code{downward} to pad below, or @code{none} to inhibit padding.
3929 The @emph{amount} of padding is always just enough to reach the next
3930 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
3933 This macro has a default definition which is right for most systems.
3934 For little-endian machines, the default is to pad upward. For
3935 big-endian machines, the default is to pad downward for an argument of
3936 constant size shorter than an @code{int}, and upward otherwise.
3939 @defmac PAD_VARARGS_DOWN
3940 If defined, a C expression which determines whether the default
3941 implementation of va_arg will attempt to pad down before reading the
3942 next argument, if that argument is smaller than its aligned space as
3943 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
3944 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
3947 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
3948 Specify padding for the last element of a block move between registers and
3949 memory. @var{first} is nonzero if this is the only element. Defining this
3950 macro allows better control of register function parameters on big-endian
3951 machines, without using @code{PARALLEL} rtl. In particular,
3952 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
3953 registers, as there is no longer a "wrong" part of a register; For example,
3954 a three byte aggregate may be passed in the high part of a register if so
3958 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
3959 If defined, a C expression that gives the alignment boundary, in bits,
3960 of an argument with the specified mode and type. If it is not defined,
3961 @code{PARM_BOUNDARY} is used for all arguments.
3964 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
3965 A C expression that is nonzero if @var{regno} is the number of a hard
3966 register in which function arguments are sometimes passed. This does
3967 @emph{not} include implicit arguments such as the static chain and
3968 the structure-value address. On many machines, no registers can be
3969 used for this purpose since all function arguments are pushed on the
3973 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type})
3974 This hook should return true if parameter of type @var{type} are passed
3975 as two scalar parameters. By default, GCC will attempt to pack complex
3976 arguments into the target's word size. Some ABIs require complex arguments
3977 to be split and treated as their individual components. For example, on
3978 AIX64, complex floats should be passed in a pair of floating point
3979 registers, even though a complex float would fit in one 64-bit floating
3982 The default value of this hook is @code{NULL}, which is treated as always
3986 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
3987 This hook returns a type node for @code{va_list} for the target.
3988 The default version of the hook returns @code{void*}.
3991 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, tree *@var{pre_p}, tree *@var{post_p})
3992 This hook performs target-specific gimplification of
3993 @code{VA_ARG_EXPR}. The first two parameters correspond to the
3994 arguments to @code{va_arg}; the latter two are as in
3995 @code{gimplify.c:gimplify_expr}.
3998 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
3999 Define this to return nonzero if the port can handle pointers
4000 with machine mode @var{mode}. The default version of this
4001 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4004 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4005 Define this to return nonzero if the port is prepared to handle
4006 insns involving scalar mode @var{mode}. For a scalar mode to be
4007 considered supported, all the basic arithmetic and comparisons
4010 The default version of this hook returns true for any mode
4011 required to handle the basic C types (as defined by the port).
4012 Included here are the double-word arithmetic supported by the
4013 code in @file{optabs.c}.
4016 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4017 Define this to return nonzero if the port is prepared to handle
4018 insns involving vector mode @var{mode}. At the very least, it
4019 must have move patterns for this mode.
4023 @subsection How Scalar Function Values Are Returned
4024 @cindex return values in registers
4025 @cindex values, returned by functions
4026 @cindex scalars, returned as values
4028 This section discusses the macros that control returning scalars as
4029 values---values that can fit in registers.
4031 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (tree @var{ret_type}, tree @var{fn_decl_or_type}, bool @var{outgoing})
4033 Define this to return an RTX representing the place where a function
4034 returns or receives a value of data type @var{ret_type}, a tree node
4035 node representing a data type. @var{fn_decl_or_type} is a tree node
4036 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4037 function being called. If @var{outgoing} is false, the hook should
4038 compute the register in which the caller will see the return value.
4039 Otherwise, the hook should return an RTX representing the place where
4040 a function returns a value.
4042 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4043 (Actually, on most machines, scalar values are returned in the same
4044 place regardless of mode.) The value of the expression is usually a
4045 @code{reg} RTX for the hard register where the return value is stored.
4046 The value can also be a @code{parallel} RTX, if the return value is in
4047 multiple places. See @code{FUNCTION_ARG} for an explanation of the
4048 @code{parallel} form.
4050 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4051 the same promotion rules specified in @code{PROMOTE_MODE} if
4052 @var{valtype} is a scalar type.
4054 If the precise function being called is known, @var{func} is a tree
4055 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4056 pointer. This makes it possible to use a different value-returning
4057 convention for specific functions when all their calls are
4060 Some target machines have ``register windows'' so that the register in
4061 which a function returns its value is not the same as the one in which
4062 the caller sees the value. For such machines, you should return
4063 different RTX depending on @var{outgoing}.
4065 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4066 aggregate data types, because these are returned in another way. See
4067 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4070 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4071 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4072 a new target instead.
4075 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
4076 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4077 a new target instead.
4080 @defmac LIBCALL_VALUE (@var{mode})
4081 A C expression to create an RTX representing the place where a library
4082 function returns a value of mode @var{mode}. If the precise function
4083 being called is known, @var{func} is a tree node
4084 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4085 pointer. This makes it possible to use a different value-returning
4086 convention for specific functions when all their calls are
4089 Note that ``library function'' in this context means a compiler
4090 support routine, used to perform arithmetic, whose name is known
4091 specially by the compiler and was not mentioned in the C code being
4094 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
4095 data types, because none of the library functions returns such types.
4098 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4099 A C expression that is nonzero if @var{regno} is the number of a hard
4100 register in which the values of called function may come back.
4102 A register whose use for returning values is limited to serving as the
4103 second of a pair (for a value of type @code{double}, say) need not be
4104 recognized by this macro. So for most machines, this definition
4108 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4111 If the machine has register windows, so that the caller and the called
4112 function use different registers for the return value, this macro
4113 should recognize only the caller's register numbers.
4116 @defmac APPLY_RESULT_SIZE
4117 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4118 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4119 saving and restoring an arbitrary return value.
4122 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type})
4123 This hook should return true if values of type @var{type} are returned
4124 at the most significant end of a register (in other words, if they are
4125 padded at the least significant end). You can assume that @var{type}
4126 is returned in a register; the caller is required to check this.
4128 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4129 be able to hold the complete return value. For example, if a 1-, 2-
4130 or 3-byte structure is returned at the most significant end of a
4131 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4135 @node Aggregate Return
4136 @subsection How Large Values Are Returned
4137 @cindex aggregates as return values
4138 @cindex large return values
4139 @cindex returning aggregate values
4140 @cindex structure value address
4142 When a function value's mode is @code{BLKmode} (and in some other
4143 cases), the value is not returned according to
4144 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4145 caller passes the address of a block of memory in which the value
4146 should be stored. This address is called the @dfn{structure value
4149 This section describes how to control returning structure values in
4152 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype})
4153 This target hook should return a nonzero value to say to return the
4154 function value in memory, just as large structures are always returned.
4155 Here @var{type} will be the data type of the value, and @var{fntype}
4156 will be the type of the function doing the returning, or @code{NULL} for
4159 Note that values of mode @code{BLKmode} must be explicitly handled
4160 by this function. Also, the option @option{-fpcc-struct-return}
4161 takes effect regardless of this macro. On most systems, it is
4162 possible to leave the hook undefined; this causes a default
4163 definition to be used, whose value is the constant 1 for @code{BLKmode}
4164 values, and 0 otherwise.
4166 Do not use this hook to indicate that structures and unions should always
4167 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4171 @defmac DEFAULT_PCC_STRUCT_RETURN
4172 Define this macro to be 1 if all structure and union return values must be
4173 in memory. Since this results in slower code, this should be defined
4174 only if needed for compatibility with other compilers or with an ABI@.
4175 If you define this macro to be 0, then the conventions used for structure
4176 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4179 If not defined, this defaults to the value 1.
4182 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4183 This target hook should return the location of the structure value
4184 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4185 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4186 be @code{NULL}, for libcalls. You do not need to define this target
4187 hook if the address is always passed as an ``invisible'' first
4190 On some architectures the place where the structure value address
4191 is found by the called function is not the same place that the
4192 caller put it. This can be due to register windows, or it could
4193 be because the function prologue moves it to a different place.
4194 @var{incoming} is @code{1} or @code{2} when the location is needed in
4195 the context of the called function, and @code{0} in the context of
4198 If @var{incoming} is nonzero and the address is to be found on the
4199 stack, return a @code{mem} which refers to the frame pointer. If
4200 @var{incoming} is @code{2}, the result is being used to fetch the
4201 structure value address at the beginning of a function. If you need
4202 to emit adjusting code, you should do it at this point.
4205 @defmac PCC_STATIC_STRUCT_RETURN
4206 Define this macro if the usual system convention on the target machine
4207 for returning structures and unions is for the called function to return
4208 the address of a static variable containing the value.
4210 Do not define this if the usual system convention is for the caller to
4211 pass an address to the subroutine.
4213 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4214 nothing when you use @option{-freg-struct-return} mode.
4218 @subsection Caller-Saves Register Allocation
4220 If you enable it, GCC can save registers around function calls. This
4221 makes it possible to use call-clobbered registers to hold variables that
4222 must live across calls.
4224 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4225 A C expression to determine whether it is worthwhile to consider placing
4226 a pseudo-register in a call-clobbered hard register and saving and
4227 restoring it around each function call. The expression should be 1 when
4228 this is worth doing, and 0 otherwise.
4230 If you don't define this macro, a default is used which is good on most
4231 machines: @code{4 * @var{calls} < @var{refs}}.
4234 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4235 A C expression specifying which mode is required for saving @var{nregs}
4236 of a pseudo-register in call-clobbered hard register @var{regno}. If
4237 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4238 returned. For most machines this macro need not be defined since GCC
4239 will select the smallest suitable mode.
4242 @node Function Entry
4243 @subsection Function Entry and Exit
4244 @cindex function entry and exit
4248 This section describes the macros that output function entry
4249 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4251 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4252 If defined, a function that outputs the assembler code for entry to a
4253 function. The prologue is responsible for setting up the stack frame,
4254 initializing the frame pointer register, saving registers that must be
4255 saved, and allocating @var{size} additional bytes of storage for the
4256 local variables. @var{size} is an integer. @var{file} is a stdio
4257 stream to which the assembler code should be output.
4259 The label for the beginning of the function need not be output by this
4260 macro. That has already been done when the macro is run.
4262 @findex regs_ever_live
4263 To determine which registers to save, the macro can refer to the array
4264 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4265 @var{r} is used anywhere within the function. This implies the function
4266 prologue should save register @var{r}, provided it is not one of the
4267 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4268 @code{regs_ever_live}.)
4270 On machines that have ``register windows'', the function entry code does
4271 not save on the stack the registers that are in the windows, even if
4272 they are supposed to be preserved by function calls; instead it takes
4273 appropriate steps to ``push'' the register stack, if any non-call-used
4274 registers are used in the function.
4276 @findex frame_pointer_needed
4277 On machines where functions may or may not have frame-pointers, the
4278 function entry code must vary accordingly; it must set up the frame
4279 pointer if one is wanted, and not otherwise. To determine whether a
4280 frame pointer is in wanted, the macro can refer to the variable
4281 @code{frame_pointer_needed}. The variable's value will be 1 at run
4282 time in a function that needs a frame pointer. @xref{Elimination}.
4284 The function entry code is responsible for allocating any stack space
4285 required for the function. This stack space consists of the regions
4286 listed below. In most cases, these regions are allocated in the
4287 order listed, with the last listed region closest to the top of the
4288 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4289 the highest address if it is not defined). You can use a different order
4290 for a machine if doing so is more convenient or required for
4291 compatibility reasons. Except in cases where required by standard
4292 or by a debugger, there is no reason why the stack layout used by GCC
4293 need agree with that used by other compilers for a machine.
4296 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4297 If defined, a function that outputs assembler code at the end of a
4298 prologue. This should be used when the function prologue is being
4299 emitted as RTL, and you have some extra assembler that needs to be
4300 emitted. @xref{prologue instruction pattern}.
4303 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4304 If defined, a function that outputs assembler code at the start of an
4305 epilogue. This should be used when the function epilogue is being
4306 emitted as RTL, and you have some extra assembler that needs to be
4307 emitted. @xref{epilogue instruction pattern}.
4310 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4311 If defined, a function that outputs the assembler code for exit from a
4312 function. The epilogue is responsible for restoring the saved
4313 registers and stack pointer to their values when the function was
4314 called, and returning control to the caller. This macro takes the
4315 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4316 registers to restore are determined from @code{regs_ever_live} and
4317 @code{CALL_USED_REGISTERS} in the same way.
4319 On some machines, there is a single instruction that does all the work
4320 of returning from the function. On these machines, give that
4321 instruction the name @samp{return} and do not define the macro
4322 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4324 Do not define a pattern named @samp{return} if you want the
4325 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4326 switches to control whether return instructions or epilogues are used,
4327 define a @samp{return} pattern with a validity condition that tests the
4328 target switches appropriately. If the @samp{return} pattern's validity
4329 condition is false, epilogues will be used.
4331 On machines where functions may or may not have frame-pointers, the
4332 function exit code must vary accordingly. Sometimes the code for these
4333 two cases is completely different. To determine whether a frame pointer
4334 is wanted, the macro can refer to the variable
4335 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4336 a function that needs a frame pointer.
4338 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4339 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4340 The C variable @code{current_function_is_leaf} is nonzero for such a
4341 function. @xref{Leaf Functions}.
4343 On some machines, some functions pop their arguments on exit while
4344 others leave that for the caller to do. For example, the 68020 when
4345 given @option{-mrtd} pops arguments in functions that take a fixed
4346 number of arguments.
4348 @findex current_function_pops_args
4349 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4350 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4351 needs to know what was decided. The variable that is called
4352 @code{current_function_pops_args} is the number of bytes of its
4353 arguments that a function should pop. @xref{Scalar Return}.
4354 @c what is the "its arguments" in the above sentence referring to, pray
4355 @c tell? --mew 5feb93
4360 @findex current_function_pretend_args_size
4361 A region of @code{current_function_pretend_args_size} bytes of
4362 uninitialized space just underneath the first argument arriving on the
4363 stack. (This may not be at the very start of the allocated stack region
4364 if the calling sequence has pushed anything else since pushing the stack
4365 arguments. But usually, on such machines, nothing else has been pushed
4366 yet, because the function prologue itself does all the pushing.) This
4367 region is used on machines where an argument may be passed partly in
4368 registers and partly in memory, and, in some cases to support the
4369 features in @code{<stdarg.h>}.
4372 An area of memory used to save certain registers used by the function.
4373 The size of this area, which may also include space for such things as
4374 the return address and pointers to previous stack frames, is
4375 machine-specific and usually depends on which registers have been used
4376 in the function. Machines with register windows often do not require
4380 A region of at least @var{size} bytes, possibly rounded up to an allocation
4381 boundary, to contain the local variables of the function. On some machines,
4382 this region and the save area may occur in the opposite order, with the
4383 save area closer to the top of the stack.
4386 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4387 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4388 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4389 argument lists of the function. @xref{Stack Arguments}.
4392 @defmac EXIT_IGNORE_STACK
4393 Define this macro as a C expression that is nonzero if the return
4394 instruction or the function epilogue ignores the value of the stack
4395 pointer; in other words, if it is safe to delete an instruction to
4396 adjust the stack pointer before a return from the function. The
4399 Note that this macro's value is relevant only for functions for which
4400 frame pointers are maintained. It is never safe to delete a final
4401 stack adjustment in a function that has no frame pointer, and the
4402 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4405 @defmac EPILOGUE_USES (@var{regno})
4406 Define this macro as a C expression that is nonzero for registers that are
4407 used by the epilogue or the @samp{return} pattern. The stack and frame
4408 pointer registers are already assumed to be used as needed.
4411 @defmac EH_USES (@var{regno})
4412 Define this macro as a C expression that is nonzero for registers that are
4413 used by the exception handling mechanism, and so should be considered live
4414 on entry to an exception edge.
4417 @defmac DELAY_SLOTS_FOR_EPILOGUE
4418 Define this macro if the function epilogue contains delay slots to which
4419 instructions from the rest of the function can be ``moved''. The
4420 definition should be a C expression whose value is an integer
4421 representing the number of delay slots there.
4424 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4425 A C expression that returns 1 if @var{insn} can be placed in delay
4426 slot number @var{n} of the epilogue.
4428 The argument @var{n} is an integer which identifies the delay slot now
4429 being considered (since different slots may have different rules of
4430 eligibility). It is never negative and is always less than the number
4431 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4432 If you reject a particular insn for a given delay slot, in principle, it
4433 may be reconsidered for a subsequent delay slot. Also, other insns may
4434 (at least in principle) be considered for the so far unfilled delay
4437 @findex current_function_epilogue_delay_list
4438 @findex final_scan_insn
4439 The insns accepted to fill the epilogue delay slots are put in an RTL
4440 list made with @code{insn_list} objects, stored in the variable
4441 @code{current_function_epilogue_delay_list}. The insn for the first
4442 delay slot comes first in the list. Your definition of the macro
4443 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4444 outputting the insns in this list, usually by calling
4445 @code{final_scan_insn}.
4447 You need not define this macro if you did not define
4448 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4451 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function})
4452 A function that outputs the assembler code for a thunk
4453 function, used to implement C++ virtual function calls with multiple
4454 inheritance. The thunk acts as a wrapper around a virtual function,
4455 adjusting the implicit object parameter before handing control off to
4458 First, emit code to add the integer @var{delta} to the location that
4459 contains the incoming first argument. Assume that this argument
4460 contains a pointer, and is the one used to pass the @code{this} pointer
4461 in C++. This is the incoming argument @emph{before} the function prologue,
4462 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4463 all other incoming arguments.
4465 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4466 made after adding @code{delta}. In particular, if @var{p} is the
4467 adjusted pointer, the following adjustment should be made:
4470 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4473 After the additions, emit code to jump to @var{function}, which is a
4474 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4475 not touch the return address. Hence returning from @var{FUNCTION} will
4476 return to whoever called the current @samp{thunk}.
4478 The effect must be as if @var{function} had been called directly with
4479 the adjusted first argument. This macro is responsible for emitting all
4480 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4481 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4483 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4484 have already been extracted from it.) It might possibly be useful on
4485 some targets, but probably not.
4487 If you do not define this macro, the target-independent code in the C++
4488 front end will generate a less efficient heavyweight thunk that calls
4489 @var{function} instead of jumping to it. The generic approach does
4490 not support varargs.
4493 @deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function})
4494 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4495 to output the assembler code for the thunk function specified by the
4496 arguments it is passed, and false otherwise. In the latter case, the
4497 generic approach will be used by the C++ front end, with the limitations
4502 @subsection Generating Code for Profiling
4503 @cindex profiling, code generation
4505 These macros will help you generate code for profiling.
4507 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4508 A C statement or compound statement to output to @var{file} some
4509 assembler code to call the profiling subroutine @code{mcount}.
4512 The details of how @code{mcount} expects to be called are determined by
4513 your operating system environment, not by GCC@. To figure them out,
4514 compile a small program for profiling using the system's installed C
4515 compiler and look at the assembler code that results.
4517 Older implementations of @code{mcount} expect the address of a counter
4518 variable to be loaded into some register. The name of this variable is
4519 @samp{LP} followed by the number @var{labelno}, so you would generate
4520 the name using @samp{LP%d} in a @code{fprintf}.
4523 @defmac PROFILE_HOOK
4524 A C statement or compound statement to output to @var{file} some assembly
4525 code to call the profiling subroutine @code{mcount} even the target does
4526 not support profiling.
4529 @defmac NO_PROFILE_COUNTERS
4530 Define this macro if the @code{mcount} subroutine on your system does
4531 not need a counter variable allocated for each function. This is true
4532 for almost all modern implementations. If you define this macro, you
4533 must not use the @var{labelno} argument to @code{FUNCTION_PROFILER}.
4536 @defmac PROFILE_BEFORE_PROLOGUE
4537 Define this macro if the code for function profiling should come before
4538 the function prologue. Normally, the profiling code comes after.
4542 @subsection Permitting tail calls
4545 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4546 True if it is ok to do sibling call optimization for the specified
4547 call expression @var{exp}. @var{decl} will be the called function,
4548 or @code{NULL} if this is an indirect call.
4550 It is not uncommon for limitations of calling conventions to prevent
4551 tail calls to functions outside the current unit of translation, or
4552 during PIC compilation. The hook is used to enforce these restrictions,
4553 as the @code{sibcall} md pattern can not fail, or fall over to a
4554 ``normal'' call. The criteria for successful sibling call optimization
4555 may vary greatly between different architectures.
4558 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap *@var{regs})
4559 Add any hard registers to @var{regs} that are live on entry to the
4560 function. This hook only needs to be defined to provide registers that
4561 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4562 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4563 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4564 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4567 @node Stack Smashing Protection
4568 @subsection Stack smashing protection
4569 @cindex stack smashing protection
4571 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4572 This hook returns a @code{DECL} node for the external variable to use
4573 for the stack protection guard. This variable is initialized by the
4574 runtime to some random value and is used to initialize the guard value
4575 that is placed at the top of the local stack frame. The type of this
4576 variable must be @code{ptr_type_node}.
4578 The default version of this hook creates a variable called
4579 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4582 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4583 This hook returns a tree expression that alerts the runtime that the
4584 stack protect guard variable has been modified. This expression should
4585 involve a call to a @code{noreturn} function.
4587 The default version of this hook invokes a function called
4588 @samp{__stack_chk_fail}, taking no arguments. This function is
4589 normally defined in @file{libgcc2.c}.
4593 @section Implementing the Varargs Macros
4594 @cindex varargs implementation
4596 GCC comes with an implementation of @code{<varargs.h>} and
4597 @code{<stdarg.h>} that work without change on machines that pass arguments
4598 on the stack. Other machines require their own implementations of
4599 varargs, and the two machine independent header files must have
4600 conditionals to include it.
4602 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4603 the calling convention for @code{va_start}. The traditional
4604 implementation takes just one argument, which is the variable in which
4605 to store the argument pointer. The ISO implementation of
4606 @code{va_start} takes an additional second argument. The user is
4607 supposed to write the last named argument of the function here.
4609 However, @code{va_start} should not use this argument. The way to find
4610 the end of the named arguments is with the built-in functions described
4613 @defmac __builtin_saveregs ()
4614 Use this built-in function to save the argument registers in memory so
4615 that the varargs mechanism can access them. Both ISO and traditional
4616 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4617 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4619 On some machines, @code{__builtin_saveregs} is open-coded under the
4620 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4621 other machines, it calls a routine written in assembler language,
4622 found in @file{libgcc2.c}.
4624 Code generated for the call to @code{__builtin_saveregs} appears at the
4625 beginning of the function, as opposed to where the call to
4626 @code{__builtin_saveregs} is written, regardless of what the code is.
4627 This is because the registers must be saved before the function starts
4628 to use them for its own purposes.
4629 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4633 @defmac __builtin_args_info (@var{category})
4634 Use this built-in function to find the first anonymous arguments in
4637 In general, a machine may have several categories of registers used for
4638 arguments, each for a particular category of data types. (For example,
4639 on some machines, floating-point registers are used for floating-point
4640 arguments while other arguments are passed in the general registers.)
4641 To make non-varargs functions use the proper calling convention, you
4642 have defined the @code{CUMULATIVE_ARGS} data type to record how many
4643 registers in each category have been used so far
4645 @code{__builtin_args_info} accesses the same data structure of type
4646 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
4647 with it, with @var{category} specifying which word to access. Thus, the
4648 value indicates the first unused register in a given category.
4650 Normally, you would use @code{__builtin_args_info} in the implementation
4651 of @code{va_start}, accessing each category just once and storing the
4652 value in the @code{va_list} object. This is because @code{va_list} will
4653 have to update the values, and there is no way to alter the
4654 values accessed by @code{__builtin_args_info}.
4657 @defmac __builtin_next_arg (@var{lastarg})
4658 This is the equivalent of @code{__builtin_args_info}, for stack
4659 arguments. It returns the address of the first anonymous stack
4660 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4661 returns the address of the location above the first anonymous stack
4662 argument. Use it in @code{va_start} to initialize the pointer for
4663 fetching arguments from the stack. Also use it in @code{va_start} to
4664 verify that the second parameter @var{lastarg} is the last named argument
4665 of the current function.
4668 @defmac __builtin_classify_type (@var{object})
4669 Since each machine has its own conventions for which data types are
4670 passed in which kind of register, your implementation of @code{va_arg}
4671 has to embody these conventions. The easiest way to categorize the
4672 specified data type is to use @code{__builtin_classify_type} together
4673 with @code{sizeof} and @code{__alignof__}.
4675 @code{__builtin_classify_type} ignores the value of @var{object},
4676 considering only its data type. It returns an integer describing what
4677 kind of type that is---integer, floating, pointer, structure, and so on.
4679 The file @file{typeclass.h} defines an enumeration that you can use to
4680 interpret the values of @code{__builtin_classify_type}.
4683 These machine description macros help implement varargs:
4685 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
4686 If defined, this hook produces the machine-specific code for a call to
4687 @code{__builtin_saveregs}. This code will be moved to the very
4688 beginning of the function, before any parameter access are made. The
4689 return value of this function should be an RTX that contains the value
4690 to use as the return of @code{__builtin_saveregs}.
4693 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (CUMULATIVE_ARGS *@var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time})
4694 This target hook offers an alternative to using
4695 @code{__builtin_saveregs} and defining the hook
4696 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4697 register arguments into the stack so that all the arguments appear to
4698 have been passed consecutively on the stack. Once this is done, you can
4699 use the standard implementation of varargs that works for machines that
4700 pass all their arguments on the stack.
4702 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4703 structure, containing the values that are obtained after processing the
4704 named arguments. The arguments @var{mode} and @var{type} describe the
4705 last named argument---its machine mode and its data type as a tree node.
4707 The target hook should do two things: first, push onto the stack all the
4708 argument registers @emph{not} used for the named arguments, and second,
4709 store the size of the data thus pushed into the @code{int}-valued
4710 variable pointed to by @var{pretend_args_size}. The value that you
4711 store here will serve as additional offset for setting up the stack
4714 Because you must generate code to push the anonymous arguments at
4715 compile time without knowing their data types,
4716 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4717 have just a single category of argument register and use it uniformly
4720 If the argument @var{second_time} is nonzero, it means that the
4721 arguments of the function are being analyzed for the second time. This
4722 happens for an inline function, which is not actually compiled until the
4723 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4724 not generate any instructions in this case.
4727 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca})
4728 Define this hook to return @code{true} if the location where a function
4729 argument is passed depends on whether or not it is a named argument.
4731 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
4732 is set for varargs and stdarg functions. If this hook returns
4733 @code{true}, the @var{named} argument is always true for named
4734 arguments, and false for unnamed arguments. If it returns @code{false},
4735 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
4736 then all arguments are treated as named. Otherwise, all named arguments
4737 except the last are treated as named.
4739 You need not define this hook if it always returns zero.
4742 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED
4743 If you need to conditionally change ABIs so that one works with
4744 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
4745 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
4746 defined, then define this hook to return @code{true} if
4747 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
4748 Otherwise, you should not define this hook.
4752 @section Trampolines for Nested Functions
4753 @cindex trampolines for nested functions
4754 @cindex nested functions, trampolines for
4756 A @dfn{trampoline} is a small piece of code that is created at run time
4757 when the address of a nested function is taken. It normally resides on
4758 the stack, in the stack frame of the containing function. These macros
4759 tell GCC how to generate code to allocate and initialize a
4762 The instructions in the trampoline must do two things: load a constant
4763 address into the static chain register, and jump to the real address of
4764 the nested function. On CISC machines such as the m68k, this requires
4765 two instructions, a move immediate and a jump. Then the two addresses
4766 exist in the trampoline as word-long immediate operands. On RISC
4767 machines, it is often necessary to load each address into a register in
4768 two parts. Then pieces of each address form separate immediate
4771 The code generated to initialize the trampoline must store the variable
4772 parts---the static chain value and the function address---into the
4773 immediate operands of the instructions. On a CISC machine, this is
4774 simply a matter of copying each address to a memory reference at the
4775 proper offset from the start of the trampoline. On a RISC machine, it
4776 may be necessary to take out pieces of the address and store them
4779 @defmac TRAMPOLINE_TEMPLATE (@var{file})
4780 A C statement to output, on the stream @var{file}, assembler code for a
4781 block of data that contains the constant parts of a trampoline. This
4782 code should not include a label---the label is taken care of
4785 If you do not define this macro, it means no template is needed
4786 for the target. Do not define this macro on systems where the block move
4787 code to copy the trampoline into place would be larger than the code
4788 to generate it on the spot.
4791 @defmac TRAMPOLINE_SECTION
4792 Return the section into which the trampoline template is to be placed
4793 (@pxref{Sections}). The default value is @code{readonly_data_section}.
4796 @defmac TRAMPOLINE_SIZE
4797 A C expression for the size in bytes of the trampoline, as an integer.
4800 @defmac TRAMPOLINE_ALIGNMENT
4801 Alignment required for trampolines, in bits.
4803 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
4804 is used for aligning trampolines.
4807 @defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
4808 A C statement to initialize the variable parts of a trampoline.
4809 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
4810 an RTX for the address of the nested function; @var{static_chain} is an
4811 RTX for the static chain value that should be passed to the function
4815 @defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr})
4816 A C statement that should perform any machine-specific adjustment in
4817 the address of the trampoline. Its argument contains the address that
4818 was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be
4819 used for a function call should be different from the address in which
4820 the template was stored, the different address should be assigned to
4821 @var{addr}. If this macro is not defined, @var{addr} will be used for
4824 @cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines
4825 @cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines
4826 If this macro is not defined, by default the trampoline is allocated as
4827 a stack slot. This default is right for most machines. The exceptions
4828 are machines where it is impossible to execute instructions in the stack
4829 area. On such machines, you may have to implement a separate stack,
4830 using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE}
4831 and @code{TARGET_ASM_FUNCTION_EPILOGUE}.
4833 @var{fp} points to a data structure, a @code{struct function}, which
4834 describes the compilation status of the immediate containing function of
4835 the function which the trampoline is for. The stack slot for the
4836 trampoline is in the stack frame of this containing function. Other
4837 allocation strategies probably must do something analogous with this
4841 Implementing trampolines is difficult on many machines because they have
4842 separate instruction and data caches. Writing into a stack location
4843 fails to clear the memory in the instruction cache, so when the program
4844 jumps to that location, it executes the old contents.
4846 Here are two possible solutions. One is to clear the relevant parts of
4847 the instruction cache whenever a trampoline is set up. The other is to
4848 make all trampolines identical, by having them jump to a standard
4849 subroutine. The former technique makes trampoline execution faster; the
4850 latter makes initialization faster.
4852 To clear the instruction cache when a trampoline is initialized, define
4853 the following macro.
4855 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
4856 If defined, expands to a C expression clearing the @emph{instruction
4857 cache} in the specified interval. The definition of this macro would
4858 typically be a series of @code{asm} statements. Both @var{beg} and
4859 @var{end} are both pointer expressions.
4862 The operating system may also require the stack to be made executable
4863 before calling the trampoline. To implement this requirement, define
4864 the following macro.
4866 @defmac ENABLE_EXECUTE_STACK
4867 Define this macro if certain operations must be performed before executing
4868 code located on the stack. The macro should expand to a series of C
4869 file-scope constructs (e.g.@: functions) and provide a unique entry point
4870 named @code{__enable_execute_stack}. The target is responsible for
4871 emitting calls to the entry point in the code, for example from the
4872 @code{INITIALIZE_TRAMPOLINE} macro.
4875 To use a standard subroutine, define the following macro. In addition,
4876 you must make sure that the instructions in a trampoline fill an entire
4877 cache line with identical instructions, or else ensure that the
4878 beginning of the trampoline code is always aligned at the same point in
4879 its cache line. Look in @file{m68k.h} as a guide.
4881 @defmac TRANSFER_FROM_TRAMPOLINE
4882 Define this macro if trampolines need a special subroutine to do their
4883 work. The macro should expand to a series of @code{asm} statements
4884 which will be compiled with GCC@. They go in a library function named
4885 @code{__transfer_from_trampoline}.
4887 If you need to avoid executing the ordinary prologue code of a compiled
4888 C function when you jump to the subroutine, you can do so by placing a
4889 special label of your own in the assembler code. Use one @code{asm}
4890 statement to generate an assembler label, and another to make the label
4891 global. Then trampolines can use that label to jump directly to your
4892 special assembler code.
4896 @section Implicit Calls to Library Routines
4897 @cindex library subroutine names
4898 @cindex @file{libgcc.a}
4900 @c prevent bad page break with this line
4901 Here is an explanation of implicit calls to library routines.
4903 @defmac DECLARE_LIBRARY_RENAMES
4904 This macro, if defined, should expand to a piece of C code that will get
4905 expanded when compiling functions for libgcc.a. It can be used to
4906 provide alternate names for GCC's internal library functions if there
4907 are ABI-mandated names that the compiler should provide.
4910 @findex init_one_libfunc
4911 @findex set_optab_libfunc
4912 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
4913 This hook should declare additional library routines or rename
4914 existing ones, using the functions @code{set_optab_libfunc} and
4915 @code{init_one_libfunc} defined in @file{optabs.c}.
4916 @code{init_optabs} calls this macro after initializing all the normal
4919 The default is to do nothing. Most ports don't need to define this hook.
4922 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
4923 This macro should return @code{true} if the library routine that
4924 implements the floating point comparison operator @var{comparison} in
4925 mode @var{mode} will return a boolean, and @var{false} if it will
4928 GCC's own floating point libraries return tristates from the
4929 comparison operators, so the default returns false always. Most ports
4930 don't need to define this macro.
4933 @defmac TARGET_LIB_INT_CMP_BIASED
4934 This macro should evaluate to @code{true} if the integer comparison
4935 functions (like @code{__cmpdi2}) return 0 to indicate that the first
4936 operand is smaller than the second, 1 to indicate that they are equal,
4937 and 2 to indicate that the first operand is greater than the second.
4938 If this macro evaluates to @code{false} the comparison functions return
4939 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
4940 in @file{libgcc.a}, you do not need to define this macro.
4943 @cindex US Software GOFAST, floating point emulation library
4944 @cindex floating point emulation library, US Software GOFAST
4945 @cindex GOFAST, floating point emulation library
4946 @findex gofast_maybe_init_libfuncs
4947 @defmac US_SOFTWARE_GOFAST
4948 Define this macro if your system C library uses the US Software GOFAST
4949 library to provide floating point emulation.
4951 In addition to defining this macro, your architecture must set
4952 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
4953 else call that function from its version of that hook. It is defined
4954 in @file{config/gofast.h}, which must be included by your
4955 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
4958 If this macro is defined, the
4959 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
4960 false for @code{SFmode} and @code{DFmode} comparisons.
4963 @cindex @code{EDOM}, implicit usage
4966 The value of @code{EDOM} on the target machine, as a C integer constant
4967 expression. If you don't define this macro, GCC does not attempt to
4968 deposit the value of @code{EDOM} into @code{errno} directly. Look in
4969 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
4972 If you do not define @code{TARGET_EDOM}, then compiled code reports
4973 domain errors by calling the library function and letting it report the
4974 error. If mathematical functions on your system use @code{matherr} when
4975 there is an error, then you should leave @code{TARGET_EDOM} undefined so
4976 that @code{matherr} is used normally.
4979 @cindex @code{errno}, implicit usage
4980 @defmac GEN_ERRNO_RTX
4981 Define this macro as a C expression to create an rtl expression that
4982 refers to the global ``variable'' @code{errno}. (On certain systems,
4983 @code{errno} may not actually be a variable.) If you don't define this
4984 macro, a reasonable default is used.
4987 @cindex C99 math functions, implicit usage
4988 @defmac TARGET_C99_FUNCTIONS
4989 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
4990 @code{sinf} and similarly for other functions defined by C99 standard. The
4991 default is nonzero that should be proper value for most modern systems, however
4992 number of existing systems lacks support for these functions in the runtime so
4993 they needs this macro to be redefined to 0.
4996 @defmac NEXT_OBJC_RUNTIME
4997 Define this macro to generate code for Objective-C message sending using
4998 the calling convention of the NeXT system. This calling convention
4999 involves passing the object, the selector and the method arguments all
5000 at once to the method-lookup library function.
5002 The default calling convention passes just the object and the selector
5003 to the lookup function, which returns a pointer to the method.
5006 @node Addressing Modes
5007 @section Addressing Modes
5008 @cindex addressing modes
5010 @c prevent bad page break with this line
5011 This is about addressing modes.
5013 @defmac HAVE_PRE_INCREMENT
5014 @defmacx HAVE_PRE_DECREMENT
5015 @defmacx HAVE_POST_INCREMENT
5016 @defmacx HAVE_POST_DECREMENT
5017 A C expression that is nonzero if the machine supports pre-increment,
5018 pre-decrement, post-increment, or post-decrement addressing respectively.
5021 @defmac HAVE_PRE_MODIFY_DISP
5022 @defmacx HAVE_POST_MODIFY_DISP
5023 A C expression that is nonzero if the machine supports pre- or
5024 post-address side-effect generation involving constants other than
5025 the size of the memory operand.
5028 @defmac HAVE_PRE_MODIFY_REG
5029 @defmacx HAVE_POST_MODIFY_REG
5030 A C expression that is nonzero if the machine supports pre- or
5031 post-address side-effect generation involving a register displacement.
5034 @defmac CONSTANT_ADDRESS_P (@var{x})
5035 A C expression that is 1 if the RTX @var{x} is a constant which
5036 is a valid address. On most machines, this can be defined as
5037 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
5038 in which constant addresses are supported.
5041 @defmac CONSTANT_P (@var{x})
5042 @code{CONSTANT_P}, which is defined by target-independent code,
5043 accepts integer-values expressions whose values are not explicitly
5044 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5045 expressions and @code{const} arithmetic expressions, in addition to
5046 @code{const_int} and @code{const_double} expressions.
5049 @defmac MAX_REGS_PER_ADDRESS
5050 A number, the maximum number of registers that can appear in a valid
5051 memory address. Note that it is up to you to specify a value equal to
5052 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
5056 @defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5057 A C compound statement with a conditional @code{goto @var{label};}
5058 executed if @var{x} (an RTX) is a legitimate memory address on the
5059 target machine for a memory operand of mode @var{mode}.
5061 It usually pays to define several simpler macros to serve as
5062 subroutines for this one. Otherwise it may be too complicated to
5065 This macro must exist in two variants: a strict variant and a
5066 non-strict one. The strict variant is used in the reload pass. It
5067 must be defined so that any pseudo-register that has not been
5068 allocated a hard register is considered a memory reference. In
5069 contexts where some kind of register is required, a pseudo-register
5070 with no hard register must be rejected.
5072 The non-strict variant is used in other passes. It must be defined to
5073 accept all pseudo-registers in every context where some kind of
5074 register is required.
5076 @findex REG_OK_STRICT
5077 Compiler source files that want to use the strict variant of this
5078 macro define the macro @code{REG_OK_STRICT}. You should use an
5079 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
5080 in that case and the non-strict variant otherwise.
5082 Subroutines to check for acceptable registers for various purposes (one
5083 for base registers, one for index registers, and so on) are typically
5084 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
5085 Then only these subroutine macros need have two variants; the higher
5086 levels of macros may be the same whether strict or not.
5088 Normally, constant addresses which are the sum of a @code{symbol_ref}
5089 and an integer are stored inside a @code{const} RTX to mark them as
5090 constant. Therefore, there is no need to recognize such sums
5091 specifically as legitimate addresses. Normally you would simply
5092 recognize any @code{const} as legitimate.
5094 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5095 sums that are not marked with @code{const}. It assumes that a naked
5096 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5097 naked constant sums as illegitimate addresses, so that none of them will
5098 be given to @code{PRINT_OPERAND_ADDRESS}.
5100 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5101 On some machines, whether a symbolic address is legitimate depends on
5102 the section that the address refers to. On these machines, define the
5103 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5104 into the @code{symbol_ref}, and then check for it here. When you see a
5105 @code{const}, you will have to look inside it to find the
5106 @code{symbol_ref} in order to determine the section. @xref{Assembler
5110 @defmac FIND_BASE_TERM (@var{x})
5111 A C expression to determine the base term of address @var{x}.
5112 This macro is used in only one place: `find_base_term' in alias.c.
5114 It is always safe for this macro to not be defined. It exists so
5115 that alias analysis can understand machine-dependent addresses.
5117 The typical use of this macro is to handle addresses containing
5118 a label_ref or symbol_ref within an UNSPEC@.
5121 @defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
5122 A C compound statement that attempts to replace @var{x} with a valid
5123 memory address for an operand of mode @var{mode}. @var{win} will be a
5124 C statement label elsewhere in the code; the macro definition may use
5127 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
5131 to avoid further processing if the address has become legitimate.
5133 @findex break_out_memory_refs
5134 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5135 and @var{oldx} will be the operand that was given to that function to produce
5138 The code generated by this macro should not alter the substructure of
5139 @var{x}. If it transforms @var{x} into a more legitimate form, it
5140 should assign @var{x} (which will always be a C variable) a new value.
5142 It is not necessary for this macro to come up with a legitimate
5143 address. The compiler has standard ways of doing so in all cases. In
5144 fact, it is safe to omit this macro. But often a
5145 machine-dependent strategy can generate better code.
5148 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5149 A C compound statement that attempts to replace @var{x}, which is an address
5150 that needs reloading, with a valid memory address for an operand of mode
5151 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5152 It is not necessary to define this macro, but it might be useful for
5153 performance reasons.
5155 For example, on the i386, it is sometimes possible to use a single
5156 reload register instead of two by reloading a sum of two pseudo
5157 registers into a register. On the other hand, for number of RISC
5158 processors offsets are limited so that often an intermediate address
5159 needs to be generated in order to address a stack slot. By defining
5160 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5161 generated for adjacent some stack slots can be made identical, and thus
5164 @emph{Note}: This macro should be used with caution. It is necessary
5165 to know something of how reload works in order to effectively use this,
5166 and it is quite easy to produce macros that build in too much knowledge
5167 of reload internals.
5169 @emph{Note}: This macro must be able to reload an address created by a
5170 previous invocation of this macro. If it fails to handle such addresses
5171 then the compiler may generate incorrect code or abort.
5174 The macro definition should use @code{push_reload} to indicate parts that
5175 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5176 suitable to be passed unaltered to @code{push_reload}.
5178 The code generated by this macro must not alter the substructure of
5179 @var{x}. If it transforms @var{x} into a more legitimate form, it
5180 should assign @var{x} (which will always be a C variable) a new value.
5181 This also applies to parts that you change indirectly by calling
5184 @findex strict_memory_address_p
5185 The macro definition may use @code{strict_memory_address_p} to test if
5186 the address has become legitimate.
5189 If you want to change only a part of @var{x}, one standard way of doing
5190 this is to use @code{copy_rtx}. Note, however, that is unshares only a
5191 single level of rtl. Thus, if the part to be changed is not at the
5192 top level, you'll need to replace first the top level.
5193 It is not necessary for this macro to come up with a legitimate
5194 address; but often a machine-dependent strategy can generate better code.
5197 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5198 A C statement or compound statement with a conditional @code{goto
5199 @var{label};} executed if memory address @var{x} (an RTX) can have
5200 different meanings depending on the machine mode of the memory
5201 reference it is used for or if the address is valid for some modes
5204 Autoincrement and autodecrement addresses typically have mode-dependent
5205 effects because the amount of the increment or decrement is the size
5206 of the operand being addressed. Some machines have other mode-dependent
5207 addresses. Many RISC machines have no mode-dependent addresses.
5209 You may assume that @var{addr} is a valid address for the machine.
5212 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5213 A C expression that is nonzero if @var{x} is a legitimate constant for
5214 an immediate operand on the target machine. You can assume that
5215 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5216 @samp{1} is a suitable definition for this macro on machines where
5217 anything @code{CONSTANT_P} is valid.
5220 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5221 This hook is used to undo the possibly obfuscating effects of the
5222 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5223 macros. Some backend implementations of these macros wrap symbol
5224 references inside an @code{UNSPEC} rtx to represent PIC or similar
5225 addressing modes. This target hook allows GCC's optimizers to understand
5226 the semantics of these opaque @code{UNSPEC}s by converting them back
5227 into their original form.
5230 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x})
5231 This hook should return true if @var{x} is of a form that cannot (or
5232 should not) be spilled to the constant pool. The default version of
5233 this hook returns false.
5235 The primary reason to define this hook is to prevent reload from
5236 deciding that a non-legitimate constant would be better reloaded
5237 from the constant pool instead of spilling and reloading a register
5238 holding the constant. This restriction is often true of addresses
5239 of TLS symbols for various targets.
5242 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x})
5243 This hook should return true if pool entries for constant @var{x} can
5244 be placed in an @code{object_block} structure. @var{mode} is the mode
5247 The default version returns false for all constants.
5250 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5251 This hook should return the DECL of a function @var{f} that given an
5252 address @var{addr} as an argument returns a mask @var{m} that can be
5253 used to extract from two vectors the relevant data that resides in
5254 @var{addr} in case @var{addr} is not properly aligned.
5256 The autovectrizer, when vectorizing a load operation from an address
5257 @var{addr} that may be unaligned, will generate two vector loads from
5258 the two aligned addresses around @var{addr}. It then generates a
5259 @code{REALIGN_LOAD} operation to extract the relevant data from the
5260 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5261 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5262 the third argument, @var{OFF}, defines how the data will be extracted
5263 from these two vectors: if @var{OFF} is 0, then the returned vector is
5264 @var{v2}; otherwise, the returned vector is composed from the last
5265 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5266 @var{OFF} elements of @var{v2}.
5268 If this hook is defined, the autovectorizer will generate a call
5269 to @var{f} (using the DECL tree that this hook returns) and will
5270 use the return value of @var{f} as the argument @var{OFF} to
5271 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5272 should comply with the semantics expected by @code{REALIGN_LOAD}
5274 If this hook is not defined, then @var{addr} will be used as
5275 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5276 log2(@var{VS})-1 bits of @var{addr} will be considered.
5279 @node Anchored Addresses
5280 @section Anchored Addresses
5281 @cindex anchored addresses
5282 @cindex @option{-fsection-anchors}
5284 GCC usually addresses every static object as a separate entity.
5285 For example, if we have:
5289 int foo (void) @{ return a + b + c; @}
5292 the code for @code{foo} will usually calculate three separate symbolic
5293 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5294 it would be better to calculate just one symbolic address and access
5295 the three variables relative to it. The equivalent pseudocode would
5301 register int *xr = &x;
5302 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5306 (which isn't valid C). We refer to shared addresses like @code{x} as
5307 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5309 The hooks below describe the target properties that GCC needs to know
5310 in order to make effective use of section anchors. It won't use
5311 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5312 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5314 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5315 The minimum offset that should be applied to a section anchor.
5316 On most targets, it should be the smallest offset that can be
5317 applied to a base register while still giving a legitimate address
5318 for every mode. The default value is 0.
5321 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5322 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5323 offset that should be applied to section anchors. The default
5327 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5328 Write the assembly code to define section anchor @var{x}, which is a
5329 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5330 The hook is called with the assembly output position set to the beginning
5331 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5333 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5334 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5335 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5336 is @code{NULL}, which disables the use of section anchors altogether.
5339 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (rtx @var{x})
5340 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5341 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5342 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5344 The default version is correct for most targets, but you might need to
5345 intercept this hook to handle things like target-specific attributes
5346 or target-specific sections.
5349 @node Condition Code
5350 @section Condition Code Status
5351 @cindex condition code status
5353 @c prevent bad page break with this line
5354 This describes the condition code status.
5357 The file @file{conditions.h} defines a variable @code{cc_status} to
5358 describe how the condition code was computed (in case the interpretation of
5359 the condition code depends on the instruction that it was set by). This
5360 variable contains the RTL expressions on which the condition code is
5361 currently based, and several standard flags.
5363 Sometimes additional machine-specific flags must be defined in the machine
5364 description header file. It can also add additional machine-specific
5365 information by defining @code{CC_STATUS_MDEP}.
5367 @defmac CC_STATUS_MDEP
5368 C code for a data type which is used for declaring the @code{mdep}
5369 component of @code{cc_status}. It defaults to @code{int}.
5371 This macro is not used on machines that do not use @code{cc0}.
5374 @defmac CC_STATUS_MDEP_INIT
5375 A C expression to initialize the @code{mdep} field to ``empty''.
5376 The default definition does nothing, since most machines don't use
5377 the field anyway. If you want to use the field, you should probably
5378 define this macro to initialize it.
5380 This macro is not used on machines that do not use @code{cc0}.
5383 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5384 A C compound statement to set the components of @code{cc_status}
5385 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5386 this macro's responsibility to recognize insns that set the condition
5387 code as a byproduct of other activity as well as those that explicitly
5390 This macro is not used on machines that do not use @code{cc0}.
5392 If there are insns that do not set the condition code but do alter
5393 other machine registers, this macro must check to see whether they
5394 invalidate the expressions that the condition code is recorded as
5395 reflecting. For example, on the 68000, insns that store in address
5396 registers do not set the condition code, which means that usually
5397 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5398 insns. But suppose that the previous insn set the condition code
5399 based on location @samp{a4@@(102)} and the current insn stores a new
5400 value in @samp{a4}. Although the condition code is not changed by
5401 this, it will no longer be true that it reflects the contents of
5402 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5403 @code{cc_status} in this case to say that nothing is known about the
5404 condition code value.
5406 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5407 with the results of peephole optimization: insns whose patterns are
5408 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5409 constants which are just the operands. The RTL structure of these
5410 insns is not sufficient to indicate what the insns actually do. What
5411 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5412 @code{CC_STATUS_INIT}.
5414 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5415 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5416 @samp{cc}. This avoids having detailed information about patterns in
5417 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5420 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5421 Returns a mode from class @code{MODE_CC} to be used when comparison
5422 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
5423 example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see
5424 @pxref{Jump Patterns} for a description of the reason for this
5428 #define SELECT_CC_MODE(OP,X,Y) \
5429 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5430 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5431 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5432 || GET_CODE (X) == NEG) \
5433 ? CC_NOOVmode : CCmode))
5436 You should define this macro if and only if you define extra CC modes
5437 in @file{@var{machine}-modes.def}.
5440 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5441 On some machines not all possible comparisons are defined, but you can
5442 convert an invalid comparison into a valid one. For example, the Alpha
5443 does not have a @code{GT} comparison, but you can use an @code{LT}
5444 comparison instead and swap the order of the operands.
5446 On such machines, define this macro to be a C statement to do any
5447 required conversions. @var{code} is the initial comparison code
5448 and @var{op0} and @var{op1} are the left and right operands of the
5449 comparison, respectively. You should modify @var{code}, @var{op0}, and
5450 @var{op1} as required.
5452 GCC will not assume that the comparison resulting from this macro is
5453 valid but will see if the resulting insn matches a pattern in the
5456 You need not define this macro if it would never change the comparison
5460 @defmac REVERSIBLE_CC_MODE (@var{mode})
5461 A C expression whose value is one if it is always safe to reverse a
5462 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5463 can ever return @var{mode} for a floating-point inequality comparison,
5464 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5466 You need not define this macro if it would always returns zero or if the
5467 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5468 For example, here is the definition used on the SPARC, where floating-point
5469 inequality comparisons are always given @code{CCFPEmode}:
5472 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5476 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5477 A C expression whose value is reversed condition code of the @var{code} for
5478 comparison done in CC_MODE @var{mode}. The macro is used only in case
5479 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5480 machine has some non-standard way how to reverse certain conditionals. For
5481 instance in case all floating point conditions are non-trapping, compiler may
5482 freely convert unordered compares to ordered one. Then definition may look
5486 #define REVERSE_CONDITION(CODE, MODE) \
5487 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5488 : reverse_condition_maybe_unordered (CODE))
5492 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
5493 A C expression that returns true if the conditional execution predicate
5494 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
5495 versa. Define this to return 0 if the target has conditional execution
5496 predicates that cannot be reversed safely. There is no need to validate
5497 that the arguments of op1 and op2 are the same, this is done separately.
5498 If no expansion is specified, this macro is defined as follows:
5501 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
5502 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
5506 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *)
5507 On targets which do not use @code{(cc0)}, and which use a hard
5508 register rather than a pseudo-register to hold condition codes, the
5509 regular CSE passes are often not able to identify cases in which the
5510 hard register is set to a common value. Use this hook to enable a
5511 small pass which optimizes such cases. This hook should return true
5512 to enable this pass, and it should set the integers to which its
5513 arguments point to the hard register numbers used for condition codes.
5514 When there is only one such register, as is true on most systems, the
5515 integer pointed to by the second argument should be set to
5516 @code{INVALID_REGNUM}.
5518 The default version of this hook returns false.
5521 @deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode)
5522 On targets which use multiple condition code modes in class
5523 @code{MODE_CC}, it is sometimes the case that a comparison can be
5524 validly done in more than one mode. On such a system, define this
5525 target hook to take two mode arguments and to return a mode in which
5526 both comparisons may be validly done. If there is no such mode,
5527 return @code{VOIDmode}.
5529 The default version of this hook checks whether the modes are the
5530 same. If they are, it returns that mode. If they are different, it
5531 returns @code{VOIDmode}.
5535 @section Describing Relative Costs of Operations
5536 @cindex costs of instructions
5537 @cindex relative costs
5538 @cindex speed of instructions
5540 These macros let you describe the relative speed of various operations
5541 on the target machine.
5543 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
5544 A C expression for the cost of moving data of mode @var{mode} from a
5545 register in class @var{from} to one in class @var{to}. The classes are
5546 expressed using the enumeration values such as @code{GENERAL_REGS}. A
5547 value of 2 is the default; other values are interpreted relative to
5550 It is not required that the cost always equal 2 when @var{from} is the
5551 same as @var{to}; on some machines it is expensive to move between
5552 registers if they are not general registers.
5554 If reload sees an insn consisting of a single @code{set} between two
5555 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
5556 classes returns a value of 2, reload does not check to ensure that the
5557 constraints of the insn are met. Setting a cost of other than 2 will
5558 allow reload to verify that the constraints are met. You should do this
5559 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
5562 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
5563 A C expression for the cost of moving data of mode @var{mode} between a
5564 register of class @var{class} and memory; @var{in} is zero if the value
5565 is to be written to memory, nonzero if it is to be read in. This cost
5566 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
5567 registers and memory is more expensive than between two registers, you
5568 should define this macro to express the relative cost.
5570 If you do not define this macro, GCC uses a default cost of 4 plus
5571 the cost of copying via a secondary reload register, if one is
5572 needed. If your machine requires a secondary reload register to copy
5573 between memory and a register of @var{class} but the reload mechanism is
5574 more complex than copying via an intermediate, define this macro to
5575 reflect the actual cost of the move.
5577 GCC defines the function @code{memory_move_secondary_cost} if
5578 secondary reloads are needed. It computes the costs due to copying via
5579 a secondary register. If your machine copies from memory using a
5580 secondary register in the conventional way but the default base value of
5581 4 is not correct for your machine, define this macro to add some other
5582 value to the result of that function. The arguments to that function
5583 are the same as to this macro.
5587 A C expression for the cost of a branch instruction. A value of 1 is
5588 the default; other values are interpreted relative to that.
5591 Here are additional macros which do not specify precise relative costs,
5592 but only that certain actions are more expensive than GCC would
5595 @defmac SLOW_BYTE_ACCESS
5596 Define this macro as a C expression which is nonzero if accessing less
5597 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
5598 faster than accessing a word of memory, i.e., if such access
5599 require more than one instruction or if there is no difference in cost
5600 between byte and (aligned) word loads.
5602 When this macro is not defined, the compiler will access a field by
5603 finding the smallest containing object; when it is defined, a fullword
5604 load will be used if alignment permits. Unless bytes accesses are
5605 faster than word accesses, using word accesses is preferable since it
5606 may eliminate subsequent memory access if subsequent accesses occur to
5607 other fields in the same word of the structure, but to different bytes.
5610 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
5611 Define this macro to be the value 1 if memory accesses described by the
5612 @var{mode} and @var{alignment} parameters have a cost many times greater
5613 than aligned accesses, for example if they are emulated in a trap
5616 When this macro is nonzero, the compiler will act as if
5617 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
5618 moves. This can cause significantly more instructions to be produced.
5619 Therefore, do not set this macro nonzero if unaligned accesses only add a
5620 cycle or two to the time for a memory access.
5622 If the value of this macro is always zero, it need not be defined. If
5623 this macro is defined, it should produce a nonzero value when
5624 @code{STRICT_ALIGNMENT} is nonzero.
5628 The threshold of number of scalar memory-to-memory move insns, @emph{below}
5629 which a sequence of insns should be generated instead of a
5630 string move insn or a library call. Increasing the value will always
5631 make code faster, but eventually incurs high cost in increased code size.
5633 Note that on machines where the corresponding move insn is a
5634 @code{define_expand} that emits a sequence of insns, this macro counts
5635 the number of such sequences.
5637 If you don't define this, a reasonable default is used.
5640 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
5641 A C expression used to determine whether @code{move_by_pieces} will be used to
5642 copy a chunk of memory, or whether some other block move mechanism
5643 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5644 than @code{MOVE_RATIO}.
5647 @defmac MOVE_MAX_PIECES
5648 A C expression used by @code{move_by_pieces} to determine the largest unit
5649 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
5653 The threshold of number of scalar move insns, @emph{below} which a sequence
5654 of insns should be generated to clear memory instead of a string clear insn
5655 or a library call. Increasing the value will always make code faster, but
5656 eventually incurs high cost in increased code size.
5658 If you don't define this, a reasonable default is used.
5661 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
5662 A C expression used to determine whether @code{clear_by_pieces} will be used
5663 to clear a chunk of memory, or whether some other block clear mechanism
5664 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5665 than @code{CLEAR_RATIO}.
5668 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
5669 A C expression used to determine whether @code{store_by_pieces} will be
5670 used to set a chunk of memory to a constant value, or whether some other
5671 mechanism will be used. Used by @code{__builtin_memset} when storing
5672 values other than constant zero and by @code{__builtin_strcpy} when
5673 when called with a constant source string.
5674 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5675 than @code{MOVE_RATIO}.
5678 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
5679 A C expression used to determine whether a load postincrement is a good
5680 thing to use for a given mode. Defaults to the value of
5681 @code{HAVE_POST_INCREMENT}.
5684 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
5685 A C expression used to determine whether a load postdecrement is a good
5686 thing to use for a given mode. Defaults to the value of
5687 @code{HAVE_POST_DECREMENT}.
5690 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
5691 A C expression used to determine whether a load preincrement is a good
5692 thing to use for a given mode. Defaults to the value of
5693 @code{HAVE_PRE_INCREMENT}.
5696 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
5697 A C expression used to determine whether a load predecrement is a good
5698 thing to use for a given mode. Defaults to the value of
5699 @code{HAVE_PRE_DECREMENT}.
5702 @defmac USE_STORE_POST_INCREMENT (@var{mode})
5703 A C expression used to determine whether a store postincrement is a good
5704 thing to use for a given mode. Defaults to the value of
5705 @code{HAVE_POST_INCREMENT}.
5708 @defmac USE_STORE_POST_DECREMENT (@var{mode})
5709 A C expression used to determine whether a store postdecrement is a good
5710 thing to use for a given mode. Defaults to the value of
5711 @code{HAVE_POST_DECREMENT}.
5714 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
5715 This macro is used to determine whether a store preincrement is a good
5716 thing to use for a given mode. Defaults to the value of
5717 @code{HAVE_PRE_INCREMENT}.
5720 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
5721 This macro is used to determine whether a store predecrement is a good
5722 thing to use for a given mode. Defaults to the value of
5723 @code{HAVE_PRE_DECREMENT}.
5726 @defmac NO_FUNCTION_CSE
5727 Define this macro if it is as good or better to call a constant
5728 function address than to call an address kept in a register.
5731 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
5732 Define this macro if a non-short-circuit operation produced by
5733 @samp{fold_range_test ()} is optimal. This macro defaults to true if
5734 @code{BRANCH_COST} is greater than or equal to the value 2.
5737 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total})
5738 This target hook describes the relative costs of RTL expressions.
5740 The cost may depend on the precise form of the expression, which is
5741 available for examination in @var{x}, and the rtx code of the expression
5742 in which it is contained, found in @var{outer_code}. @var{code} is the
5743 expression code---redundant, since it can be obtained with
5744 @code{GET_CODE (@var{x})}.
5746 In implementing this hook, you can use the construct
5747 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
5750 On entry to the hook, @code{*@var{total}} contains a default estimate
5751 for the cost of the expression. The hook should modify this value as
5752 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
5753 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
5754 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
5756 When optimizing for code size, i.e.@: when @code{optimize_size} is
5757 nonzero, this target hook should be used to estimate the relative
5758 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
5760 The hook returns true when all subexpressions of @var{x} have been
5761 processed, and false when @code{rtx_cost} should recurse.
5764 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address})
5765 This hook computes the cost of an addressing mode that contains
5766 @var{address}. If not defined, the cost is computed from
5767 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
5769 For most CISC machines, the default cost is a good approximation of the
5770 true cost of the addressing mode. However, on RISC machines, all
5771 instructions normally have the same length and execution time. Hence
5772 all addresses will have equal costs.
5774 In cases where more than one form of an address is known, the form with
5775 the lowest cost will be used. If multiple forms have the same, lowest,
5776 cost, the one that is the most complex will be used.
5778 For example, suppose an address that is equal to the sum of a register
5779 and a constant is used twice in the same basic block. When this macro
5780 is not defined, the address will be computed in a register and memory
5781 references will be indirect through that register. On machines where
5782 the cost of the addressing mode containing the sum is no higher than
5783 that of a simple indirect reference, this will produce an additional
5784 instruction and possibly require an additional register. Proper
5785 specification of this macro eliminates this overhead for such machines.
5787 This hook is never called with an invalid address.
5789 On machines where an address involving more than one register is as
5790 cheap as an address computation involving only one register, defining
5791 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
5792 be live over a region of code where only one would have been if
5793 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
5794 should be considered in the definition of this macro. Equivalent costs
5795 should probably only be given to addresses with different numbers of
5796 registers on machines with lots of registers.
5800 @section Adjusting the Instruction Scheduler
5802 The instruction scheduler may need a fair amount of machine-specific
5803 adjustment in order to produce good code. GCC provides several target
5804 hooks for this purpose. It is usually enough to define just a few of
5805 them: try the first ones in this list first.
5807 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
5808 This hook returns the maximum number of instructions that can ever
5809 issue at the same time on the target machine. The default is one.
5810 Although the insn scheduler can define itself the possibility of issue
5811 an insn on the same cycle, the value can serve as an additional
5812 constraint to issue insns on the same simulated processor cycle (see
5813 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
5814 This value must be constant over the entire compilation. If you need
5815 it to vary depending on what the instructions are, you must use
5816 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
5819 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
5820 This hook is executed by the scheduler after it has scheduled an insn
5821 from the ready list. It should return the number of insns which can
5822 still be issued in the current cycle. The default is
5823 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
5824 @code{USE}, which normally are not counted against the issue rate.
5825 You should define this hook if some insns take more machine resources
5826 than others, so that fewer insns can follow them in the same cycle.
5827 @var{file} is either a null pointer, or a stdio stream to write any
5828 debug output to. @var{verbose} is the verbose level provided by
5829 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
5833 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
5834 This function corrects the value of @var{cost} based on the
5835 relationship between @var{insn} and @var{dep_insn} through the
5836 dependence @var{link}. It should return the new value. The default
5837 is to make no adjustment to @var{cost}. This can be used for example
5838 to specify to the scheduler using the traditional pipeline description
5839 that an output- or anti-dependence does not incur the same cost as a
5840 data-dependence. If the scheduler using the automaton based pipeline
5841 description, the cost of anti-dependence is zero and the cost of
5842 output-dependence is maximum of one and the difference of latency
5843 times of the first and the second insns. If these values are not
5844 acceptable, you could use the hook to modify them too. See also
5845 @pxref{Processor pipeline description}.
5848 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
5849 This hook adjusts the integer scheduling priority @var{priority} of
5850 @var{insn}. It should return the new priority. Increase the priority to
5851 execute @var{insn} earlier, reduce the priority to execute @var{insn}
5852 later. Do not define this hook if you do not need to adjust the
5853 scheduling priorities of insns.
5856 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
5857 This hook is executed by the scheduler after it has scheduled the ready
5858 list, to allow the machine description to reorder it (for example to
5859 combine two small instructions together on @samp{VLIW} machines).
5860 @var{file} is either a null pointer, or a stdio stream to write any
5861 debug output to. @var{verbose} is the verbose level provided by
5862 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
5863 list of instructions that are ready to be scheduled. @var{n_readyp} is
5864 a pointer to the number of elements in the ready list. The scheduler
5865 reads the ready list in reverse order, starting with
5866 @var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock}
5867 is the timer tick of the scheduler. You may modify the ready list and
5868 the number of ready insns. The return value is the number of insns that
5869 can issue this cycle; normally this is just @code{issue_rate}. See also
5870 @samp{TARGET_SCHED_REORDER2}.
5873 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock})
5874 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
5875 function is called whenever the scheduler starts a new cycle. This one
5876 is called once per iteration over a cycle, immediately after
5877 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
5878 return the number of insns to be scheduled in the same cycle. Defining
5879 this hook can be useful if there are frequent situations where
5880 scheduling one insn causes other insns to become ready in the same
5881 cycle. These other insns can then be taken into account properly.
5884 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
5885 This hook is called after evaluation forward dependencies of insns in
5886 chain given by two parameter values (@var{head} and @var{tail}
5887 correspondingly) but before insns scheduling of the insn chain. For
5888 example, it can be used for better insn classification if it requires
5889 analysis of dependencies. This hook can use backward and forward
5890 dependencies of the insn scheduler because they are already
5894 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
5895 This hook is executed by the scheduler at the beginning of each block of
5896 instructions that are to be scheduled. @var{file} is either a null
5897 pointer, or a stdio stream to write any debug output to. @var{verbose}
5898 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5899 @var{max_ready} is the maximum number of insns in the current scheduling
5900 region that can be live at the same time. This can be used to allocate
5901 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
5904 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
5905 This hook is executed by the scheduler at the end of each block of
5906 instructions that are to be scheduled. It can be used to perform
5907 cleanup of any actions done by the other scheduling hooks. @var{file}
5908 is either a null pointer, or a stdio stream to write any debug output
5909 to. @var{verbose} is the verbose level provided by
5910 @option{-fsched-verbose-@var{n}}.
5913 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
5914 This hook is executed by the scheduler after function level initializations.
5915 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5916 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5917 @var{old_max_uid} is the maximum insn uid when scheduling begins.
5920 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
5921 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
5922 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5923 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5926 @deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
5927 The hook returns an RTL insn. The automaton state used in the
5928 pipeline hazard recognizer is changed as if the insn were scheduled
5929 when the new simulated processor cycle starts. Usage of the hook may
5930 simplify the automaton pipeline description for some @acronym{VLIW}
5931 processors. If the hook is defined, it is used only for the automaton
5932 based pipeline description. The default is not to change the state
5933 when the new simulated processor cycle starts.
5936 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
5937 The hook can be used to initialize data used by the previous hook.
5940 @deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
5941 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
5942 to changed the state as if the insn were scheduled when the new
5943 simulated processor cycle finishes.
5946 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
5947 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
5948 used to initialize data used by the previous hook.
5951 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
5952 This hook controls better choosing an insn from the ready insn queue
5953 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
5954 chooses the first insn from the queue. If the hook returns a positive
5955 value, an additional scheduler code tries all permutations of
5956 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
5957 subsequent ready insns to choose an insn whose issue will result in
5958 maximal number of issued insns on the same cycle. For the
5959 @acronym{VLIW} processor, the code could actually solve the problem of
5960 packing simple insns into the @acronym{VLIW} insn. Of course, if the
5961 rules of @acronym{VLIW} packing are described in the automaton.
5963 This code also could be used for superscalar @acronym{RISC}
5964 processors. Let us consider a superscalar @acronym{RISC} processor
5965 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
5966 @var{B}, some insns can be executed only in pipelines @var{B} or
5967 @var{C}, and one insn can be executed in pipeline @var{B}. The
5968 processor may issue the 1st insn into @var{A} and the 2nd one into
5969 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
5970 until the next cycle. If the scheduler issues the 3rd insn the first,
5971 the processor could issue all 3 insns per cycle.
5973 Actually this code demonstrates advantages of the automaton based
5974 pipeline hazard recognizer. We try quickly and easy many insn
5975 schedules to choose the best one.
5977 The default is no multipass scheduling.
5980 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
5982 This hook controls what insns from the ready insn queue will be
5983 considered for the multipass insn scheduling. If the hook returns
5984 zero for insn passed as the parameter, the insn will be not chosen to
5987 The default is that any ready insns can be chosen to be issued.
5990 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *)
5992 This hook is called by the insn scheduler before issuing insn passed
5993 as the third parameter on given cycle. If the hook returns nonzero,
5994 the insn is not issued on given processors cycle. Instead of that,
5995 the processor cycle is advanced. If the value passed through the last
5996 parameter is zero, the insn ready queue is not sorted on the new cycle
5997 start as usually. The first parameter passes file for debugging
5998 output. The second one passes the scheduler verbose level of the
5999 debugging output. The forth and the fifth parameter values are
6000 correspondingly processor cycle on which the previous insn has been
6001 issued and the current processor cycle.
6004 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (rtx @var{insn1}, rtx @var{insn2}, rtx @var{dep_link}, int @var{dep_cost}, int @var{distance})
6005 This hook is used to define which dependences are considered costly by
6006 the target, so costly that it is not advisable to schedule the insns that
6007 are involved in the dependence too close to one another. The parameters
6008 to this hook are as follows: The second parameter @var{insn2} is dependent
6009 upon the first parameter @var{insn1}. The dependence between @var{insn1}
6010 and @var{insn2} is represented by the third parameter @var{dep_link}. The
6011 fourth parameter @var{cost} is the cost of the dependence, and the fifth
6012 parameter @var{distance} is the distance in cycles between the two insns.
6013 The hook returns @code{true} if considering the distance between the two
6014 insns the dependence between them is considered costly by the target,
6015 and @code{false} otherwise.
6017 Defining this hook can be useful in multiple-issue out-of-order machines,
6018 where (a) it's practically hopeless to predict the actual data/resource
6019 delays, however: (b) there's a better chance to predict the actual grouping
6020 that will be formed, and (c) correctly emulating the grouping can be very
6021 important. In such targets one may want to allow issuing dependent insns
6022 closer to one another---i.e., closer than the dependence distance; however,
6023 not in cases of "costly dependences", which this hooks allows to define.
6026 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST_2 (rtx @var{insn}, int @var{dep_type}, rtx @var{dep_insn}, int @var{cost})
6027 This hook is a modified version of @samp{TARGET_SCHED_ADJUST_COST}. Instead
6028 of passing dependence as a second parameter, it passes a type of that
6029 dependence. This is useful to calculate cost of dependence between insns
6030 not having the corresponding link. If @samp{TARGET_SCHED_ADJUST_COST_2} is
6031 defined it is used instead of @samp{TARGET_SCHED_ADJUST_COST}.
6034 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6035 This hook is called by the insn scheduler after emitting a new instruction to
6036 the instruction stream. The hook notifies a target backend to extend its
6037 per instruction data structures.
6040 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat})
6041 This hook is called by the insn scheduler when @var{insn} has only
6042 speculative dependencies and therefore can be scheduled speculatively.
6043 The hook is used to check if the pattern of @var{insn} has a speculative
6044 version and, in case of successful check, to generate that speculative
6045 pattern. The hook should return 1, if the instruction has a speculative form,
6046 or -1, if it doesn't. @var{request} describes the type of requested
6047 speculation. If the return value equals 1 then @var{new_pat} is assigned
6048 the generated speculative pattern.
6051 @deftypefn {Target Hook} int TARGET_SCHED_NEEDS_BLOCK_P (rtx @var{insn})
6052 This hook is called by the insn scheduler during generation of recovery code
6053 for @var{insn}. It should return nonzero, if the corresponding check
6054 instruction should branch to recovery code, or zero otherwise.
6057 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p})
6058 This hook is called by the insn scheduler to generate a pattern for recovery
6059 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6060 speculative instruction for which the check should be generated.
6061 @var{label} is either a label of a basic block, where recovery code should
6062 be emitted, or a null pointer, when requested check doesn't branch to
6063 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6064 a pattern for a branchy check corresponding to a simple check denoted by
6065 @var{insn} should be generated. In this case @var{label} can't be null.
6068 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (rtx @var{insn})
6069 This hook is used as a workaround for
6070 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6071 called on the first instruction of the ready list. The hook is used to
6072 discard speculative instruction that stand first in the ready list from
6073 being scheduled on the current cycle. For non-speculative instructions,
6074 the hook should always return nonzero. For example, in the ia64 backend
6075 the hook is used to cancel data speculative insns when the ALAT table
6079 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (unsigned int *@var{flags}, spec_info_t @var{spec_info})
6080 This hook is used by the insn scheduler to find out what features should be
6081 enabled/used. @var{flags} initially may have either the SCHED_RGN or SCHED_EBB
6082 bit set. This denotes the scheduler pass for which the data should be
6083 provided. The target backend should modify @var{flags} by modifying
6084 the bits corresponding to the following features: USE_DEPS_LIST, USE_GLAT,
6085 DETACH_LIFE_INFO, and DO_SPECULATION. For the DO_SPECULATION feature
6086 an additional structure @var{spec_info} should be filled by the target.
6087 The structure describes speculation types that can be used in the scheduler.
6091 @section Dividing the Output into Sections (Texts, Data, @dots{})
6092 @c the above section title is WAY too long. maybe cut the part between
6093 @c the (...)? --mew 10feb93
6095 An object file is divided into sections containing different types of
6096 data. In the most common case, there are three sections: the @dfn{text
6097 section}, which holds instructions and read-only data; the @dfn{data
6098 section}, which holds initialized writable data; and the @dfn{bss
6099 section}, which holds uninitialized data. Some systems have other kinds
6102 @file{varasm.c} provides several well-known sections, such as
6103 @code{text_section}, @code{data_section} and @code{bss_section}.
6104 The normal way of controlling a @code{@var{foo}_section} variable
6105 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6106 as described below. The macros are only read once, when @file{varasm.c}
6107 initializes itself, so their values must be run-time constants.
6108 They may however depend on command-line flags.
6110 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6111 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6112 to be string literals.
6114 Some assemblers require a different string to be written every time a
6115 section is selected. If your assembler falls into this category, you
6116 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6117 @code{get_unnamed_section} to set up the sections.
6119 You must always create a @code{text_section}, either by defining
6120 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6121 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6122 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6123 create a distinct @code{readonly_data_section}, the default is to
6124 reuse @code{text_section}.
6126 All the other @file{varasm.c} sections are optional, and are null
6127 if the target does not provide them.
6129 @defmac TEXT_SECTION_ASM_OP
6130 A C expression whose value is a string, including spacing, containing the
6131 assembler operation that should precede instructions and read-only data.
6132 Normally @code{"\t.text"} is right.
6135 @defmac HOT_TEXT_SECTION_NAME
6136 If defined, a C string constant for the name of the section containing most
6137 frequently executed functions of the program. If not defined, GCC will provide
6138 a default definition if the target supports named sections.
6141 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6142 If defined, a C string constant for the name of the section containing unlikely
6143 executed functions in the program.
6146 @defmac DATA_SECTION_ASM_OP
6147 A C expression whose value is a string, including spacing, containing the
6148 assembler operation to identify the following data as writable initialized
6149 data. Normally @code{"\t.data"} is right.
6152 @defmac SDATA_SECTION_ASM_OP
6153 If defined, a C expression whose value is a string, including spacing,
6154 containing the assembler operation to identify the following data as
6155 initialized, writable small data.
6158 @defmac READONLY_DATA_SECTION_ASM_OP
6159 A C expression whose value is a string, including spacing, containing the
6160 assembler operation to identify the following data as read-only initialized
6164 @defmac BSS_SECTION_ASM_OP
6165 If defined, a C expression whose value is a string, including spacing,
6166 containing the assembler operation to identify the following data as
6167 uninitialized global data. If not defined, and neither
6168 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6169 uninitialized global data will be output in the data section if
6170 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6174 @defmac SBSS_SECTION_ASM_OP
6175 If defined, a C expression whose value is a string, including spacing,
6176 containing the assembler operation to identify the following data as
6177 uninitialized, writable small data.
6180 @defmac INIT_SECTION_ASM_OP
6181 If defined, a C expression whose value is a string, including spacing,
6182 containing the assembler operation to identify the following data as
6183 initialization code. If not defined, GCC will assume such a section does
6184 not exist. This section has no corresponding @code{init_section}
6185 variable; it is used entirely in runtime code.
6188 @defmac FINI_SECTION_ASM_OP
6189 If defined, a C expression whose value is a string, including spacing,
6190 containing the assembler operation to identify the following data as
6191 finalization code. If not defined, GCC will assume such a section does
6192 not exist. This section has no corresponding @code{fini_section}
6193 variable; it is used entirely in runtime code.
6196 @defmac INIT_ARRAY_SECTION_ASM_OP
6197 If defined, a C expression whose value is a string, including spacing,
6198 containing the assembler operation to identify the following data as
6199 part of the @code{.init_array} (or equivalent) section. If not
6200 defined, GCC will assume such a section does not exist. Do not define
6201 both this macro and @code{INIT_SECTION_ASM_OP}.
6204 @defmac FINI_ARRAY_SECTION_ASM_OP
6205 If defined, a C expression whose value is a string, including spacing,
6206 containing the assembler operation to identify the following data as
6207 part of the @code{.fini_array} (or equivalent) section. If not
6208 defined, GCC will assume such a section does not exist. Do not define
6209 both this macro and @code{FINI_SECTION_ASM_OP}.
6212 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6213 If defined, an ASM statement that switches to a different section
6214 via @var{section_op}, calls @var{function}, and switches back to
6215 the text section. This is used in @file{crtstuff.c} if
6216 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6217 to initialization and finalization functions from the init and fini
6218 sections. By default, this macro uses a simple function call. Some
6219 ports need hand-crafted assembly code to avoid dependencies on
6220 registers initialized in the function prologue or to ensure that
6221 constant pools don't end up too far way in the text section.
6224 @defmac TARGET_LIBGCC_SDATA_SECTION
6225 If defined, a string which names the section into which small
6226 variables defined in crtstuff and libgcc should go. This is useful
6227 when the target has options for optimizing access to small data, and
6228 you want the crtstuff and libgcc routines to be conservative in what
6229 they expect of your application yet liberal in what your application
6230 expects. For example, for targets with a @code{.sdata} section (like
6231 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6232 require small data support from your application, but use this macro
6233 to put small data into @code{.sdata} so that your application can
6234 access these variables whether it uses small data or not.
6237 @defmac FORCE_CODE_SECTION_ALIGN
6238 If defined, an ASM statement that aligns a code section to some
6239 arbitrary boundary. This is used to force all fragments of the
6240 @code{.init} and @code{.fini} sections to have to same alignment
6241 and thus prevent the linker from having to add any padding.
6244 @defmac JUMP_TABLES_IN_TEXT_SECTION
6245 Define this macro to be an expression with a nonzero value if jump
6246 tables (for @code{tablejump} insns) should be output in the text
6247 section, along with the assembler instructions. Otherwise, the
6248 readonly data section is used.
6250 This macro is irrelevant if there is no separate readonly data section.
6253 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
6254 Define this hook if you need to do something special to set up the
6255 @file{varasm.c} sections, or if your target has some special sections
6256 of its own that you need to create.
6258 GCC calls this hook after processing the command line, but before writing
6259 any assembly code, and before calling any of the section-returning hooks
6263 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
6264 Return the section into which @var{exp} should be placed. You can
6265 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6266 some sort. @var{reloc} indicates whether the initial value of @var{exp}
6267 requires link-time relocations. Bit 0 is set when variable contains
6268 local relocations only, while bit 1 is set for global relocations.
6269 @var{align} is the constant alignment in bits.
6271 The default version of this function takes care of putting read-only
6272 variables in @code{readonly_data_section}.
6274 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
6277 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
6278 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
6279 for @code{FUNCTION_DECL}s as well as for variables and constants.
6281 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
6282 function has been determined to be likely to be called, and nonzero if
6283 it is unlikely to be called.
6286 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
6287 Build up a unique section name, expressed as a @code{STRING_CST} node,
6288 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
6289 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
6290 the initial value of @var{exp} requires link-time relocations.
6292 The default version of this function appends the symbol name to the
6293 ELF section name that would normally be used for the symbol. For
6294 example, the function @code{foo} would be placed in @code{.text.foo}.
6295 Whatever the actual target object format, this is often good enough.
6298 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
6299 Return the readonly data section associated with
6300 @samp{DECL_SECTION_NAME (@var{decl})}.
6301 The default version of this function selects @code{.gnu.linkonce.r.name} if
6302 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
6303 if function is in @code{.text.name}, and the normal readonly-data section
6307 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
6308 Return the section into which a constant @var{x}, of mode @var{mode},
6309 should be placed. You can assume that @var{x} is some kind of
6310 constant in RTL@. The argument @var{mode} is redundant except in the
6311 case of a @code{const_int} rtx. @var{align} is the constant alignment
6314 The default version of this function takes care of putting symbolic
6315 constants in @code{flag_pic} mode in @code{data_section} and everything
6316 else in @code{readonly_data_section}.
6319 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
6320 Define this hook if references to a symbol or a constant must be
6321 treated differently depending on something about the variable or
6322 function named by the symbol (such as what section it is in).
6324 The hook is executed immediately after rtl has been created for
6325 @var{decl}, which may be a variable or function declaration or
6326 an entry in the constant pool. In either case, @var{rtl} is the
6327 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
6328 in this hook; that field may not have been initialized yet.
6330 In the case of a constant, it is safe to assume that the rtl is
6331 a @code{mem} whose address is a @code{symbol_ref}. Most decls
6332 will also have this form, but that is not guaranteed. Global
6333 register variables, for instance, will have a @code{reg} for their
6334 rtl. (Normally the right thing to do with such unusual rtl is
6337 The @var{new_decl_p} argument will be true if this is the first time
6338 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
6339 be false for subsequent invocations, which will happen for duplicate
6340 declarations. Whether or not anything must be done for the duplicate
6341 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
6342 @var{new_decl_p} is always true when the hook is called for a constant.
6344 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
6345 The usual thing for this hook to do is to record flags in the
6346 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
6347 Historically, the name string was modified if it was necessary to
6348 encode more than one bit of information, but this practice is now
6349 discouraged; use @code{SYMBOL_REF_FLAGS}.
6351 The default definition of this hook, @code{default_encode_section_info}
6352 in @file{varasm.c}, sets a number of commonly-useful bits in
6353 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
6354 before overriding it.
6357 @deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name)
6358 Decode @var{name} and return the real name part, sans
6359 the characters that @code{TARGET_ENCODE_SECTION_INFO}
6363 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp})
6364 Returns true if @var{exp} should be placed into a ``small data'' section.
6365 The default version of this hook always returns false.
6368 @deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
6369 Contains the value true if the target places read-only
6370 ``small data'' into a separate section. The default value is false.
6373 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp})
6374 Returns true if @var{exp} names an object for which name resolution
6375 rules must resolve to the current ``module'' (dynamic shared library
6376 or executable image).
6378 The default version of this hook implements the name resolution rules
6379 for ELF, which has a looser model of global name binding than other
6380 currently supported object file formats.
6383 @deftypevar {Target Hook} bool TARGET_HAVE_TLS
6384 Contains the value true if the target supports thread-local storage.
6385 The default value is false.
6390 @section Position Independent Code
6391 @cindex position independent code
6394 This section describes macros that help implement generation of position
6395 independent code. Simply defining these macros is not enough to
6396 generate valid PIC; you must also add support to the macros
6397 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
6398 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
6399 @samp{movsi} to do something appropriate when the source operand
6400 contains a symbolic address. You may also need to alter the handling of
6401 switch statements so that they use relative addresses.
6402 @c i rearranged the order of the macros above to try to force one of
6403 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
6405 @defmac PIC_OFFSET_TABLE_REGNUM
6406 The register number of the register used to address a table of static
6407 data addresses in memory. In some cases this register is defined by a
6408 processor's ``application binary interface'' (ABI)@. When this macro
6409 is defined, RTL is generated for this register once, as with the stack
6410 pointer and frame pointer registers. If this macro is not defined, it
6411 is up to the machine-dependent files to allocate such a register (if
6412 necessary). Note that this register must be fixed when in use (e.g.@:
6413 when @code{flag_pic} is true).
6416 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
6417 Define this macro if the register defined by
6418 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
6419 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
6422 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
6423 A C expression that is nonzero if @var{x} is a legitimate immediate
6424 operand on the target machine when generating position independent code.
6425 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6426 check this. You can also assume @var{flag_pic} is true, so you need not
6427 check it either. You need not define this macro if all constants
6428 (including @code{SYMBOL_REF}) can be immediate operands when generating
6429 position independent code.
6432 @node Assembler Format
6433 @section Defining the Output Assembler Language
6435 This section describes macros whose principal purpose is to describe how
6436 to write instructions in assembler language---rather than what the
6440 * File Framework:: Structural information for the assembler file.
6441 * Data Output:: Output of constants (numbers, strings, addresses).
6442 * Uninitialized Data:: Output of uninitialized variables.
6443 * Label Output:: Output and generation of labels.
6444 * Initialization:: General principles of initialization
6445 and termination routines.
6446 * Macros for Initialization::
6447 Specific macros that control the handling of
6448 initialization and termination routines.
6449 * Instruction Output:: Output of actual instructions.
6450 * Dispatch Tables:: Output of jump tables.
6451 * Exception Region Output:: Output of exception region code.
6452 * Alignment Output:: Pseudo ops for alignment and skipping data.
6455 @node File Framework
6456 @subsection The Overall Framework of an Assembler File
6457 @cindex assembler format
6458 @cindex output of assembler code
6460 @c prevent bad page break with this line
6461 This describes the overall framework of an assembly file.
6463 @deftypefn {Target Hook} void TARGET_ASM_FILE_START ()
6464 @findex default_file_start
6465 Output to @code{asm_out_file} any text which the assembler expects to
6466 find at the beginning of a file. The default behavior is controlled
6467 by two flags, documented below. Unless your target's assembler is
6468 quite unusual, if you override the default, you should call
6469 @code{default_file_start} at some point in your target hook. This
6470 lets other target files rely on these variables.
6473 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
6474 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
6475 printed as the very first line in the assembly file, unless
6476 @option{-fverbose-asm} is in effect. (If that macro has been defined
6477 to the empty string, this variable has no effect.) With the normal
6478 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
6479 assembler that it need not bother stripping comments or extra
6480 whitespace from its input. This allows it to work a bit faster.
6482 The default is false. You should not set it to true unless you have
6483 verified that your port does not generate any extra whitespace or
6484 comments that will cause GAS to issue errors in NO_APP mode.
6487 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
6488 If this flag is true, @code{output_file_directive} will be called
6489 for the primary source file, immediately after printing
6490 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
6491 this to be done. The default is false.
6494 @deftypefn {Target Hook} void TARGET_ASM_FILE_END ()
6495 Output to @code{asm_out_file} any text which the assembler expects
6496 to find at the end of a file. The default is to output nothing.
6499 @deftypefun void file_end_indicate_exec_stack ()
6500 Some systems use a common convention, the @samp{.note.GNU-stack}
6501 special section, to indicate whether or not an object file relies on
6502 the stack being executable. If your system uses this convention, you
6503 should define @code{TARGET_ASM_FILE_END} to this function. If you
6504 need to do other things in that hook, have your hook function call
6508 @defmac ASM_COMMENT_START
6509 A C string constant describing how to begin a comment in the target
6510 assembler language. The compiler assumes that the comment will end at
6511 the end of the line.
6515 A C string constant for text to be output before each @code{asm}
6516 statement or group of consecutive ones. Normally this is
6517 @code{"#APP"}, which is a comment that has no effect on most
6518 assemblers but tells the GNU assembler that it must check the lines
6519 that follow for all valid assembler constructs.
6523 A C string constant for text to be output after each @code{asm}
6524 statement or group of consecutive ones. Normally this is
6525 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6526 time-saving assumptions that are valid for ordinary compiler output.
6529 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6530 A C statement to output COFF information or DWARF debugging information
6531 which indicates that filename @var{name} is the current source file to
6532 the stdio stream @var{stream}.
6534 This macro need not be defined if the standard form of output
6535 for the file format in use is appropriate.
6538 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
6539 A C statement to output the string @var{string} to the stdio stream
6540 @var{stream}. If you do not call the function @code{output_quoted_string}
6541 in your config files, GCC will only call it to output filenames to
6542 the assembler source. So you can use it to canonicalize the format
6543 of the filename using this macro.
6546 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
6547 A C statement to output something to the assembler file to handle a
6548 @samp{#ident} directive containing the text @var{string}. If this
6549 macro is not defined, nothing is output for a @samp{#ident} directive.
6552 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align})
6553 Output assembly directives to switch to section @var{name}. The section
6554 should have attributes as specified by @var{flags}, which is a bit mask
6555 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align}
6556 is nonzero, it contains an alignment in bytes to be used for the section,
6557 otherwise some target default should be used. Only targets that must
6558 specify an alignment within the section directive need pay attention to
6559 @var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}.
6562 @deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
6563 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
6566 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
6567 @deftypefn {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
6568 This flag is true if we can create zeroed data by switching to a BSS
6569 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
6570 This is true on most ELF targets.
6573 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
6574 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
6575 based on a variable or function decl, a section name, and whether or not the
6576 declaration's initializer may contain runtime relocations. @var{decl} may be
6577 null, in which case read-write data should be assumed.
6579 The default version of this function handles choosing code vs data,
6580 read-only vs read-write data, and @code{flag_pic}. You should only
6581 need to override this if your target has special flags that might be
6582 set via @code{__attribute__}.
6587 @subsection Output of Data
6590 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
6591 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
6592 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
6593 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
6594 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
6595 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
6596 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
6597 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
6598 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
6599 These hooks specify assembly directives for creating certain kinds
6600 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
6601 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
6602 aligned two-byte object, and so on. Any of the hooks may be
6603 @code{NULL}, indicating that no suitable directive is available.
6605 The compiler will print these strings at the start of a new line,
6606 followed immediately by the object's initial value. In most cases,
6607 the string should contain a tab, a pseudo-op, and then another tab.
6610 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
6611 The @code{assemble_integer} function uses this hook to output an
6612 integer object. @var{x} is the object's value, @var{size} is its size
6613 in bytes and @var{aligned_p} indicates whether it is aligned. The
6614 function should return @code{true} if it was able to output the
6615 object. If it returns false, @code{assemble_integer} will try to
6616 split the object into smaller parts.
6618 The default implementation of this hook will use the
6619 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
6620 when the relevant string is @code{NULL}.
6623 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
6624 A C statement to recognize @var{rtx} patterns that
6625 @code{output_addr_const} can't deal with, and output assembly code to
6626 @var{stream} corresponding to the pattern @var{x}. This may be used to
6627 allow machine-dependent @code{UNSPEC}s to appear within constants.
6629 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
6630 @code{goto fail}, so that a standard error message is printed. If it
6631 prints an error message itself, by calling, for example,
6632 @code{output_operand_lossage}, it may just complete normally.
6635 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
6636 A C statement to output to the stdio stream @var{stream} an assembler
6637 instruction to assemble a string constant containing the @var{len}
6638 bytes at @var{ptr}. @var{ptr} will be a C expression of type
6639 @code{char *} and @var{len} a C expression of type @code{int}.
6641 If the assembler has a @code{.ascii} pseudo-op as found in the
6642 Berkeley Unix assembler, do not define the macro
6643 @code{ASM_OUTPUT_ASCII}.
6646 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
6647 A C statement to output word @var{n} of a function descriptor for
6648 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
6649 is defined, and is otherwise unused.
6652 @defmac CONSTANT_POOL_BEFORE_FUNCTION
6653 You may define this macro as a C expression. You should define the
6654 expression to have a nonzero value if GCC should output the constant
6655 pool for a function before the code for the function, or a zero value if
6656 GCC should output the constant pool after the function. If you do
6657 not define this macro, the usual case, GCC will output the constant
6658 pool before the function.
6661 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
6662 A C statement to output assembler commands to define the start of the
6663 constant pool for a function. @var{funname} is a string giving
6664 the name of the function. Should the return type of the function
6665 be required, it can be obtained via @var{fundecl}. @var{size}
6666 is the size, in bytes, of the constant pool that will be written
6667 immediately after this call.
6669 If no constant-pool prefix is required, the usual case, this macro need
6673 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
6674 A C statement (with or without semicolon) to output a constant in the
6675 constant pool, if it needs special treatment. (This macro need not do
6676 anything for RTL expressions that can be output normally.)
6678 The argument @var{file} is the standard I/O stream to output the
6679 assembler code on. @var{x} is the RTL expression for the constant to
6680 output, and @var{mode} is the machine mode (in case @var{x} is a
6681 @samp{const_int}). @var{align} is the required alignment for the value
6682 @var{x}; you should output an assembler directive to force this much
6685 The argument @var{labelno} is a number to use in an internal label for
6686 the address of this pool entry. The definition of this macro is
6687 responsible for outputting the label definition at the proper place.
6688 Here is how to do this:
6691 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
6694 When you output a pool entry specially, you should end with a
6695 @code{goto} to the label @var{jumpto}. This will prevent the same pool
6696 entry from being output a second time in the usual manner.
6698 You need not define this macro if it would do nothing.
6701 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
6702 A C statement to output assembler commands to at the end of the constant
6703 pool for a function. @var{funname} is a string giving the name of the
6704 function. Should the return type of the function be required, you can
6705 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
6706 constant pool that GCC wrote immediately before this call.
6708 If no constant-pool epilogue is required, the usual case, you need not
6712 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C})
6713 Define this macro as a C expression which is nonzero if @var{C} is
6714 used as a logical line separator by the assembler.
6716 If you do not define this macro, the default is that only
6717 the character @samp{;} is treated as a logical line separator.
6720 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
6721 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
6722 These target hooks are C string constants, describing the syntax in the
6723 assembler for grouping arithmetic expressions. If not overridden, they
6724 default to normal parentheses, which is correct for most assemblers.
6727 These macros are provided by @file{real.h} for writing the definitions
6728 of @code{ASM_OUTPUT_DOUBLE} and the like:
6730 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
6731 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
6732 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
6733 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
6734 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
6735 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
6736 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
6737 target's floating point representation, and store its bit pattern in
6738 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
6739 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
6740 simple @code{long int}. For the others, it should be an array of
6741 @code{long int}. The number of elements in this array is determined
6742 by the size of the desired target floating point data type: 32 bits of
6743 it go in each @code{long int} array element. Each array element holds
6744 32 bits of the result, even if @code{long int} is wider than 32 bits
6745 on the host machine.
6747 The array element values are designed so that you can print them out
6748 using @code{fprintf} in the order they should appear in the target
6752 @node Uninitialized Data
6753 @subsection Output of Uninitialized Variables
6755 Each of the macros in this section is used to do the whole job of
6756 outputting a single uninitialized variable.
6758 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6759 A C statement (sans semicolon) to output to the stdio stream
6760 @var{stream} the assembler definition of a common-label named
6761 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6762 is the size rounded up to whatever alignment the caller wants.
6764 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6765 output the name itself; before and after that, output the additional
6766 assembler syntax for defining the name, and a newline.
6768 This macro controls how the assembler definitions of uninitialized
6769 common global variables are output.
6772 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
6773 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
6774 separate, explicit argument. If you define this macro, it is used in
6775 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
6776 handling the required alignment of the variable. The alignment is specified
6777 as the number of bits.
6780 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6781 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
6782 variable to be output, if there is one, or @code{NULL_TREE} if there
6783 is no corresponding variable. If you define this macro, GCC will use it
6784 in place of both @code{ASM_OUTPUT_COMMON} and
6785 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
6786 the variable's decl in order to chose what to output.
6789 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6790 A C statement (sans semicolon) to output to the stdio stream
6791 @var{stream} the assembler definition of uninitialized global @var{decl} named
6792 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6793 is the size rounded up to whatever alignment the caller wants.
6795 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
6796 defining this macro. If unable, use the expression
6797 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
6798 before and after that, output the additional assembler syntax for defining
6799 the name, and a newline.
6801 There are two ways of handling global BSS. One is to define either
6802 this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
6803 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
6804 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
6805 You do not need to do both.
6807 Some languages do not have @code{common} data, and require a
6808 non-common form of global BSS in order to handle uninitialized globals
6809 efficiently. C++ is one example of this. However, if the target does
6810 not support global BSS, the front end may choose to make globals
6811 common in order to save space in the object file.
6814 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6815 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
6816 separate, explicit argument. If you define this macro, it is used in
6817 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
6818 handling the required alignment of the variable. The alignment is specified
6819 as the number of bits.
6821 Try to use function @code{asm_output_aligned_bss} defined in file
6822 @file{varasm.c} when defining this macro.
6825 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6826 A C statement (sans semicolon) to output to the stdio stream
6827 @var{stream} the assembler definition of a local-common-label named
6828 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6829 is the size rounded up to whatever alignment the caller wants.
6831 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6832 output the name itself; before and after that, output the additional
6833 assembler syntax for defining the name, and a newline.
6835 This macro controls how the assembler definitions of uninitialized
6836 static variables are output.
6839 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
6840 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
6841 separate, explicit argument. If you define this macro, it is used in
6842 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
6843 handling the required alignment of the variable. The alignment is specified
6844 as the number of bits.
6847 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6848 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
6849 variable to be output, if there is one, or @code{NULL_TREE} if there
6850 is no corresponding variable. If you define this macro, GCC will use it
6851 in place of both @code{ASM_OUTPUT_DECL} and
6852 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
6853 the variable's decl in order to chose what to output.
6857 @subsection Output and Generation of Labels
6859 @c prevent bad page break with this line
6860 This is about outputting labels.
6862 @findex assemble_name
6863 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
6864 A C statement (sans semicolon) to output to the stdio stream
6865 @var{stream} the assembler definition of a label named @var{name}.
6866 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6867 output the name itself; before and after that, output the additional
6868 assembler syntax for defining the name, and a newline. A default
6869 definition of this macro is provided which is correct for most systems.
6872 @findex assemble_name_raw
6873 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
6874 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
6875 to refer to a compiler-generated label. The default definition uses
6876 @code{assemble_name_raw}, which is like @code{assemble_name} except
6877 that it is more efficient.
6881 A C string containing the appropriate assembler directive to specify the
6882 size of a symbol, without any arguments. On systems that use ELF, the
6883 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
6884 systems, the default is not to define this macro.
6886 Define this macro only if it is correct to use the default definitions
6887 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
6888 for your system. If you need your own custom definitions of those
6889 macros, or if you do not need explicit symbol sizes at all, do not
6893 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
6894 A C statement (sans semicolon) to output to the stdio stream
6895 @var{stream} a directive telling the assembler that the size of the
6896 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
6897 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6901 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
6902 A C statement (sans semicolon) to output to the stdio stream
6903 @var{stream} a directive telling the assembler to calculate the size of
6904 the symbol @var{name} by subtracting its address from the current
6907 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6908 provided. The default assumes that the assembler recognizes a special
6909 @samp{.} symbol as referring to the current address, and can calculate
6910 the difference between this and another symbol. If your assembler does
6911 not recognize @samp{.} or cannot do calculations with it, you will need
6912 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
6916 A C string containing the appropriate assembler directive to specify the
6917 type of a symbol, without any arguments. On systems that use ELF, the
6918 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
6919 systems, the default is not to define this macro.
6921 Define this macro only if it is correct to use the default definition of
6922 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6923 custom definition of this macro, or if you do not need explicit symbol
6924 types at all, do not define this macro.
6927 @defmac TYPE_OPERAND_FMT
6928 A C string which specifies (using @code{printf} syntax) the format of
6929 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
6930 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
6931 the default is not to define this macro.
6933 Define this macro only if it is correct to use the default definition of
6934 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6935 custom definition of this macro, or if you do not need explicit symbol
6936 types at all, do not define this macro.
6939 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
6940 A C statement (sans semicolon) to output to the stdio stream
6941 @var{stream} a directive telling the assembler that the type of the
6942 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
6943 that string is always either @samp{"function"} or @samp{"object"}, but
6944 you should not count on this.
6946 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
6947 definition of this macro is provided.
6950 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
6951 A C statement (sans semicolon) to output to the stdio stream
6952 @var{stream} any text necessary for declaring the name @var{name} of a
6953 function which is being defined. This macro is responsible for
6954 outputting the label definition (perhaps using
6955 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
6956 @code{FUNCTION_DECL} tree node representing the function.
6958 If this macro is not defined, then the function name is defined in the
6959 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6961 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6965 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
6966 A C statement (sans semicolon) to output to the stdio stream
6967 @var{stream} any text necessary for declaring the size of a function
6968 which is being defined. The argument @var{name} is the name of the
6969 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
6970 representing the function.
6972 If this macro is not defined, then the function size is not defined.
6974 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
6978 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
6979 A C statement (sans semicolon) to output to the stdio stream
6980 @var{stream} any text necessary for declaring the name @var{name} of an
6981 initialized variable which is being defined. This macro must output the
6982 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
6983 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
6985 If this macro is not defined, then the variable name is defined in the
6986 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6988 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
6989 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
6992 @defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size})
6993 A C statement (sans semicolon) to output to the stdio stream
6994 @var{stream} any text necessary for declaring the name @var{name} of a
6995 constant which is being defined. This macro is responsible for
6996 outputting the label definition (perhaps using
6997 @code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the
6998 value of the constant, and @var{size} is the size of the constant
6999 in bytes. @var{name} will be an internal label.
7001 If this macro is not defined, then the @var{name} is defined in the
7002 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7004 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7008 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7009 A C statement (sans semicolon) to output to the stdio stream
7010 @var{stream} any text necessary for claiming a register @var{regno}
7011 for a global variable @var{decl} with name @var{name}.
7013 If you don't define this macro, that is equivalent to defining it to do
7017 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7018 A C statement (sans semicolon) to finish up declaring a variable name
7019 once the compiler has processed its initializer fully and thus has had a
7020 chance to determine the size of an array when controlled by an
7021 initializer. This is used on systems where it's necessary to declare
7022 something about the size of the object.
7024 If you don't define this macro, that is equivalent to defining it to do
7027 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7028 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7031 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
7032 This target hook is a function to output to the stdio stream
7033 @var{stream} some commands that will make the label @var{name} global;
7034 that is, available for reference from other files.
7036 The default implementation relies on a proper definition of
7037 @code{GLOBAL_ASM_OP}.
7040 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7041 A C statement (sans semicolon) to output to the stdio stream
7042 @var{stream} some commands that will make the label @var{name} weak;
7043 that is, available for reference from other files but only used if
7044 no other definition is available. Use the expression
7045 @code{assemble_name (@var{stream}, @var{name})} to output the name
7046 itself; before and after that, output the additional assembler syntax
7047 for making that name weak, and a newline.
7049 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7050 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7054 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7055 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7056 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7057 or variable decl. If @var{value} is not @code{NULL}, this C statement
7058 should output to the stdio stream @var{stream} assembler code which
7059 defines (equates) the weak symbol @var{name} to have the value
7060 @var{value}. If @var{value} is @code{NULL}, it should output commands
7061 to make @var{name} weak.
7064 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7065 Outputs a directive that enables @var{name} to be used to refer to
7066 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7067 declaration of @code{name}.
7070 @defmac SUPPORTS_WEAK
7071 A C expression which evaluates to true if the target supports weak symbols.
7073 If you don't define this macro, @file{defaults.h} provides a default
7074 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7075 is defined, the default definition is @samp{1}; otherwise, it is
7076 @samp{0}. Define this macro if you want to control weak symbol support
7077 with a compiler flag such as @option{-melf}.
7080 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
7081 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7082 public symbol such that extra copies in multiple translation units will
7083 be discarded by the linker. Define this macro if your object file
7084 format provides support for this concept, such as the @samp{COMDAT}
7085 section flags in the Microsoft Windows PE/COFF format, and this support
7086 requires changes to @var{decl}, such as putting it in a separate section.
7089 @defmac SUPPORTS_ONE_ONLY
7090 A C expression which evaluates to true if the target supports one-only
7093 If you don't define this macro, @file{varasm.c} provides a default
7094 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7095 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7096 you want to control one-only symbol support with a compiler flag, or if
7097 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7098 be emitted as one-only.
7101 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility})
7102 This target hook is a function to output to @var{asm_out_file} some
7103 commands that will make the symbol(s) associated with @var{decl} have
7104 hidden, protected or internal visibility as specified by @var{visibility}.
7107 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7108 A C expression that evaluates to true if the target's linker expects
7109 that weak symbols do not appear in a static archive's table of contents.
7110 The default is @code{0}.
7112 Leaving weak symbols out of an archive's table of contents means that,
7113 if a symbol will only have a definition in one translation unit and
7114 will have undefined references from other translation units, that
7115 symbol should not be weak. Defining this macro to be nonzero will
7116 thus have the effect that certain symbols that would normally be weak
7117 (explicit template instantiations, and vtables for polymorphic classes
7118 with noninline key methods) will instead be nonweak.
7120 The C++ ABI requires this macro to be zero. Define this macro for
7121 targets where full C++ ABI compliance is impossible and where linker
7122 restrictions require weak symbols to be left out of a static archive's
7126 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7127 A C statement (sans semicolon) to output to the stdio stream
7128 @var{stream} any text necessary for declaring the name of an external
7129 symbol named @var{name} which is referenced in this compilation but
7130 not defined. The value of @var{decl} is the tree node for the
7133 This macro need not be defined if it does not need to output anything.
7134 The GNU assembler and most Unix assemblers don't require anything.
7137 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
7138 This target hook is a function to output to @var{asm_out_file} an assembler
7139 pseudo-op to declare a library function name external. The name of the
7140 library function is given by @var{symref}, which is a @code{symbol_ref}.
7143 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (tree @var{decl})
7144 This target hook is a function to output to @var{asm_out_file} an assembler
7145 directive to annotate used symbol. Darwin target use .no_dead_code_strip
7149 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7150 A C statement (sans semicolon) to output to the stdio stream
7151 @var{stream} a reference in assembler syntax to a label named
7152 @var{name}. This should add @samp{_} to the front of the name, if that
7153 is customary on your operating system, as it is in most Berkeley Unix
7154 systems. This macro is used in @code{assemble_name}.
7157 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
7158 A C statement (sans semicolon) to output a reference to
7159 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
7160 will be used to output the name of the symbol. This macro may be used
7161 to modify the way a symbol is referenced depending on information
7162 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
7165 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
7166 A C statement (sans semicolon) to output a reference to @var{buf}, the
7167 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
7168 @code{assemble_name} will be used to output the name of the symbol.
7169 This macro is not used by @code{output_asm_label}, or the @code{%l}
7170 specifier that calls it; the intention is that this macro should be set
7171 when it is necessary to output a label differently when its address is
7175 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
7176 A function to output to the stdio stream @var{stream} a label whose
7177 name is made from the string @var{prefix} and the number @var{labelno}.
7179 It is absolutely essential that these labels be distinct from the labels
7180 used for user-level functions and variables. Otherwise, certain programs
7181 will have name conflicts with internal labels.
7183 It is desirable to exclude internal labels from the symbol table of the
7184 object file. Most assemblers have a naming convention for labels that
7185 should be excluded; on many systems, the letter @samp{L} at the
7186 beginning of a label has this effect. You should find out what
7187 convention your system uses, and follow it.
7189 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
7192 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
7193 A C statement to output to the stdio stream @var{stream} a debug info
7194 label whose name is made from the string @var{prefix} and the number
7195 @var{num}. This is useful for VLIW targets, where debug info labels
7196 may need to be treated differently than branch target labels. On some
7197 systems, branch target labels must be at the beginning of instruction
7198 bundles, but debug info labels can occur in the middle of instruction
7201 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
7205 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
7206 A C statement to store into the string @var{string} a label whose name
7207 is made from the string @var{prefix} and the number @var{num}.
7209 This string, when output subsequently by @code{assemble_name}, should
7210 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
7211 with the same @var{prefix} and @var{num}.
7213 If the string begins with @samp{*}, then @code{assemble_name} will
7214 output the rest of the string unchanged. It is often convenient for
7215 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
7216 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
7217 to output the string, and may change it. (Of course,
7218 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
7219 you should know what it does on your machine.)
7222 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7223 A C expression to assign to @var{outvar} (which is a variable of type
7224 @code{char *}) a newly allocated string made from the string
7225 @var{name} and the number @var{number}, with some suitable punctuation
7226 added. Use @code{alloca} to get space for the string.
7228 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
7229 produce an assembler label for an internal static variable whose name is
7230 @var{name}. Therefore, the string must be such as to result in valid
7231 assembler code. The argument @var{number} is different each time this
7232 macro is executed; it prevents conflicts between similarly-named
7233 internal static variables in different scopes.
7235 Ideally this string should not be a valid C identifier, to prevent any
7236 conflict with the user's own symbols. Most assemblers allow periods
7237 or percent signs in assembler symbols; putting at least one of these
7238 between the name and the number will suffice.
7240 If this macro is not defined, a default definition will be provided
7241 which is correct for most systems.
7244 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
7245 A C statement to output to the stdio stream @var{stream} assembler code
7246 which defines (equates) the symbol @var{name} to have the value @var{value}.
7249 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7250 correct for most systems.
7253 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
7254 A C statement to output to the stdio stream @var{stream} assembler code
7255 which defines (equates) the symbol whose tree node is @var{decl_of_name}
7256 to have the value of the tree node @var{decl_of_value}. This macro will
7257 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
7258 the tree nodes are available.
7261 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7262 correct for most systems.
7265 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
7266 A C statement that evaluates to true if the assembler code which defines
7267 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
7268 of the tree node @var{decl_of_value} should be emitted near the end of the
7269 current compilation unit. The default is to not defer output of defines.
7270 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
7271 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
7274 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
7275 A C statement to output to the stdio stream @var{stream} assembler code
7276 which defines (equates) the weak symbol @var{name} to have the value
7277 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
7278 an undefined weak symbol.
7280 Define this macro if the target only supports weak aliases; define
7281 @code{ASM_OUTPUT_DEF} instead if possible.
7284 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
7285 Define this macro to override the default assembler names used for
7286 Objective-C methods.
7288 The default name is a unique method number followed by the name of the
7289 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
7290 the category is also included in the assembler name (e.g.@:
7293 These names are safe on most systems, but make debugging difficult since
7294 the method's selector is not present in the name. Therefore, particular
7295 systems define other ways of computing names.
7297 @var{buf} is an expression of type @code{char *} which gives you a
7298 buffer in which to store the name; its length is as long as
7299 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
7300 50 characters extra.
7302 The argument @var{is_inst} specifies whether the method is an instance
7303 method or a class method; @var{class_name} is the name of the class;
7304 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
7305 in a category); and @var{sel_name} is the name of the selector.
7307 On systems where the assembler can handle quoted names, you can use this
7308 macro to provide more human-readable names.
7311 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
7312 A C statement (sans semicolon) to output to the stdio stream
7313 @var{stream} commands to declare that the label @var{name} is an
7314 Objective-C class reference. This is only needed for targets whose
7315 linkers have special support for NeXT-style runtimes.
7318 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
7319 A C statement (sans semicolon) to output to the stdio stream
7320 @var{stream} commands to declare that the label @var{name} is an
7321 unresolved Objective-C class reference. This is only needed for targets
7322 whose linkers have special support for NeXT-style runtimes.
7325 @node Initialization
7326 @subsection How Initialization Functions Are Handled
7327 @cindex initialization routines
7328 @cindex termination routines
7329 @cindex constructors, output of
7330 @cindex destructors, output of
7332 The compiled code for certain languages includes @dfn{constructors}
7333 (also called @dfn{initialization routines})---functions to initialize
7334 data in the program when the program is started. These functions need
7335 to be called before the program is ``started''---that is to say, before
7336 @code{main} is called.
7338 Compiling some languages generates @dfn{destructors} (also called
7339 @dfn{termination routines}) that should be called when the program
7342 To make the initialization and termination functions work, the compiler
7343 must output something in the assembler code to cause those functions to
7344 be called at the appropriate time. When you port the compiler to a new
7345 system, you need to specify how to do this.
7347 There are two major ways that GCC currently supports the execution of
7348 initialization and termination functions. Each way has two variants.
7349 Much of the structure is common to all four variations.
7351 @findex __CTOR_LIST__
7352 @findex __DTOR_LIST__
7353 The linker must build two lists of these functions---a list of
7354 initialization functions, called @code{__CTOR_LIST__}, and a list of
7355 termination functions, called @code{__DTOR_LIST__}.
7357 Each list always begins with an ignored function pointer (which may hold
7358 0, @minus{}1, or a count of the function pointers after it, depending on
7359 the environment). This is followed by a series of zero or more function
7360 pointers to constructors (or destructors), followed by a function
7361 pointer containing zero.
7363 Depending on the operating system and its executable file format, either
7364 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
7365 time and exit time. Constructors are called in reverse order of the
7366 list; destructors in forward order.
7368 The best way to handle static constructors works only for object file
7369 formats which provide arbitrarily-named sections. A section is set
7370 aside for a list of constructors, and another for a list of destructors.
7371 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
7372 object file that defines an initialization function also puts a word in
7373 the constructor section to point to that function. The linker
7374 accumulates all these words into one contiguous @samp{.ctors} section.
7375 Termination functions are handled similarly.
7377 This method will be chosen as the default by @file{target-def.h} if
7378 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
7379 support arbitrary sections, but does support special designated
7380 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
7381 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
7383 When arbitrary sections are available, there are two variants, depending
7384 upon how the code in @file{crtstuff.c} is called. On systems that
7385 support a @dfn{.init} section which is executed at program startup,
7386 parts of @file{crtstuff.c} are compiled into that section. The
7387 program is linked by the @command{gcc} driver like this:
7390 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
7393 The prologue of a function (@code{__init}) appears in the @code{.init}
7394 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
7395 for the function @code{__fini} in the @dfn{.fini} section. Normally these
7396 files are provided by the operating system or by the GNU C library, but
7397 are provided by GCC for a few targets.
7399 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
7400 compiled from @file{crtstuff.c}. They contain, among other things, code
7401 fragments within the @code{.init} and @code{.fini} sections that branch
7402 to routines in the @code{.text} section. The linker will pull all parts
7403 of a section together, which results in a complete @code{__init} function
7404 that invokes the routines we need at startup.
7406 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
7409 If no init section is available, when GCC compiles any function called
7410 @code{main} (or more accurately, any function designated as a program
7411 entry point by the language front end calling @code{expand_main_function}),
7412 it inserts a procedure call to @code{__main} as the first executable code
7413 after the function prologue. The @code{__main} function is defined
7414 in @file{libgcc2.c} and runs the global constructors.
7416 In file formats that don't support arbitrary sections, there are again
7417 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
7418 and an `a.out' format must be used. In this case,
7419 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
7420 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
7421 and with the address of the void function containing the initialization
7422 code as its value. The GNU linker recognizes this as a request to add
7423 the value to a @dfn{set}; the values are accumulated, and are eventually
7424 placed in the executable as a vector in the format described above, with
7425 a leading (ignored) count and a trailing zero element.
7426 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
7427 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
7428 the compilation of @code{main} to call @code{__main} as above, starting
7429 the initialization process.
7431 The last variant uses neither arbitrary sections nor the GNU linker.
7432 This is preferable when you want to do dynamic linking and when using
7433 file formats which the GNU linker does not support, such as `ECOFF'@. In
7434 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
7435 termination functions are recognized simply by their names. This requires
7436 an extra program in the linkage step, called @command{collect2}. This program
7437 pretends to be the linker, for use with GCC; it does its job by running
7438 the ordinary linker, but also arranges to include the vectors of
7439 initialization and termination functions. These functions are called
7440 via @code{__main} as described above. In order to use this method,
7441 @code{use_collect2} must be defined in the target in @file{config.gcc}.
7444 The following section describes the specific macros that control and
7445 customize the handling of initialization and termination functions.
7448 @node Macros for Initialization
7449 @subsection Macros Controlling Initialization Routines
7451 Here are the macros that control how the compiler handles initialization
7452 and termination functions:
7454 @defmac INIT_SECTION_ASM_OP
7455 If defined, a C string constant, including spacing, for the assembler
7456 operation to identify the following data as initialization code. If not
7457 defined, GCC will assume such a section does not exist. When you are
7458 using special sections for initialization and termination functions, this
7459 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
7460 run the initialization functions.
7463 @defmac HAS_INIT_SECTION
7464 If defined, @code{main} will not call @code{__main} as described above.
7465 This macro should be defined for systems that control start-up code
7466 on a symbol-by-symbol basis, such as OSF/1, and should not
7467 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
7470 @defmac LD_INIT_SWITCH
7471 If defined, a C string constant for a switch that tells the linker that
7472 the following symbol is an initialization routine.
7475 @defmac LD_FINI_SWITCH
7476 If defined, a C string constant for a switch that tells the linker that
7477 the following symbol is a finalization routine.
7480 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
7481 If defined, a C statement that will write a function that can be
7482 automatically called when a shared library is loaded. The function
7483 should call @var{func}, which takes no arguments. If not defined, and
7484 the object format requires an explicit initialization function, then a
7485 function called @code{_GLOBAL__DI} will be generated.
7487 This function and the following one are used by collect2 when linking a
7488 shared library that needs constructors or destructors, or has DWARF2
7489 exception tables embedded in the code.
7492 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
7493 If defined, a C statement that will write a function that can be
7494 automatically called when a shared library is unloaded. The function
7495 should call @var{func}, which takes no arguments. If not defined, and
7496 the object format requires an explicit finalization function, then a
7497 function called @code{_GLOBAL__DD} will be generated.
7500 @defmac INVOKE__main
7501 If defined, @code{main} will call @code{__main} despite the presence of
7502 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
7503 where the init section is not actually run automatically, but is still
7504 useful for collecting the lists of constructors and destructors.
7507 @defmac SUPPORTS_INIT_PRIORITY
7508 If nonzero, the C++ @code{init_priority} attribute is supported and the
7509 compiler should emit instructions to control the order of initialization
7510 of objects. If zero, the compiler will issue an error message upon
7511 encountering an @code{init_priority} attribute.
7514 @deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS
7515 This value is true if the target supports some ``native'' method of
7516 collecting constructors and destructors to be run at startup and exit.
7517 It is false if we must use @command{collect2}.
7520 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
7521 If defined, a function that outputs assembler code to arrange to call
7522 the function referenced by @var{symbol} at initialization time.
7524 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
7525 no arguments and with no return value. If the target supports initialization
7526 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
7527 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
7529 If this macro is not defined by the target, a suitable default will
7530 be chosen if (1) the target supports arbitrary section names, (2) the
7531 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
7535 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
7536 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
7537 functions rather than initialization functions.
7540 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
7541 generated for the generated object file will have static linkage.
7543 If your system uses @command{collect2} as the means of processing
7544 constructors, then that program normally uses @command{nm} to scan
7545 an object file for constructor functions to be called.
7547 On certain kinds of systems, you can define this macro to make
7548 @command{collect2} work faster (and, in some cases, make it work at all):
7550 @defmac OBJECT_FORMAT_COFF
7551 Define this macro if the system uses COFF (Common Object File Format)
7552 object files, so that @command{collect2} can assume this format and scan
7553 object files directly for dynamic constructor/destructor functions.
7555 This macro is effective only in a native compiler; @command{collect2} as
7556 part of a cross compiler always uses @command{nm} for the target machine.
7559 @defmac REAL_NM_FILE_NAME
7560 Define this macro as a C string constant containing the file name to use
7561 to execute @command{nm}. The default is to search the path normally for
7564 If your system supports shared libraries and has a program to list the
7565 dynamic dependencies of a given library or executable, you can define
7566 these macros to enable support for running initialization and
7567 termination functions in shared libraries:
7571 Define this macro to a C string constant containing the name of the program
7572 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
7575 @defmac PARSE_LDD_OUTPUT (@var{ptr})
7576 Define this macro to be C code that extracts filenames from the output
7577 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
7578 of type @code{char *} that points to the beginning of a line of output
7579 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
7580 code must advance @var{ptr} to the beginning of the filename on that
7581 line. Otherwise, it must set @var{ptr} to @code{NULL}.
7584 @node Instruction Output
7585 @subsection Output of Assembler Instructions
7587 @c prevent bad page break with this line
7588 This describes assembler instruction output.
7590 @defmac REGISTER_NAMES
7591 A C initializer containing the assembler's names for the machine
7592 registers, each one as a C string constant. This is what translates
7593 register numbers in the compiler into assembler language.
7596 @defmac ADDITIONAL_REGISTER_NAMES
7597 If defined, a C initializer for an array of structures containing a name
7598 and a register number. This macro defines additional names for hard
7599 registers, thus allowing the @code{asm} option in declarations to refer
7600 to registers using alternate names.
7603 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7604 Define this macro if you are using an unusual assembler that
7605 requires different names for the machine instructions.
7607 The definition is a C statement or statements which output an
7608 assembler instruction opcode to the stdio stream @var{stream}. The
7609 macro-operand @var{ptr} is a variable of type @code{char *} which
7610 points to the opcode name in its ``internal'' form---the form that is
7611 written in the machine description. The definition should output the
7612 opcode name to @var{stream}, performing any translation you desire, and
7613 increment the variable @var{ptr} to point at the end of the opcode
7614 so that it will not be output twice.
7616 In fact, your macro definition may process less than the entire opcode
7617 name, or more than the opcode name; but if you want to process text
7618 that includes @samp{%}-sequences to substitute operands, you must take
7619 care of the substitution yourself. Just be sure to increment
7620 @var{ptr} over whatever text should not be output normally.
7622 @findex recog_data.operand
7623 If you need to look at the operand values, they can be found as the
7624 elements of @code{recog_data.operand}.
7626 If the macro definition does nothing, the instruction is output
7630 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7631 If defined, a C statement to be executed just prior to the output of
7632 assembler code for @var{insn}, to modify the extracted operands so
7633 they will be output differently.
7635 Here the argument @var{opvec} is the vector containing the operands
7636 extracted from @var{insn}, and @var{noperands} is the number of
7637 elements of the vector which contain meaningful data for this insn.
7638 The contents of this vector are what will be used to convert the insn
7639 template into assembler code, so you can change the assembler output
7640 by changing the contents of the vector.
7642 This macro is useful when various assembler syntaxes share a single
7643 file of instruction patterns; by defining this macro differently, you
7644 can cause a large class of instructions to be output differently (such
7645 as with rearranged operands). Naturally, variations in assembler
7646 syntax affecting individual insn patterns ought to be handled by
7647 writing conditional output routines in those patterns.
7649 If this macro is not defined, it is equivalent to a null statement.
7652 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7653 A C compound statement to output to stdio stream @var{stream} the
7654 assembler syntax for an instruction operand @var{x}. @var{x} is an
7657 @var{code} is a value that can be used to specify one of several ways
7658 of printing the operand. It is used when identical operands must be
7659 printed differently depending on the context. @var{code} comes from
7660 the @samp{%} specification that was used to request printing of the
7661 operand. If the specification was just @samp{%@var{digit}} then
7662 @var{code} is 0; if the specification was @samp{%@var{ltr}
7663 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7666 If @var{x} is a register, this macro should print the register's name.
7667 The names can be found in an array @code{reg_names} whose type is
7668 @code{char *[]}. @code{reg_names} is initialized from
7669 @code{REGISTER_NAMES}.
7671 When the machine description has a specification @samp{%@var{punct}}
7672 (a @samp{%} followed by a punctuation character), this macro is called
7673 with a null pointer for @var{x} and the punctuation character for
7677 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
7678 A C expression which evaluates to true if @var{code} is a valid
7679 punctuation character for use in the @code{PRINT_OPERAND} macro. If
7680 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
7681 punctuation characters (except for the standard one, @samp{%}) are used
7685 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7686 A C compound statement to output to stdio stream @var{stream} the
7687 assembler syntax for an instruction operand that is a memory reference
7688 whose address is @var{x}. @var{x} is an RTL expression.
7690 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
7691 On some machines, the syntax for a symbolic address depends on the
7692 section that the address refers to. On these machines, define the hook
7693 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
7694 @code{symbol_ref}, and then check for it here. @xref{Assembler
7698 @findex dbr_sequence_length
7699 @defmac DBR_OUTPUT_SEQEND (@var{file})
7700 A C statement, to be executed after all slot-filler instructions have
7701 been output. If necessary, call @code{dbr_sequence_length} to
7702 determine the number of slots filled in a sequence (zero if not
7703 currently outputting a sequence), to decide how many no-ops to output,
7706 Don't define this macro if it has nothing to do, but it is helpful in
7707 reading assembly output if the extent of the delay sequence is made
7708 explicit (e.g.@: with white space).
7711 @findex final_sequence
7712 Note that output routines for instructions with delay slots must be
7713 prepared to deal with not being output as part of a sequence
7714 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
7715 found.) The variable @code{final_sequence} is null when not
7716 processing a sequence, otherwise it contains the @code{sequence} rtx
7720 @defmac REGISTER_PREFIX
7721 @defmacx LOCAL_LABEL_PREFIX
7722 @defmacx USER_LABEL_PREFIX
7723 @defmacx IMMEDIATE_PREFIX
7724 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
7725 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
7726 @file{final.c}). These are useful when a single @file{md} file must
7727 support multiple assembler formats. In that case, the various @file{tm.h}
7728 files can define these macros differently.
7731 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
7732 If defined this macro should expand to a series of @code{case}
7733 statements which will be parsed inside the @code{switch} statement of
7734 the @code{asm_fprintf} function. This allows targets to define extra
7735 printf formats which may useful when generating their assembler
7736 statements. Note that uppercase letters are reserved for future
7737 generic extensions to asm_fprintf, and so are not available to target
7738 specific code. The output file is given by the parameter @var{file}.
7739 The varargs input pointer is @var{argptr} and the rest of the format
7740 string, starting the character after the one that is being switched
7741 upon, is pointed to by @var{format}.
7744 @defmac ASSEMBLER_DIALECT
7745 If your target supports multiple dialects of assembler language (such as
7746 different opcodes), define this macro as a C expression that gives the
7747 numeric index of the assembler language dialect to use, with zero as the
7750 If this macro is defined, you may use constructs of the form
7752 @samp{@{option0|option1|option2@dots{}@}}
7755 in the output templates of patterns (@pxref{Output Template}) or in the
7756 first argument of @code{asm_fprintf}. This construct outputs
7757 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
7758 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
7759 within these strings retain their usual meaning. If there are fewer
7760 alternatives within the braces than the value of
7761 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
7763 If you do not define this macro, the characters @samp{@{}, @samp{|} and
7764 @samp{@}} do not have any special meaning when used in templates or
7765 operands to @code{asm_fprintf}.
7767 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
7768 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
7769 the variations in assembler language syntax with that mechanism. Define
7770 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
7771 if the syntax variant are larger and involve such things as different
7772 opcodes or operand order.
7775 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
7776 A C expression to output to @var{stream} some assembler code
7777 which will push hard register number @var{regno} onto the stack.
7778 The code need not be optimal, since this macro is used only when
7782 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
7783 A C expression to output to @var{stream} some assembler code
7784 which will pop hard register number @var{regno} off of the stack.
7785 The code need not be optimal, since this macro is used only when
7789 @node Dispatch Tables
7790 @subsection Output of Dispatch Tables
7792 @c prevent bad page break with this line
7793 This concerns dispatch tables.
7795 @cindex dispatch table
7796 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
7797 A C statement to output to the stdio stream @var{stream} an assembler
7798 pseudo-instruction to generate a difference between two labels.
7799 @var{value} and @var{rel} are the numbers of two internal labels. The
7800 definitions of these labels are output using
7801 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
7802 way here. For example,
7805 fprintf (@var{stream}, "\t.word L%d-L%d\n",
7806 @var{value}, @var{rel})
7809 You must provide this macro on machines where the addresses in a
7810 dispatch table are relative to the table's own address. If defined, GCC
7811 will also use this macro on all machines when producing PIC@.
7812 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
7813 mode and flags can be read.
7816 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
7817 This macro should be provided on machines where the addresses
7818 in a dispatch table are absolute.
7820 The definition should be a C statement to output to the stdio stream
7821 @var{stream} an assembler pseudo-instruction to generate a reference to
7822 a label. @var{value} is the number of an internal label whose
7823 definition is output using @code{(*targetm.asm_out.internal_label)}.
7827 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
7831 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
7832 Define this if the label before a jump-table needs to be output
7833 specially. The first three arguments are the same as for
7834 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
7835 jump-table which follows (a @code{jump_insn} containing an
7836 @code{addr_vec} or @code{addr_diff_vec}).
7838 This feature is used on system V to output a @code{swbeg} statement
7841 If this macro is not defined, these labels are output with
7842 @code{(*targetm.asm_out.internal_label)}.
7845 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7846 Define this if something special must be output at the end of a
7847 jump-table. The definition should be a C statement to be executed
7848 after the assembler code for the table is written. It should write
7849 the appropriate code to stdio stream @var{stream}. The argument
7850 @var{table} is the jump-table insn, and @var{num} is the label-number
7851 of the preceding label.
7853 If this macro is not defined, nothing special is output at the end of
7857 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (@var{stream}, @var{decl}, @var{for_eh}, @var{empty})
7858 This target hook emits a label at the beginning of each FDE@. It
7859 should be defined on targets where FDEs need special labels, and it
7860 should write the appropriate label, for the FDE associated with the
7861 function declaration @var{decl}, to the stdio stream @var{stream}.
7862 The third argument, @var{for_eh}, is a boolean: true if this is for an
7863 exception table. The fourth argument, @var{empty}, is a boolean:
7864 true if this is a placeholder label for an omitted FDE@.
7866 The default is that FDEs are not given nonlocal labels.
7869 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (@var{stream})
7870 This target hook emits a label at the beginning of the exception table.
7871 It should be defined on targets where it is desirable for the table
7872 to be broken up according to function.
7874 The default is that no label is emitted.
7877 @deftypefn {Target Hook} void TARGET_UNWIND_EMIT (FILE * @var{stream}, rtx @var{insn})
7878 This target hook emits and assembly directives required to unwind the
7879 given instruction. This is only used when TARGET_UNWIND_INFO is set.
7882 @node Exception Region Output
7883 @subsection Assembler Commands for Exception Regions
7885 @c prevent bad page break with this line
7887 This describes commands marking the start and the end of an exception
7890 @defmac EH_FRAME_SECTION_NAME
7891 If defined, a C string constant for the name of the section containing
7892 exception handling frame unwind information. If not defined, GCC will
7893 provide a default definition if the target supports named sections.
7894 @file{crtstuff.c} uses this macro to switch to the appropriate section.
7896 You should define this symbol if your target supports DWARF 2 frame
7897 unwind information and the default definition does not work.
7900 @defmac EH_FRAME_IN_DATA_SECTION
7901 If defined, DWARF 2 frame unwind information will be placed in the
7902 data section even though the target supports named sections. This
7903 might be necessary, for instance, if the system linker does garbage
7904 collection and sections cannot be marked as not to be collected.
7906 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
7910 @defmac EH_TABLES_CAN_BE_READ_ONLY
7911 Define this macro to 1 if your target is such that no frame unwind
7912 information encoding used with non-PIC code will ever require a
7913 runtime relocation, but the linker may not support merging read-only
7914 and read-write sections into a single read-write section.
7917 @defmac MASK_RETURN_ADDR
7918 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
7919 that it does not contain any extraneous set bits in it.
7922 @defmac DWARF2_UNWIND_INFO
7923 Define this macro to 0 if your target supports DWARF 2 frame unwind
7924 information, but it does not yet work with exception handling.
7925 Otherwise, if your target supports this information (if it defines
7926 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
7927 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of
7930 If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder
7931 will be used in all cases. Defining this macro will enable the generation
7932 of DWARF 2 frame debugging information.
7934 If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1,
7935 the DWARF 2 unwinder will be the default exception handling mechanism;
7936 otherwise, @code{setjmp}/@code{longjmp} will be used by default.
7939 @defmac TARGET_UNWIND_INFO
7940 Define this macro if your target has ABI specified unwind tables. Usually
7941 these will be output by @code{TARGET_UNWIND_EMIT}.
7944 @deftypevar {Target Hook} bool TARGET_UNWID_TABLES_DEFAULT
7945 This variable should be set to @code{true} if the target ABI requires unwinding
7946 tables even when exceptions are not used.
7949 @defmac MUST_USE_SJLJ_EXCEPTIONS
7950 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
7951 runtime-variable. In that case, @file{except.h} cannot correctly
7952 determine the corresponding definition of
7953 @code{MUST_USE_SJLJ_EXCEPTIONS}, so the target must provide it directly.
7956 @defmac DWARF_CIE_DATA_ALIGNMENT
7957 This macro need only be defined if the target might save registers in the
7958 function prologue at an offset to the stack pointer that is not aligned to
7959 @code{UNITS_PER_WORD}. The definition should be the negative minimum
7960 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
7961 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
7962 the target supports DWARF 2 frame unwind information.
7965 @deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
7966 Contains the value true if the target should add a zero word onto the
7967 end of a Dwarf-2 frame info section when used for exception handling.
7968 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
7972 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
7973 Given a register, this hook should return a parallel of registers to
7974 represent where to find the register pieces. Define this hook if the
7975 register and its mode are represented in Dwarf in non-contiguous
7976 locations, or if the register should be represented in more than one
7977 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
7978 If not defined, the default is to return @code{NULL_RTX}.
7981 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
7982 This hook is used to output a reference from a frame unwinding table to
7983 the type_info object identified by @var{sym}. It should return @code{true}
7984 if the reference was output. Returning @code{false} will cause the
7985 reference to be output using the normal Dwarf2 routines.
7988 @deftypefn {Target Hook} bool TARGET_ARM_EABI_UNWINDER
7989 This hook should be set to @code{true} on targets that use an ARM EABI
7990 based unwinding library, and @code{false} on other targets. This effects
7991 the format of unwinding tables, and how the unwinder in entered after
7992 running a cleanup. The default is @code{false}.
7995 @node Alignment Output
7996 @subsection Assembler Commands for Alignment
7998 @c prevent bad page break with this line
7999 This describes commands for alignment.
8001 @defmac JUMP_ALIGN (@var{label})
8002 The alignment (log base 2) to put in front of @var{label}, which is
8003 a common destination of jumps and has no fallthru incoming edge.
8005 This macro need not be defined if you don't want any special alignment
8006 to be done at such a time. Most machine descriptions do not currently
8009 Unless it's necessary to inspect the @var{label} parameter, it is better
8010 to set the variable @var{align_jumps} in the target's
8011 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8012 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8015 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8016 The alignment (log base 2) to put in front of @var{label}, which follows
8019 This macro need not be defined if you don't want any special alignment
8020 to be done at such a time. Most machine descriptions do not currently
8024 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8025 The maximum number of bytes to skip when applying
8026 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8027 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8030 @defmac LOOP_ALIGN (@var{label})
8031 The alignment (log base 2) to put in front of @var{label}, which follows
8032 a @code{NOTE_INSN_LOOP_BEG} note.
8034 This macro need not be defined if you don't want any special alignment
8035 to be done at such a time. Most machine descriptions do not currently
8038 Unless it's necessary to inspect the @var{label} parameter, it is better
8039 to set the variable @code{align_loops} in the target's
8040 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8041 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8044 @defmac LOOP_ALIGN_MAX_SKIP
8045 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
8046 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8049 @defmac LABEL_ALIGN (@var{label})
8050 The alignment (log base 2) to put in front of @var{label}.
8051 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8052 the maximum of the specified values is used.
8054 Unless it's necessary to inspect the @var{label} parameter, it is better
8055 to set the variable @code{align_labels} in the target's
8056 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8057 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8060 @defmac LABEL_ALIGN_MAX_SKIP
8061 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
8062 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8065 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8066 A C statement to output to the stdio stream @var{stream} an assembler
8067 instruction to advance the location counter by @var{nbytes} bytes.
8068 Those bytes should be zero when loaded. @var{nbytes} will be a C
8069 expression of type @code{int}.
8072 @defmac ASM_NO_SKIP_IN_TEXT
8073 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
8074 text section because it fails to put zeros in the bytes that are skipped.
8075 This is true on many Unix systems, where the pseudo--op to skip bytes
8076 produces no-op instructions rather than zeros when used in the text
8080 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
8081 A C statement to output to the stdio stream @var{stream} an assembler
8082 command to advance the location counter to a multiple of 2 to the
8083 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
8086 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
8087 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
8088 for padding, if necessary.
8091 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
8092 A C statement to output to the stdio stream @var{stream} an assembler
8093 command to advance the location counter to a multiple of 2 to the
8094 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
8095 satisfy the alignment request. @var{power} and @var{max_skip} will be
8096 a C expression of type @code{int}.
8100 @node Debugging Info
8101 @section Controlling Debugging Information Format
8103 @c prevent bad page break with this line
8104 This describes how to specify debugging information.
8107 * All Debuggers:: Macros that affect all debugging formats uniformly.
8108 * DBX Options:: Macros enabling specific options in DBX format.
8109 * DBX Hooks:: Hook macros for varying DBX format.
8110 * File Names and DBX:: Macros controlling output of file names in DBX format.
8111 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
8112 * VMS Debug:: Macros for VMS debug format.
8116 @subsection Macros Affecting All Debugging Formats
8118 @c prevent bad page break with this line
8119 These macros affect all debugging formats.
8121 @defmac DBX_REGISTER_NUMBER (@var{regno})
8122 A C expression that returns the DBX register number for the compiler
8123 register number @var{regno}. In the default macro provided, the value
8124 of this expression will be @var{regno} itself. But sometimes there are
8125 some registers that the compiler knows about and DBX does not, or vice
8126 versa. In such cases, some register may need to have one number in the
8127 compiler and another for DBX@.
8129 If two registers have consecutive numbers inside GCC, and they can be
8130 used as a pair to hold a multiword value, then they @emph{must} have
8131 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
8132 Otherwise, debuggers will be unable to access such a pair, because they
8133 expect register pairs to be consecutive in their own numbering scheme.
8135 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
8136 does not preserve register pairs, then what you must do instead is
8137 redefine the actual register numbering scheme.
8140 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
8141 A C expression that returns the integer offset value for an automatic
8142 variable having address @var{x} (an RTL expression). The default
8143 computation assumes that @var{x} is based on the frame-pointer and
8144 gives the offset from the frame-pointer. This is required for targets
8145 that produce debugging output for DBX or COFF-style debugging output
8146 for SDB and allow the frame-pointer to be eliminated when the
8147 @option{-g} options is used.
8150 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
8151 A C expression that returns the integer offset value for an argument
8152 having address @var{x} (an RTL expression). The nominal offset is
8156 @defmac PREFERRED_DEBUGGING_TYPE
8157 A C expression that returns the type of debugging output GCC should
8158 produce when the user specifies just @option{-g}. Define
8159 this if you have arranged for GCC to support more than one format of
8160 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
8161 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
8162 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
8164 When the user specifies @option{-ggdb}, GCC normally also uses the
8165 value of this macro to select the debugging output format, but with two
8166 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
8167 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
8168 defined, GCC uses @code{DBX_DEBUG}.
8170 The value of this macro only affects the default debugging output; the
8171 user can always get a specific type of output by using @option{-gstabs},
8172 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
8176 @subsection Specific Options for DBX Output
8178 @c prevent bad page break with this line
8179 These are specific options for DBX output.
8181 @defmac DBX_DEBUGGING_INFO
8182 Define this macro if GCC should produce debugging output for DBX
8183 in response to the @option{-g} option.
8186 @defmac XCOFF_DEBUGGING_INFO
8187 Define this macro if GCC should produce XCOFF format debugging output
8188 in response to the @option{-g} option. This is a variant of DBX format.
8191 @defmac DEFAULT_GDB_EXTENSIONS
8192 Define this macro to control whether GCC should by default generate
8193 GDB's extended version of DBX debugging information (assuming DBX-format
8194 debugging information is enabled at all). If you don't define the
8195 macro, the default is 1: always generate the extended information
8196 if there is any occasion to.
8199 @defmac DEBUG_SYMS_TEXT
8200 Define this macro if all @code{.stabs} commands should be output while
8201 in the text section.
8204 @defmac ASM_STABS_OP
8205 A C string constant, including spacing, naming the assembler pseudo op to
8206 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
8207 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
8208 applies only to DBX debugging information format.
8211 @defmac ASM_STABD_OP
8212 A C string constant, including spacing, naming the assembler pseudo op to
8213 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
8214 value is the current location. If you don't define this macro,
8215 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
8219 @defmac ASM_STABN_OP
8220 A C string constant, including spacing, naming the assembler pseudo op to
8221 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
8222 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
8223 macro applies only to DBX debugging information format.
8226 @defmac DBX_NO_XREFS
8227 Define this macro if DBX on your system does not support the construct
8228 @samp{xs@var{tagname}}. On some systems, this construct is used to
8229 describe a forward reference to a structure named @var{tagname}.
8230 On other systems, this construct is not supported at all.
8233 @defmac DBX_CONTIN_LENGTH
8234 A symbol name in DBX-format debugging information is normally
8235 continued (split into two separate @code{.stabs} directives) when it
8236 exceeds a certain length (by default, 80 characters). On some
8237 operating systems, DBX requires this splitting; on others, splitting
8238 must not be done. You can inhibit splitting by defining this macro
8239 with the value zero. You can override the default splitting-length by
8240 defining this macro as an expression for the length you desire.
8243 @defmac DBX_CONTIN_CHAR
8244 Normally continuation is indicated by adding a @samp{\} character to
8245 the end of a @code{.stabs} string when a continuation follows. To use
8246 a different character instead, define this macro as a character
8247 constant for the character you want to use. Do not define this macro
8248 if backslash is correct for your system.
8251 @defmac DBX_STATIC_STAB_DATA_SECTION
8252 Define this macro if it is necessary to go to the data section before
8253 outputting the @samp{.stabs} pseudo-op for a non-global static
8257 @defmac DBX_TYPE_DECL_STABS_CODE
8258 The value to use in the ``code'' field of the @code{.stabs} directive
8259 for a typedef. The default is @code{N_LSYM}.
8262 @defmac DBX_STATIC_CONST_VAR_CODE
8263 The value to use in the ``code'' field of the @code{.stabs} directive
8264 for a static variable located in the text section. DBX format does not
8265 provide any ``right'' way to do this. The default is @code{N_FUN}.
8268 @defmac DBX_REGPARM_STABS_CODE
8269 The value to use in the ``code'' field of the @code{.stabs} directive
8270 for a parameter passed in registers. DBX format does not provide any
8271 ``right'' way to do this. The default is @code{N_RSYM}.
8274 @defmac DBX_REGPARM_STABS_LETTER
8275 The letter to use in DBX symbol data to identify a symbol as a parameter
8276 passed in registers. DBX format does not customarily provide any way to
8277 do this. The default is @code{'P'}.
8280 @defmac DBX_FUNCTION_FIRST
8281 Define this macro if the DBX information for a function and its
8282 arguments should precede the assembler code for the function. Normally,
8283 in DBX format, the debugging information entirely follows the assembler
8287 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
8288 Define this macro, with value 1, if the value of a symbol describing
8289 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
8290 relative to the start of the enclosing function. Normally, GCC uses
8291 an absolute address.
8294 @defmac DBX_LINES_FUNCTION_RELATIVE
8295 Define this macro, with value 1, if the value of a symbol indicating
8296 the current line number (@code{N_SLINE}) should be relative to the
8297 start of the enclosing function. Normally, GCC uses an absolute address.
8300 @defmac DBX_USE_BINCL
8301 Define this macro if GCC should generate @code{N_BINCL} and
8302 @code{N_EINCL} stabs for included header files, as on Sun systems. This
8303 macro also directs GCC to output a type number as a pair of a file
8304 number and a type number within the file. Normally, GCC does not
8305 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
8306 number for a type number.
8310 @subsection Open-Ended Hooks for DBX Format
8312 @c prevent bad page break with this line
8313 These are hooks for DBX format.
8315 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
8316 Define this macro to say how to output to @var{stream} the debugging
8317 information for the start of a scope level for variable names. The
8318 argument @var{name} is the name of an assembler symbol (for use with
8319 @code{assemble_name}) whose value is the address where the scope begins.
8322 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
8323 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
8326 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
8327 Define this macro if the target machine requires special handling to
8328 output an @code{N_FUN} entry for the function @var{decl}.
8331 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
8332 A C statement to output DBX debugging information before code for line
8333 number @var{line} of the current source file to the stdio stream
8334 @var{stream}. @var{counter} is the number of time the macro was
8335 invoked, including the current invocation; it is intended to generate
8336 unique labels in the assembly output.
8338 This macro should not be defined if the default output is correct, or
8339 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
8342 @defmac NO_DBX_FUNCTION_END
8343 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
8344 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
8345 On those machines, define this macro to turn this feature off without
8346 disturbing the rest of the gdb extensions.
8349 @defmac NO_DBX_BNSYM_ENSYM
8350 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
8351 extension construct. On those machines, define this macro to turn this
8352 feature off without disturbing the rest of the gdb extensions.
8355 @node File Names and DBX
8356 @subsection File Names in DBX Format
8358 @c prevent bad page break with this line
8359 This describes file names in DBX format.
8361 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
8362 A C statement to output DBX debugging information to the stdio stream
8363 @var{stream}, which indicates that file @var{name} is the main source
8364 file---the file specified as the input file for compilation.
8365 This macro is called only once, at the beginning of compilation.
8367 This macro need not be defined if the standard form of output
8368 for DBX debugging information is appropriate.
8370 It may be necessary to refer to a label equal to the beginning of the
8371 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
8372 to do so. If you do this, you must also set the variable
8373 @var{used_ltext_label_name} to @code{true}.
8376 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
8377 Define this macro, with value 1, if GCC should not emit an indication
8378 of the current directory for compilation and current source language at
8379 the beginning of the file.
8382 @defmac NO_DBX_GCC_MARKER
8383 Define this macro, with value 1, if GCC should not emit an indication
8384 that this object file was compiled by GCC@. The default is to emit
8385 an @code{N_OPT} stab at the beginning of every source file, with
8386 @samp{gcc2_compiled.} for the string and value 0.
8389 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
8390 A C statement to output DBX debugging information at the end of
8391 compilation of the main source file @var{name}. Output should be
8392 written to the stdio stream @var{stream}.
8394 If you don't define this macro, nothing special is output at the end
8395 of compilation, which is correct for most machines.
8398 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
8399 Define this macro @emph{instead of} defining
8400 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
8401 the end of compilation is a @code{N_SO} stab with an empty string,
8402 whose value is the highest absolute text address in the file.
8407 @subsection Macros for SDB and DWARF Output
8409 @c prevent bad page break with this line
8410 Here are macros for SDB and DWARF output.
8412 @defmac SDB_DEBUGGING_INFO
8413 Define this macro if GCC should produce COFF-style debugging output
8414 for SDB in response to the @option{-g} option.
8417 @defmac DWARF2_DEBUGGING_INFO
8418 Define this macro if GCC should produce dwarf version 2 format
8419 debugging output in response to the @option{-g} option.
8421 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (tree @var{function})
8422 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
8423 be emitted for each function. Instead of an integer return the enum
8424 value for the @code{DW_CC_} tag.
8427 To support optional call frame debugging information, you must also
8428 define @code{INCOMING_RETURN_ADDR_RTX} and either set
8429 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
8430 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
8431 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
8434 @defmac DWARF2_FRAME_INFO
8435 Define this macro to a nonzero value if GCC should always output
8436 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
8437 (@pxref{Exception Region Output} is nonzero, GCC will output this
8438 information not matter how you define @code{DWARF2_FRAME_INFO}.
8441 @defmac DWARF2_ASM_LINE_DEBUG_INFO
8442 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
8443 line debug info sections. This will result in much more compact line number
8444 tables, and hence is desirable if it works.
8447 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
8448 A C statement to issue assembly directives that create a difference
8449 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
8452 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
8453 A C statement to issue assembly directives that create a
8454 section-relative reference to the given @var{label}, using an integer of the
8455 given @var{size}. The label is known to be defined in the given @var{section}.
8458 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
8459 A C statement to issue assembly directives that create a self-relative
8460 reference to the given @var{label}, using an integer of the given @var{size}.
8463 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{FILE}, int @var{size}, rtx @var{x})
8464 If defined, this target hook is a function which outputs a DTP-relative
8465 reference to the given TLS symbol of the specified size.
8468 @defmac PUT_SDB_@dots{}
8469 Define these macros to override the assembler syntax for the special
8470 SDB assembler directives. See @file{sdbout.c} for a list of these
8471 macros and their arguments. If the standard syntax is used, you need
8472 not define them yourself.
8476 Some assemblers do not support a semicolon as a delimiter, even between
8477 SDB assembler directives. In that case, define this macro to be the
8478 delimiter to use (usually @samp{\n}). It is not necessary to define
8479 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
8483 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
8484 Define this macro to allow references to unknown structure,
8485 union, or enumeration tags to be emitted. Standard COFF does not
8486 allow handling of unknown references, MIPS ECOFF has support for
8490 @defmac SDB_ALLOW_FORWARD_REFERENCES
8491 Define this macro to allow references to structure, union, or
8492 enumeration tags that have not yet been seen to be handled. Some
8493 assemblers choke if forward tags are used, while some require it.
8496 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
8497 A C statement to output SDB debugging information before code for line
8498 number @var{line} of the current source file to the stdio stream
8499 @var{stream}. The default is to emit an @code{.ln} directive.
8504 @subsection Macros for VMS Debug Format
8506 @c prevent bad page break with this line
8507 Here are macros for VMS debug format.
8509 @defmac VMS_DEBUGGING_INFO
8510 Define this macro if GCC should produce debugging output for VMS
8511 in response to the @option{-g} option. The default behavior for VMS
8512 is to generate minimal debug info for a traceback in the absence of
8513 @option{-g} unless explicitly overridden with @option{-g0}. This
8514 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
8515 @code{OVERRIDE_OPTIONS}.
8518 @node Floating Point
8519 @section Cross Compilation and Floating Point
8520 @cindex cross compilation and floating point
8521 @cindex floating point and cross compilation
8523 While all modern machines use twos-complement representation for integers,
8524 there are a variety of representations for floating point numbers. This
8525 means that in a cross-compiler the representation of floating point numbers
8526 in the compiled program may be different from that used in the machine
8527 doing the compilation.
8529 Because different representation systems may offer different amounts of
8530 range and precision, all floating point constants must be represented in
8531 the target machine's format. Therefore, the cross compiler cannot
8532 safely use the host machine's floating point arithmetic; it must emulate
8533 the target's arithmetic. To ensure consistency, GCC always uses
8534 emulation to work with floating point values, even when the host and
8535 target floating point formats are identical.
8537 The following macros are provided by @file{real.h} for the compiler to
8538 use. All parts of the compiler which generate or optimize
8539 floating-point calculations must use these macros. They may evaluate
8540 their operands more than once, so operands must not have side effects.
8542 @defmac REAL_VALUE_TYPE
8543 The C data type to be used to hold a floating point value in the target
8544 machine's format. Typically this is a @code{struct} containing an
8545 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
8549 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8550 Compares for equality the two values, @var{x} and @var{y}. If the target
8551 floating point format supports negative zeroes and/or NaNs,
8552 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
8553 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
8556 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8557 Tests whether @var{x} is less than @var{y}.
8560 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
8561 Truncates @var{x} to a signed integer, rounding toward zero.
8564 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
8565 Truncates @var{x} to an unsigned integer, rounding toward zero. If
8566 @var{x} is negative, returns zero.
8569 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
8570 Converts @var{string} into a floating point number in the target machine's
8571 representation for mode @var{mode}. This routine can handle both
8572 decimal and hexadecimal floating point constants, using the syntax
8573 defined by the C language for both.
8576 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
8577 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
8580 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
8581 Determines whether @var{x} represents infinity (positive or negative).
8584 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
8585 Determines whether @var{x} represents a ``NaN'' (not-a-number).
8588 @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})
8589 Calculates an arithmetic operation on the two floating point values
8590 @var{x} and @var{y}, storing the result in @var{output} (which must be a
8593 The operation to be performed is specified by @var{code}. Only the
8594 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
8595 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
8597 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
8598 target's floating point format cannot represent infinity, it will call
8599 @code{abort}. Callers should check for this situation first, using
8600 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
8603 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
8604 Returns the negative of the floating point value @var{x}.
8607 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
8608 Returns the absolute value of @var{x}.
8611 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
8612 Truncates the floating point value @var{x} to fit in @var{mode}. The
8613 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
8614 appropriate bit pattern to be output asa floating constant whose
8615 precision accords with mode @var{mode}.
8618 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
8619 Converts a floating point value @var{x} into a double-precision integer
8620 which is then stored into @var{low} and @var{high}. If the value is not
8621 integral, it is truncated.
8624 @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})
8625 Converts a double-precision integer found in @var{low} and @var{high},
8626 into a floating point value which is then stored into @var{x}. The
8627 value is truncated to fit in mode @var{mode}.
8630 @node Mode Switching
8631 @section Mode Switching Instructions
8632 @cindex mode switching
8633 The following macros control mode switching optimizations:
8635 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
8636 Define this macro if the port needs extra instructions inserted for mode
8637 switching in an optimizing compilation.
8639 For an example, the SH4 can perform both single and double precision
8640 floating point operations, but to perform a single precision operation,
8641 the FPSCR PR bit has to be cleared, while for a double precision
8642 operation, this bit has to be set. Changing the PR bit requires a general
8643 purpose register as a scratch register, hence these FPSCR sets have to
8644 be inserted before reload, i.e.@: you can't put this into instruction emitting
8645 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
8647 You can have multiple entities that are mode-switched, and select at run time
8648 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
8649 return nonzero for any @var{entity} that needs mode-switching.
8650 If you define this macro, you also have to define
8651 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
8652 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
8653 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
8657 @defmac NUM_MODES_FOR_MODE_SWITCHING
8658 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
8659 initializer for an array of integers. Each initializer element
8660 N refers to an entity that needs mode switching, and specifies the number
8661 of different modes that might need to be set for this entity.
8662 The position of the initializer in the initializer---starting counting at
8663 zero---determines the integer that is used to refer to the mode-switched
8665 In macros that take mode arguments / yield a mode result, modes are
8666 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
8667 switch is needed / supplied.
8670 @defmac MODE_NEEDED (@var{entity}, @var{insn})
8671 @var{entity} is an integer specifying a mode-switched entity. If
8672 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
8673 return an integer value not larger than the corresponding element in
8674 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
8675 be switched into prior to the execution of @var{insn}.
8678 @defmac MODE_AFTER (@var{mode}, @var{insn})
8679 If this macro is defined, it is evaluated for every @var{insn} during
8680 mode switching. It determines the mode that an insn results in (if
8681 different from the incoming mode).
8684 @defmac MODE_ENTRY (@var{entity})
8685 If this macro is defined, it is evaluated for every @var{entity} that needs
8686 mode switching. It should evaluate to an integer, which is a mode that
8687 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
8688 is defined then @code{MODE_EXIT} must be defined.
8691 @defmac MODE_EXIT (@var{entity})
8692 If this macro is defined, it is evaluated for every @var{entity} that needs
8693 mode switching. It should evaluate to an integer, which is a mode that
8694 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
8695 is defined then @code{MODE_ENTRY} must be defined.
8698 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
8699 This macro specifies the order in which modes for @var{entity} are processed.
8700 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
8701 lowest. The value of the macro should be an integer designating a mode
8702 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
8703 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
8704 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
8707 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
8708 Generate one or more insns to set @var{entity} to @var{mode}.
8709 @var{hard_reg_live} is the set of hard registers live at the point where
8710 the insn(s) are to be inserted.
8713 @node Target Attributes
8714 @section Defining target-specific uses of @code{__attribute__}
8715 @cindex target attributes
8716 @cindex machine attributes
8717 @cindex attributes, target-specific
8719 Target-specific attributes may be defined for functions, data and types.
8720 These are described using the following target hooks; they also need to
8721 be documented in @file{extend.texi}.
8723 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
8724 If defined, this target hook points to an array of @samp{struct
8725 attribute_spec} (defined in @file{tree.h}) specifying the machine
8726 specific attributes for this target and some of the restrictions on the
8727 entities to which these attributes are applied and the arguments they
8731 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8732 If defined, this target hook is a function which returns zero if the attributes on
8733 @var{type1} and @var{type2} are incompatible, one if they are compatible,
8734 and two if they are nearly compatible (which causes a warning to be
8735 generated). If this is not defined, machine-specific attributes are
8736 supposed always to be compatible.
8739 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
8740 If defined, this target hook is a function which assigns default attributes to
8741 newly defined @var{type}.
8744 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8745 Define this target hook if the merging of type attributes needs special
8746 handling. If defined, the result is a list of the combined
8747 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
8748 that @code{comptypes} has already been called and returned 1. This
8749 function may call @code{merge_attributes} to handle machine-independent
8753 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
8754 Define this target hook if the merging of decl attributes needs special
8755 handling. If defined, the result is a list of the combined
8756 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
8757 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
8758 when this is needed are when one attribute overrides another, or when an
8759 attribute is nullified by a subsequent definition. This function may
8760 call @code{merge_attributes} to handle machine-independent merging.
8762 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
8763 If the only target-specific handling you require is @samp{dllimport}
8764 for Microsoft Windows targets, you should define the macro
8765 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
8766 will then define a function called
8767 @code{merge_dllimport_decl_attributes} which can then be defined as
8768 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
8769 add @code{handle_dll_attribute} in the attribute table for your port
8770 to perform initial processing of the @samp{dllimport} and
8771 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
8772 @file{i386/i386.c}, for example.
8775 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree @var{decl})
8776 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
8777 specified. Use this hook if the target needs to add extra validation
8778 checks to @code{handle_dll_attribute}.
8781 @defmac TARGET_DECLSPEC
8782 Define this macro to a nonzero value if you want to treat
8783 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
8784 default, this behavior is enabled only for targets that define
8785 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
8786 of @code{__declspec} is via a built-in macro, but you should not rely
8787 on this implementation detail.
8790 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
8791 Define this target hook if you want to be able to add attributes to a decl
8792 when it is being created. This is normally useful for back ends which
8793 wish to implement a pragma by using the attributes which correspond to
8794 the pragma's effect. The @var{node} argument is the decl which is being
8795 created. The @var{attr_ptr} argument is a pointer to the attribute list
8796 for this decl. The list itself should not be modified, since it may be
8797 shared with other decls, but attributes may be chained on the head of
8798 the list and @code{*@var{attr_ptr}} modified to point to the new
8799 attributes, or a copy of the list may be made if further changes are
8803 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl})
8805 This target hook returns @code{true} if it is ok to inline @var{fndecl}
8806 into the current function, despite its having target-specific
8807 attributes, @code{false} otherwise. By default, if a function has a
8808 target specific attribute attached to it, it will not be inlined.
8811 @node MIPS Coprocessors
8812 @section Defining coprocessor specifics for MIPS targets.
8813 @cindex MIPS coprocessor-definition macros
8815 The MIPS specification allows MIPS implementations to have as many as 4
8816 coprocessors, each with as many as 32 private registers. GCC supports
8817 accessing these registers and transferring values between the registers
8818 and memory using asm-ized variables. For example:
8821 register unsigned int cp0count asm ("c0r1");
8827 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
8828 names may be added as described below, or the default names may be
8829 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
8831 Coprocessor registers are assumed to be epilogue-used; sets to them will
8832 be preserved even if it does not appear that the register is used again
8833 later in the function.
8835 Another note: according to the MIPS spec, coprocessor 1 (if present) is
8836 the FPU@. One accesses COP1 registers through standard mips
8837 floating-point support; they are not included in this mechanism.
8839 There is one macro used in defining the MIPS coprocessor interface which
8840 you may want to override in subtargets; it is described below.
8842 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
8843 A comma-separated list (with leading comma) of pairs describing the
8844 alternate names of coprocessor registers. The format of each entry should be
8846 @{ @var{alternatename}, @var{register_number}@}
8852 @section Parameters for Precompiled Header Validity Checking
8853 @cindex parameters, precompiled headers
8855 @deftypefn {Target Hook} void *TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
8856 This hook returns the data needed by @code{TARGET_PCH_VALID_P} and sets
8857 @samp{*@var{sz}} to the size of the data in bytes.
8860 @deftypefn {Target Hook} const char *TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
8861 This hook checks whether the options used to create a PCH file are
8862 compatible with the current settings. It returns @code{NULL}
8863 if so and a suitable error message if not. Error messages will
8864 be presented to the user and must be localized using @samp{_(@var{msg})}.
8866 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
8867 when the PCH file was created and @var{sz} is the size of that data in bytes.
8868 It's safe to assume that the data was created by the same version of the
8869 compiler, so no format checking is needed.
8871 The default definition of @code{default_pch_valid_p} should be
8872 suitable for most targets.
8875 @deftypefn {Target Hook} const char *TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
8876 If this hook is nonnull, the default implementation of
8877 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
8878 of @code{target_flags}. @var{pch_flags} specifies the value that
8879 @code{target_flags} had when the PCH file was created. The return
8880 value is the same as for @code{TARGET_PCH_VALID_P}.
8884 @section C++ ABI parameters
8885 @cindex parameters, c++ abi
8887 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
8888 Define this hook to override the integer type used for guard variables.
8889 These are used to implement one-time construction of static objects. The
8890 default is long_long_integer_type_node.
8893 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
8894 This hook determines how guard variables are used. It should return
8895 @code{false} (the default) if first byte should be used. A return value of
8896 @code{true} indicates the least significant bit should be used.
8899 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
8900 This hook returns the size of the cookie to use when allocating an array
8901 whose elements have the indicated @var{type}. Assumes that it is already
8902 known that a cookie is needed. The default is
8903 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
8904 IA64/Generic C++ ABI@.
8907 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
8908 This hook should return @code{true} if the element size should be stored in
8909 array cookies. The default is to return @code{false}.
8912 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
8913 If defined by a backend this hook allows the decision made to export
8914 class @var{type} to be overruled. Upon entry @var{import_export}
8915 will contain 1 if the class is going to be exported, @minus{}1 if it is going
8916 to be imported and 0 otherwise. This function should return the
8917 modified value and perform any other actions necessary to support the
8918 backend's targeted operating system.
8921 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
8922 This hook should return @code{true} if constructors and destructors return
8923 the address of the object created/destroyed. The default is to return
8927 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
8928 This hook returns true if the key method for a class (i.e., the method
8929 which, if defined in the current translation unit, causes the virtual
8930 table to be emitted) may be an inline function. Under the standard
8931 Itanium C++ ABI the key method may be an inline function so long as
8932 the function is not declared inline in the class definition. Under
8933 some variants of the ABI, an inline function can never be the key
8934 method. The default is to return @code{true}.
8937 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
8938 @var{decl} is a virtual table, virtual table table, typeinfo object,
8939 or other similar implicit class data object that will be emitted with
8940 external linkage in this translation unit. No ELF visibility has been
8941 explicitly specified. If the target needs to specify a visibility
8942 other than that of the containing class, use this hook to set
8943 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
8946 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
8947 This hook returns true (the default) if virtual tables and other
8948 similar implicit class data objects are always COMDAT if they have
8949 external linkage. If this hook returns false, then class data for
8950 classes whose virtual table will be emitted in only one translation
8951 unit will not be COMDAT.
8954 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
8955 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
8956 should be used to register static destructors when @option{-fuse-cxa-atexit}
8957 is in effect. The default is to return false to use @code{__cxa_atexit}.
8960 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
8961 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been
8962 defined. Use this hook to make adjustments to the class (eg, tweak
8963 visibility or perform any other required target modifications).
8967 @section Miscellaneous Parameters
8968 @cindex parameters, miscellaneous
8970 @c prevent bad page break with this line
8971 Here are several miscellaneous parameters.
8973 @defmac HAS_LONG_COND_BRANCH
8974 Define this boolean macro to indicate whether or not your architecture
8975 has conditional branches that can span all of memory. It is used in
8976 conjunction with an optimization that partitions hot and cold basic
8977 blocks into separate sections of the executable. If this macro is
8978 set to false, gcc will convert any conditional branches that attempt
8979 to cross between sections into unconditional branches or indirect jumps.
8982 @defmac HAS_LONG_UNCOND_BRANCH
8983 Define this boolean macro to indicate whether or not your architecture
8984 has unconditional branches that can span all of memory. It is used in
8985 conjunction with an optimization that partitions hot and cold basic
8986 blocks into separate sections of the executable. If this macro is
8987 set to false, gcc will convert any unconditional branches that attempt
8988 to cross between sections into indirect jumps.
8991 @defmac CASE_VECTOR_MODE
8992 An alias for a machine mode name. This is the machine mode that
8993 elements of a jump-table should have.
8996 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
8997 Optional: return the preferred mode for an @code{addr_diff_vec}
8998 when the minimum and maximum offset are known. If you define this,
8999 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
9000 To make this work, you also have to define @code{INSN_ALIGN} and
9001 make the alignment for @code{addr_diff_vec} explicit.
9002 The @var{body} argument is provided so that the offset_unsigned and scale
9003 flags can be updated.
9006 @defmac CASE_VECTOR_PC_RELATIVE
9007 Define this macro to be a C expression to indicate when jump-tables
9008 should contain relative addresses. You need not define this macro if
9009 jump-tables never contain relative addresses, or jump-tables should
9010 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
9014 @defmac CASE_VALUES_THRESHOLD
9015 Define this to be the smallest number of different values for which it
9016 is best to use a jump-table instead of a tree of conditional branches.
9017 The default is four for machines with a @code{casesi} instruction and
9018 five otherwise. This is best for most machines.
9021 @defmac CASE_USE_BIT_TESTS
9022 Define this macro to be a C expression to indicate whether C switch
9023 statements may be implemented by a sequence of bit tests. This is
9024 advantageous on processors that can efficiently implement left shift
9025 of 1 by the number of bits held in a register, but inappropriate on
9026 targets that would require a loop. By default, this macro returns
9027 @code{true} if the target defines an @code{ashlsi3} pattern, and
9028 @code{false} otherwise.
9031 @defmac WORD_REGISTER_OPERATIONS
9032 Define this macro if operations between registers with integral mode
9033 smaller than a word are always performed on the entire register.
9034 Most RISC machines have this property and most CISC machines do not.
9037 @defmac LOAD_EXTEND_OP (@var{mem_mode})
9038 Define this macro to be a C expression indicating when insns that read
9039 memory in @var{mem_mode}, an integral mode narrower than a word, set the
9040 bits outside of @var{mem_mode} to be either the sign-extension or the
9041 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
9042 of @var{mem_mode} for which the
9043 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
9044 @code{UNKNOWN} for other modes.
9046 This macro is not called with @var{mem_mode} non-integral or with a width
9047 greater than or equal to @code{BITS_PER_WORD}, so you may return any
9048 value in this case. Do not define this macro if it would always return
9049 @code{UNKNOWN}. On machines where this macro is defined, you will normally
9050 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
9052 You may return a non-@code{UNKNOWN} value even if for some hard registers
9053 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
9054 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
9055 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
9056 integral mode larger than this but not larger than @code{word_mode}.
9058 You must return @code{UNKNOWN} if for some hard registers that allow this
9059 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
9060 @code{word_mode}, but that they can change to another integral mode that
9061 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
9064 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
9065 Define this macro if loading short immediate values into registers sign
9069 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
9070 Define this macro if the same instructions that convert a floating
9071 point number to a signed fixed point number also convert validly to an
9075 @deftypefn {Target Hook} int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
9076 When @option{-ffast-math} is in effect, GCC tries to optimize
9077 divisions by the same divisor, by turning them into multiplications by
9078 the reciprocal. This target hook specifies the minimum number of divisions
9079 that should be there for GCC to perform the optimization for a variable
9080 of mode @var{mode}. The default implementation returns 3 if the machine
9081 has an instruction for the division, and 2 if it does not.
9085 The maximum number of bytes that a single instruction can move quickly
9086 between memory and registers or between two memory locations.
9089 @defmac MAX_MOVE_MAX
9090 The maximum number of bytes that a single instruction can move quickly
9091 between memory and registers or between two memory locations. If this
9092 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
9093 constant value that is the largest value that @code{MOVE_MAX} can have
9097 @defmac SHIFT_COUNT_TRUNCATED
9098 A C expression that is nonzero if on this machine the number of bits
9099 actually used for the count of a shift operation is equal to the number
9100 of bits needed to represent the size of the object being shifted. When
9101 this macro is nonzero, the compiler will assume that it is safe to omit
9102 a sign-extend, zero-extend, and certain bitwise `and' instructions that
9103 truncates the count of a shift operation. On machines that have
9104 instructions that act on bit-fields at variable positions, which may
9105 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
9106 also enables deletion of truncations of the values that serve as
9107 arguments to bit-field instructions.
9109 If both types of instructions truncate the count (for shifts) and
9110 position (for bit-field operations), or if no variable-position bit-field
9111 instructions exist, you should define this macro.
9113 However, on some machines, such as the 80386 and the 680x0, truncation
9114 only applies to shift operations and not the (real or pretended)
9115 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
9116 such machines. Instead, add patterns to the @file{md} file that include
9117 the implied truncation of the shift instructions.
9119 You need not define this macro if it would always have the value of zero.
9122 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
9123 @deftypefn {Target Hook} int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
9124 This function describes how the standard shift patterns for @var{mode}
9125 deal with shifts by negative amounts or by more than the width of the mode.
9126 @xref{shift patterns}.
9128 On many machines, the shift patterns will apply a mask @var{m} to the
9129 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
9130 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
9131 this is true for mode @var{mode}, the function should return @var{m},
9132 otherwise it should return 0. A return value of 0 indicates that no
9133 particular behavior is guaranteed.
9135 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
9136 @emph{not} apply to general shift rtxes; it applies only to instructions
9137 that are generated by the named shift patterns.
9139 The default implementation of this function returns
9140 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
9141 and 0 otherwise. This definition is always safe, but if
9142 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
9143 nevertheless truncate the shift count, you may get better code
9147 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9148 A C expression which is nonzero if on this machine it is safe to
9149 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9150 bits (where @var{outprec} is smaller than @var{inprec}) by merely
9151 operating on it as if it had only @var{outprec} bits.
9153 On many machines, this expression can be 1.
9155 @c rearranged this, removed the phrase "it is reported that". this was
9156 @c to fix an overfull hbox. --mew 10feb93
9157 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
9158 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
9159 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
9160 such cases may improve things.
9163 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode})
9164 The representation of an integral mode can be such that the values
9165 are always extended to a wider integral mode. Return
9166 @code{SIGN_EXTEND} if values of @var{mode} are represented in
9167 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
9168 otherwise. (Currently, none of the targets use zero-extended
9169 representation this way so unlike @code{LOAD_EXTEND_OP},
9170 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
9171 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
9172 @var{mode} to @var{mode_rep} so that @var{mode_rep} is not the next
9173 widest integral mode and currently we take advantage of this fact.)
9175 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
9176 value even if the extension is not performed on certain hard registers
9177 as long as for the @code{REGNO_REG_CLASS} of these hard registers
9178 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
9180 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
9181 describe two related properties. If you define
9182 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
9183 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
9186 In order to enforce the representation of @code{mode},
9187 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
9191 @defmac STORE_FLAG_VALUE
9192 A C expression describing the value returned by a comparison operator
9193 with an integral mode and stored by a store-flag instruction
9194 (@samp{s@var{cond}}) when the condition is true. This description must
9195 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
9196 comparison operators whose results have a @code{MODE_INT} mode.
9198 A value of 1 or @minus{}1 means that the instruction implementing the
9199 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
9200 and 0 when the comparison is false. Otherwise, the value indicates
9201 which bits of the result are guaranteed to be 1 when the comparison is
9202 true. This value is interpreted in the mode of the comparison
9203 operation, which is given by the mode of the first operand in the
9204 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
9205 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
9208 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
9209 generate code that depends only on the specified bits. It can also
9210 replace comparison operators with equivalent operations if they cause
9211 the required bits to be set, even if the remaining bits are undefined.
9212 For example, on a machine whose comparison operators return an
9213 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
9214 @samp{0x80000000}, saying that just the sign bit is relevant, the
9218 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
9225 (ashift:SI @var{x} (const_int @var{n}))
9229 where @var{n} is the appropriate shift count to move the bit being
9230 tested into the sign bit.
9232 There is no way to describe a machine that always sets the low-order bit
9233 for a true value, but does not guarantee the value of any other bits,
9234 but we do not know of any machine that has such an instruction. If you
9235 are trying to port GCC to such a machine, include an instruction to
9236 perform a logical-and of the result with 1 in the pattern for the
9237 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
9239 Often, a machine will have multiple instructions that obtain a value
9240 from a comparison (or the condition codes). Here are rules to guide the
9241 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
9246 Use the shortest sequence that yields a valid definition for
9247 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
9248 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
9249 comparison operators to do so because there may be opportunities to
9250 combine the normalization with other operations.
9253 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
9254 slightly preferred on machines with expensive jumps and 1 preferred on
9258 As a second choice, choose a value of @samp{0x80000001} if instructions
9259 exist that set both the sign and low-order bits but do not define the
9263 Otherwise, use a value of @samp{0x80000000}.
9266 Many machines can produce both the value chosen for
9267 @code{STORE_FLAG_VALUE} and its negation in the same number of
9268 instructions. On those machines, you should also define a pattern for
9269 those cases, e.g., one matching
9272 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
9275 Some machines can also perform @code{and} or @code{plus} operations on
9276 condition code values with less instructions than the corresponding
9277 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
9278 machines, define the appropriate patterns. Use the names @code{incscc}
9279 and @code{decscc}, respectively, for the patterns which perform
9280 @code{plus} or @code{minus} operations on condition code values. See
9281 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
9282 find such instruction sequences on other machines.
9284 If this macro is not defined, the default value, 1, is used. You need
9285 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9286 instructions, or if the value generated by these instructions is 1.
9289 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
9290 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
9291 returned when comparison operators with floating-point results are true.
9292 Define this macro on machines that have comparison operations that return
9293 floating-point values. If there are no such operations, do not define
9297 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
9298 A C expression that gives a rtx representing the nonzero true element
9299 for vector comparisons. The returned rtx should be valid for the inner
9300 mode of @var{mode} which is guaranteed to be a vector mode. Define
9301 this macro on machines that have vector comparison operations that
9302 return a vector result. If there are no such operations, do not define
9303 this macro. Typically, this macro is defined as @code{const1_rtx} or
9304 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
9305 the compiler optimizing such vector comparison operations for the
9309 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9310 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9311 A C expression that evaluates to true if the architecture defines a value
9312 for @code{clz} or @code{ctz} with a zero operand. If so, @var{value}
9313 should be set to this value. If this macro is not defined, the value of
9314 @code{clz} or @code{ctz} is assumed to be undefined.
9316 This macro must be defined if the target's expansion for @code{ffs}
9317 relies on a particular value to get correct results. Otherwise it
9318 is not necessary, though it may be used to optimize some corner cases.
9320 Note that regardless of this macro the ``definedness'' of @code{clz}
9321 and @code{ctz} at zero do @emph{not} extend to the builtin functions
9322 visible to the user. Thus one may be free to adjust the value at will
9323 to match the target expansion of these operations without fear of
9328 An alias for the machine mode for pointers. On most machines, define
9329 this to be the integer mode corresponding to the width of a hardware
9330 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
9331 On some machines you must define this to be one of the partial integer
9332 modes, such as @code{PSImode}.
9334 The width of @code{Pmode} must be at least as large as the value of
9335 @code{POINTER_SIZE}. If it is not equal, you must define the macro
9336 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
9340 @defmac FUNCTION_MODE
9341 An alias for the machine mode used for memory references to functions
9342 being called, in @code{call} RTL expressions. On most machines this
9343 should be @code{QImode}.
9346 @defmac STDC_0_IN_SYSTEM_HEADERS
9347 In normal operation, the preprocessor expands @code{__STDC__} to the
9348 constant 1, to signify that GCC conforms to ISO Standard C@. On some
9349 hosts, like Solaris, the system compiler uses a different convention,
9350 where @code{__STDC__} is normally 0, but is 1 if the user specifies
9351 strict conformance to the C Standard.
9353 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
9354 convention when processing system header files, but when processing user
9355 files @code{__STDC__} will always expand to 1.
9358 @defmac NO_IMPLICIT_EXTERN_C
9359 Define this macro if the system header files support C++ as well as C@.
9360 This macro inhibits the usual method of using system header files in
9361 C++, which is to pretend that the file's contents are enclosed in
9362 @samp{extern "C" @{@dots{}@}}.
9367 @defmac REGISTER_TARGET_PRAGMAS ()
9368 Define this macro if you want to implement any target-specific pragmas.
9369 If defined, it is a C expression which makes a series of calls to
9370 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
9371 for each pragma. The macro may also do any
9372 setup required for the pragmas.
9374 The primary reason to define this macro is to provide compatibility with
9375 other compilers for the same target. In general, we discourage
9376 definition of target-specific pragmas for GCC@.
9378 If the pragma can be implemented by attributes then you should consider
9379 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
9381 Preprocessor macros that appear on pragma lines are not expanded. All
9382 @samp{#pragma} directives that do not match any registered pragma are
9383 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
9386 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9387 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9389 Each call to @code{c_register_pragma} or
9390 @code{c_register_pragma_with_expansion} establishes one pragma. The
9391 @var{callback} routine will be called when the preprocessor encounters a
9395 #pragma [@var{space}] @var{name} @dots{}
9398 @var{space} is the case-sensitive namespace of the pragma, or
9399 @code{NULL} to put the pragma in the global namespace. The callback
9400 routine receives @var{pfile} as its first argument, which can be passed
9401 on to cpplib's functions if necessary. You can lex tokens after the
9402 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
9403 callback will be silently ignored. The end of the line is indicated by
9404 a token of type @code{CPP_EOF}. Macro expansion occurs on the
9405 arguments of pragmas registered with
9406 @code{c_register_pragma_with_expansion} but not on the arguments of
9407 pragmas registered with @code{c_register_pragma}.
9409 For an example use of this routine, see @file{c4x.h} and the callback
9410 routines defined in @file{c4x-c.c}.
9412 Note that the use of @code{pragma_lex} is specific to the C and C++
9413 compilers. It will not work in the Java or Fortran compilers, or any
9414 other language compilers for that matter. Thus if @code{pragma_lex} is going
9415 to be called from target-specific code, it must only be done so when
9416 building the C and C++ compilers. This can be done by defining the
9417 variables @code{c_target_objs} and @code{cxx_target_objs} in the
9418 target entry in the @file{config.gcc} file. These variables should name
9419 the target-specific, language-specific object file which contains the
9420 code that uses @code{pragma_lex}. Note it will also be necessary to add a
9421 rule to the makefile fragment pointed to by @code{tmake_file} that shows
9422 how to build this object file.
9427 @defmac HANDLE_SYSV_PRAGMA
9428 Define this macro (to a value of 1) if you want the System V style
9429 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
9430 [=<value>]} to be supported by gcc.
9432 The pack pragma specifies the maximum alignment (in bytes) of fields
9433 within a structure, in much the same way as the @samp{__aligned__} and
9434 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
9435 the behavior to the default.
9437 A subtlety for Microsoft Visual C/C++ style bit-field packing
9438 (e.g.@: -mms-bitfields) for targets that support it:
9439 When a bit-field is inserted into a packed record, the whole size
9440 of the underlying type is used by one or more same-size adjacent
9441 bit-fields (that is, if its long:3, 32 bits is used in the record,
9442 and any additional adjacent long bit-fields are packed into the same
9443 chunk of 32 bits. However, if the size changes, a new field of that
9446 If both MS bit-fields and @samp{__attribute__((packed))} are used,
9447 the latter will take precedence. If @samp{__attribute__((packed))} is
9448 used on a single field when MS bit-fields are in use, it will take
9449 precedence for that field, but the alignment of the rest of the structure
9450 may affect its placement.
9452 The weak pragma only works if @code{SUPPORTS_WEAK} and
9453 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
9454 of specifically named weak labels, optionally with a value.
9459 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
9460 Define this macro (to a value of 1) if you want to support the Win32
9461 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
9462 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
9463 alignment (in bytes) of fields within a structure, in much the same way as
9464 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
9465 pack value of zero resets the behavior to the default. Successive
9466 invocations of this pragma cause the previous values to be stacked, so
9467 that invocations of @samp{#pragma pack(pop)} will return to the previous
9471 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
9472 Define this macro, as well as
9473 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
9474 arguments of @samp{#pragma pack}.
9477 @defmac TARGET_DEFAULT_PACK_STRUCT
9478 If your target requires a structure packing default other than 0 (meaning
9479 the machine default), define this macro to the necessary value (in bytes).
9480 This must be a value that would also be valid to use with
9481 @samp{#pragma pack()} (that is, a small power of two).
9484 @defmac DOLLARS_IN_IDENTIFIERS
9485 Define this macro to control use of the character @samp{$} in
9486 identifier names for the C family of languages. 0 means @samp{$} is
9487 not allowed by default; 1 means it is allowed. 1 is the default;
9488 there is no need to define this macro in that case.
9491 @defmac NO_DOLLAR_IN_LABEL
9492 Define this macro if the assembler does not accept the character
9493 @samp{$} in label names. By default constructors and destructors in
9494 G++ have @samp{$} in the identifiers. If this macro is defined,
9495 @samp{.} is used instead.
9498 @defmac NO_DOT_IN_LABEL
9499 Define this macro if the assembler does not accept the character
9500 @samp{.} in label names. By default constructors and destructors in G++
9501 have names that use @samp{.}. If this macro is defined, these names
9502 are rewritten to avoid @samp{.}.
9505 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
9506 Define this macro as a C expression that is nonzero if it is safe for the
9507 delay slot scheduler to place instructions in the delay slot of @var{insn},
9508 even if they appear to use a resource set or clobbered in @var{insn}.
9509 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
9510 every @code{call_insn} has this behavior. On machines where some @code{insn}
9511 or @code{jump_insn} is really a function call and hence has this behavior,
9512 you should define this macro.
9514 You need not define this macro if it would always return zero.
9517 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
9518 Define this macro as a C expression that is nonzero if it is safe for the
9519 delay slot scheduler to place instructions in the delay slot of @var{insn},
9520 even if they appear to set or clobber a resource referenced in @var{insn}.
9521 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
9522 some @code{insn} or @code{jump_insn} is really a function call and its operands
9523 are registers whose use is actually in the subroutine it calls, you should
9524 define this macro. Doing so allows the delay slot scheduler to move
9525 instructions which copy arguments into the argument registers into the delay
9528 You need not define this macro if it would always return zero.
9531 @defmac MULTIPLE_SYMBOL_SPACES
9532 Define this macro as a C expression that is nonzero if, in some cases,
9533 global symbols from one translation unit may not be bound to undefined
9534 symbols in another translation unit without user intervention. For
9535 instance, under Microsoft Windows symbols must be explicitly imported
9536 from shared libraries (DLLs).
9538 You need not define this macro if it would always evaluate to zero.
9541 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
9542 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
9543 any hard regs the port wishes to automatically clobber for an asm.
9544 It should return the result of the last @code{tree_cons} used to add a
9545 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
9546 corresponding parameters to the asm and may be inspected to avoid
9547 clobbering a register that is an input or output of the asm. You can use
9548 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
9549 for overlap with regards to asm-declared registers.
9552 @defmac MATH_LIBRARY
9553 Define this macro as a C string constant for the linker argument to link
9554 in the system math library, or @samp{""} if the target does not have a
9555 separate math library.
9557 You need only define this macro if the default of @samp{"-lm"} is wrong.
9560 @defmac LIBRARY_PATH_ENV
9561 Define this macro as a C string constant for the environment variable that
9562 specifies where the linker should look for libraries.
9564 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
9568 @defmac TARGET_POSIX_IO
9569 Define this macro if the target supports the following POSIX@ file
9570 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
9571 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
9572 to use file locking when exiting a program, which avoids race conditions
9573 if the program has forked. It will also create directories at run-time
9574 for cross-profiling.
9577 @defmac MAX_CONDITIONAL_EXECUTE
9579 A C expression for the maximum number of instructions to execute via
9580 conditional execution instructions instead of a branch. A value of
9581 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
9582 1 if it does use cc0.
9585 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
9586 Used if the target needs to perform machine-dependent modifications on the
9587 conditionals used for turning basic blocks into conditionally executed code.
9588 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
9589 contains information about the currently processed blocks. @var{true_expr}
9590 and @var{false_expr} are the tests that are used for converting the
9591 then-block and the else-block, respectively. Set either @var{true_expr} or
9592 @var{false_expr} to a null pointer if the tests cannot be converted.
9595 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
9596 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
9597 if-statements into conditions combined by @code{and} and @code{or} operations.
9598 @var{bb} contains the basic block that contains the test that is currently
9599 being processed and about to be turned into a condition.
9602 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
9603 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
9604 be converted to conditional execution format. @var{ce_info} points to
9605 a data structure, @code{struct ce_if_block}, which contains information
9606 about the currently processed blocks.
9609 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
9610 A C expression to perform any final machine dependent modifications in
9611 converting code to conditional execution. The involved basic blocks
9612 can be found in the @code{struct ce_if_block} structure that is pointed
9613 to by @var{ce_info}.
9616 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
9617 A C expression to cancel any machine dependent modifications in
9618 converting code to conditional execution. The involved basic blocks
9619 can be found in the @code{struct ce_if_block} structure that is pointed
9620 to by @var{ce_info}.
9623 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
9624 A C expression to initialize any extra fields in a @code{struct ce_if_block}
9625 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
9628 @defmac IFCVT_EXTRA_FIELDS
9629 If defined, it should expand to a set of field declarations that will be
9630 added to the @code{struct ce_if_block} structure. These should be initialized
9631 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
9634 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG ()
9635 If non-null, this hook performs a target-specific pass over the
9636 instruction stream. The compiler will run it at all optimization levels,
9637 just before the point at which it normally does delayed-branch scheduling.
9639 The exact purpose of the hook varies from target to target. Some use
9640 it to do transformations that are necessary for correctness, such as
9641 laying out in-function constant pools or avoiding hardware hazards.
9642 Others use it as an opportunity to do some machine-dependent optimizations.
9644 You need not implement the hook if it has nothing to do. The default
9648 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS ()
9649 Define this hook if you have any machine-specific built-in functions
9650 that need to be defined. It should be a function that performs the
9653 Machine specific built-in functions can be useful to expand special machine
9654 instructions that would otherwise not normally be generated because
9655 they have no equivalent in the source language (for example, SIMD vector
9656 instructions or prefetch instructions).
9658 To create a built-in function, call the function
9659 @code{lang_hooks.builtin_function}
9660 which is defined by the language front end. You can use any type nodes set
9661 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
9662 only language front ends that use those two functions will call
9663 @samp{TARGET_INIT_BUILTINS}.
9666 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
9668 Expand a call to a machine specific built-in function that was set up by
9669 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
9670 function call; the result should go to @var{target} if that is
9671 convenient, and have mode @var{mode} if that is convenient.
9672 @var{subtarget} may be used as the target for computing one of
9673 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
9674 ignored. This function should return the result of the call to the
9678 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree @var{fndecl}, tree @var{arglist})
9680 Select a replacement for a machine specific built-in function that
9681 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
9682 @emph{before} regular type checking, and so allows the target to
9683 implement a crude form of function overloading. @var{fndecl} is the
9684 declaration of the built-in function. @var{arglist} is the list of
9685 arguments passed to the built-in function. The result is a
9686 complete expression that implements the operation, usually
9687 another @code{CALL_EXPR}.
9690 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore})
9692 Fold a call to a machine specific built-in function that was set up by
9693 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
9694 built-in function. @var{arglist} is the list of arguments passed to
9695 the built-in function. The result is another tree containing a
9696 simplified expression for the call's result. If @var{ignore} is true
9697 the value will be ignored.
9700 @deftypefn {Target Hook} const char * TARGET_INVALID_WITHIN_DOLOOP (rtx @var{insn})
9702 Take an instruction in @var{insn} and return NULL if it is valid within a
9703 low-overhead loop, otherwise return a string why doloop could not be applied.
9705 Many targets use special registers for low-overhead looping. For any
9706 instruction that clobbers these this function should return a string indicating
9707 the reason why the doloop could not be applied.
9708 By default, the RTL loop optimizer does not use a present doloop pattern for
9709 loops containing function calls or branch on table instructions.
9712 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
9714 Take a branch insn in @var{branch1} and another in @var{branch2}.
9715 Return true if redirecting @var{branch1} to the destination of
9716 @var{branch2} is possible.
9718 On some targets, branches may have a limited range. Optimizing the
9719 filling of delay slots can result in branches being redirected, and this
9720 may in turn cause a branch offset to overflow.
9723 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (rtx @var{x}, @var{outer_code})
9724 This target hook returns @code{true} if @var{x} is considered to be commutative.
9725 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
9726 PLUS to be commutative inside a MEM. @var{outer_code} is the rtx code
9727 of the enclosing rtl, if known, otherwise it is UNKNOWN.
9730 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
9732 When the initial value of a hard register has been copied in a pseudo
9733 register, it is often not necessary to actually allocate another register
9734 to this pseudo register, because the original hard register or a stack slot
9735 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
9736 is called at the start of register allocation once for each hard register
9737 that had its initial value copied by using
9738 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
9739 Possible values are @code{NULL_RTX}, if you don't want
9740 to do any special allocation, a @code{REG} rtx---that would typically be
9741 the hard register itself, if it is known not to be clobbered---or a
9743 If you are returning a @code{MEM}, this is only a hint for the allocator;
9744 it might decide to use another register anyways.
9745 You may use @code{current_function_leaf_function} in the hook, functions
9746 that use @code{REG_N_SETS}, to determine if the hard
9747 register in question will not be clobbered.
9748 The default value of this hook is @code{NULL}, which disables any special
9752 @defmac TARGET_OBJECT_SUFFIX
9753 Define this macro to be a C string representing the suffix for object
9754 files on your target machine. If you do not define this macro, GCC will
9755 use @samp{.o} as the suffix for object files.
9758 @defmac TARGET_EXECUTABLE_SUFFIX
9759 Define this macro to be a C string representing the suffix to be
9760 automatically added to executable files on your target machine. If you
9761 do not define this macro, GCC will use the null string as the suffix for
9765 @defmac COLLECT_EXPORT_LIST
9766 If defined, @code{collect2} will scan the individual object files
9767 specified on its command line and create an export list for the linker.
9768 Define this macro for systems like AIX, where the linker discards
9769 object files that are not referenced from @code{main} and uses export
9773 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
9774 Define this macro to a C expression representing a variant of the
9775 method call @var{mdecl}, if Java Native Interface (JNI) methods
9776 must be invoked differently from other methods on your target.
9777 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
9778 the @code{stdcall} calling convention and this macro is then
9779 defined as this expression:
9782 build_type_attribute_variant (@var{mdecl},
9784 (get_identifier ("stdcall"),
9789 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
9790 This target hook returns @code{true} past the point in which new jump
9791 instructions could be created. On machines that require a register for
9792 every jump such as the SHmedia ISA of SH5, this point would typically be
9793 reload, so this target hook should be defined to a function such as:
9797 cannot_modify_jumps_past_reload_p ()
9799 return (reload_completed || reload_in_progress);
9804 @deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
9805 This target hook returns a register class for which branch target register
9806 optimizations should be applied. All registers in this class should be
9807 usable interchangeably. After reload, registers in this class will be
9808 re-allocated and loads will be hoisted out of loops and be subjected
9809 to inter-block scheduling.
9812 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
9813 Branch target register optimization will by default exclude callee-saved
9815 that are not already live during the current function; if this target hook
9816 returns true, they will be included. The target code must than make sure
9817 that all target registers in the class returned by
9818 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
9819 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
9820 epilogues have already been generated. Note, even if you only return
9821 true when @var{after_prologue_epilogue_gen} is false, you still are likely
9822 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
9823 to reserve space for caller-saved target registers.
9826 @defmac POWI_MAX_MULTS
9827 If defined, this macro is interpreted as a signed integer C expression
9828 that specifies the maximum number of floating point multiplications
9829 that should be emitted when expanding exponentiation by an integer
9830 constant inline. When this value is defined, exponentiation requiring
9831 more than this number of multiplications is implemented by calling the
9832 system library's @code{pow}, @code{powf} or @code{powl} routines.
9833 The default value places no upper bound on the multiplication count.
9836 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9837 This target hook should register any extra include files for the
9838 target. The parameter @var{stdinc} indicates if normal include files
9839 are present. The parameter @var{sysroot} is the system root directory.
9840 The parameter @var{iprefix} is the prefix for the gcc directory.
9843 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9844 This target hook should register any extra include files for the
9845 target before any standard headers. The parameter @var{stdinc}
9846 indicates if normal include files are present. The parameter
9847 @var{sysroot} is the system root directory. The parameter
9848 @var{iprefix} is the prefix for the gcc directory.
9851 @deftypefn Macro void TARGET_OPTF (char *@var{path})
9852 This target hook should register special include paths for the target.
9853 The parameter @var{path} is the include to register. On Darwin
9854 systems, this is used for Framework includes, which have semantics
9855 that are different from @option{-I}.
9858 @deftypefn {Target Hook} bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
9859 This target hook returns @code{true} if it is safe to use a local alias
9860 for a virtual function @var{fndecl} when constructing thunks,
9861 @code{false} otherwise. By default, the hook returns @code{true} for all
9862 functions, if a target supports aliases (i.e.@: defines
9863 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
9866 @defmac TARGET_FORMAT_TYPES
9867 If defined, this macro is the name of a global variable containing
9868 target-specific format checking information for the @option{-Wformat}
9869 option. The default is to have no target-specific format checks.
9872 @defmac TARGET_N_FORMAT_TYPES
9873 If defined, this macro is the number of entries in
9874 @code{TARGET_FORMAT_TYPES}.
9877 @deftypefn {Target Hook} bool TARGET_RELAXED_ORDERING
9878 If set to @code{true}, means that the target's memory model does not
9879 guarantee that loads which do not depend on one another will access
9880 main memory in the order of the instruction stream; if ordering is
9881 important, an explicit memory barrier must be used. This is true of
9882 many recent processors which implement a policy of ``relaxed,''
9883 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
9884 and ia64. The default is @code{false}.
9887 @deftypefn {Target Hook} const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (tree @var{typelist}, tree @var{funcdecl}, tree @var{val})
9888 If defined, this macro returns the diagnostic message when it is
9889 illegal to pass argument @var{val} to function @var{funcdecl}
9890 with prototype @var{typelist}.
9893 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (tree @var{fromtype}, tree @var{totype})
9894 If defined, this macro returns the diagnostic message when it is
9895 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
9896 if validity should be determined by the front end.
9899 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, tree @var{type})
9900 If defined, this macro returns the diagnostic message when it is
9901 invalid to apply operation @var{op} (where unary plus is denoted by
9902 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
9903 if validity should be determined by the front end.
9906 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, tree @var{type1}, tree @var{type2})
9907 If defined, this macro returns the diagnostic message when it is
9908 invalid to apply operation @var{op} to operands of types @var{type1}
9909 and @var{type2}, or @code{NULL} if validity should be determined by
9913 @defmac TARGET_USE_JCR_SECTION
9914 This macro determines whether to use the JCR section to register Java
9915 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
9916 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.