1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005 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 * Stack and Calling:: Defining which way the stack grows and by how much.
37 * Varargs:: Defining the varargs macros.
38 * Trampolines:: Code set up at run time to enter a nested function.
39 * Library Calls:: Controlling how library routines are implicitly called.
40 * Addressing Modes:: Defining addressing modes valid for memory operands.
41 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
42 * Condition Code:: Defining how insns update the condition code.
43 * Costs:: Defining relative costs of different operations.
44 * Scheduling:: Adjusting the behavior of the instruction scheduler.
45 * Sections:: Dividing storage into text, data, and other sections.
46 * PIC:: Macros for position independent code.
47 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
48 * Debugging Info:: Defining the format of debugging output.
49 * Floating Point:: Handling floating point for cross-compilers.
50 * Mode Switching:: Insertion of mode-switching instructions.
51 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
52 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
53 * PCH Target:: Validity checking for precompiled headers.
54 * C++ ABI:: Controlling C++ ABI changes.
55 * Misc:: Everything else.
58 @node Target Structure
59 @section The Global @code{targetm} Variable
61 @cindex target functions
63 @deftypevar {struct gcc_target} targetm
64 The target @file{.c} file must define the global @code{targetm} variable
65 which contains pointers to functions and data relating to the target
66 machine. The variable is declared in @file{target.h};
67 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
68 used to initialize the variable, and macros for the default initializers
69 for elements of the structure. The @file{.c} file should override those
70 macros for which the default definition is inappropriate. For example:
73 #include "target-def.h"
75 /* @r{Initialize the GCC target structure.} */
77 #undef TARGET_COMP_TYPE_ATTRIBUTES
78 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
80 struct gcc_target targetm = TARGET_INITIALIZER;
84 Where a macro should be defined in the @file{.c} file in this manner to
85 form part of the @code{targetm} structure, it is documented below as a
86 ``Target Hook'' with a prototype. Many macros will change in future
87 from being defined in the @file{.h} file to being part of the
88 @code{targetm} structure.
91 @section Controlling the Compilation Driver, @file{gcc}
93 @cindex controlling the compilation driver
95 @c prevent bad page break with this line
96 You can control the compilation driver.
98 @defmac SWITCH_TAKES_ARG (@var{char})
99 A C expression which determines whether the option @option{-@var{char}}
100 takes arguments. The value should be the number of arguments that
101 option takes--zero, for many options.
103 By default, this macro is defined as
104 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
105 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
106 wish to add additional options which take arguments. Any redefinition
107 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
111 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
112 A C expression which determines whether the option @option{-@var{name}}
113 takes arguments. The value should be the number of arguments that
114 option takes--zero, for many options. This macro rather than
115 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
117 By default, this macro is defined as
118 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
119 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
120 wish to add additional options which take arguments. Any redefinition
121 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
125 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
126 A C expression which determines whether the option @option{-@var{char}}
127 stops compilation before the generation of an executable. The value is
128 boolean, nonzero if the option does stop an executable from being
129 generated, zero otherwise.
131 By default, this macro is defined as
132 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
133 options properly. You need not define
134 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
135 options which affect the generation of an executable. Any redefinition
136 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
137 for additional options.
140 @defmac SWITCHES_NEED_SPACES
141 A string-valued C expression which enumerates the options for which
142 the linker needs a space between the option and its argument.
144 If this macro is not defined, the default value is @code{""}.
147 @defmac TARGET_OPTION_TRANSLATE_TABLE
148 If defined, a list of pairs of strings, the first of which is a
149 potential command line target to the @file{gcc} driver program, and the
150 second of which is a space-separated (tabs and other whitespace are not
151 supported) list of options with which to replace the first option. The
152 target defining this list is responsible for assuring that the results
153 are valid. Replacement options may not be the @code{--opt} style, they
154 must be the @code{-opt} style. It is the intention of this macro to
155 provide a mechanism for substitution that affects the multilibs chosen,
156 such as one option that enables many options, some of which select
157 multilibs. Example nonsensical definition, where @option{-malt-abi},
158 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
161 #define TARGET_OPTION_TRANSLATE_TABLE \
162 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
163 @{ "-compat", "-EB -malign=4 -mspoo" @}
167 @defmac DRIVER_SELF_SPECS
168 A list of specs for the driver itself. It should be a suitable
169 initializer for an array of strings, with no surrounding braces.
171 The driver applies these specs to its own command line between loading
172 default @file{specs} files (but not command-line specified ones) and
173 choosing the multilib directory or running any subcommands. It
174 applies them in the order given, so each spec can depend on the
175 options added by earlier ones. It is also possible to remove options
176 using @samp{%<@var{option}} in the usual way.
178 This macro can be useful when a port has several interdependent target
179 options. It provides a way of standardizing the command line so
180 that the other specs are easier to write.
182 Do not define this macro if it does not need to do anything.
185 @defmac OPTION_DEFAULT_SPECS
186 A list of specs used to support configure-time default options (i.e.@:
187 @option{--with} options) in the driver. It should be a suitable initializer
188 for an array of structures, each containing two strings, without the
189 outermost pair of surrounding braces.
191 The first item in the pair is the name of the default. This must match
192 the code in @file{config.gcc} for the target. The second item is a spec
193 to apply if a default with this name was specified. The string
194 @samp{%(VALUE)} in the spec will be replaced by the value of the default
195 everywhere it occurs.
197 The driver will apply these specs to its own command line between loading
198 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
199 the same mechanism as @code{DRIVER_SELF_SPECS}.
201 Do not define this macro if it does not need to do anything.
205 A C string constant that tells the GCC driver program options to
206 pass to CPP@. It can also specify how to translate options you
207 give to GCC into options for GCC to pass to the CPP@.
209 Do not define this macro if it does not need to do anything.
212 @defmac CPLUSPLUS_CPP_SPEC
213 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
214 than C@. If you do not define this macro, then the value of
215 @code{CPP_SPEC} (if any) will be used instead.
219 A C string constant that tells the GCC driver program options to
220 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
222 It can also specify how to translate options you give to GCC into options
223 for GCC to pass to front ends.
225 Do not define this macro if it does not need to do anything.
229 A C string constant that tells the GCC driver program options to
230 pass to @code{cc1plus}. It can also specify how to translate options you
231 give to GCC into options for GCC to pass to the @code{cc1plus}.
233 Do not define this macro if it does not need to do anything.
234 Note that everything defined in CC1_SPEC is already passed to
235 @code{cc1plus} so there is no need to duplicate the contents of
236 CC1_SPEC in CC1PLUS_SPEC@.
240 A C string constant that tells the GCC driver program options to
241 pass to the assembler. It can also specify how to translate options
242 you give to GCC into options for GCC to pass to the assembler.
243 See the file @file{sun3.h} for an example of this.
245 Do not define this macro if it does not need to do anything.
248 @defmac ASM_FINAL_SPEC
249 A C string constant that tells the GCC driver program how to
250 run any programs which cleanup after the normal assembler.
251 Normally, this is not needed. See the file @file{mips.h} for
254 Do not define this macro if it does not need to do anything.
257 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
258 Define this macro, with no value, if the driver should give the assembler
259 an argument consisting of a single dash, @option{-}, to instruct it to
260 read from its standard input (which will be a pipe connected to the
261 output of the compiler proper). This argument is given after any
262 @option{-o} option specifying the name of the output file.
264 If you do not define this macro, the assembler is assumed to read its
265 standard input if given no non-option arguments. If your assembler
266 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
267 see @file{mips.h} for instance.
271 A C string constant that tells the GCC driver program options to
272 pass to the linker. It can also specify how to translate options you
273 give to GCC into options for GCC to pass to the linker.
275 Do not define this macro if it does not need to do anything.
279 Another C string constant used much like @code{LINK_SPEC}. The difference
280 between the two is that @code{LIB_SPEC} is used at the end of the
281 command given to the linker.
283 If this macro is not defined, a default is provided that
284 loads the standard C library from the usual place. See @file{gcc.c}.
288 Another C string constant that tells the GCC driver program
289 how and when to place a reference to @file{libgcc.a} into the
290 linker command line. This constant is placed both before and after
291 the value of @code{LIB_SPEC}.
293 If this macro is not defined, the GCC driver provides a default that
294 passes the string @option{-lgcc} to the linker.
297 @defmac REAL_LIBGCC_SPEC
298 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
299 @code{LIBGCC_SPEC} is not directly used by the driver program but is
300 instead modified to refer to different versions of @file{libgcc.a}
301 depending on the values of the command line flags @option{-static},
302 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
303 targets where these modifications are inappropriate, define
304 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
305 driver how to place a reference to @file{libgcc} on the link command
306 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
309 @defmac USE_LD_AS_NEEDED
310 A macro that controls the modifications to @code{LIBGCC_SPEC}
311 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
312 generated that uses --as-needed and the shared libgcc in place of the
313 static exception handler library, when linking without any of
314 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
318 If defined, this C string constant is added to @code{LINK_SPEC}.
319 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
320 the modifications to @code{LIBGCC_SPEC} mentioned in
321 @code{REAL_LIBGCC_SPEC}.
324 @defmac STARTFILE_SPEC
325 Another C string constant used much like @code{LINK_SPEC}. The
326 difference between the two is that @code{STARTFILE_SPEC} is used at
327 the very beginning of the command given to the linker.
329 If this macro is not defined, a default is provided that loads the
330 standard C startup file from the usual place. See @file{gcc.c}.
334 Another C string constant used much like @code{LINK_SPEC}. The
335 difference between the two is that @code{ENDFILE_SPEC} is used at
336 the very end of the command given to the linker.
338 Do not define this macro if it does not need to do anything.
341 @defmac THREAD_MODEL_SPEC
342 GCC @code{-v} will print the thread model GCC was configured to use.
343 However, this doesn't work on platforms that are multilibbed on thread
344 models, such as AIX 4.3. On such platforms, define
345 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
346 blanks that names one of the recognized thread models. @code{%*}, the
347 default value of this macro, will expand to the value of
348 @code{thread_file} set in @file{config.gcc}.
351 @defmac SYSROOT_SUFFIX_SPEC
352 Define this macro to add a suffix to the target sysroot when GCC is
353 configured with a sysroot. This will cause GCC to search for usr/lib,
354 et al, within sysroot+suffix.
357 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
358 Define this macro to add a headers_suffix to the target sysroot when
359 GCC is configured with a sysroot. This will cause GCC to pass the
360 updated sysroot+headers_suffix to CPP, causing it to search for
361 usr/include, et al, within sysroot+headers_suffix.
365 Define this macro to provide additional specifications to put in the
366 @file{specs} file that can be used in various specifications like
369 The definition should be an initializer for an array of structures,
370 containing a string constant, that defines the specification name, and a
371 string constant that provides the specification.
373 Do not define this macro if it does not need to do anything.
375 @code{EXTRA_SPECS} is useful when an architecture contains several
376 related targets, which have various @code{@dots{}_SPECS} which are similar
377 to each other, and the maintainer would like one central place to keep
380 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
381 define either @code{_CALL_SYSV} when the System V calling sequence is
382 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
385 The @file{config/rs6000/rs6000.h} target file defines:
388 #define EXTRA_SPECS \
389 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
391 #define CPP_SYS_DEFAULT ""
394 The @file{config/rs6000/sysv.h} target file defines:
398 "%@{posix: -D_POSIX_SOURCE @} \
399 %@{mcall-sysv: -D_CALL_SYSV @} \
400 %@{!mcall-sysv: %(cpp_sysv_default) @} \
401 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
403 #undef CPP_SYSV_DEFAULT
404 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
407 while the @file{config/rs6000/eabiaix.h} target file defines
408 @code{CPP_SYSV_DEFAULT} as:
411 #undef CPP_SYSV_DEFAULT
412 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
416 @defmac LINK_LIBGCC_SPECIAL_1
417 Define this macro if the driver program should find the library
418 @file{libgcc.a}. If you do not define this macro, the driver program will pass
419 the argument @option{-lgcc} to tell the linker to do the search.
422 @defmac LINK_GCC_C_SEQUENCE_SPEC
423 The sequence in which libgcc and libc are specified to the linker.
424 By default this is @code{%G %L %G}.
427 @defmac LINK_COMMAND_SPEC
428 A C string constant giving the complete command line need to execute the
429 linker. When you do this, you will need to update your port each time a
430 change is made to the link command line within @file{gcc.c}. Therefore,
431 define this macro only if you need to completely redefine the command
432 line for invoking the linker and there is no other way to accomplish
433 the effect you need. Overriding this macro may be avoidable by overriding
434 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
437 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
438 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
439 directories from linking commands. Do not give it a nonzero value if
440 removing duplicate search directories changes the linker's semantics.
443 @defmac MULTILIB_DEFAULTS
444 Define this macro as a C expression for the initializer of an array of
445 string to tell the driver program which options are defaults for this
446 target and thus do not need to be handled specially when using
447 @code{MULTILIB_OPTIONS}.
449 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
450 the target makefile fragment or if none of the options listed in
451 @code{MULTILIB_OPTIONS} are set by default.
452 @xref{Target Fragment}.
455 @defmac RELATIVE_PREFIX_NOT_LINKDIR
456 Define this macro to tell @command{gcc} that it should only translate
457 a @option{-B} prefix into a @option{-L} linker option if the prefix
458 indicates an absolute file name.
461 @defmac MD_EXEC_PREFIX
462 If defined, this macro is an additional prefix to try after
463 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
464 when the @option{-b} option is used, or the compiler is built as a cross
465 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
466 to the list of directories used to find the assembler in @file{configure.in}.
469 @defmac STANDARD_STARTFILE_PREFIX
470 Define this macro as a C string constant if you wish to override the
471 standard choice of @code{libdir} as the default prefix to
472 try when searching for startup files such as @file{crt0.o}.
473 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
474 is built as a cross compiler.
477 @defmac STANDARD_STARTFILE_PREFIX_1
478 Define this macro as a C string constant if you wish to override the
479 standard choice of @code{/lib} as a prefix to try after the default prefix
480 when searching for startup files such as @file{crt0.o}.
481 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
482 is built as a cross compiler.
485 @defmac STANDARD_STARTFILE_PREFIX_2
486 Define this macro as a C string constant if you wish to override the
487 standard choice of @code{/lib} as yet another prefix to try after the
488 default prefix when searching for startup files such as @file{crt0.o}.
489 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
490 is built as a cross compiler.
493 @defmac MD_STARTFILE_PREFIX
494 If defined, this macro supplies an additional prefix to try after the
495 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
496 @option{-b} option is used, or when the compiler is built as a cross
500 @defmac MD_STARTFILE_PREFIX_1
501 If defined, this macro supplies yet another prefix to try after the
502 standard prefixes. It is not searched when the @option{-b} option is
503 used, or when the compiler is built as a cross compiler.
506 @defmac INIT_ENVIRONMENT
507 Define this macro as a C string constant if you wish to set environment
508 variables for programs called by the driver, such as the assembler and
509 loader. The driver passes the value of this macro to @code{putenv} to
510 initialize the necessary environment variables.
513 @defmac LOCAL_INCLUDE_DIR
514 Define this macro as a C string constant if you wish to override the
515 standard choice of @file{/usr/local/include} as the default prefix to
516 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
517 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
519 Cross compilers do not search either @file{/usr/local/include} or its
523 @defmac MODIFY_TARGET_NAME
524 Define this macro if you wish to define command-line switches that
525 modify the default target name.
527 For each switch, you can include a string to be appended to the first
528 part of the configuration name or a string to be deleted from the
529 configuration name, if present. The definition should be an initializer
530 for an array of structures. Each array element should have three
531 elements: the switch name (a string constant, including the initial
532 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
533 indicate whether the string should be inserted or deleted, and the string
534 to be inserted or deleted (a string constant).
536 For example, on a machine where @samp{64} at the end of the
537 configuration name denotes a 64-bit target and you want the @option{-32}
538 and @option{-64} switches to select between 32- and 64-bit targets, you would
542 #define MODIFY_TARGET_NAME \
543 @{ @{ "-32", DELETE, "64"@}, \
544 @{"-64", ADD, "64"@}@}
548 @defmac SYSTEM_INCLUDE_DIR
549 Define this macro as a C string constant if you wish to specify a
550 system-specific directory to search for header files before the standard
551 directory. @code{SYSTEM_INCLUDE_DIR} comes before
552 @code{STANDARD_INCLUDE_DIR} in the search order.
554 Cross compilers do not use this macro and do not search the directory
558 @defmac STANDARD_INCLUDE_DIR
559 Define this macro as a C string constant if you wish to override the
560 standard choice of @file{/usr/include} as the default prefix to
561 try when searching for header files.
563 Cross compilers ignore this macro and do not search either
564 @file{/usr/include} or its replacement.
567 @defmac STANDARD_INCLUDE_COMPONENT
568 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
569 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
570 If you do not define this macro, no component is used.
573 @defmac INCLUDE_DEFAULTS
574 Define this macro if you wish to override the entire default search path
575 for include files. For a native compiler, the default search path
576 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
577 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
578 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
579 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
580 and specify private search areas for GCC@. The directory
581 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
583 The definition should be an initializer for an array of structures.
584 Each array element should have four elements: the directory name (a
585 string constant), the component name (also a string constant), a flag
586 for C++-only directories,
587 and a flag showing that the includes in the directory don't need to be
588 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
589 the array with a null element.
591 The component name denotes what GNU package the include file is part of,
592 if any, in all uppercase letters. For example, it might be @samp{GCC}
593 or @samp{BINUTILS}. If the package is part of a vendor-supplied
594 operating system, code the component name as @samp{0}.
596 For example, here is the definition used for VAX/VMS:
599 #define INCLUDE_DEFAULTS \
601 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
602 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
603 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
610 Here is the order of prefixes tried for exec files:
614 Any prefixes specified by the user with @option{-B}.
617 The environment variable @code{GCC_EXEC_PREFIX}, if any.
620 The directories specified by the environment variable @code{COMPILER_PATH}.
623 The macro @code{STANDARD_EXEC_PREFIX}.
626 @file{/usr/lib/gcc/}.
629 The macro @code{MD_EXEC_PREFIX}, if any.
632 Here is the order of prefixes tried for startfiles:
636 Any prefixes specified by the user with @option{-B}.
639 The environment variable @code{GCC_EXEC_PREFIX}, if any.
642 The directories specified by the environment variable @code{LIBRARY_PATH}
643 (or port-specific name; native only, cross compilers do not use this).
646 The macro @code{STANDARD_EXEC_PREFIX}.
649 @file{/usr/lib/gcc/}.
652 The macro @code{MD_EXEC_PREFIX}, if any.
655 The macro @code{MD_STARTFILE_PREFIX}, if any.
658 The macro @code{STANDARD_STARTFILE_PREFIX}.
667 @node Run-time Target
668 @section Run-time Target Specification
669 @cindex run-time target specification
670 @cindex predefined macros
671 @cindex target specifications
673 @c prevent bad page break with this line
674 Here are run-time target specifications.
676 @defmac TARGET_CPU_CPP_BUILTINS ()
677 This function-like macro expands to a block of code that defines
678 built-in preprocessor macros and assertions for the target cpu, using
679 the functions @code{builtin_define}, @code{builtin_define_std} and
680 @code{builtin_assert}. When the front end
681 calls this macro it provides a trailing semicolon, and since it has
682 finished command line option processing your code can use those
685 @code{builtin_assert} takes a string in the form you pass to the
686 command-line option @option{-A}, such as @code{cpu=mips}, and creates
687 the assertion. @code{builtin_define} takes a string in the form
688 accepted by option @option{-D} and unconditionally defines the macro.
690 @code{builtin_define_std} takes a string representing the name of an
691 object-like macro. If it doesn't lie in the user's namespace,
692 @code{builtin_define_std} defines it unconditionally. Otherwise, it
693 defines a version with two leading underscores, and another version
694 with two leading and trailing underscores, and defines the original
695 only if an ISO standard was not requested on the command line. For
696 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
697 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
698 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
699 defines only @code{_ABI64}.
701 You can also test for the C dialect being compiled. The variable
702 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
703 or @code{clk_objective_c}. Note that if we are preprocessing
704 assembler, this variable will be @code{clk_c} but the function-like
705 macro @code{preprocessing_asm_p()} will return true, so you might want
706 to check for that first. If you need to check for strict ANSI, the
707 variable @code{flag_iso} can be used. The function-like macro
708 @code{preprocessing_trad_p()} can be used to check for traditional
712 @defmac TARGET_OS_CPP_BUILTINS ()
713 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
714 and is used for the target operating system instead.
717 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
718 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
719 and is used for the target object format. @file{elfos.h} uses this
720 macro to define @code{__ELF__}, so you probably do not need to define
724 @deftypevar {extern int} target_flags
725 This variable is declared in @file{options.h}, which is included before
726 any target-specific headers.
729 @deftypevar {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
730 This variable specifies the initial value of @code{target_flags}.
731 Its default setting is 0.
734 @cindex optional hardware or system features
735 @cindex features, optional, in system conventions
737 @deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
738 This hook is called whenever the user specifies one of the
739 target-specific options described by the @file{.opt} definition files
740 (@pxref{Options}). It has the opportunity to do some option-specific
741 processing and should return true if the option is valid. The default
742 definition does nothing but return true.
744 @var{code} specifies the @code{OPT_@var{name}} enumeration value
745 associated with the selected option; @var{name} is just a rendering of
746 the option name in which non-alphanumeric characters are replaced by
747 underscores. @var{arg} specifies the string argument and is null if
748 no argument was given. If the option is flagged as a @code{UInteger}
749 (@pxref{Option properties}), @var{value} is the numeric value of the
750 argument. Otherwise @var{value} is 1 if the positive form of the
751 option was used and 0 if the ``no-'' form was.
754 @defmac TARGET_VERSION
755 This macro is a C statement to print on @code{stderr} a string
756 describing the particular machine description choice. Every machine
757 description should define @code{TARGET_VERSION}. For example:
761 #define TARGET_VERSION \
762 fprintf (stderr, " (68k, Motorola syntax)");
764 #define TARGET_VERSION \
765 fprintf (stderr, " (68k, MIT syntax)");
770 @defmac OVERRIDE_OPTIONS
771 Sometimes certain combinations of command options do not make sense on
772 a particular target machine. You can define a macro
773 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
774 defined, is executed once just after all the command options have been
777 Don't use this macro to turn on various extra optimizations for
778 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
781 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
782 Some machines may desire to change what optimizations are performed for
783 various optimization levels. This macro, if defined, is executed once
784 just after the optimization level is determined and before the remainder
785 of the command options have been parsed. Values set in this macro are
786 used as the default values for the other command line options.
788 @var{level} is the optimization level specified; 2 if @option{-O2} is
789 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
791 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
793 You should not use this macro to change options that are not
794 machine-specific. These should uniformly selected by the same
795 optimization level on all supported machines. Use this macro to enable
796 machine-specific optimizations.
798 @strong{Do not examine @code{write_symbols} in
799 this macro!} The debugging options are not supposed to alter the
803 @defmac CAN_DEBUG_WITHOUT_FP
804 Define this macro if debugging can be performed even without a frame
805 pointer. If this macro is defined, GCC will turn on the
806 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
809 @node Per-Function Data
810 @section Defining data structures for per-function information.
811 @cindex per-function data
812 @cindex data structures
814 If the target needs to store information on a per-function basis, GCC
815 provides a macro and a couple of variables to allow this. Note, just
816 using statics to store the information is a bad idea, since GCC supports
817 nested functions, so you can be halfway through encoding one function
818 when another one comes along.
820 GCC defines a data structure called @code{struct function} which
821 contains all of the data specific to an individual function. This
822 structure contains a field called @code{machine} whose type is
823 @code{struct machine_function *}, which can be used by targets to point
824 to their own specific data.
826 If a target needs per-function specific data it should define the type
827 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
828 This macro should be used to initialize the function pointer
829 @code{init_machine_status}. This pointer is explained below.
831 One typical use of per-function, target specific data is to create an
832 RTX to hold the register containing the function's return address. This
833 RTX can then be used to implement the @code{__builtin_return_address}
834 function, for level 0.
836 Note---earlier implementations of GCC used a single data area to hold
837 all of the per-function information. Thus when processing of a nested
838 function began the old per-function data had to be pushed onto a
839 stack, and when the processing was finished, it had to be popped off the
840 stack. GCC used to provide function pointers called
841 @code{save_machine_status} and @code{restore_machine_status} to handle
842 the saving and restoring of the target specific information. Since the
843 single data area approach is no longer used, these pointers are no
846 @defmac INIT_EXPANDERS
847 Macro called to initialize any target specific information. This macro
848 is called once per function, before generation of any RTL has begun.
849 The intention of this macro is to allow the initialization of the
850 function pointer @code{init_machine_status}.
853 @deftypevar {void (*)(struct function *)} init_machine_status
854 If this function pointer is non-@code{NULL} it will be called once per
855 function, before function compilation starts, in order to allow the
856 target to perform any target specific initialization of the
857 @code{struct function} structure. It is intended that this would be
858 used to initialize the @code{machine} of that structure.
860 @code{struct machine_function} structures are expected to be freed by GC@.
861 Generally, any memory that they reference must be allocated by using
862 @code{ggc_alloc}, including the structure itself.
866 @section Storage Layout
867 @cindex storage layout
869 Note that the definitions of the macros in this table which are sizes or
870 alignments measured in bits do not need to be constant. They can be C
871 expressions that refer to static variables, such as the @code{target_flags}.
872 @xref{Run-time Target}.
874 @defmac BITS_BIG_ENDIAN
875 Define this macro to have the value 1 if the most significant bit in a
876 byte has the lowest number; otherwise define it to have the value zero.
877 This means that bit-field instructions count from the most significant
878 bit. If the machine has no bit-field instructions, then this must still
879 be defined, but it doesn't matter which value it is defined to. This
880 macro need not be a constant.
882 This macro does not affect the way structure fields are packed into
883 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
886 @defmac BYTES_BIG_ENDIAN
887 Define this macro to have the value 1 if the most significant byte in a
888 word has the lowest number. This macro need not be a constant.
891 @defmac WORDS_BIG_ENDIAN
892 Define this macro to have the value 1 if, in a multiword object, the
893 most significant word has the lowest number. This applies to both
894 memory locations and registers; GCC fundamentally assumes that the
895 order of words in memory is the same as the order in registers. This
896 macro need not be a constant.
899 @defmac LIBGCC2_WORDS_BIG_ENDIAN
900 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
901 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
902 used only when compiling @file{libgcc2.c}. Typically the value will be set
903 based on preprocessor defines.
906 @defmac FLOAT_WORDS_BIG_ENDIAN
907 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
908 @code{TFmode} floating point numbers are stored in memory with the word
909 containing the sign bit at the lowest address; otherwise define it to
910 have the value 0. This macro need not be a constant.
912 You need not define this macro if the ordering is the same as for
916 @defmac BITS_PER_UNIT
917 Define this macro to be the number of bits in an addressable storage
918 unit (byte). If you do not define this macro the default is 8.
921 @defmac BITS_PER_WORD
922 Number of bits in a word. If you do not define this macro, the default
923 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
926 @defmac MAX_BITS_PER_WORD
927 Maximum number of bits in a word. If this is undefined, the default is
928 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
929 largest value that @code{BITS_PER_WORD} can have at run-time.
932 @defmac UNITS_PER_WORD
933 Number of storage units in a word; normally the size of a general-purpose
934 register, a power of two from 1 or 8.
937 @defmac MIN_UNITS_PER_WORD
938 Minimum number of units in a word. If this is undefined, the default is
939 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
940 smallest value that @code{UNITS_PER_WORD} can have at run-time.
943 @defmac UNITS_PER_SIMD_WORD
944 Number of units in the vectors that the vectorizer can produce.
945 The default is equal to @code{UNITS_PER_WORD}, because the vectorizer
946 can do some transformations even in absence of specialized @acronym{SIMD}
951 Width of a pointer, in bits. You must specify a value no wider than the
952 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
953 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
954 a value the default is @code{BITS_PER_WORD}.
957 @defmac POINTERS_EXTEND_UNSIGNED
958 A C expression whose value is greater than zero if pointers that need to be
959 extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to
960 be zero-extended and zero if they are to be sign-extended. If the value
961 is less then zero then there must be an "ptr_extend" instruction that
962 extends a pointer from @code{POINTER_SIZE} to @code{Pmode}.
964 You need not define this macro if the @code{POINTER_SIZE} is equal
965 to the width of @code{Pmode}.
968 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
969 A macro to update @var{m} and @var{unsignedp} when an object whose type
970 is @var{type} and which has the specified mode and signedness is to be
971 stored in a register. This macro is only called when @var{type} is a
974 On most RISC machines, which only have operations that operate on a full
975 register, define this macro to set @var{m} to @code{word_mode} if
976 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
977 cases, only integer modes should be widened because wider-precision
978 floating-point operations are usually more expensive than their narrower
981 For most machines, the macro definition does not change @var{unsignedp}.
982 However, some machines, have instructions that preferentially handle
983 either signed or unsigned quantities of certain modes. For example, on
984 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
985 sign-extend the result to 64 bits. On such machines, set
986 @var{unsignedp} according to which kind of extension is more efficient.
988 Do not define this macro if it would never modify @var{m}.
991 @defmac PROMOTE_FUNCTION_MODE
992 Like @code{PROMOTE_MODE}, but is applied to outgoing function arguments or
993 function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS}
994 and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively.
996 The default is @code{PROMOTE_MODE}.
999 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype})
1000 This target hook should return @code{true} if the promotion described by
1001 @code{PROMOTE_FUNCTION_MODE} should be done for outgoing function
1005 @deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype})
1006 This target hook should return @code{true} if the promotion described by
1007 @code{PROMOTE_FUNCTION_MODE} should be done for the return value of
1010 If this target hook returns @code{true}, @code{FUNCTION_VALUE} must
1011 perform the same promotions done by @code{PROMOTE_FUNCTION_MODE}.
1014 @defmac PARM_BOUNDARY
1015 Normal alignment required for function parameters on the stack, in
1016 bits. All stack parameters receive at least this much alignment
1017 regardless of data type. On most machines, this is the same as the
1021 @defmac STACK_BOUNDARY
1022 Define this macro to the minimum alignment enforced by hardware for the
1023 stack pointer on this machine. The definition is a C expression for the
1024 desired alignment (measured in bits). This value is used as a default
1025 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1026 this should be the same as @code{PARM_BOUNDARY}.
1029 @defmac PREFERRED_STACK_BOUNDARY
1030 Define this macro if you wish to preserve a certain alignment for the
1031 stack pointer, greater than what the hardware enforces. The definition
1032 is a C expression for the desired alignment (measured in bits). This
1033 macro must evaluate to a value equal to or larger than
1034 @code{STACK_BOUNDARY}.
1037 @defmac FUNCTION_BOUNDARY
1038 Alignment required for a function entry point, in bits.
1041 @defmac BIGGEST_ALIGNMENT
1042 Biggest alignment that any data type can require on this machine, in bits.
1045 @defmac MINIMUM_ATOMIC_ALIGNMENT
1046 If defined, the smallest alignment, in bits, that can be given to an
1047 object that can be referenced in one operation, without disturbing any
1048 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1049 on machines that don't have byte or half-word store operations.
1052 @defmac BIGGEST_FIELD_ALIGNMENT
1053 Biggest alignment that any structure or union field can require on this
1054 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1055 structure and union fields only, unless the field alignment has been set
1056 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1059 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1060 An expression for the alignment of a structure field @var{field} if the
1061 alignment computed in the usual way (including applying of
1062 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1063 alignment) is @var{computed}. It overrides alignment only if the
1064 field alignment has not been set by the
1065 @code{__attribute__ ((aligned (@var{n})))} construct.
1068 @defmac MAX_OFILE_ALIGNMENT
1069 Biggest alignment supported by the object file format of this machine.
1070 Use this macro to limit the alignment which can be specified using the
1071 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1072 the default value is @code{BIGGEST_ALIGNMENT}.
1075 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1076 If defined, a C expression to compute the alignment for a variable in
1077 the static store. @var{type} is the data type, and @var{basic-align} is
1078 the alignment that the object would ordinarily have. The value of this
1079 macro is used instead of that alignment to align the object.
1081 If this macro is not defined, then @var{basic-align} is used.
1084 One use of this macro is to increase alignment of medium-size data to
1085 make it all fit in fewer cache lines. Another is to cause character
1086 arrays to be word-aligned so that @code{strcpy} calls that copy
1087 constants to character arrays can be done inline.
1090 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1091 If defined, a C expression to compute the alignment given to a constant
1092 that is being placed in memory. @var{constant} is the constant and
1093 @var{basic-align} is the alignment that the object would ordinarily
1094 have. The value of this macro is used instead of that alignment to
1097 If this macro is not defined, then @var{basic-align} is used.
1099 The typical use of this macro is to increase alignment for string
1100 constants to be word aligned so that @code{strcpy} calls that copy
1101 constants can be done inline.
1104 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1105 If defined, a C expression to compute the alignment for a variable in
1106 the local store. @var{type} is the data type, and @var{basic-align} is
1107 the alignment that the object would ordinarily have. The value of this
1108 macro is used instead of that alignment to align the object.
1110 If this macro is not defined, then @var{basic-align} is used.
1112 One use of this macro is to increase alignment of medium-size data to
1113 make it all fit in fewer cache lines.
1116 @defmac EMPTY_FIELD_BOUNDARY
1117 Alignment in bits to be given to a structure bit-field that follows an
1118 empty field such as @code{int : 0;}.
1120 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1123 @defmac STRUCTURE_SIZE_BOUNDARY
1124 Number of bits which any structure or union's size must be a multiple of.
1125 Each structure or union's size is rounded up to a multiple of this.
1127 If you do not define this macro, the default is the same as
1128 @code{BITS_PER_UNIT}.
1131 @defmac STRICT_ALIGNMENT
1132 Define this macro to be the value 1 if instructions will fail to work
1133 if given data not on the nominal alignment. If instructions will merely
1134 go slower in that case, define this macro as 0.
1137 @defmac PCC_BITFIELD_TYPE_MATTERS
1138 Define this if you wish to imitate the way many other C compilers handle
1139 alignment of bit-fields and the structures that contain them.
1141 The behavior is that the type written for a named bit-field (@code{int},
1142 @code{short}, or other integer type) imposes an alignment for the entire
1143 structure, as if the structure really did contain an ordinary field of
1144 that type. In addition, the bit-field is placed within the structure so
1145 that it would fit within such a field, not crossing a boundary for it.
1147 Thus, on most machines, a named bit-field whose type is written as
1148 @code{int} would not cross a four-byte boundary, and would force
1149 four-byte alignment for the whole structure. (The alignment used may
1150 not be four bytes; it is controlled by the other alignment parameters.)
1152 An unnamed bit-field will not affect the alignment of the containing
1155 If the macro is defined, its definition should be a C expression;
1156 a nonzero value for the expression enables this behavior.
1158 Note that if this macro is not defined, or its value is zero, some
1159 bit-fields may cross more than one alignment boundary. The compiler can
1160 support such references if there are @samp{insv}, @samp{extv}, and
1161 @samp{extzv} insns that can directly reference memory.
1163 The other known way of making bit-fields work is to define
1164 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1165 Then every structure can be accessed with fullwords.
1167 Unless the machine has bit-field instructions or you define
1168 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1169 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1171 If your aim is to make GCC use the same conventions for laying out
1172 bit-fields as are used by another compiler, here is how to investigate
1173 what the other compiler does. Compile and run this program:
1192 printf ("Size of foo1 is %d\n",
1193 sizeof (struct foo1));
1194 printf ("Size of foo2 is %d\n",
1195 sizeof (struct foo2));
1200 If this prints 2 and 5, then the compiler's behavior is what you would
1201 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1204 @defmac BITFIELD_NBYTES_LIMITED
1205 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1206 to aligning a bit-field within the structure.
1209 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELDS (void)
1210 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1211 whether unnamed bitfields affect the alignment of the containing
1212 structure. The hook should return true if the structure should inherit
1213 the alignment requirements of an unnamed bitfield's type.
1216 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1217 Return 1 if a structure or array containing @var{field} should be accessed using
1220 If @var{field} is the only field in the structure, @var{mode} is its
1221 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1222 case where structures of one field would require the structure's mode to
1223 retain the field's mode.
1225 Normally, this is not needed. See the file @file{c4x.h} for an example
1226 of how to use this macro to prevent a structure having a floating point
1227 field from being accessed in an integer mode.
1230 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1231 Define this macro as an expression for the alignment of a type (given
1232 by @var{type} as a tree node) if the alignment computed in the usual
1233 way is @var{computed} and the alignment explicitly specified was
1236 The default is to use @var{specified} if it is larger; otherwise, use
1237 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1240 @defmac MAX_FIXED_MODE_SIZE
1241 An integer expression for the size in bits of the largest integer
1242 machine mode that should actually be used. All integer machine modes of
1243 this size or smaller can be used for structures and unions with the
1244 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1245 (DImode)} is assumed.
1248 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1249 If defined, an expression of type @code{enum machine_mode} that
1250 specifies the mode of the save area operand of a
1251 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1252 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1253 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1254 having its mode specified.
1256 You need not define this macro if it always returns @code{Pmode}. You
1257 would most commonly define this macro if the
1258 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1262 @defmac STACK_SIZE_MODE
1263 If defined, an expression of type @code{enum machine_mode} that
1264 specifies the mode of the size increment operand of an
1265 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1267 You need not define this macro if it always returns @code{word_mode}.
1268 You would most commonly define this macro if the @code{allocate_stack}
1269 pattern needs to support both a 32- and a 64-bit mode.
1272 @defmac TARGET_FLOAT_FORMAT
1273 A code distinguishing the floating point format of the target machine.
1274 There are four defined values:
1277 @item IEEE_FLOAT_FORMAT
1278 This code indicates IEEE floating point. It is the default; there is no
1279 need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@.
1281 @item VAX_FLOAT_FORMAT
1282 This code indicates the ``F float'' (for @code{float}) and ``D float''
1283 or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@.
1285 @item IBM_FLOAT_FORMAT
1286 This code indicates the format used on the IBM System/370.
1288 @item C4X_FLOAT_FORMAT
1289 This code indicates the format used on the TMS320C3x/C4x.
1292 If your target uses a floating point format other than these, you must
1293 define a new @var{name}_FLOAT_FORMAT code for it, and add support for
1294 it to @file{real.c}.
1296 The ordering of the component words of floating point values stored in
1297 memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}.
1300 @defmac MODE_HAS_NANS (@var{mode})
1301 When defined, this macro should be true if @var{mode} has a NaN
1302 representation. The compiler assumes that NaNs are not equal to
1303 anything (including themselves) and that addition, subtraction,
1304 multiplication and division all return NaNs when one operand is
1307 By default, this macro is true if @var{mode} is a floating-point
1308 mode and the target floating-point format is IEEE@.
1311 @defmac MODE_HAS_INFINITIES (@var{mode})
1312 This macro should be true if @var{mode} can represent infinity. At
1313 present, the compiler uses this macro to decide whether @samp{x - x}
1314 is always defined. By default, the macro is true when @var{mode}
1315 is a floating-point mode and the target format is IEEE@.
1318 @defmac MODE_HAS_SIGNED_ZEROS (@var{mode})
1319 True if @var{mode} distinguishes between positive and negative zero.
1320 The rules are expected to follow the IEEE standard:
1324 @samp{x + x} has the same sign as @samp{x}.
1327 If the sum of two values with opposite sign is zero, the result is
1328 positive for all rounding modes expect towards @minus{}infinity, for
1329 which it is negative.
1332 The sign of a product or quotient is negative when exactly one
1333 of the operands is negative.
1336 The default definition is true if @var{mode} is a floating-point
1337 mode and the target format is IEEE@.
1340 @defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode})
1341 If defined, this macro should be true for @var{mode} if it has at
1342 least one rounding mode in which @samp{x} and @samp{-x} can be
1343 rounded to numbers of different magnitude. Two such modes are
1344 towards @minus{}infinity and towards +infinity.
1346 The default definition of this macro is true if @var{mode} is
1347 a floating-point mode and the target format is IEEE@.
1350 @defmac ROUND_TOWARDS_ZERO
1351 If defined, this macro should be true if the prevailing rounding
1352 mode is towards zero. A true value has the following effects:
1356 @code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes.
1359 @file{libgcc.a}'s floating-point emulator will round towards zero
1360 rather than towards nearest.
1363 The compiler's floating-point emulator will round towards zero after
1364 doing arithmetic, and when converting from the internal float format to
1368 The macro does not affect the parsing of string literals. When the
1369 primary rounding mode is towards zero, library functions like
1370 @code{strtod} might still round towards nearest, and the compiler's
1371 parser should behave like the target's @code{strtod} where possible.
1373 Not defining this macro is equivalent to returning zero.
1376 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1377 This macro should return true if floats with @var{size}
1378 bits do not have a NaN or infinity representation, but use the largest
1379 exponent for normal numbers instead.
1381 Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS}
1382 and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes.
1383 It also affects the way @file{libgcc.a} and @file{real.c} emulate
1384 floating-point arithmetic.
1386 The default definition of this macro returns false for all sizes.
1389 @deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type})
1390 This target hook should return @code{true} a vector is opaque. That
1391 is, if no cast is needed when copying a vector value of type
1392 @var{type} into another vector lvalue of the same size. Vector opaque
1393 types cannot be initialized. The default is that there are no such
1397 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type})
1398 This target hook returns @code{true} if bit-fields in the given
1399 @var{record_type} are to be laid out following the rules of Microsoft
1400 Visual C/C++, namely: (i) a bit-field won't share the same storage
1401 unit with the previous bit-field if their underlying types have
1402 different sizes, and the bit-field will be aligned to the highest
1403 alignment of the underlying types of itself and of the previous
1404 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1405 the whole enclosing structure, even if it is unnamed; except that
1406 (iii) a zero-sized bit-field will be disregarded unless it follows
1407 another bit-field of nonzero size. If this hook returns @code{true},
1408 other macros that control bit-field layout are ignored.
1410 When a bit-field is inserted into a packed record, the whole size
1411 of the underlying type is used by one or more same-size adjacent
1412 bit-fields (that is, if its long:3, 32 bits is used in the record,
1413 and any additional adjacent long bit-fields are packed into the same
1414 chunk of 32 bits. However, if the size changes, a new field of that
1415 size is allocated). In an unpacked record, this is the same as using
1416 alignment, but not equivalent when packing.
1418 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1419 the latter will take precedence. If @samp{__attribute__((packed))} is
1420 used on a single field when MS bit-fields are in use, it will take
1421 precedence for that field, but the alignment of the rest of the structure
1422 may affect its placement.
1425 @deftypefn {Target Hook} {bool} TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1426 Returns true if the target supports decimal floating point.
1429 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_FUNDAMENTAL_TYPE (tree @var{type})
1430 If your target defines any fundamental types, define this hook to
1431 return the appropriate encoding for these types as part of a C++
1432 mangled name. The @var{type} argument is the tree structure
1433 representing the type to be mangled. The hook may be applied to trees
1434 which are not target-specific fundamental types; it should return
1435 @code{NULL} for all such types, as well as arguments it does not
1436 recognize. If the return value is not @code{NULL}, it must point to
1437 a statically-allocated string constant.
1439 Target-specific fundamental types might be new fundamental types or
1440 qualified versions of ordinary fundamental types. Encode new
1441 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1442 is the name used for the type in source code, and @var{n} is the
1443 length of @var{name} in decimal. Encode qualified versions of
1444 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1445 @var{name} is the name used for the type qualifier in source code,
1446 @var{n} is the length of @var{name} as above, and @var{code} is the
1447 code used to represent the unqualified version of this type. (See
1448 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1449 codes.) In both cases the spaces are for clarity; do not include any
1450 spaces in your string.
1452 The default version of this hook always returns @code{NULL}, which is
1453 appropriate for a target that does not define any new fundamental
1458 @section Layout of Source Language Data Types
1460 These macros define the sizes and other characteristics of the standard
1461 basic data types used in programs being compiled. Unlike the macros in
1462 the previous section, these apply to specific features of C and related
1463 languages, rather than to fundamental aspects of storage layout.
1465 @defmac INT_TYPE_SIZE
1466 A C expression for the size in bits of the type @code{int} on the
1467 target machine. If you don't define this, the default is one word.
1470 @defmac SHORT_TYPE_SIZE
1471 A C expression for the size in bits of the type @code{short} on the
1472 target machine. If you don't define this, the default is half a word.
1473 (If this would be less than one storage unit, it is rounded up to one
1477 @defmac LONG_TYPE_SIZE
1478 A C expression for the size in bits of the type @code{long} on the
1479 target machine. If you don't define this, the default is one word.
1482 @defmac ADA_LONG_TYPE_SIZE
1483 On some machines, the size used for the Ada equivalent of the type
1484 @code{long} by a native Ada compiler differs from that used by C@. In
1485 that situation, define this macro to be a C expression to be used for
1486 the size of that type. If you don't define this, the default is the
1487 value of @code{LONG_TYPE_SIZE}.
1490 @defmac LONG_LONG_TYPE_SIZE
1491 A C expression for the size in bits of the type @code{long long} on the
1492 target machine. If you don't define this, the default is two
1493 words. If you want to support GNU Ada on your machine, the value of this
1494 macro must be at least 64.
1497 @defmac CHAR_TYPE_SIZE
1498 A C expression for the size in bits of the type @code{char} on the
1499 target machine. If you don't define this, the default is
1500 @code{BITS_PER_UNIT}.
1503 @defmac BOOL_TYPE_SIZE
1504 A C expression for the size in bits of the C++ type @code{bool} and
1505 C99 type @code{_Bool} on the target machine. If you don't define
1506 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1509 @defmac FLOAT_TYPE_SIZE
1510 A C expression for the size in bits of the type @code{float} on the
1511 target machine. If you don't define this, the default is one word.
1514 @defmac DOUBLE_TYPE_SIZE
1515 A C expression for the size in bits of the type @code{double} on the
1516 target machine. If you don't define this, the default is two
1520 @defmac LONG_DOUBLE_TYPE_SIZE
1521 A C expression for the size in bits of the type @code{long double} on
1522 the target machine. If you don't define this, the default is two
1526 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1527 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1528 if you want routines in @file{libgcc2.a} for a size other than
1529 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1530 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1533 @defmac LIBGCC2_HAS_DF_MODE
1534 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1535 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1536 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1537 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1538 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1542 @defmac LIBGCC2_HAS_XF_MODE
1543 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1544 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1545 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1546 is 80 then the default is 1, otherwise it is 0.
1549 @defmac LIBGCC2_HAS_TF_MODE
1550 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1551 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1552 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1553 is 128 then the default is 1, otherwise it is 0.
1560 Define these macros to be the size in bits of the mantissa of
1561 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1562 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1563 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1564 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1565 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1566 @code{LIBGCC2_DOUBLE_TYPE_SIZE} or
1567 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1570 @defmac TARGET_FLT_EVAL_METHOD
1571 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1572 assuming, if applicable, that the floating-point control word is in its
1573 default state. If you do not define this macro the value of
1574 @code{FLT_EVAL_METHOD} will be zero.
1577 @defmac WIDEST_HARDWARE_FP_SIZE
1578 A C expression for the size in bits of the widest floating-point format
1579 supported by the hardware. If you define this macro, you must specify a
1580 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1581 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1585 @defmac DEFAULT_SIGNED_CHAR
1586 An expression whose value is 1 or 0, according to whether the type
1587 @code{char} should be signed or unsigned by default. The user can
1588 always override this default with the options @option{-fsigned-char}
1589 and @option{-funsigned-char}.
1592 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1593 This target hook should return true if the compiler should give an
1594 @code{enum} type only as many bytes as it takes to represent the range
1595 of possible values of that type. It should return false if all
1596 @code{enum} types should be allocated like @code{int}.
1598 The default is to return false.
1602 A C expression for a string describing the name of the data type to use
1603 for size values. The typedef name @code{size_t} is defined using the
1604 contents of the string.
1606 The string can contain more than one keyword. If so, separate them with
1607 spaces, and write first any length keyword, then @code{unsigned} if
1608 appropriate, and finally @code{int}. The string must exactly match one
1609 of the data type names defined in the function
1610 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1611 omit @code{int} or change the order---that would cause the compiler to
1614 If you don't define this macro, the default is @code{"long unsigned
1618 @defmac PTRDIFF_TYPE
1619 A C expression for a string describing the name of the data type to use
1620 for the result of subtracting two pointers. The typedef name
1621 @code{ptrdiff_t} is defined using the contents of the string. See
1622 @code{SIZE_TYPE} above for more information.
1624 If you don't define this macro, the default is @code{"long int"}.
1628 A C expression for a string describing the name of the data type to use
1629 for wide characters. The typedef name @code{wchar_t} is defined using
1630 the contents of the string. See @code{SIZE_TYPE} above for more
1633 If you don't define this macro, the default is @code{"int"}.
1636 @defmac WCHAR_TYPE_SIZE
1637 A C expression for the size in bits of the data type for wide
1638 characters. This is used in @code{cpp}, which cannot make use of
1643 A C expression for a string describing the name of the data type to
1644 use for wide characters passed to @code{printf} and returned from
1645 @code{getwc}. The typedef name @code{wint_t} is defined using the
1646 contents of the string. See @code{SIZE_TYPE} above for more
1649 If you don't define this macro, the default is @code{"unsigned int"}.
1653 A C expression for a string describing the name of the data type that
1654 can represent any value of any standard or extended signed integer type.
1655 The typedef name @code{intmax_t} is defined using the contents of the
1656 string. See @code{SIZE_TYPE} above for more information.
1658 If you don't define this macro, the default is the first of
1659 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1660 much precision as @code{long long int}.
1663 @defmac UINTMAX_TYPE
1664 A C expression for a string describing the name of the data type that
1665 can represent any value of any standard or extended unsigned integer
1666 type. The typedef name @code{uintmax_t} is defined using the contents
1667 of the string. See @code{SIZE_TYPE} above for more information.
1669 If you don't define this macro, the default is the first of
1670 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1671 unsigned int"} that has as much precision as @code{long long unsigned
1675 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1676 The C++ compiler represents a pointer-to-member-function with a struct
1683 ptrdiff_t vtable_index;
1690 The C++ compiler must use one bit to indicate whether the function that
1691 will be called through a pointer-to-member-function is virtual.
1692 Normally, we assume that the low-order bit of a function pointer must
1693 always be zero. Then, by ensuring that the vtable_index is odd, we can
1694 distinguish which variant of the union is in use. But, on some
1695 platforms function pointers can be odd, and so this doesn't work. In
1696 that case, we use the low-order bit of the @code{delta} field, and shift
1697 the remainder of the @code{delta} field to the left.
1699 GCC will automatically make the right selection about where to store
1700 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1701 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1702 set such that functions always start at even addresses, but the lowest
1703 bit of pointers to functions indicate whether the function at that
1704 address is in ARM or Thumb mode. If this is the case of your
1705 architecture, you should define this macro to
1706 @code{ptrmemfunc_vbit_in_delta}.
1708 In general, you should not have to define this macro. On architectures
1709 in which function addresses are always even, according to
1710 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1711 @code{ptrmemfunc_vbit_in_pfn}.
1714 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1715 Normally, the C++ compiler uses function pointers in vtables. This
1716 macro allows the target to change to use ``function descriptors''
1717 instead. Function descriptors are found on targets for whom a
1718 function pointer is actually a small data structure. Normally the
1719 data structure consists of the actual code address plus a data
1720 pointer to which the function's data is relative.
1722 If vtables are used, the value of this macro should be the number
1723 of words that the function descriptor occupies.
1726 @defmac TARGET_VTABLE_ENTRY_ALIGN
1727 By default, the vtable entries are void pointers, the so the alignment
1728 is the same as pointer alignment. The value of this macro specifies
1729 the alignment of the vtable entry in bits. It should be defined only
1730 when special alignment is necessary. */
1733 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1734 There are a few non-descriptor entries in the vtable at offsets below
1735 zero. If these entries must be padded (say, to preserve the alignment
1736 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1737 of words in each data entry.
1741 @section Register Usage
1742 @cindex register usage
1744 This section explains how to describe what registers the target machine
1745 has, and how (in general) they can be used.
1747 The description of which registers a specific instruction can use is
1748 done with register classes; see @ref{Register Classes}. For information
1749 on using registers to access a stack frame, see @ref{Frame Registers}.
1750 For passing values in registers, see @ref{Register Arguments}.
1751 For returning values in registers, see @ref{Scalar Return}.
1754 * Register Basics:: Number and kinds of registers.
1755 * Allocation Order:: Order in which registers are allocated.
1756 * Values in Registers:: What kinds of values each reg can hold.
1757 * Leaf Functions:: Renumbering registers for leaf functions.
1758 * Stack Registers:: Handling a register stack such as 80387.
1761 @node Register Basics
1762 @subsection Basic Characteristics of Registers
1764 @c prevent bad page break with this line
1765 Registers have various characteristics.
1767 @defmac FIRST_PSEUDO_REGISTER
1768 Number of hardware registers known to the compiler. They receive
1769 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1770 pseudo register's number really is assigned the number
1771 @code{FIRST_PSEUDO_REGISTER}.
1774 @defmac FIXED_REGISTERS
1775 @cindex fixed register
1776 An initializer that says which registers are used for fixed purposes
1777 all throughout the compiled code and are therefore not available for
1778 general allocation. These would include the stack pointer, the frame
1779 pointer (except on machines where that can be used as a general
1780 register when no frame pointer is needed), the program counter on
1781 machines where that is considered one of the addressable registers,
1782 and any other numbered register with a standard use.
1784 This information is expressed as a sequence of numbers, separated by
1785 commas and surrounded by braces. The @var{n}th number is 1 if
1786 register @var{n} is fixed, 0 otherwise.
1788 The table initialized from this macro, and the table initialized by
1789 the following one, may be overridden at run time either automatically,
1790 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1791 the user with the command options @option{-ffixed-@var{reg}},
1792 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1795 @defmac CALL_USED_REGISTERS
1796 @cindex call-used register
1797 @cindex call-clobbered register
1798 @cindex call-saved register
1799 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1800 clobbered (in general) by function calls as well as for fixed
1801 registers. This macro therefore identifies the registers that are not
1802 available for general allocation of values that must live across
1805 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1806 automatically saves it on function entry and restores it on function
1807 exit, if the register is used within the function.
1810 @defmac CALL_REALLY_USED_REGISTERS
1811 @cindex call-used register
1812 @cindex call-clobbered register
1813 @cindex call-saved register
1814 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1815 that the entire set of @code{FIXED_REGISTERS} be included.
1816 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1817 This macro is optional. If not specified, it defaults to the value
1818 of @code{CALL_USED_REGISTERS}.
1821 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1822 @cindex call-used register
1823 @cindex call-clobbered register
1824 @cindex call-saved register
1825 A C expression that is nonzero if it is not permissible to store a
1826 value of mode @var{mode} in hard register number @var{regno} across a
1827 call without some part of it being clobbered. For most machines this
1828 macro need not be defined. It is only required for machines that do not
1829 preserve the entire contents of a register across a call.
1833 @findex call_used_regs
1836 @findex reg_class_contents
1837 @defmac CONDITIONAL_REGISTER_USAGE
1838 Zero or more C statements that may conditionally modify five variables
1839 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1840 @code{reg_names}, and @code{reg_class_contents}, to take into account
1841 any dependence of these register sets on target flags. The first three
1842 of these are of type @code{char []} (interpreted as Boolean vectors).
1843 @code{global_regs} is a @code{const char *[]}, and
1844 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1845 called, @code{fixed_regs}, @code{call_used_regs},
1846 @code{reg_class_contents}, and @code{reg_names} have been initialized
1847 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1848 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1849 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1850 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1851 command options have been applied.
1853 You need not define this macro if it has no work to do.
1855 @cindex disabling certain registers
1856 @cindex controlling register usage
1857 If the usage of an entire class of registers depends on the target
1858 flags, you may indicate this to GCC by using this macro to modify
1859 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1860 registers in the classes which should not be used by GCC@. Also define
1861 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1862 to return @code{NO_REGS} if it
1863 is called with a letter for a class that shouldn't be used.
1865 (However, if this class is not included in @code{GENERAL_REGS} and all
1866 of the insn patterns whose constraints permit this class are
1867 controlled by target switches, then GCC will automatically avoid using
1868 these registers when the target switches are opposed to them.)
1871 @defmac INCOMING_REGNO (@var{out})
1872 Define this macro if the target machine has register windows. This C
1873 expression returns the register number as seen by the called function
1874 corresponding to the register number @var{out} as seen by the calling
1875 function. Return @var{out} if register number @var{out} is not an
1879 @defmac OUTGOING_REGNO (@var{in})
1880 Define this macro if the target machine has register windows. This C
1881 expression returns the register number as seen by the calling function
1882 corresponding to the register number @var{in} as seen by the called
1883 function. Return @var{in} if register number @var{in} is not an inbound
1887 @defmac LOCAL_REGNO (@var{regno})
1888 Define this macro if the target machine has register windows. This C
1889 expression returns true if the register is call-saved but is in the
1890 register window. Unlike most call-saved registers, such registers
1891 need not be explicitly restored on function exit or during non-local
1896 If the program counter has a register number, define this as that
1897 register number. Otherwise, do not define it.
1900 @node Allocation Order
1901 @subsection Order of Allocation of Registers
1902 @cindex order of register allocation
1903 @cindex register allocation order
1905 @c prevent bad page break with this line
1906 Registers are allocated in order.
1908 @defmac REG_ALLOC_ORDER
1909 If defined, an initializer for a vector of integers, containing the
1910 numbers of hard registers in the order in which GCC should prefer
1911 to use them (from most preferred to least).
1913 If this macro is not defined, registers are used lowest numbered first
1914 (all else being equal).
1916 One use of this macro is on machines where the highest numbered
1917 registers must always be saved and the save-multiple-registers
1918 instruction supports only sequences of consecutive registers. On such
1919 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1920 the highest numbered allocable register first.
1923 @defmac ORDER_REGS_FOR_LOCAL_ALLOC
1924 A C statement (sans semicolon) to choose the order in which to allocate
1925 hard registers for pseudo-registers local to a basic block.
1927 Store the desired register order in the array @code{reg_alloc_order}.
1928 Element 0 should be the register to allocate first; element 1, the next
1929 register; and so on.
1931 The macro body should not assume anything about the contents of
1932 @code{reg_alloc_order} before execution of the macro.
1934 On most machines, it is not necessary to define this macro.
1937 @node Values in Registers
1938 @subsection How Values Fit in Registers
1940 This section discusses the macros that describe which kinds of values
1941 (specifically, which machine modes) each register can hold, and how many
1942 consecutive registers are needed for a given mode.
1944 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
1945 A C expression for the number of consecutive hard registers, starting
1946 at register number @var{regno}, required to hold a value of mode
1949 On a machine where all registers are exactly one word, a suitable
1950 definition of this macro is
1953 #define HARD_REGNO_NREGS(REGNO, MODE) \
1954 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
1959 @defmac REGMODE_NATURAL_SIZE (@var{mode})
1960 Define this macro if the natural size of registers that hold values
1961 of mode @var{mode} is not the word size. It is a C expression that
1962 should give the natural size in bytes for the specified mode. It is
1963 used by the register allocator to try to optimize its results. This
1964 happens for example on SPARC 64-bit where the natural size of
1965 floating-point registers is still 32-bit.
1968 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
1969 A C expression that is nonzero if it is permissible to store a value
1970 of mode @var{mode} in hard register number @var{regno} (or in several
1971 registers starting with that one). For a machine where all registers
1972 are equivalent, a suitable definition is
1975 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
1978 You need not include code to check for the numbers of fixed registers,
1979 because the allocation mechanism considers them to be always occupied.
1981 @cindex register pairs
1982 On some machines, double-precision values must be kept in even/odd
1983 register pairs. You can implement that by defining this macro to reject
1984 odd register numbers for such modes.
1986 The minimum requirement for a mode to be OK in a register is that the
1987 @samp{mov@var{mode}} instruction pattern support moves between the
1988 register and other hard register in the same class and that moving a
1989 value into the register and back out not alter it.
1991 Since the same instruction used to move @code{word_mode} will work for
1992 all narrower integer modes, it is not necessary on any machine for
1993 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
1994 you define patterns @samp{movhi}, etc., to take advantage of this. This
1995 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
1996 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
1999 Many machines have special registers for floating point arithmetic.
2000 Often people assume that floating point machine modes are allowed only
2001 in floating point registers. This is not true. Any registers that
2002 can hold integers can safely @emph{hold} a floating point machine
2003 mode, whether or not floating arithmetic can be done on it in those
2004 registers. Integer move instructions can be used to move the values.
2006 On some machines, though, the converse is true: fixed-point machine
2007 modes may not go in floating registers. This is true if the floating
2008 registers normalize any value stored in them, because storing a
2009 non-floating value there would garble it. In this case,
2010 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2011 floating registers. But if the floating registers do not automatically
2012 normalize, if you can store any bit pattern in one and retrieve it
2013 unchanged without a trap, then any machine mode may go in a floating
2014 register, so you can define this macro to say so.
2016 The primary significance of special floating registers is rather that
2017 they are the registers acceptable in floating point arithmetic
2018 instructions. However, this is of no concern to
2019 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2020 constraints for those instructions.
2022 On some machines, the floating registers are especially slow to access,
2023 so that it is better to store a value in a stack frame than in such a
2024 register if floating point arithmetic is not being done. As long as the
2025 floating registers are not in class @code{GENERAL_REGS}, they will not
2026 be used unless some pattern's constraint asks for one.
2029 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2030 A C expression that is nonzero if it is OK to rename a hard register
2031 @var{from} to another hard register @var{to}.
2033 One common use of this macro is to prevent renaming of a register to
2034 another register that is not saved by a prologue in an interrupt
2037 The default is always nonzero.
2040 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2041 A C expression that is nonzero if a value of mode
2042 @var{mode1} is accessible in mode @var{mode2} without copying.
2044 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2045 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2046 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2047 should be nonzero. If they differ for any @var{r}, you should define
2048 this macro to return zero unless some other mechanism ensures the
2049 accessibility of the value in a narrower mode.
2051 You should define this macro to return nonzero in as many cases as
2052 possible since doing so will allow GCC to perform better register
2056 @defmac AVOID_CCMODE_COPIES
2057 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2058 registers. You should only define this macro if support for copying to/from
2059 @code{CCmode} is incomplete.
2062 @node Leaf Functions
2063 @subsection Handling Leaf Functions
2065 @cindex leaf functions
2066 @cindex functions, leaf
2067 On some machines, a leaf function (i.e., one which makes no calls) can run
2068 more efficiently if it does not make its own register window. Often this
2069 means it is required to receive its arguments in the registers where they
2070 are passed by the caller, instead of the registers where they would
2073 The special treatment for leaf functions generally applies only when
2074 other conditions are met; for example, often they may use only those
2075 registers for its own variables and temporaries. We use the term ``leaf
2076 function'' to mean a function that is suitable for this special
2077 handling, so that functions with no calls are not necessarily ``leaf
2080 GCC assigns register numbers before it knows whether the function is
2081 suitable for leaf function treatment. So it needs to renumber the
2082 registers in order to output a leaf function. The following macros
2085 @defmac LEAF_REGISTERS
2086 Name of a char vector, indexed by hard register number, which
2087 contains 1 for a register that is allowable in a candidate for leaf
2090 If leaf function treatment involves renumbering the registers, then the
2091 registers marked here should be the ones before renumbering---those that
2092 GCC would ordinarily allocate. The registers which will actually be
2093 used in the assembler code, after renumbering, should not be marked with 1
2096 Define this macro only if the target machine offers a way to optimize
2097 the treatment of leaf functions.
2100 @defmac LEAF_REG_REMAP (@var{regno})
2101 A C expression whose value is the register number to which @var{regno}
2102 should be renumbered, when a function is treated as a leaf function.
2104 If @var{regno} is a register number which should not appear in a leaf
2105 function before renumbering, then the expression should yield @minus{}1, which
2106 will cause the compiler to abort.
2108 Define this macro only if the target machine offers a way to optimize the
2109 treatment of leaf functions, and registers need to be renumbered to do
2113 @findex current_function_is_leaf
2114 @findex current_function_uses_only_leaf_regs
2115 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2116 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2117 specially. They can test the C variable @code{current_function_is_leaf}
2118 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2119 set prior to local register allocation and is valid for the remaining
2120 compiler passes. They can also test the C variable
2121 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2122 functions which only use leaf registers.
2123 @code{current_function_uses_only_leaf_regs} is valid after all passes
2124 that modify the instructions have been run and is only useful if
2125 @code{LEAF_REGISTERS} is defined.
2126 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2127 @c of the next paragraph?! --mew 2feb93
2129 @node Stack Registers
2130 @subsection Registers That Form a Stack
2132 There are special features to handle computers where some of the
2133 ``registers'' form a stack. Stack registers are normally written by
2134 pushing onto the stack, and are numbered relative to the top of the
2137 Currently, GCC can only handle one group of stack-like registers, and
2138 they must be consecutively numbered. Furthermore, the existing
2139 support for stack-like registers is specific to the 80387 floating
2140 point coprocessor. If you have a new architecture that uses
2141 stack-like registers, you will need to do substantial work on
2142 @file{reg-stack.c} and write your machine description to cooperate
2143 with it, as well as defining these macros.
2146 Define this if the machine has any stack-like registers.
2149 @defmac FIRST_STACK_REG
2150 The number of the first stack-like register. This one is the top
2154 @defmac LAST_STACK_REG
2155 The number of the last stack-like register. This one is the bottom of
2159 @node Register Classes
2160 @section Register Classes
2161 @cindex register class definitions
2162 @cindex class definitions, register
2164 On many machines, the numbered registers are not all equivalent.
2165 For example, certain registers may not be allowed for indexed addressing;
2166 certain registers may not be allowed in some instructions. These machine
2167 restrictions are described to the compiler using @dfn{register classes}.
2169 You define a number of register classes, giving each one a name and saying
2170 which of the registers belong to it. Then you can specify register classes
2171 that are allowed as operands to particular instruction patterns.
2175 In general, each register will belong to several classes. In fact, one
2176 class must be named @code{ALL_REGS} and contain all the registers. Another
2177 class must be named @code{NO_REGS} and contain no registers. Often the
2178 union of two classes will be another class; however, this is not required.
2180 @findex GENERAL_REGS
2181 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2182 terribly special about the name, but the operand constraint letters
2183 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2184 the same as @code{ALL_REGS}, just define it as a macro which expands
2187 Order the classes so that if class @var{x} is contained in class @var{y}
2188 then @var{x} has a lower class number than @var{y}.
2190 The way classes other than @code{GENERAL_REGS} are specified in operand
2191 constraints is through machine-dependent operand constraint letters.
2192 You can define such letters to correspond to various classes, then use
2193 them in operand constraints.
2195 You should define a class for the union of two classes whenever some
2196 instruction allows both classes. For example, if an instruction allows
2197 either a floating point (coprocessor) register or a general register for a
2198 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2199 which includes both of them. Otherwise you will get suboptimal code.
2201 You must also specify certain redundant information about the register
2202 classes: for each class, which classes contain it and which ones are
2203 contained in it; for each pair of classes, the largest class contained
2206 When a value occupying several consecutive registers is expected in a
2207 certain class, all the registers used must belong to that class.
2208 Therefore, register classes cannot be used to enforce a requirement for
2209 a register pair to start with an even-numbered register. The way to
2210 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2212 Register classes used for input-operands of bitwise-and or shift
2213 instructions have a special requirement: each such class must have, for
2214 each fixed-point machine mode, a subclass whose registers can transfer that
2215 mode to or from memory. For example, on some machines, the operations for
2216 single-byte values (@code{QImode}) are limited to certain registers. When
2217 this is so, each register class that is used in a bitwise-and or shift
2218 instruction must have a subclass consisting of registers from which
2219 single-byte values can be loaded or stored. This is so that
2220 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2222 @deftp {Data type} {enum reg_class}
2223 An enumerated type that must be defined with all the register class names
2224 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2225 must be the last register class, followed by one more enumerated value,
2226 @code{LIM_REG_CLASSES}, which is not a register class but rather
2227 tells how many classes there are.
2229 Each register class has a number, which is the value of casting
2230 the class name to type @code{int}. The number serves as an index
2231 in many of the tables described below.
2234 @defmac N_REG_CLASSES
2235 The number of distinct register classes, defined as follows:
2238 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2242 @defmac REG_CLASS_NAMES
2243 An initializer containing the names of the register classes as C string
2244 constants. These names are used in writing some of the debugging dumps.
2247 @defmac REG_CLASS_CONTENTS
2248 An initializer containing the contents of the register classes, as integers
2249 which are bit masks. The @var{n}th integer specifies the contents of class
2250 @var{n}. The way the integer @var{mask} is interpreted is that
2251 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2253 When the machine has more than 32 registers, an integer does not suffice.
2254 Then the integers are replaced by sub-initializers, braced groupings containing
2255 several integers. Each sub-initializer must be suitable as an initializer
2256 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2257 In this situation, the first integer in each sub-initializer corresponds to
2258 registers 0 through 31, the second integer to registers 32 through 63, and
2262 @defmac REGNO_REG_CLASS (@var{regno})
2263 A C expression whose value is a register class containing hard register
2264 @var{regno}. In general there is more than one such class; choose a class
2265 which is @dfn{minimal}, meaning that no smaller class also contains the
2269 @defmac BASE_REG_CLASS
2270 A macro whose definition is the name of the class to which a valid
2271 base register must belong. A base register is one used in an address
2272 which is the register value plus a displacement.
2275 @defmac MODE_BASE_REG_CLASS (@var{mode})
2276 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2277 the selection of a base register in a mode dependent manner. If
2278 @var{mode} is VOIDmode then it should return the same value as
2279 @code{BASE_REG_CLASS}.
2282 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2283 A C expression whose value is the register class to which a valid
2284 base register must belong in order to be used in a base plus index
2285 register address. You should define this macro if base plus index
2286 addresses have different requirements than other base register uses.
2289 @defmac INDEX_REG_CLASS
2290 A macro whose definition is the name of the class to which a valid
2291 index register must belong. An index register is one used in an
2292 address where its value is either multiplied by a scale factor or
2293 added to another register (as well as added to a displacement).
2296 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2297 For the constraint at the start of @var{str}, which starts with the letter
2298 @var{c}, return the length. This allows you to have register class /
2299 constant / extra constraints that are longer than a single letter;
2300 you don't need to define this macro if you can do with single-letter
2301 constraints only. The definition of this macro should use
2302 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2303 to handle specially.
2304 There are some sanity checks in genoutput.c that check the constraint lengths
2305 for the md file, so you can also use this macro to help you while you are
2306 transitioning from a byzantine single-letter-constraint scheme: when you
2307 return a negative length for a constraint you want to re-use, genoutput
2308 will complain about every instance where it is used in the md file.
2311 @defmac REG_CLASS_FROM_LETTER (@var{char})
2312 A C expression which defines the machine-dependent operand constraint
2313 letters for register classes. If @var{char} is such a letter, the
2314 value should be the register class corresponding to it. Otherwise,
2315 the value should be @code{NO_REGS}. The register letter @samp{r},
2316 corresponding to class @code{GENERAL_REGS}, will not be passed
2317 to this macro; you do not need to handle it.
2320 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2321 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2322 passed in @var{str}, so that you can use suffixes to distinguish between
2326 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2327 A C expression which is nonzero if register number @var{num} is
2328 suitable for use as a base register in operand addresses. It may be
2329 either a suitable hard register or a pseudo register that has been
2330 allocated such a hard register.
2333 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2334 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2335 that expression may examine the mode of the memory reference in
2336 @var{mode}. You should define this macro if the mode of the memory
2337 reference affects whether a register may be used as a base register. If
2338 you define this macro, the compiler will use it instead of
2339 @code{REGNO_OK_FOR_BASE_P}.
2342 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2343 A C expression which is nonzero if register number @var{num} is suitable for
2344 use as a base register in base plus index operand addresses, accessing
2345 memory in mode @var{mode}. It may be either a suitable hard register or a
2346 pseudo register that has been allocated such a hard register. You should
2347 define this macro if base plus index addresses have different requirements
2348 than other base register uses.
2351 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2352 A C expression which is nonzero if register number @var{num} is
2353 suitable for use as an index register in operand addresses. It may be
2354 either a suitable hard register or a pseudo register that has been
2355 allocated such a hard register.
2357 The difference between an index register and a base register is that
2358 the index register may be scaled. If an address involves the sum of
2359 two registers, neither one of them scaled, then either one may be
2360 labeled the ``base'' and the other the ``index''; but whichever
2361 labeling is used must fit the machine's constraints of which registers
2362 may serve in each capacity. The compiler will try both labelings,
2363 looking for one that is valid, and will reload one or both registers
2364 only if neither labeling works.
2367 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2368 A C expression that places additional restrictions on the register class
2369 to use when it is necessary to copy value @var{x} into a register in class
2370 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2371 another, smaller class. On many machines, the following definition is
2375 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2378 Sometimes returning a more restrictive class makes better code. For
2379 example, on the 68000, when @var{x} is an integer constant that is in range
2380 for a @samp{moveq} instruction, the value of this macro is always
2381 @code{DATA_REGS} as long as @var{class} includes the data registers.
2382 Requiring a data register guarantees that a @samp{moveq} will be used.
2384 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2385 @var{class} is if @var{x} is a legitimate constant which cannot be
2386 loaded into some register class. By returning @code{NO_REGS} you can
2387 force @var{x} into a memory location. For example, rs6000 can load
2388 immediate values into general-purpose registers, but does not have an
2389 instruction for loading an immediate value into a floating-point
2390 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2391 @var{x} is a floating-point constant. If the constant can't be loaded
2392 into any kind of register, code generation will be better if
2393 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2394 of using @code{PREFERRED_RELOAD_CLASS}.
2397 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2398 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2399 input reloads. If you don't define this macro, the default is to use
2400 @var{class}, unchanged.
2403 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2404 A C expression that places additional restrictions on the register class
2405 to use when it is necessary to be able to hold a value of mode
2406 @var{mode} in a reload register for which class @var{class} would
2409 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2410 there are certain modes that simply can't go in certain reload classes.
2412 The value is a register class; perhaps @var{class}, or perhaps another,
2415 Don't define this macro unless the target machine has limitations which
2416 require the macro to do something nontrivial.
2419 @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})
2420 Many machines have some registers that cannot be copied directly to or
2421 from memory or even from other types of registers. An example is the
2422 @samp{MQ} register, which on most machines, can only be copied to or
2423 from general registers, but not memory. Below, we shall be using the
2424 term 'intermediate register' when a move operation cannot be performed
2425 directly, but has to be done by copying the source into the intermediate
2426 register first, and then copying the intermediate register to the
2427 destination. An intermediate register always has the same mode as
2428 source and destination. Since it holds the actual value being copied,
2429 reload might apply optimizations to re-use an intermediate register
2430 and eliding the copy from the source when it can determine that the
2431 intermediate register still holds the required value.
2433 Another kind of secondary reload is required on some machines which
2434 allow copying all registers to and from memory, but require a scratch
2435 register for stores to some memory locations (e.g., those with symbolic
2436 address on the RT, and those with certain symbolic address on the SPARC
2437 when compiling PIC)@. Scratch registers need not have the same mode
2438 as the value being copied, and usually hold a different value that
2439 that being copied. Special patterns in the md file are needed to
2440 describe how the copy is performed with the help of the scratch register;
2441 these patterns also describe the number, register class(es) and mode(s)
2442 of the scratch register(s).
2444 In some cases, both an intermediate and a scratch register are required.
2446 For input reloads, this target hook is called with nonzero @var{in_p},
2447 and @var{x} is an rtx that needs to be copied to a register in of class
2448 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2449 hook is called with zero @var{in_p}, and a register of class @var{reload_mode}
2450 needs to be copied to rtx @var{x} in @var{reload_mode}.
2452 If copying a register of @var{reload_class} from/to @var{x} requires
2453 an intermediate register, the hook @code{secondary_reload} should
2454 return the register class required for this intermediate register.
2455 If no intermediate register is required, it should return NO_REGS.
2456 If more than one intermediate register is required, describe the one
2457 that is closest in the copy chain to the reload register.
2459 If scratch registers are needed, you also have to describe how to
2460 perform the copy from/to the reload register to/from this
2461 closest intermediate register. Or if no intermediate register is
2462 required, but still a scratch register is needed, describe the
2463 copy from/to the reload register to/from the reload operand @var{x}.
2465 You do this by setting @code{sri->icode} to the instruction code of a pattern
2466 in the md file which performs the move. Operands 0 and 1 are the output
2467 and input of this copy, respectively. Operands from operand 2 onward are
2468 for scratch operands. These scratch operands must have a mode, and a
2469 single-register-class
2470 @c [later: or memory]
2473 When an intermediate register is used, the @code{secondary_reload}
2474 hook will be called again to determine how to copy the intermediate
2475 register to/from the reload operand @var{x}, so your hook must also
2476 have code to handle the register class of the intermediate operand.
2478 @c [For later: maybe we'll allow multi-alternative reload patterns -
2479 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2480 @c and match the constraints of input and output to determine the required
2481 @c alternative. A restriction would be that constraints used to match
2482 @c against reloads registers would have to be written as register class
2483 @c constraints, or we need a new target macro / hook that tells us if an
2484 @c arbitrary constraint can match an unknown register of a given class.
2485 @c Such a macro / hook would also be useful in other places.]
2488 @var{x} might be a pseudo-register or a @code{subreg} of a
2489 pseudo-register, which could either be in a hard register or in memory.
2490 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2491 in memory and the hard register number if it is in a register.
2493 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2494 currently not supported. For the time being, you will have to continue
2495 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2497 @code{copy_cost} also uses this target hook to find out how values are
2498 copied. If you want it to include some extra cost for the need to allocate
2499 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2500 Or if two dependent moves are supposed to have a lower cost than the sum
2501 of the individual moves due to expected fortuitous scheduling and/or special
2502 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2505 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2506 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2507 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2508 These macros are obsolete, new ports should use the target hook
2509 @code{TARGET_SECONDARY_RELOAD} instead.
2511 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2512 target hook. Older ports still define these macros to indicate to the
2513 reload phase that it may
2514 need to allocate at least one register for a reload in addition to the
2515 register to contain the data. Specifically, if copying @var{x} to a
2516 register @var{class} in @var{mode} requires an intermediate register,
2517 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2518 largest register class all of whose registers can be used as
2519 intermediate registers or scratch registers.
2521 If copying a register @var{class} in @var{mode} to @var{x} requires an
2522 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2523 was supposed to be defined be defined to return the largest register
2524 class required. If the
2525 requirements for input and output reloads were the same, the macro
2526 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2529 The values returned by these macros are often @code{GENERAL_REGS}.
2530 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2531 can be directly copied to or from a register of @var{class} in
2532 @var{mode} without requiring a scratch register. Do not define this
2533 macro if it would always return @code{NO_REGS}.
2535 If a scratch register is required (either with or without an
2536 intermediate register), you were supposed to define patterns for
2537 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2538 (@pxref{Standard Names}. These patterns, which were normally
2539 implemented with a @code{define_expand}, should be similar to the
2540 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2543 These patterns need constraints for the reload register and scratch
2545 contain a single register class. If the original reload register (whose
2546 class is @var{class}) can meet the constraint given in the pattern, the
2547 value returned by these macros is used for the class of the scratch
2548 register. Otherwise, two additional reload registers are required.
2549 Their classes are obtained from the constraints in the insn pattern.
2551 @var{x} might be a pseudo-register or a @code{subreg} of a
2552 pseudo-register, which could either be in a hard register or in memory.
2553 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2554 in memory and the hard register number if it is in a register.
2556 These macros should not be used in the case where a particular class of
2557 registers can only be copied to memory and not to another class of
2558 registers. In that case, secondary reload registers are not needed and
2559 would not be helpful. Instead, a stack location must be used to perform
2560 the copy and the @code{mov@var{m}} pattern should use memory as an
2561 intermediate storage. This case often occurs between floating-point and
2565 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2566 Certain machines have the property that some registers cannot be copied
2567 to some other registers without using memory. Define this macro on
2568 those machines to be a C expression that is nonzero if objects of mode
2569 @var{m} in registers of @var{class1} can only be copied to registers of
2570 class @var{class2} by storing a register of @var{class1} into memory
2571 and loading that memory location into a register of @var{class2}.
2573 Do not define this macro if its value would always be zero.
2576 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2577 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2578 allocates a stack slot for a memory location needed for register copies.
2579 If this macro is defined, the compiler instead uses the memory location
2580 defined by this macro.
2582 Do not define this macro if you do not define
2583 @code{SECONDARY_MEMORY_NEEDED}.
2586 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2587 When the compiler needs a secondary memory location to copy between two
2588 registers of mode @var{mode}, it normally allocates sufficient memory to
2589 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2590 load operations in a mode that many bits wide and whose class is the
2591 same as that of @var{mode}.
2593 This is right thing to do on most machines because it ensures that all
2594 bits of the register are copied and prevents accesses to the registers
2595 in a narrower mode, which some machines prohibit for floating-point
2598 However, this default behavior is not correct on some machines, such as
2599 the DEC Alpha, that store short integers in floating-point registers
2600 differently than in integer registers. On those machines, the default
2601 widening will not work correctly and you must define this macro to
2602 suppress that widening in some cases. See the file @file{alpha.h} for
2605 Do not define this macro if you do not define
2606 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2607 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2610 @defmac SMALL_REGISTER_CLASSES
2611 On some machines, it is risky to let hard registers live across arbitrary
2612 insns. Typically, these machines have instructions that require values
2613 to be in specific registers (like an accumulator), and reload will fail
2614 if the required hard register is used for another purpose across such an
2617 Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero
2618 value on these machines. When this macro has a nonzero value, the
2619 compiler will try to minimize the lifetime of hard registers.
2621 It is always safe to define this macro with a nonzero value, but if you
2622 unnecessarily define it, you will reduce the amount of optimizations
2623 that can be performed in some cases. If you do not define this macro
2624 with a nonzero value when it is required, the compiler will run out of
2625 spill registers and print a fatal error message. For most machines, you
2626 should not define this macro at all.
2629 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2630 A C expression whose value is nonzero if pseudos that have been assigned
2631 to registers of class @var{class} would likely be spilled because
2632 registers of @var{class} are needed for spill registers.
2634 The default value of this macro returns 1 if @var{class} has exactly one
2635 register and zero otherwise. On most machines, this default should be
2636 used. Only define this macro to some other expression if pseudos
2637 allocated by @file{local-alloc.c} end up in memory because their hard
2638 registers were needed for spill registers. If this macro returns nonzero
2639 for those classes, those pseudos will only be allocated by
2640 @file{global.c}, which knows how to reallocate the pseudo to another
2641 register. If there would not be another register available for
2642 reallocation, you should not change the definition of this macro since
2643 the only effect of such a definition would be to slow down register
2647 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2648 A C expression for the maximum number of consecutive registers
2649 of class @var{class} needed to hold a value of mode @var{mode}.
2651 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2652 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2653 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2654 @var{mode})} for all @var{regno} values in the class @var{class}.
2656 This macro helps control the handling of multiple-word values
2660 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2661 If defined, a C expression that returns nonzero for a @var{class} for which
2662 a change from mode @var{from} to mode @var{to} is invalid.
2664 For the example, loading 32-bit integer or floating-point objects into
2665 floating-point registers on the Alpha extends them to 64 bits.
2666 Therefore loading a 64-bit object and then storing it as a 32-bit object
2667 does not store the low-order 32 bits, as would be the case for a normal
2668 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2672 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2673 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2674 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2678 Three other special macros describe which operands fit which constraint
2681 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2682 A C expression that defines the machine-dependent operand constraint
2683 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2684 particular ranges of integer values. If @var{c} is one of those
2685 letters, the expression should check that @var{value}, an integer, is in
2686 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2687 not one of those letters, the value should be 0 regardless of
2691 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2692 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2693 string passed in @var{str}, so that you can use suffixes to distinguish
2694 between different variants.
2697 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2698 A C expression that defines the machine-dependent operand constraint
2699 letters that specify particular ranges of @code{const_double} values
2700 (@samp{G} or @samp{H}).
2702 If @var{c} is one of those letters, the expression should check that
2703 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2704 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2705 letters, the value should be 0 regardless of @var{value}.
2707 @code{const_double} is used for all floating-point constants and for
2708 @code{DImode} fixed-point constants. A given letter can accept either
2709 or both kinds of values. It can use @code{GET_MODE} to distinguish
2710 between these kinds.
2713 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2714 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2715 string passed in @var{str}, so that you can use suffixes to distinguish
2716 between different variants.
2719 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2720 A C expression that defines the optional machine-dependent constraint
2721 letters that can be used to segregate specific types of operands, usually
2722 memory references, for the target machine. Any letter that is not
2723 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2724 @code{REG_CLASS_FROM_CONSTRAINT}
2725 may be used. Normally this macro will not be defined.
2727 If it is required for a particular target machine, it should return 1
2728 if @var{value} corresponds to the operand type represented by the
2729 constraint letter @var{c}. If @var{c} is not defined as an extra
2730 constraint, the value returned should be 0 regardless of @var{value}.
2732 For example, on the ROMP, load instructions cannot have their output
2733 in r0 if the memory reference contains a symbolic address. Constraint
2734 letter @samp{Q} is defined as representing a memory address that does
2735 @emph{not} contain a symbolic address. An alternative is specified with
2736 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2737 alternative specifies @samp{m} on the input and a register class that
2738 does not include r0 on the output.
2741 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2742 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2743 in @var{str}, so that you can use suffixes to distinguish between different
2747 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2748 A C expression that defines the optional machine-dependent constraint
2749 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2750 be treated like memory constraints by the reload pass.
2752 It should return 1 if the operand type represented by the constraint
2753 at the start of @var{str}, the first letter of which is the letter @var{c},
2754 comprises a subset of all memory references including
2755 all those whose address is simply a base register. This allows the reload
2756 pass to reload an operand, if it does not directly correspond to the operand
2757 type of @var{c}, by copying its address into a base register.
2759 For example, on the S/390, some instructions do not accept arbitrary
2760 memory references, but only those that do not make use of an index
2761 register. The constraint letter @samp{Q} is defined via
2762 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2763 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2764 a @samp{Q} constraint can handle any memory operand, because the
2765 reload pass knows it can be reloaded by copying the memory address
2766 into a base register if required. This is analogous to the way
2767 a @samp{o} constraint can handle any memory operand.
2770 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
2771 A C expression that defines the optional machine-dependent constraint
2772 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
2773 @code{EXTRA_CONSTRAINT_STR}, that should
2774 be treated like address constraints by the reload pass.
2776 It should return 1 if the operand type represented by the constraint
2777 at the start of @var{str}, which starts with the letter @var{c}, comprises
2778 a subset of all memory addresses including
2779 all those that consist of just a base register. This allows the reload
2780 pass to reload an operand, if it does not directly correspond to the operand
2781 type of @var{str}, by copying it into a base register.
2783 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
2784 be used with the @code{address_operand} predicate. It is treated
2785 analogously to the @samp{p} constraint.
2788 @node Stack and Calling
2789 @section Stack Layout and Calling Conventions
2790 @cindex calling conventions
2792 @c prevent bad page break with this line
2793 This describes the stack layout and calling conventions.
2797 * Exception Handling::
2802 * Register Arguments::
2804 * Aggregate Return::
2809 * Stack Smashing Protection::
2813 @subsection Basic Stack Layout
2814 @cindex stack frame layout
2815 @cindex frame layout
2817 @c prevent bad page break with this line
2818 Here is the basic stack layout.
2820 @defmac STACK_GROWS_DOWNWARD
2821 Define this macro if pushing a word onto the stack moves the stack
2822 pointer to a smaller address.
2824 When we say, ``define this macro if @dots{}'', it means that the
2825 compiler checks this macro only with @code{#ifdef} so the precise
2826 definition used does not matter.
2829 @defmac STACK_PUSH_CODE
2830 This macro defines the operation used when something is pushed
2831 on the stack. In RTL, a push operation will be
2832 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
2834 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
2835 and @code{POST_INC}. Which of these is correct depends on
2836 the stack direction and on whether the stack pointer points
2837 to the last item on the stack or whether it points to the
2838 space for the next item on the stack.
2840 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
2841 defined, which is almost always right, and @code{PRE_INC} otherwise,
2842 which is often wrong.
2845 @defmac FRAME_GROWS_DOWNWARD
2846 Define this macro to nonzero value if the addresses of local variable slots
2847 are at negative offsets from the frame pointer.
2850 @defmac ARGS_GROW_DOWNWARD
2851 Define this macro if successive arguments to a function occupy decreasing
2852 addresses on the stack.
2855 @defmac STARTING_FRAME_OFFSET
2856 Offset from the frame pointer to the first local variable slot to be allocated.
2858 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
2859 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
2860 Otherwise, it is found by adding the length of the first slot to the
2861 value @code{STARTING_FRAME_OFFSET}.
2862 @c i'm not sure if the above is still correct.. had to change it to get
2863 @c rid of an overfull. --mew 2feb93
2866 @defmac STACK_ALIGNMENT_NEEDED
2867 Define to zero to disable final alignment of the stack during reload.
2868 The nonzero default for this macro is suitable for most ports.
2870 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
2871 is a register save block following the local block that doesn't require
2872 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
2873 stack alignment and do it in the backend.
2876 @defmac STACK_POINTER_OFFSET
2877 Offset from the stack pointer register to the first location at which
2878 outgoing arguments are placed. If not specified, the default value of
2879 zero is used. This is the proper value for most machines.
2881 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2882 the first location at which outgoing arguments are placed.
2885 @defmac FIRST_PARM_OFFSET (@var{fundecl})
2886 Offset from the argument pointer register to the first argument's
2887 address. On some machines it may depend on the data type of the
2890 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
2891 the first argument's address.
2894 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
2895 Offset from the stack pointer register to an item dynamically allocated
2896 on the stack, e.g., by @code{alloca}.
2898 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
2899 length of the outgoing arguments. The default is correct for most
2900 machines. See @file{function.c} for details.
2903 @defmac INITIAL_FRAME_ADDRESS_RTX
2904 A C expression whose value is RTL representing the address of the initial
2905 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
2906 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
2907 default value will be used. Define this macro in order to make frame pointer
2908 elimination work in the presence of @code{__builtin_frame_address (count)} and
2909 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
2912 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
2913 A C expression whose value is RTL representing the address in a stack
2914 frame where the pointer to the caller's frame is stored. Assume that
2915 @var{frameaddr} is an RTL expression for the address of the stack frame
2918 If you don't define this macro, the default is to return the value
2919 of @var{frameaddr}---that is, the stack frame address is also the
2920 address of the stack word that points to the previous frame.
2923 @defmac SETUP_FRAME_ADDRESSES
2924 If defined, a C expression that produces the machine-specific code to
2925 setup the stack so that arbitrary frames can be accessed. For example,
2926 on the SPARC, we must flush all of the register windows to the stack
2927 before we can access arbitrary stack frames. You will seldom need to
2931 @deftypefn {Target Hook} bool TARGET_BUILTIN_SETJMP_FRAME_VALUE ()
2932 This target hook should return an rtx that is used to store
2933 the address of the current frame into the built in @code{setjmp} buffer.
2934 The default value, @code{virtual_stack_vars_rtx}, is correct for most
2935 machines. One reason you may need to define this target hook is if
2936 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
2939 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
2940 A C expression whose value is RTL representing the value of the return
2941 address for the frame @var{count} steps up from the current frame, after
2942 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
2943 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
2944 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
2946 The value of the expression must always be the correct address when
2947 @var{count} is zero, but may be @code{NULL_RTX} if there is not way to
2948 determine the return address of other frames.
2951 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
2952 Define this if the return address of a particular stack frame is accessed
2953 from the frame pointer of the previous stack frame.
2956 @defmac INCOMING_RETURN_ADDR_RTX
2957 A C expression whose value is RTL representing the location of the
2958 incoming return address at the beginning of any function, before the
2959 prologue. This RTL is either a @code{REG}, indicating that the return
2960 value is saved in @samp{REG}, or a @code{MEM} representing a location in
2963 You only need to define this macro if you want to support call frame
2964 debugging information like that provided by DWARF 2.
2966 If this RTL is a @code{REG}, you should also define
2967 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
2970 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
2971 A C expression whose value is an integer giving a DWARF 2 column
2972 number that may be used as an alternate return column. This should
2973 be defined only if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
2974 general register, but an alternate column needs to be used for
2978 @defmac DWARF_ZERO_REG
2979 A C expression whose value is an integer giving a DWARF 2 register
2980 number that is considered to always have the value zero. This should
2981 only be defined if the target has an architected zero register, and
2982 someone decided it was a good idea to use that register number to
2983 terminate the stack backtrace. New ports should avoid this.
2986 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
2987 This target hook allows the backend to emit frame-related insns that
2988 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
2989 info engine will invoke it on insns of the form
2991 (set (reg) (unspec [...] UNSPEC_INDEX))
2995 (set (reg) (unspec_volatile [...] UNSPECV_INDEX)).
2997 to let the backend emit the call frame instructions. @var{label} is
2998 the CFI label attached to the insn, @var{pattern} is the pattern of
2999 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3002 @defmac INCOMING_FRAME_SP_OFFSET
3003 A C expression whose value is an integer giving the offset, in bytes,
3004 from the value of the stack pointer register to the top of the stack
3005 frame at the beginning of any function, before the prologue. The top of
3006 the frame is defined to be the value of the stack pointer in the
3007 previous frame, just before the call instruction.
3009 You only need to define this macro if you want to support call frame
3010 debugging information like that provided by DWARF 2.
3013 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3014 A C expression whose value is an integer giving the offset, in bytes,
3015 from the argument pointer to the canonical frame address (cfa). The
3016 final value should coincide with that calculated by
3017 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3018 during virtual register instantiation.
3020 The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)},
3021 which is correct for most machines; in general, the arguments are found
3022 immediately before the stack frame. Note that this is not the case on
3023 some targets that save registers into the caller's frame, such as SPARC
3024 and rs6000, and so such targets need to define this macro.
3026 You only need to define this macro if the default is incorrect, and you
3027 want to support call frame debugging information like that provided by
3031 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3032 If defined, a C expression whose value is an integer giving the offset
3033 in bytes from the frame pointer to the canonical frame address (cfa).
3034 The final value should conincide with that calculated by
3035 @code{INCOMING_FRAME_SP_OFFSET}.
3037 Normally the CFA is calculated as an offset from the argument pointer,
3038 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3039 variable due to the ABI, this may not be possible. If this macro is
3040 defined, it implies that the virtual register instantiation should be
3041 based on the frame pointer instead of the argument pointer. Only one
3042 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3046 @node Exception Handling
3047 @subsection Exception Handling Support
3048 @cindex exception handling
3050 @defmac EH_RETURN_DATA_REGNO (@var{N})
3051 A C expression whose value is the @var{N}th register number used for
3052 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3053 @var{N} registers are usable.
3055 The exception handling library routines communicate with the exception
3056 handlers via a set of agreed upon registers. Ideally these registers
3057 should be call-clobbered; it is possible to use call-saved registers,
3058 but may negatively impact code size. The target must support at least
3059 2 data registers, but should define 4 if there are enough free registers.
3061 You must define this macro if you want to support call frame exception
3062 handling like that provided by DWARF 2.
3065 @defmac EH_RETURN_STACKADJ_RTX
3066 A C expression whose value is RTL representing a location in which
3067 to store a stack adjustment to be applied before function return.
3068 This is used to unwind the stack to an exception handler's call frame.
3069 It will be assigned zero on code paths that return normally.
3071 Typically this is a call-clobbered hard register that is otherwise
3072 untouched by the epilogue, but could also be a stack slot.
3074 Do not define this macro if the stack pointer is saved and restored
3075 by the regular prolog and epilog code in the call frame itself; in
3076 this case, the exception handling library routines will update the
3077 stack location to be restored in place. Otherwise, you must define
3078 this macro if you want to support call frame exception handling like
3079 that provided by DWARF 2.
3082 @defmac EH_RETURN_HANDLER_RTX
3083 A C expression whose value is RTL representing a location in which
3084 to store the address of an exception handler to which we should
3085 return. It will not be assigned on code paths that return normally.
3087 Typically this is the location in the call frame at which the normal
3088 return address is stored. For targets that return by popping an
3089 address off the stack, this might be a memory address just below
3090 the @emph{target} call frame rather than inside the current call
3091 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3092 been assigned, so it may be used to calculate the location of the
3095 Some targets have more complex requirements than storing to an
3096 address calculable during initial code generation. In that case
3097 the @code{eh_return} instruction pattern should be used instead.
3099 If you want to support call frame exception handling, you must
3100 define either this macro or the @code{eh_return} instruction pattern.
3103 @defmac RETURN_ADDR_OFFSET
3104 If defined, an integer-valued C expression for which rtl will be generated
3105 to add it to the exception handler address before it is searched in the
3106 exception handling tables, and to subtract it again from the address before
3107 using it to return to the exception handler.
3110 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3111 This macro chooses the encoding of pointers embedded in the exception
3112 handling sections. If at all possible, this should be defined such
3113 that the exception handling section will not require dynamic relocations,
3114 and so may be read-only.
3116 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3117 @var{global} is true if the symbol may be affected by dynamic relocations.
3118 The macro should return a combination of the @code{DW_EH_PE_*} defines
3119 as found in @file{dwarf2.h}.
3121 If this macro is not defined, pointers will not be encoded but
3122 represented directly.
3125 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3126 This macro allows the target to emit whatever special magic is required
3127 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3128 Generic code takes care of pc-relative and indirect encodings; this must
3129 be defined if the target uses text-relative or data-relative encodings.
3131 This is a C statement that branches to @var{done} if the format was
3132 handled. @var{encoding} is the format chosen, @var{size} is the number
3133 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3137 @defmac MD_UNWIND_SUPPORT
3138 A string specifying a file to be #include'd in unwind-dw2.c. The file
3139 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3142 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3143 This macro allows the target to add cpu and operating system specific
3144 code to the call-frame unwinder for use when there is no unwind data
3145 available. The most common reason to implement this macro is to unwind
3146 through signal frames.
3148 This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c}
3149 and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3150 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3151 for the address of the code being executed and @code{context->cfa} for
3152 the stack pointer value. If the frame can be decoded, the register save
3153 addresses should be updated in @var{fs} and the macro should evaluate to
3154 @code{_URC_NO_REASON}. If the frame cannot be decoded, the macro should
3155 evaluate to @code{_URC_END_OF_STACK}.
3157 For proper signal handling in Java this macro is accompanied by
3158 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3161 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3162 This macro allows the target to add operating system specific code to the
3163 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3164 usually used for signal or interrupt frames.
3166 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3167 @var{context} is an @code{_Unwind_Context};
3168 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3169 for the abi and context in the @code{.unwabi} directive. If the
3170 @code{.unwabi} directive can be handled, the register save addresses should
3171 be updated in @var{fs}.
3174 @defmac TARGET_USES_WEAK_UNWIND_INFO
3175 A C expression that evaluates to true if the target requires unwind
3176 info to be given comdat linkage. Define it to be @code{1} if comdat
3177 linkage is necessary. The default is @code{0}.
3180 @node Stack Checking
3181 @subsection Specifying How Stack Checking is Done
3183 GCC will check that stack references are within the boundaries of
3184 the stack, if the @option{-fstack-check} is specified, in one of three ways:
3188 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3189 will assume that you have arranged for stack checking to be done at
3190 appropriate places in the configuration files, e.g., in
3191 @code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special
3195 If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern
3196 called @code{check_stack} in your @file{md} file, GCC will call that
3197 pattern with one argument which is the address to compare the stack
3198 value against. You must arrange for this pattern to report an error if
3199 the stack pointer is out of range.
3202 If neither of the above are true, GCC will generate code to periodically
3203 ``probe'' the stack pointer using the values of the macros defined below.
3206 Normally, you will use the default values of these macros, so GCC
3207 will use the third approach.
3209 @defmac STACK_CHECK_BUILTIN
3210 A nonzero value if stack checking is done by the configuration files in a
3211 machine-dependent manner. You should define this macro if stack checking
3212 is require by the ABI of your machine or if you would like to have to stack
3213 checking in some more efficient way than GCC's portable approach.
3214 The default value of this macro is zero.
3217 @defmac STACK_CHECK_PROBE_INTERVAL
3218 An integer representing the interval at which GCC must generate stack
3219 probe instructions. You will normally define this macro to be no larger
3220 than the size of the ``guard pages'' at the end of a stack area. The
3221 default value of 4096 is suitable for most systems.
3224 @defmac STACK_CHECK_PROBE_LOAD
3225 A integer which is nonzero if GCC should perform the stack probe
3226 as a load instruction and zero if GCC should use a store instruction.
3227 The default is zero, which is the most efficient choice on most systems.
3230 @defmac STACK_CHECK_PROTECT
3231 The number of bytes of stack needed to recover from a stack overflow,
3232 for languages where such a recovery is supported. The default value of
3233 75 words should be adequate for most machines.
3236 @defmac STACK_CHECK_MAX_FRAME_SIZE
3237 The maximum size of a stack frame, in bytes. GCC will generate probe
3238 instructions in non-leaf functions to ensure at least this many bytes of
3239 stack are available. If a stack frame is larger than this size, stack
3240 checking will not be reliable and GCC will issue a warning. The
3241 default is chosen so that GCC only generates one instruction on most
3242 systems. You should normally not change the default value of this macro.
3245 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3246 GCC uses this value to generate the above warning message. It
3247 represents the amount of fixed frame used by a function, not including
3248 space for any callee-saved registers, temporaries and user variables.
3249 You need only specify an upper bound for this amount and will normally
3250 use the default of four words.
3253 @defmac STACK_CHECK_MAX_VAR_SIZE
3254 The maximum size, in bytes, of an object that GCC will place in the
3255 fixed area of the stack frame when the user specifies
3256 @option{-fstack-check}.
3257 GCC computed the default from the values of the above macros and you will
3258 normally not need to override that default.
3262 @node Frame Registers
3263 @subsection Registers That Address the Stack Frame
3265 @c prevent bad page break with this line
3266 This discusses registers that address the stack frame.
3268 @defmac STACK_POINTER_REGNUM
3269 The register number of the stack pointer register, which must also be a
3270 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3271 the hardware determines which register this is.
3274 @defmac FRAME_POINTER_REGNUM
3275 The register number of the frame pointer register, which is used to
3276 access automatic variables in the stack frame. On some machines, the
3277 hardware determines which register this is. On other machines, you can
3278 choose any register you wish for this purpose.
3281 @defmac HARD_FRAME_POINTER_REGNUM
3282 On some machines the offset between the frame pointer and starting
3283 offset of the automatic variables is not known until after register
3284 allocation has been done (for example, because the saved registers are
3285 between these two locations). On those machines, define
3286 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3287 be used internally until the offset is known, and define
3288 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3289 used for the frame pointer.
3291 You should define this macro only in the very rare circumstances when it
3292 is not possible to calculate the offset between the frame pointer and
3293 the automatic variables until after register allocation has been
3294 completed. When this macro is defined, you must also indicate in your
3295 definition of @code{ELIMINABLE_REGS} how to eliminate
3296 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3297 or @code{STACK_POINTER_REGNUM}.
3299 Do not define this macro if it would be the same as
3300 @code{FRAME_POINTER_REGNUM}.
3303 @defmac ARG_POINTER_REGNUM
3304 The register number of the arg pointer register, which is used to access
3305 the function's argument list. On some machines, this is the same as the
3306 frame pointer register. On some machines, the hardware determines which
3307 register this is. On other machines, you can choose any register you
3308 wish for this purpose. If this is not the same register as the frame
3309 pointer register, then you must mark it as a fixed register according to
3310 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3311 (@pxref{Elimination}).
3314 @defmac RETURN_ADDRESS_POINTER_REGNUM
3315 The register number of the return address pointer register, which is used to
3316 access the current function's return address from the stack. On some
3317 machines, the return address is not at a fixed offset from the frame
3318 pointer or stack pointer or argument pointer. This register can be defined
3319 to point to the return address on the stack, and then be converted by
3320 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3322 Do not define this macro unless there is no other way to get the return
3323 address from the stack.
3326 @defmac STATIC_CHAIN_REGNUM
3327 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3328 Register numbers used for passing a function's static chain pointer. If
3329 register windows are used, the register number as seen by the called
3330 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3331 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3332 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3335 The static chain register need not be a fixed register.
3337 If the static chain is passed in memory, these macros should not be
3338 defined; instead, the next two macros should be defined.
3341 @defmac STATIC_CHAIN
3342 @defmacx STATIC_CHAIN_INCOMING
3343 If the static chain is passed in memory, these macros provide rtx giving
3344 @code{mem} expressions that denote where they are stored.
3345 @code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations
3346 as seen by the calling and called functions, respectively. Often the former
3347 will be at an offset from the stack pointer and the latter at an offset from
3350 @findex stack_pointer_rtx
3351 @findex frame_pointer_rtx
3352 @findex arg_pointer_rtx
3353 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3354 @code{arg_pointer_rtx} will have been initialized prior to the use of these
3355 macros and should be used to refer to those items.
3357 If the static chain is passed in a register, the two previous macros should
3361 @defmac DWARF_FRAME_REGISTERS
3362 This macro specifies the maximum number of hard registers that can be
3363 saved in a call frame. This is used to size data structures used in
3364 DWARF2 exception handling.
3366 Prior to GCC 3.0, this macro was needed in order to establish a stable
3367 exception handling ABI in the face of adding new hard registers for ISA
3368 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3369 in the number of hard registers. Nevertheless, this macro can still be
3370 used to reduce the runtime memory requirements of the exception handling
3371 routines, which can be substantial if the ISA contains a lot of
3372 registers that are not call-saved.
3374 If this macro is not defined, it defaults to
3375 @code{FIRST_PSEUDO_REGISTER}.
3378 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3380 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3381 for backward compatibility in pre GCC 3.0 compiled code.
3383 If this macro is not defined, it defaults to
3384 @code{DWARF_FRAME_REGISTERS}.
3387 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3389 Define this macro if the target's representation for dwarf registers
3390 is different than the internal representation for unwind column.
3391 Given a dwarf register, this macro should return the internal unwind
3392 column number to use instead.
3394 See the PowerPC's SPE target for an example.
3397 @defmac DWARF_FRAME_REGNUM (@var{regno})
3399 Define this macro if the target's representation for dwarf registers
3400 used in .eh_frame or .debug_frame is different from that used in other
3401 debug info sections. Given a GCC hard register number, this macro
3402 should return the .eh_frame register number. The default is
3403 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3407 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3409 Define this macro to map register numbers held in the call frame info
3410 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3411 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3412 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3413 return @code{@var{regno}}.
3418 @subsection Eliminating Frame Pointer and Arg Pointer
3420 @c prevent bad page break with this line
3421 This is about eliminating the frame pointer and arg pointer.
3423 @defmac FRAME_POINTER_REQUIRED
3424 A C expression which is nonzero if a function must have and use a frame
3425 pointer. This expression is evaluated in the reload pass. If its value is
3426 nonzero the function will have a frame pointer.
3428 The expression can in principle examine the current function and decide
3429 according to the facts, but on most machines the constant 0 or the
3430 constant 1 suffices. Use 0 when the machine allows code to be generated
3431 with no frame pointer, and doing so saves some time or space. Use 1
3432 when there is no possible advantage to avoiding a frame pointer.
3434 In certain cases, the compiler does not know how to produce valid code
3435 without a frame pointer. The compiler recognizes those cases and
3436 automatically gives the function a frame pointer regardless of what
3437 @code{FRAME_POINTER_REQUIRED} says. You don't need to worry about
3440 In a function that does not require a frame pointer, the frame pointer
3441 register can be allocated for ordinary usage, unless you mark it as a
3442 fixed register. See @code{FIXED_REGISTERS} for more information.
3445 @findex get_frame_size
3446 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3447 A C statement to store in the variable @var{depth-var} the difference
3448 between the frame pointer and the stack pointer values immediately after
3449 the function prologue. The value would be computed from information
3450 such as the result of @code{get_frame_size ()} and the tables of
3451 registers @code{regs_ever_live} and @code{call_used_regs}.
3453 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3454 need not be defined. Otherwise, it must be defined even if
3455 @code{FRAME_POINTER_REQUIRED} is defined to always be true; in that
3456 case, you may set @var{depth-var} to anything.
3459 @defmac ELIMINABLE_REGS
3460 If defined, this macro specifies a table of register pairs used to
3461 eliminate unneeded registers that point into the stack frame. If it is not
3462 defined, the only elimination attempted by the compiler is to replace
3463 references to the frame pointer with references to the stack pointer.
3465 The definition of this macro is a list of structure initializations, each
3466 of which specifies an original and replacement register.
3468 On some machines, the position of the argument pointer is not known until
3469 the compilation is completed. In such a case, a separate hard register
3470 must be used for the argument pointer. This register can be eliminated by
3471 replacing it with either the frame pointer or the argument pointer,
3472 depending on whether or not the frame pointer has been eliminated.
3474 In this case, you might specify:
3476 #define ELIMINABLE_REGS \
3477 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3478 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3479 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3482 Note that the elimination of the argument pointer with the stack pointer is
3483 specified first since that is the preferred elimination.
3486 @defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg})
3487 A C expression that returns nonzero if the compiler is allowed to try
3488 to replace register number @var{from-reg} with register number
3489 @var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS}
3490 is defined, and will usually be the constant 1, since most of the cases
3491 preventing register elimination are things that the compiler already
3495 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3496 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3497 specifies the initial difference between the specified pair of
3498 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3502 @node Stack Arguments
3503 @subsection Passing Function Arguments on the Stack
3504 @cindex arguments on stack
3505 @cindex stack arguments
3507 The macros in this section control how arguments are passed
3508 on the stack. See the following section for other macros that
3509 control passing certain arguments in registers.
3511 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype})
3512 This target hook returns @code{true} if an argument declared in a
3513 prototype as an integral type smaller than @code{int} should actually be
3514 passed as an @code{int}. In addition to avoiding errors in certain
3515 cases of mismatch, it also makes for better code on certain machines.
3516 The default is to not promote prototypes.
3520 A C expression. If nonzero, push insns will be used to pass
3522 If the target machine does not have a push instruction, set it to zero.
3523 That directs GCC to use an alternate strategy: to
3524 allocate the entire argument block and then store the arguments into
3525 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3528 @defmac PUSH_ARGS_REVERSED
3529 A C expression. If nonzero, function arguments will be evaluated from
3530 last to first, rather than from first to last. If this macro is not
3531 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3532 and args grow in opposite directions, and 0 otherwise.
3535 @defmac PUSH_ROUNDING (@var{npushed})
3536 A C expression that is the number of bytes actually pushed onto the
3537 stack when an instruction attempts to push @var{npushed} bytes.
3539 On some machines, the definition
3542 #define PUSH_ROUNDING(BYTES) (BYTES)
3546 will suffice. But on other machines, instructions that appear
3547 to push one byte actually push two bytes in an attempt to maintain
3548 alignment. Then the definition should be
3551 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3555 @findex current_function_outgoing_args_size
3556 @defmac ACCUMULATE_OUTGOING_ARGS
3557 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3558 will be computed and placed into the variable
3559 @code{current_function_outgoing_args_size}. No space will be pushed
3560 onto the stack for each call; instead, the function prologue should
3561 increase the stack frame size by this amount.
3563 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3567 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3568 Define this macro if functions should assume that stack space has been
3569 allocated for arguments even when their values are passed in
3572 The value of this macro is the size, in bytes, of the area reserved for
3573 arguments passed in registers for the function represented by @var{fndecl},
3574 which can be zero if GCC is calling a library function.
3576 This space can be allocated by the caller, or be a part of the
3577 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3580 @c above is overfull. not sure what to do. --mew 5feb93 did
3581 @c something, not sure if it looks good. --mew 10feb93
3583 @defmac OUTGOING_REG_PARM_STACK_SPACE
3584 Define this if it is the responsibility of the caller to allocate the area
3585 reserved for arguments passed in registers.
3587 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3588 whether the space for these arguments counts in the value of
3589 @code{current_function_outgoing_args_size}.
3592 @defmac STACK_PARMS_IN_REG_PARM_AREA
3593 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3594 stack parameters don't skip the area specified by it.
3595 @c i changed this, makes more sens and it should have taken care of the
3596 @c overfull.. not as specific, tho. --mew 5feb93
3598 Normally, when a parameter is not passed in registers, it is placed on the
3599 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3600 suppresses this behavior and causes the parameter to be passed on the
3601 stack in its natural location.
3604 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3605 A C expression that should indicate the number of bytes of its own
3606 arguments that a function pops on returning, or 0 if the
3607 function pops no arguments and the caller must therefore pop them all
3608 after the function returns.
3610 @var{fundecl} is a C variable whose value is a tree node that describes
3611 the function in question. Normally it is a node of type
3612 @code{FUNCTION_DECL} that describes the declaration of the function.
3613 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3615 @var{funtype} is a C variable whose value is a tree node that
3616 describes the function in question. Normally it is a node of type
3617 @code{FUNCTION_TYPE} that describes the data type of the function.
3618 From this it is possible to obtain the data types of the value and
3619 arguments (if known).
3621 When a call to a library function is being considered, @var{fundecl}
3622 will contain an identifier node for the library function. Thus, if
3623 you need to distinguish among various library functions, you can do so
3624 by their names. Note that ``library function'' in this context means
3625 a function used to perform arithmetic, whose name is known specially
3626 in the compiler and was not mentioned in the C code being compiled.
3628 @var{stack-size} is the number of bytes of arguments passed on the
3629 stack. If a variable number of bytes is passed, it is zero, and
3630 argument popping will always be the responsibility of the calling function.
3632 On the VAX, all functions always pop their arguments, so the definition
3633 of this macro is @var{stack-size}. On the 68000, using the standard
3634 calling convention, no functions pop their arguments, so the value of
3635 the macro is always 0 in this case. But an alternative calling
3636 convention is available in which functions that take a fixed number of
3637 arguments pop them but other functions (such as @code{printf}) pop
3638 nothing (the caller pops all). When this convention is in use,
3639 @var{funtype} is examined to determine whether a function takes a fixed
3640 number of arguments.
3643 @defmac CALL_POPS_ARGS (@var{cum})
3644 A C expression that should indicate the number of bytes a call sequence
3645 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3646 when compiling a function call.
3648 @var{cum} is the variable in which all arguments to the called function
3649 have been accumulated.
3651 On certain architectures, such as the SH5, a call trampoline is used
3652 that pops certain registers off the stack, depending on the arguments
3653 that have been passed to the function. Since this is a property of the
3654 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3658 @node Register Arguments
3659 @subsection Passing Arguments in Registers
3660 @cindex arguments in registers
3661 @cindex registers arguments
3663 This section describes the macros which let you control how various
3664 types of arguments are passed in registers or how they are arranged in
3667 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3668 A C expression that controls whether a function argument is passed
3669 in a register, and which register.
3671 The arguments are @var{cum}, which summarizes all the previous
3672 arguments; @var{mode}, the machine mode of the argument; @var{type},
3673 the data type of the argument as a tree node or 0 if that is not known
3674 (which happens for C support library functions); and @var{named},
3675 which is 1 for an ordinary argument and 0 for nameless arguments that
3676 correspond to @samp{@dots{}} in the called function's prototype.
3677 @var{type} can be an incomplete type if a syntax error has previously
3680 The value of the expression is usually either a @code{reg} RTX for the
3681 hard register in which to pass the argument, or zero to pass the
3682 argument on the stack.
3684 For machines like the VAX and 68000, where normally all arguments are
3685 pushed, zero suffices as a definition.
3687 The value of the expression can also be a @code{parallel} RTX@. This is
3688 used when an argument is passed in multiple locations. The mode of the
3689 @code{parallel} should be the mode of the entire argument. The
3690 @code{parallel} holds any number of @code{expr_list} pairs; each one
3691 describes where part of the argument is passed. In each
3692 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
3693 register in which to pass this part of the argument, and the mode of the
3694 register RTX indicates how large this part of the argument is. The
3695 second operand of the @code{expr_list} is a @code{const_int} which gives
3696 the offset in bytes into the entire argument of where this part starts.
3697 As a special exception the first @code{expr_list} in the @code{parallel}
3698 RTX may have a first operand of zero. This indicates that the entire
3699 argument is also stored on the stack.
3701 The last time this macro is called, it is called with @code{MODE ==
3702 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
3703 pattern as operands 2 and 3 respectively.
3705 @cindex @file{stdarg.h} and register arguments
3706 The usual way to make the ISO library @file{stdarg.h} work on a machine
3707 where some arguments are usually passed in registers, is to cause
3708 nameless arguments to be passed on the stack instead. This is done
3709 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
3711 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
3712 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
3713 You may use the hook @code{targetm.calls.must_pass_in_stack}
3714 in the definition of this macro to determine if this argument is of a
3715 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
3716 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
3717 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
3718 defined, the argument will be computed in the stack and then loaded into
3722 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, tree @var{type})
3723 This target hook should return @code{true} if we should not pass @var{type}
3724 solely in registers. The file @file{expr.h} defines a
3725 definition that is usually appropriate, refer to @file{expr.h} for additional
3729 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
3730 Define this macro if the target machine has ``register windows'', so
3731 that the register in which a function sees an arguments is not
3732 necessarily the same as the one in which the caller passed the
3735 For such machines, @code{FUNCTION_ARG} computes the register in which
3736 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
3737 be defined in a similar fashion to tell the function being called
3738 where the arguments will arrive.
3740 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
3741 serves both purposes.
3744 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3745 This target hook returns the number of bytes at the beginning of an
3746 argument that must be put in registers. The value must be zero for
3747 arguments that are passed entirely in registers or that are entirely
3748 pushed on the stack.
3750 On some machines, certain arguments must be passed partially in
3751 registers and partially in memory. On these machines, typically the
3752 first few words of arguments are passed in registers, and the rest
3753 on the stack. If a multi-word argument (a @code{double} or a
3754 structure) crosses that boundary, its first few words must be passed
3755 in registers and the rest must be pushed. This macro tells the
3756 compiler when this occurs, and how many bytes should go in registers.
3758 @code{FUNCTION_ARG} for these arguments should return the first
3759 register to be used by the caller for this argument; likewise
3760 @code{FUNCTION_INCOMING_ARG}, for the called function.
3763 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3764 This target hook should return @code{true} if an argument at the
3765 position indicated by @var{cum} should be passed by reference. This
3766 predicate is queried after target independent reasons for being
3767 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
3769 If the hook returns true, a copy of that argument is made in memory and a
3770 pointer to the argument is passed instead of the argument itself.
3771 The pointer is passed in whatever way is appropriate for passing a pointer
3775 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
3776 The function argument described by the parameters to this hook is
3777 known to be passed by reference. The hook should return true if the
3778 function argument should be copied by the callee instead of copied
3781 For any argument for which the hook returns true, if it can be
3782 determined that the argument is not modified, then a copy need
3785 The default version of this hook always returns false.
3788 @defmac CUMULATIVE_ARGS
3789 A C type for declaring a variable that is used as the first argument of
3790 @code{FUNCTION_ARG} and other related values. For some target machines,
3791 the type @code{int} suffices and can hold the number of bytes of
3794 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
3795 arguments that have been passed on the stack. The compiler has other
3796 variables to keep track of that. For target machines on which all
3797 arguments are passed on the stack, there is no need to store anything in
3798 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
3799 should not be empty, so use @code{int}.
3802 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
3803 A C statement (sans semicolon) for initializing the variable
3804 @var{cum} for the state at the beginning of the argument list. The
3805 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
3806 is the tree node for the data type of the function which will receive
3807 the args, or 0 if the args are to a compiler support library function.
3808 For direct calls that are not libcalls, @var{fndecl} contain the
3809 declaration node of the function. @var{fndecl} is also set when
3810 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
3811 being compiled. @var{n_named_args} is set to the number of named
3812 arguments, including a structure return address if it is passed as a
3813 parameter, when making a call. When processing incoming arguments,
3814 @var{n_named_args} is set to @minus{}1.
3816 When processing a call to a compiler support library function,
3817 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
3818 contains the name of the function, as a string. @var{libname} is 0 when
3819 an ordinary C function call is being processed. Thus, each time this
3820 macro is called, either @var{libname} or @var{fntype} is nonzero, but
3821 never both of them at once.
3824 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
3825 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
3826 it gets a @code{MODE} argument instead of @var{fntype}, that would be
3827 @code{NULL}. @var{indirect} would always be zero, too. If this macro
3828 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
3829 0)} is used instead.
3832 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
3833 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
3834 finding the arguments for the function being compiled. If this macro is
3835 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
3837 The value passed for @var{libname} is always 0, since library routines
3838 with special calling conventions are never compiled with GCC@. The
3839 argument @var{libname} exists for symmetry with
3840 @code{INIT_CUMULATIVE_ARGS}.
3841 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
3842 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
3845 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
3846 A C statement (sans semicolon) to update the summarizer variable
3847 @var{cum} to advance past an argument in the argument list. The
3848 values @var{mode}, @var{type} and @var{named} describe that argument.
3849 Once this is done, the variable @var{cum} is suitable for analyzing
3850 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
3852 This macro need not do anything if the argument in question was passed
3853 on the stack. The compiler knows how to track the amount of stack space
3854 used for arguments without any special help.
3857 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
3858 If defined, a C expression which determines whether, and in which direction,
3859 to pad out an argument with extra space. The value should be of type
3860 @code{enum direction}: either @code{upward} to pad above the argument,
3861 @code{downward} to pad below, or @code{none} to inhibit padding.
3863 The @emph{amount} of padding is always just enough to reach the next
3864 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
3867 This macro has a default definition which is right for most systems.
3868 For little-endian machines, the default is to pad upward. For
3869 big-endian machines, the default is to pad downward for an argument of
3870 constant size shorter than an @code{int}, and upward otherwise.
3873 @defmac PAD_VARARGS_DOWN
3874 If defined, a C expression which determines whether the default
3875 implementation of va_arg will attempt to pad down before reading the
3876 next argument, if that argument is smaller than its aligned space as
3877 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
3878 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
3881 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
3882 Specify padding for the last element of a block move between registers and
3883 memory. @var{first} is nonzero if this is the only element. Defining this
3884 macro allows better control of register function parameters on big-endian
3885 machines, without using @code{PARALLEL} rtl. In particular,
3886 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
3887 registers, as there is no longer a "wrong" part of a register; For example,
3888 a three byte aggregate may be passed in the high part of a register if so
3892 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
3893 If defined, a C expression that gives the alignment boundary, in bits,
3894 of an argument with the specified mode and type. If it is not defined,
3895 @code{PARM_BOUNDARY} is used for all arguments.
3898 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
3899 A C expression that is nonzero if @var{regno} is the number of a hard
3900 register in which function arguments are sometimes passed. This does
3901 @emph{not} include implicit arguments such as the static chain and
3902 the structure-value address. On many machines, no registers can be
3903 used for this purpose since all function arguments are pushed on the
3907 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type})
3908 This hook should return true if parameter of type @var{type} are passed
3909 as two scalar parameters. By default, GCC will attempt to pack complex
3910 arguments into the target's word size. Some ABIs require complex arguments
3911 to be split and treated as their individual components. For example, on
3912 AIX64, complex floats should be passed in a pair of floating point
3913 registers, even though a complex float would fit in one 64-bit floating
3916 The default value of this hook is @code{NULL}, which is treated as always
3920 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
3921 This hook returns a type node for @code{va_list} for the target.
3922 The default version of the hook returns @code{void*}.
3925 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, tree *@var{pre_p}, tree *@var{post_p})
3926 This hook performs target-specific gimplification of
3927 @code{VA_ARG_EXPR}. The first two parameters correspond to the
3928 arguments to @code{va_arg}; the latter two are as in
3929 @code{gimplify.c:gimplify_expr}.
3932 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
3933 Define this to return nonzero if the port can handle pointers
3934 with machine mode @var{mode}. The default version of this
3935 hook returns true for both @code{ptr_mode} and @code{Pmode}.
3938 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
3939 Define this to return nonzero if the port is prepared to handle
3940 insns involving scalar mode @var{mode}. For a scalar mode to be
3941 considered supported, all the basic arithmetic and comparisons
3944 The default version of this hook returns true for any mode
3945 required to handle the basic C types (as defined by the port).
3946 Included here are the double-word arithmetic supported by the
3947 code in @file{optabs.c}.
3950 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
3951 Define this to return nonzero if the port is prepared to handle
3952 insns involving vector mode @var{mode}. At the very least, it
3953 must have move patterns for this mode.
3957 @subsection How Scalar Function Values Are Returned
3958 @cindex return values in registers
3959 @cindex values, returned by functions
3960 @cindex scalars, returned as values
3962 This section discusses the macros that control returning scalars as
3963 values---values that can fit in registers.
3965 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
3966 A C expression to create an RTX representing the place where a
3967 function returns a value of data type @var{valtype}. @var{valtype} is
3968 a tree node representing a data type. Write @code{TYPE_MODE
3969 (@var{valtype})} to get the machine mode used to represent that type.
3970 On many machines, only the mode is relevant. (Actually, on most
3971 machines, scalar values are returned in the same place regardless of
3974 The value of the expression is usually a @code{reg} RTX for the hard
3975 register where the return value is stored. The value can also be a
3976 @code{parallel} RTX, if the return value is in multiple places. See
3977 @code{FUNCTION_ARG} for an explanation of the @code{parallel} form.
3979 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply the same
3980 promotion rules specified in @code{PROMOTE_MODE} if @var{valtype} is a
3983 If the precise function being called is known, @var{func} is a tree
3984 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
3985 pointer. This makes it possible to use a different value-returning
3986 convention for specific functions when all their calls are
3989 @code{FUNCTION_VALUE} is not used for return vales with aggregate data
3990 types, because these are returned in another way. See
3991 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
3994 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
3995 Define this macro if the target machine has ``register windows''
3996 so that the register in which a function returns its value is not
3997 the same as the one in which the caller sees the value.
3999 For such machines, @code{FUNCTION_VALUE} computes the register in which
4000 the caller will see the value. @code{FUNCTION_OUTGOING_VALUE} should be
4001 defined in a similar fashion to tell the function where to put the
4004 If @code{FUNCTION_OUTGOING_VALUE} is not defined,
4005 @code{FUNCTION_VALUE} serves both purposes.
4007 @code{FUNCTION_OUTGOING_VALUE} is not used for return vales with
4008 aggregate data types, because these are returned in another way. See
4009 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4012 @defmac LIBCALL_VALUE (@var{mode})
4013 A C expression to create an RTX representing the place where a library
4014 function returns a value of mode @var{mode}. If the precise function
4015 being called is known, @var{func} is a tree node
4016 (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4017 pointer. This makes it possible to use a different value-returning
4018 convention for specific functions when all their calls are
4021 Note that ``library function'' in this context means a compiler
4022 support routine, used to perform arithmetic, whose name is known
4023 specially by the compiler and was not mentioned in the C code being
4026 The definition of @code{LIBRARY_VALUE} need not be concerned aggregate
4027 data types, because none of the library functions returns such types.
4030 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4031 A C expression that is nonzero if @var{regno} is the number of a hard
4032 register in which the values of called function may come back.
4034 A register whose use for returning values is limited to serving as the
4035 second of a pair (for a value of type @code{double}, say) need not be
4036 recognized by this macro. So for most machines, this definition
4040 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4043 If the machine has register windows, so that the caller and the called
4044 function use different registers for the return value, this macro
4045 should recognize only the caller's register numbers.
4048 @defmac APPLY_RESULT_SIZE
4049 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4050 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4051 saving and restoring an arbitrary return value.
4054 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type})
4055 This hook should return true if values of type @var{type} are returned
4056 at the most significant end of a register (in other words, if they are
4057 padded at the least significant end). You can assume that @var{type}
4058 is returned in a register; the caller is required to check this.
4060 Note that the register provided by @code{FUNCTION_VALUE} must be able
4061 to hold the complete return value. For example, if a 1-, 2- or 3-byte
4062 structure is returned at the most significant end of a 4-byte register,
4063 @code{FUNCTION_VALUE} should provide an @code{SImode} rtx.
4066 @node Aggregate Return
4067 @subsection How Large Values Are Returned
4068 @cindex aggregates as return values
4069 @cindex large return values
4070 @cindex returning aggregate values
4071 @cindex structure value address
4073 When a function value's mode is @code{BLKmode} (and in some other
4074 cases), the value is not returned according to @code{FUNCTION_VALUE}
4075 (@pxref{Scalar Return}). Instead, the caller passes the address of a
4076 block of memory in which the value should be stored. This address
4077 is called the @dfn{structure value address}.
4079 This section describes how to control returning structure values in
4082 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype})
4083 This target hook should return a nonzero value to say to return the
4084 function value in memory, just as large structures are always returned.
4085 Here @var{type} will be the data type of the value, and @var{fntype}
4086 will be the type of the function doing the returning, or @code{NULL} for
4089 Note that values of mode @code{BLKmode} must be explicitly handled
4090 by this function. Also, the option @option{-fpcc-struct-return}
4091 takes effect regardless of this macro. On most systems, it is
4092 possible to leave the hook undefined; this causes a default
4093 definition to be used, whose value is the constant 1 for @code{BLKmode}
4094 values, and 0 otherwise.
4096 Do not use this hook to indicate that structures and unions should always
4097 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4101 @defmac DEFAULT_PCC_STRUCT_RETURN
4102 Define this macro to be 1 if all structure and union return values must be
4103 in memory. Since this results in slower code, this should be defined
4104 only if needed for compatibility with other compilers or with an ABI@.
4105 If you define this macro to be 0, then the conventions used for structure
4106 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4109 If not defined, this defaults to the value 1.
4112 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4113 This target hook should return the location of the structure value
4114 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4115 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4116 be @code{NULL}, for libcalls. You do not need to define this target
4117 hook if the address is always passed as an ``invisible'' first
4120 On some architectures the place where the structure value address
4121 is found by the called function is not the same place that the
4122 caller put it. This can be due to register windows, or it could
4123 be because the function prologue moves it to a different place.
4124 @var{incoming} is @code{true} when the location is needed in
4125 the context of the called function, and @code{false} in the context of
4128 If @var{incoming} is @code{true} and the address is to be found on the
4129 stack, return a @code{mem} which refers to the frame pointer.
4132 @defmac PCC_STATIC_STRUCT_RETURN
4133 Define this macro if the usual system convention on the target machine
4134 for returning structures and unions is for the called function to return
4135 the address of a static variable containing the value.
4137 Do not define this if the usual system convention is for the caller to
4138 pass an address to the subroutine.
4140 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4141 nothing when you use @option{-freg-struct-return} mode.
4145 @subsection Caller-Saves Register Allocation
4147 If you enable it, GCC can save registers around function calls. This
4148 makes it possible to use call-clobbered registers to hold variables that
4149 must live across calls.
4151 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4152 A C expression to determine whether it is worthwhile to consider placing
4153 a pseudo-register in a call-clobbered hard register and saving and
4154 restoring it around each function call. The expression should be 1 when
4155 this is worth doing, and 0 otherwise.
4157 If you don't define this macro, a default is used which is good on most
4158 machines: @code{4 * @var{calls} < @var{refs}}.
4161 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4162 A C expression specifying which mode is required for saving @var{nregs}
4163 of a pseudo-register in call-clobbered hard register @var{regno}. If
4164 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4165 returned. For most machines this macro need not be defined since GCC
4166 will select the smallest suitable mode.
4169 @node Function Entry
4170 @subsection Function Entry and Exit
4171 @cindex function entry and exit
4175 This section describes the macros that output function entry
4176 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4178 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4179 If defined, a function that outputs the assembler code for entry to a
4180 function. The prologue is responsible for setting up the stack frame,
4181 initializing the frame pointer register, saving registers that must be
4182 saved, and allocating @var{size} additional bytes of storage for the
4183 local variables. @var{size} is an integer. @var{file} is a stdio
4184 stream to which the assembler code should be output.
4186 The label for the beginning of the function need not be output by this
4187 macro. That has already been done when the macro is run.
4189 @findex regs_ever_live
4190 To determine which registers to save, the macro can refer to the array
4191 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4192 @var{r} is used anywhere within the function. This implies the function
4193 prologue should save register @var{r}, provided it is not one of the
4194 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4195 @code{regs_ever_live}.)
4197 On machines that have ``register windows'', the function entry code does
4198 not save on the stack the registers that are in the windows, even if
4199 they are supposed to be preserved by function calls; instead it takes
4200 appropriate steps to ``push'' the register stack, if any non-call-used
4201 registers are used in the function.
4203 @findex frame_pointer_needed
4204 On machines where functions may or may not have frame-pointers, the
4205 function entry code must vary accordingly; it must set up the frame
4206 pointer if one is wanted, and not otherwise. To determine whether a
4207 frame pointer is in wanted, the macro can refer to the variable
4208 @code{frame_pointer_needed}. The variable's value will be 1 at run
4209 time in a function that needs a frame pointer. @xref{Elimination}.
4211 The function entry code is responsible for allocating any stack space
4212 required for the function. This stack space consists of the regions
4213 listed below. In most cases, these regions are allocated in the
4214 order listed, with the last listed region closest to the top of the
4215 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4216 the highest address if it is not defined). You can use a different order
4217 for a machine if doing so is more convenient or required for
4218 compatibility reasons. Except in cases where required by standard
4219 or by a debugger, there is no reason why the stack layout used by GCC
4220 need agree with that used by other compilers for a machine.
4223 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4224 If defined, a function that outputs assembler code at the end of a
4225 prologue. This should be used when the function prologue is being
4226 emitted as RTL, and you have some extra assembler that needs to be
4227 emitted. @xref{prologue instruction pattern}.
4230 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4231 If defined, a function that outputs assembler code at the start of an
4232 epilogue. This should be used when the function epilogue is being
4233 emitted as RTL, and you have some extra assembler that needs to be
4234 emitted. @xref{epilogue instruction pattern}.
4237 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4238 If defined, a function that outputs the assembler code for exit from a
4239 function. The epilogue is responsible for restoring the saved
4240 registers and stack pointer to their values when the function was
4241 called, and returning control to the caller. This macro takes the
4242 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4243 registers to restore are determined from @code{regs_ever_live} and
4244 @code{CALL_USED_REGISTERS} in the same way.
4246 On some machines, there is a single instruction that does all the work
4247 of returning from the function. On these machines, give that
4248 instruction the name @samp{return} and do not define the macro
4249 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4251 Do not define a pattern named @samp{return} if you want the
4252 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4253 switches to control whether return instructions or epilogues are used,
4254 define a @samp{return} pattern with a validity condition that tests the
4255 target switches appropriately. If the @samp{return} pattern's validity
4256 condition is false, epilogues will be used.
4258 On machines where functions may or may not have frame-pointers, the
4259 function exit code must vary accordingly. Sometimes the code for these
4260 two cases is completely different. To determine whether a frame pointer
4261 is wanted, the macro can refer to the variable
4262 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4263 a function that needs a frame pointer.
4265 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4266 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4267 The C variable @code{current_function_is_leaf} is nonzero for such a
4268 function. @xref{Leaf Functions}.
4270 On some machines, some functions pop their arguments on exit while
4271 others leave that for the caller to do. For example, the 68020 when
4272 given @option{-mrtd} pops arguments in functions that take a fixed
4273 number of arguments.
4275 @findex current_function_pops_args
4276 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4277 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4278 needs to know what was decided. The variable that is called
4279 @code{current_function_pops_args} is the number of bytes of its
4280 arguments that a function should pop. @xref{Scalar Return}.
4281 @c what is the "its arguments" in the above sentence referring to, pray
4282 @c tell? --mew 5feb93
4287 @findex current_function_pretend_args_size
4288 A region of @code{current_function_pretend_args_size} bytes of
4289 uninitialized space just underneath the first argument arriving on the
4290 stack. (This may not be at the very start of the allocated stack region
4291 if the calling sequence has pushed anything else since pushing the stack
4292 arguments. But usually, on such machines, nothing else has been pushed
4293 yet, because the function prologue itself does all the pushing.) This
4294 region is used on machines where an argument may be passed partly in
4295 registers and partly in memory, and, in some cases to support the
4296 features in @code{<stdarg.h>}.
4299 An area of memory used to save certain registers used by the function.
4300 The size of this area, which may also include space for such things as
4301 the return address and pointers to previous stack frames, is
4302 machine-specific and usually depends on which registers have been used
4303 in the function. Machines with register windows often do not require
4307 A region of at least @var{size} bytes, possibly rounded up to an allocation
4308 boundary, to contain the local variables of the function. On some machines,
4309 this region and the save area may occur in the opposite order, with the
4310 save area closer to the top of the stack.
4313 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4314 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4315 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4316 argument lists of the function. @xref{Stack Arguments}.
4319 @defmac EXIT_IGNORE_STACK
4320 Define this macro as a C expression that is nonzero if the return
4321 instruction or the function epilogue ignores the value of the stack
4322 pointer; in other words, if it is safe to delete an instruction to
4323 adjust the stack pointer before a return from the function. The
4326 Note that this macro's value is relevant only for functions for which
4327 frame pointers are maintained. It is never safe to delete a final
4328 stack adjustment in a function that has no frame pointer, and the
4329 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4332 @defmac EPILOGUE_USES (@var{regno})
4333 Define this macro as a C expression that is nonzero for registers that are
4334 used by the epilogue or the @samp{return} pattern. The stack and frame
4335 pointer registers are already be assumed to be used as needed.
4338 @defmac EH_USES (@var{regno})
4339 Define this macro as a C expression that is nonzero for registers that are
4340 used by the exception handling mechanism, and so should be considered live
4341 on entry to an exception edge.
4344 @defmac DELAY_SLOTS_FOR_EPILOGUE
4345 Define this macro if the function epilogue contains delay slots to which
4346 instructions from the rest of the function can be ``moved''. The
4347 definition should be a C expression whose value is an integer
4348 representing the number of delay slots there.
4351 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4352 A C expression that returns 1 if @var{insn} can be placed in delay
4353 slot number @var{n} of the epilogue.
4355 The argument @var{n} is an integer which identifies the delay slot now
4356 being considered (since different slots may have different rules of
4357 eligibility). It is never negative and is always less than the number
4358 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4359 If you reject a particular insn for a given delay slot, in principle, it
4360 may be reconsidered for a subsequent delay slot. Also, other insns may
4361 (at least in principle) be considered for the so far unfilled delay
4364 @findex current_function_epilogue_delay_list
4365 @findex final_scan_insn
4366 The insns accepted to fill the epilogue delay slots are put in an RTL
4367 list made with @code{insn_list} objects, stored in the variable
4368 @code{current_function_epilogue_delay_list}. The insn for the first
4369 delay slot comes first in the list. Your definition of the macro
4370 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4371 outputting the insns in this list, usually by calling
4372 @code{final_scan_insn}.
4374 You need not define this macro if you did not define
4375 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4378 @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})
4379 A function that outputs the assembler code for a thunk
4380 function, used to implement C++ virtual function calls with multiple
4381 inheritance. The thunk acts as a wrapper around a virtual function,
4382 adjusting the implicit object parameter before handing control off to
4385 First, emit code to add the integer @var{delta} to the location that
4386 contains the incoming first argument. Assume that this argument
4387 contains a pointer, and is the one used to pass the @code{this} pointer
4388 in C++. This is the incoming argument @emph{before} the function prologue,
4389 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4390 all other incoming arguments.
4392 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4393 made after adding @code{delta}. In particular, if @var{p} is the
4394 adjusted pointer, the following adjustment should be made:
4397 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4400 After the additions, emit code to jump to @var{function}, which is a
4401 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4402 not touch the return address. Hence returning from @var{FUNCTION} will
4403 return to whoever called the current @samp{thunk}.
4405 The effect must be as if @var{function} had been called directly with
4406 the adjusted first argument. This macro is responsible for emitting all
4407 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4408 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4410 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4411 have already been extracted from it.) It might possibly be useful on
4412 some targets, but probably not.
4414 If you do not define this macro, the target-independent code in the C++
4415 front end will generate a less efficient heavyweight thunk that calls
4416 @var{function} instead of jumping to it. The generic approach does
4417 not support varargs.
4420 @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})
4421 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4422 to output the assembler code for the thunk function specified by the
4423 arguments it is passed, and false otherwise. In the latter case, the
4424 generic approach will be used by the C++ front end, with the limitations
4429 @subsection Generating Code for Profiling
4430 @cindex profiling, code generation
4432 These macros will help you generate code for profiling.
4434 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4435 A C statement or compound statement to output to @var{file} some
4436 assembler code to call the profiling subroutine @code{mcount}.
4439 The details of how @code{mcount} expects to be called are determined by
4440 your operating system environment, not by GCC@. To figure them out,
4441 compile a small program for profiling using the system's installed C
4442 compiler and look at the assembler code that results.
4444 Older implementations of @code{mcount} expect the address of a counter
4445 variable to be loaded into some register. The name of this variable is
4446 @samp{LP} followed by the number @var{labelno}, so you would generate
4447 the name using @samp{LP%d} in a @code{fprintf}.
4450 @defmac PROFILE_HOOK
4451 A C statement or compound statement to output to @var{file} some assembly
4452 code to call the profiling subroutine @code{mcount} even the target does
4453 not support profiling.
4456 @defmac NO_PROFILE_COUNTERS
4457 Define this macro if the @code{mcount} subroutine on your system does
4458 not need a counter variable allocated for each function. This is true
4459 for almost all modern implementations. If you define this macro, you
4460 must not use the @var{labelno} argument to @code{FUNCTION_PROFILER}.
4463 @defmac PROFILE_BEFORE_PROLOGUE
4464 Define this macro if the code for function profiling should come before
4465 the function prologue. Normally, the profiling code comes after.
4469 @subsection Permitting tail calls
4472 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4473 True if it is ok to do sibling call optimization for the specified
4474 call expression @var{exp}. @var{decl} will be the called function,
4475 or @code{NULL} if this is an indirect call.
4477 It is not uncommon for limitations of calling conventions to prevent
4478 tail calls to functions outside the current unit of translation, or
4479 during PIC compilation. The hook is used to enforce these restrictions,
4480 as the @code{sibcall} md pattern can not fail, or fall over to a
4481 ``normal'' call. The criteria for successful sibling call optimization
4482 may vary greatly between different architectures.
4485 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap *@var{regs})
4486 Add any hard registers to @var{regs} that are live on entry to the
4487 function. This hook only needs to be defined to provide registers that
4488 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4489 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4490 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4491 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4494 @node Stack Smashing Protection
4495 @subsection Stack smashing protection
4496 @cindex stack smashing protection
4498 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4499 This hook returns a @code{DECL} node for the external variable to use
4500 for the stack protection guard. This variable is initialized by the
4501 runtime to some random value and is used to initialize the guard value
4502 that is placed at the top of the local stack frame. The type of this
4503 variable must be @code{ptr_type_node}.
4505 The default version of this hook creates a variable called
4506 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4509 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4510 This hook returns a tree expression that alerts the runtime that the
4511 stack protect guard variable has been modified. This expression should
4512 involve a call to a @code{noreturn} function.
4514 The default version of this hook invokes a function called
4515 @samp{__stack_chk_fail}, taking no arguments. This function is
4516 normally defined in @file{libgcc2.c}.
4520 @section Implementing the Varargs Macros
4521 @cindex varargs implementation
4523 GCC comes with an implementation of @code{<varargs.h>} and
4524 @code{<stdarg.h>} that work without change on machines that pass arguments
4525 on the stack. Other machines require their own implementations of
4526 varargs, and the two machine independent header files must have
4527 conditionals to include it.
4529 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4530 the calling convention for @code{va_start}. The traditional
4531 implementation takes just one argument, which is the variable in which
4532 to store the argument pointer. The ISO implementation of
4533 @code{va_start} takes an additional second argument. The user is
4534 supposed to write the last named argument of the function here.
4536 However, @code{va_start} should not use this argument. The way to find
4537 the end of the named arguments is with the built-in functions described
4540 @defmac __builtin_saveregs ()
4541 Use this built-in function to save the argument registers in memory so
4542 that the varargs mechanism can access them. Both ISO and traditional
4543 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
4544 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
4546 On some machines, @code{__builtin_saveregs} is open-coded under the
4547 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
4548 other machines, it calls a routine written in assembler language,
4549 found in @file{libgcc2.c}.
4551 Code generated for the call to @code{__builtin_saveregs} appears at the
4552 beginning of the function, as opposed to where the call to
4553 @code{__builtin_saveregs} is written, regardless of what the code is.
4554 This is because the registers must be saved before the function starts
4555 to use them for its own purposes.
4556 @c i rewrote the first sentence above to fix an overfull hbox. --mew
4560 @defmac __builtin_args_info (@var{category})
4561 Use this built-in function to find the first anonymous arguments in
4564 In general, a machine may have several categories of registers used for
4565 arguments, each for a particular category of data types. (For example,
4566 on some machines, floating-point registers are used for floating-point
4567 arguments while other arguments are passed in the general registers.)
4568 To make non-varargs functions use the proper calling convention, you
4569 have defined the @code{CUMULATIVE_ARGS} data type to record how many
4570 registers in each category have been used so far
4572 @code{__builtin_args_info} accesses the same data structure of type
4573 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
4574 with it, with @var{category} specifying which word to access. Thus, the
4575 value indicates the first unused register in a given category.
4577 Normally, you would use @code{__builtin_args_info} in the implementation
4578 of @code{va_start}, accessing each category just once and storing the
4579 value in the @code{va_list} object. This is because @code{va_list} will
4580 have to update the values, and there is no way to alter the
4581 values accessed by @code{__builtin_args_info}.
4584 @defmac __builtin_next_arg (@var{lastarg})
4585 This is the equivalent of @code{__builtin_args_info}, for stack
4586 arguments. It returns the address of the first anonymous stack
4587 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
4588 returns the address of the location above the first anonymous stack
4589 argument. Use it in @code{va_start} to initialize the pointer for
4590 fetching arguments from the stack. Also use it in @code{va_start} to
4591 verify that the second parameter @var{lastarg} is the last named argument
4592 of the current function.
4595 @defmac __builtin_classify_type (@var{object})
4596 Since each machine has its own conventions for which data types are
4597 passed in which kind of register, your implementation of @code{va_arg}
4598 has to embody these conventions. The easiest way to categorize the
4599 specified data type is to use @code{__builtin_classify_type} together
4600 with @code{sizeof} and @code{__alignof__}.
4602 @code{__builtin_classify_type} ignores the value of @var{object},
4603 considering only its data type. It returns an integer describing what
4604 kind of type that is---integer, floating, pointer, structure, and so on.
4606 The file @file{typeclass.h} defines an enumeration that you can use to
4607 interpret the values of @code{__builtin_classify_type}.
4610 These machine description macros help implement varargs:
4612 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
4613 If defined, this hook produces the machine-specific code for a call to
4614 @code{__builtin_saveregs}. This code will be moved to the very
4615 beginning of the function, before any parameter access are made. The
4616 return value of this function should be an RTX that contains the value
4617 to use as the return of @code{__builtin_saveregs}.
4620 @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})
4621 This target hook offers an alternative to using
4622 @code{__builtin_saveregs} and defining the hook
4623 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
4624 register arguments into the stack so that all the arguments appear to
4625 have been passed consecutively on the stack. Once this is done, you can
4626 use the standard implementation of varargs that works for machines that
4627 pass all their arguments on the stack.
4629 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
4630 structure, containing the values that are obtained after processing the
4631 named arguments. The arguments @var{mode} and @var{type} describe the
4632 last named argument---its machine mode and its data type as a tree node.
4634 The target hook should do two things: first, push onto the stack all the
4635 argument registers @emph{not} used for the named arguments, and second,
4636 store the size of the data thus pushed into the @code{int}-valued
4637 variable pointed to by @var{pretend_args_size}. The value that you
4638 store here will serve as additional offset for setting up the stack
4641 Because you must generate code to push the anonymous arguments at
4642 compile time without knowing their data types,
4643 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
4644 have just a single category of argument register and use it uniformly
4647 If the argument @var{second_time} is nonzero, it means that the
4648 arguments of the function are being analyzed for the second time. This
4649 happens for an inline function, which is not actually compiled until the
4650 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
4651 not generate any instructions in this case.
4654 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca})
4655 Define this hook to return @code{true} if the location where a function
4656 argument is passed depends on whether or not it is a named argument.
4658 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
4659 is set for varargs and stdarg functions. If this hook returns
4660 @code{true}, the @var{named} argument is always true for named
4661 arguments, and false for unnamed arguments. If it returns @code{false},
4662 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
4663 then all arguments are treated as named. Otherwise, all named arguments
4664 except the last are treated as named.
4666 You need not define this hook if it always returns zero.
4669 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED
4670 If you need to conditionally change ABIs so that one works with
4671 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
4672 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
4673 defined, then define this hook to return @code{true} if
4674 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
4675 Otherwise, you should not define this hook.
4679 @section Trampolines for Nested Functions
4680 @cindex trampolines for nested functions
4681 @cindex nested functions, trampolines for
4683 A @dfn{trampoline} is a small piece of code that is created at run time
4684 when the address of a nested function is taken. It normally resides on
4685 the stack, in the stack frame of the containing function. These macros
4686 tell GCC how to generate code to allocate and initialize a
4689 The instructions in the trampoline must do two things: load a constant
4690 address into the static chain register, and jump to the real address of
4691 the nested function. On CISC machines such as the m68k, this requires
4692 two instructions, a move immediate and a jump. Then the two addresses
4693 exist in the trampoline as word-long immediate operands. On RISC
4694 machines, it is often necessary to load each address into a register in
4695 two parts. Then pieces of each address form separate immediate
4698 The code generated to initialize the trampoline must store the variable
4699 parts---the static chain value and the function address---into the
4700 immediate operands of the instructions. On a CISC machine, this is
4701 simply a matter of copying each address to a memory reference at the
4702 proper offset from the start of the trampoline. On a RISC machine, it
4703 may be necessary to take out pieces of the address and store them
4706 @defmac TRAMPOLINE_TEMPLATE (@var{file})
4707 A C statement to output, on the stream @var{file}, assembler code for a
4708 block of data that contains the constant parts of a trampoline. This
4709 code should not include a label---the label is taken care of
4712 If you do not define this macro, it means no template is needed
4713 for the target. Do not define this macro on systems where the block move
4714 code to copy the trampoline into place would be larger than the code
4715 to generate it on the spot.
4718 @defmac TRAMPOLINE_SECTION
4719 Return the section into which the trampoline template is to be placed
4720 (@pxref{Sections}). The default value is @code{readonly_data_section}.
4723 @defmac TRAMPOLINE_SIZE
4724 A C expression for the size in bytes of the trampoline, as an integer.
4727 @defmac TRAMPOLINE_ALIGNMENT
4728 Alignment required for trampolines, in bits.
4730 If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT}
4731 is used for aligning trampolines.
4734 @defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain})
4735 A C statement to initialize the variable parts of a trampoline.
4736 @var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is
4737 an RTX for the address of the nested function; @var{static_chain} is an
4738 RTX for the static chain value that should be passed to the function
4742 @defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr})
4743 A C statement that should perform any machine-specific adjustment in
4744 the address of the trampoline. Its argument contains the address that
4745 was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be
4746 used for a function call should be different from the address in which
4747 the template was stored, the different address should be assigned to
4748 @var{addr}. If this macro is not defined, @var{addr} will be used for
4751 @cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines
4752 @cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines
4753 If this macro is not defined, by default the trampoline is allocated as
4754 a stack slot. This default is right for most machines. The exceptions
4755 are machines where it is impossible to execute instructions in the stack
4756 area. On such machines, you may have to implement a separate stack,
4757 using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE}
4758 and @code{TARGET_ASM_FUNCTION_EPILOGUE}.
4760 @var{fp} points to a data structure, a @code{struct function}, which
4761 describes the compilation status of the immediate containing function of
4762 the function which the trampoline is for. The stack slot for the
4763 trampoline is in the stack frame of this containing function. Other
4764 allocation strategies probably must do something analogous with this
4768 Implementing trampolines is difficult on many machines because they have
4769 separate instruction and data caches. Writing into a stack location
4770 fails to clear the memory in the instruction cache, so when the program
4771 jumps to that location, it executes the old contents.
4773 Here are two possible solutions. One is to clear the relevant parts of
4774 the instruction cache whenever a trampoline is set up. The other is to
4775 make all trampolines identical, by having them jump to a standard
4776 subroutine. The former technique makes trampoline execution faster; the
4777 latter makes initialization faster.
4779 To clear the instruction cache when a trampoline is initialized, define
4780 the following macro.
4782 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
4783 If defined, expands to a C expression clearing the @emph{instruction
4784 cache} in the specified interval. The definition of this macro would
4785 typically be a series of @code{asm} statements. Both @var{beg} and
4786 @var{end} are both pointer expressions.
4789 The operating system may also require the stack to be made executable
4790 before calling the trampoline. To implement this requirement, define
4791 the following macro.
4793 @defmac ENABLE_EXECUTE_STACK
4794 Define this macro if certain operations must be performed before executing
4795 code located on the stack. The macro should expand to a series of C
4796 file-scope constructs (e.g.@: functions) and provide a unique entry point
4797 named @code{__enable_execute_stack}. The target is responsible for
4798 emitting calls to the entry point in the code, for example from the
4799 @code{INITIALIZE_TRAMPOLINE} macro.
4802 To use a standard subroutine, define the following macro. In addition,
4803 you must make sure that the instructions in a trampoline fill an entire
4804 cache line with identical instructions, or else ensure that the
4805 beginning of the trampoline code is always aligned at the same point in
4806 its cache line. Look in @file{m68k.h} as a guide.
4808 @defmac TRANSFER_FROM_TRAMPOLINE
4809 Define this macro if trampolines need a special subroutine to do their
4810 work. The macro should expand to a series of @code{asm} statements
4811 which will be compiled with GCC@. They go in a library function named
4812 @code{__transfer_from_trampoline}.
4814 If you need to avoid executing the ordinary prologue code of a compiled
4815 C function when you jump to the subroutine, you can do so by placing a
4816 special label of your own in the assembler code. Use one @code{asm}
4817 statement to generate an assembler label, and another to make the label
4818 global. Then trampolines can use that label to jump directly to your
4819 special assembler code.
4823 @section Implicit Calls to Library Routines
4824 @cindex library subroutine names
4825 @cindex @file{libgcc.a}
4827 @c prevent bad page break with this line
4828 Here is an explanation of implicit calls to library routines.
4830 @defmac DECLARE_LIBRARY_RENAMES
4831 This macro, if defined, should expand to a piece of C code that will get
4832 expanded when compiling functions for libgcc.a. It can be used to
4833 provide alternate names for GCC's internal library functions if there
4834 are ABI-mandated names that the compiler should provide.
4837 @findex init_one_libfunc
4838 @findex set_optab_libfunc
4839 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
4840 This hook should declare additional library routines or rename
4841 existing ones, using the functions @code{set_optab_libfunc} and
4842 @code{init_one_libfunc} defined in @file{optabs.c}.
4843 @code{init_optabs} calls this macro after initializing all the normal
4846 The default is to do nothing. Most ports don't need to define this hook.
4849 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
4850 This macro should return @code{true} if the library routine that
4851 implements the floating point comparison operator @var{comparison} in
4852 mode @var{mode} will return a boolean, and @var{false} if it will
4855 GCC's own floating point libraries return tristates from the
4856 comparison operators, so the default returns false always. Most ports
4857 don't need to define this macro.
4860 @defmac TARGET_LIB_INT_CMP_BIASED
4861 This macro should evaluate to @code{true} if the integer comparison
4862 functions (like @code{__cmpdi2}) return 0 to indicate that the first
4863 operand is smaller than the second, 1 to indicate that they are equal,
4864 and 2 to indicate that the first operand is greater than the second.
4865 If this macro evaluates to @code{false} the comparison functions return
4866 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
4867 in @file{libgcc.a}, you do not need to define this macro.
4870 @cindex US Software GOFAST, floating point emulation library
4871 @cindex floating point emulation library, US Software GOFAST
4872 @cindex GOFAST, floating point emulation library
4873 @findex gofast_maybe_init_libfuncs
4874 @defmac US_SOFTWARE_GOFAST
4875 Define this macro if your system C library uses the US Software GOFAST
4876 library to provide floating point emulation.
4878 In addition to defining this macro, your architecture must set
4879 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
4880 else call that function from its version of that hook. It is defined
4881 in @file{config/gofast.h}, which must be included by your
4882 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
4885 If this macro is defined, the
4886 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
4887 false for @code{SFmode} and @code{DFmode} comparisons.
4890 @cindex @code{EDOM}, implicit usage
4893 The value of @code{EDOM} on the target machine, as a C integer constant
4894 expression. If you don't define this macro, GCC does not attempt to
4895 deposit the value of @code{EDOM} into @code{errno} directly. Look in
4896 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
4899 If you do not define @code{TARGET_EDOM}, then compiled code reports
4900 domain errors by calling the library function and letting it report the
4901 error. If mathematical functions on your system use @code{matherr} when
4902 there is an error, then you should leave @code{TARGET_EDOM} undefined so
4903 that @code{matherr} is used normally.
4906 @cindex @code{errno}, implicit usage
4907 @defmac GEN_ERRNO_RTX
4908 Define this macro as a C expression to create an rtl expression that
4909 refers to the global ``variable'' @code{errno}. (On certain systems,
4910 @code{errno} may not actually be a variable.) If you don't define this
4911 macro, a reasonable default is used.
4914 @cindex C99 math functions, implicit usage
4915 @defmac TARGET_C99_FUNCTIONS
4916 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
4917 @code{sinf} and similarly for other functions defined by C99 standard. The
4918 default is nonzero that should be proper value for most modern systems, however
4919 number of existing systems lacks support for these functions in the runtime so
4920 they needs this macro to be redefined to 0.
4923 @defmac NEXT_OBJC_RUNTIME
4924 Define this macro to generate code for Objective-C message sending using
4925 the calling convention of the NeXT system. This calling convention
4926 involves passing the object, the selector and the method arguments all
4927 at once to the method-lookup library function.
4929 The default calling convention passes just the object and the selector
4930 to the lookup function, which returns a pointer to the method.
4933 @node Addressing Modes
4934 @section Addressing Modes
4935 @cindex addressing modes
4937 @c prevent bad page break with this line
4938 This is about addressing modes.
4940 @defmac HAVE_PRE_INCREMENT
4941 @defmacx HAVE_PRE_DECREMENT
4942 @defmacx HAVE_POST_INCREMENT
4943 @defmacx HAVE_POST_DECREMENT
4944 A C expression that is nonzero if the machine supports pre-increment,
4945 pre-decrement, post-increment, or post-decrement addressing respectively.
4948 @defmac HAVE_PRE_MODIFY_DISP
4949 @defmacx HAVE_POST_MODIFY_DISP
4950 A C expression that is nonzero if the machine supports pre- or
4951 post-address side-effect generation involving constants other than
4952 the size of the memory operand.
4955 @defmac HAVE_PRE_MODIFY_REG
4956 @defmacx HAVE_POST_MODIFY_REG
4957 A C expression that is nonzero if the machine supports pre- or
4958 post-address side-effect generation involving a register displacement.
4961 @defmac CONSTANT_ADDRESS_P (@var{x})
4962 A C expression that is 1 if the RTX @var{x} is a constant which
4963 is a valid address. On most machines, this can be defined as
4964 @code{CONSTANT_P (@var{x})}, but a few machines are more restrictive
4965 in which constant addresses are supported.
4968 @defmac CONSTANT_P (@var{x})
4969 @code{CONSTANT_P}, which is defined by target-independent code,
4970 accepts integer-values expressions whose values are not explicitly
4971 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
4972 expressions and @code{const} arithmetic expressions, in addition to
4973 @code{const_int} and @code{const_double} expressions.
4976 @defmac MAX_REGS_PER_ADDRESS
4977 A number, the maximum number of registers that can appear in a valid
4978 memory address. Note that it is up to you to specify a value equal to
4979 the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever
4983 @defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
4984 A C compound statement with a conditional @code{goto @var{label};}
4985 executed if @var{x} (an RTX) is a legitimate memory address on the
4986 target machine for a memory operand of mode @var{mode}.
4988 It usually pays to define several simpler macros to serve as
4989 subroutines for this one. Otherwise it may be too complicated to
4992 This macro must exist in two variants: a strict variant and a
4993 non-strict one. The strict variant is used in the reload pass. It
4994 must be defined so that any pseudo-register that has not been
4995 allocated a hard register is considered a memory reference. In
4996 contexts where some kind of register is required, a pseudo-register
4997 with no hard register must be rejected.
4999 The non-strict variant is used in other passes. It must be defined to
5000 accept all pseudo-registers in every context where some kind of
5001 register is required.
5003 @findex REG_OK_STRICT
5004 Compiler source files that want to use the strict variant of this
5005 macro define the macro @code{REG_OK_STRICT}. You should use an
5006 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant
5007 in that case and the non-strict variant otherwise.
5009 Subroutines to check for acceptable registers for various purposes (one
5010 for base registers, one for index registers, and so on) are typically
5011 among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}.
5012 Then only these subroutine macros need have two variants; the higher
5013 levels of macros may be the same whether strict or not.
5015 Normally, constant addresses which are the sum of a @code{symbol_ref}
5016 and an integer are stored inside a @code{const} RTX to mark them as
5017 constant. Therefore, there is no need to recognize such sums
5018 specifically as legitimate addresses. Normally you would simply
5019 recognize any @code{const} as legitimate.
5021 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5022 sums that are not marked with @code{const}. It assumes that a naked
5023 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5024 naked constant sums as illegitimate addresses, so that none of them will
5025 be given to @code{PRINT_OPERAND_ADDRESS}.
5027 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5028 On some machines, whether a symbolic address is legitimate depends on
5029 the section that the address refers to. On these machines, define the
5030 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5031 into the @code{symbol_ref}, and then check for it here. When you see a
5032 @code{const}, you will have to look inside it to find the
5033 @code{symbol_ref} in order to determine the section. @xref{Assembler
5037 @defmac REG_OK_FOR_BASE_P (@var{x})
5038 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
5039 RTX) is valid for use as a base register. For hard registers, it
5040 should always accept those which the hardware permits and reject the
5041 others. Whether the macro accepts or rejects pseudo registers must be
5042 controlled by @code{REG_OK_STRICT} as described above. This usually
5043 requires two variant definitions, of which @code{REG_OK_STRICT}
5044 controls the one actually used.
5047 @defmac REG_MODE_OK_FOR_BASE_P (@var{x}, @var{mode})
5048 A C expression that is just like @code{REG_OK_FOR_BASE_P}, except that
5049 that expression may examine the mode of the memory reference in
5050 @var{mode}. You should define this macro if the mode of the memory
5051 reference affects whether a register may be used as a base register. If
5052 you define this macro, the compiler will use it instead of
5053 @code{REG_OK_FOR_BASE_P}.
5056 @defmac REG_MODE_OK_FOR_REG_BASE_P (@var{x}, @var{mode})
5057 A C expression which is nonzero if @var{x} (assumed to be a @code{reg} RTX)
5058 is suitable for use as a base register in base plus index operand addresses,
5059 accessing memory in mode @var{mode}. It may be either a suitable hard
5060 register or a pseudo register that has been allocated such a hard register.
5061 You should define this macro if base plus index addresses have different
5062 requirements than other base register uses.
5065 @defmac REG_OK_FOR_INDEX_P (@var{x})
5066 A C expression that is nonzero if @var{x} (assumed to be a @code{reg}
5067 RTX) is valid for use as an index register.
5069 The difference between an index register and a base register is that
5070 the index register may be scaled. If an address involves the sum of
5071 two registers, neither one of them scaled, then either one may be
5072 labeled the ``base'' and the other the ``index''; but whichever
5073 labeling is used must fit the machine's constraints of which registers
5074 may serve in each capacity. The compiler will try both labelings,
5075 looking for one that is valid, and will reload one or both registers
5076 only if neither labeling works.
5079 @defmac FIND_BASE_TERM (@var{x})
5080 A C expression to determine the base term of address @var{x}.
5081 This macro is used in only one place: `find_base_term' in alias.c.
5083 It is always safe for this macro to not be defined. It exists so
5084 that alias analysis can understand machine-dependent addresses.
5086 The typical use of this macro is to handle addresses containing
5087 a label_ref or symbol_ref within an UNSPEC@.
5090 @defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win})
5091 A C compound statement that attempts to replace @var{x} with a valid
5092 memory address for an operand of mode @var{mode}. @var{win} will be a
5093 C statement label elsewhere in the code; the macro definition may use
5096 GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win});
5100 to avoid further processing if the address has become legitimate.
5102 @findex break_out_memory_refs
5103 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5104 and @var{oldx} will be the operand that was given to that function to produce
5107 The code generated by this macro should not alter the substructure of
5108 @var{x}. If it transforms @var{x} into a more legitimate form, it
5109 should assign @var{x} (which will always be a C variable) a new value.
5111 It is not necessary for this macro to come up with a legitimate
5112 address. The compiler has standard ways of doing so in all cases. In
5113 fact, it is safe to omit this macro. But often a
5114 machine-dependent strategy can generate better code.
5117 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5118 A C compound statement that attempts to replace @var{x}, which is an address
5119 that needs reloading, with a valid memory address for an operand of mode
5120 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5121 It is not necessary to define this macro, but it might be useful for
5122 performance reasons.
5124 For example, on the i386, it is sometimes possible to use a single
5125 reload register instead of two by reloading a sum of two pseudo
5126 registers into a register. On the other hand, for number of RISC
5127 processors offsets are limited so that often an intermediate address
5128 needs to be generated in order to address a stack slot. By defining
5129 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5130 generated for adjacent some stack slots can be made identical, and thus
5133 @emph{Note}: This macro should be used with caution. It is necessary
5134 to know something of how reload works in order to effectively use this,
5135 and it is quite easy to produce macros that build in too much knowledge
5136 of reload internals.
5138 @emph{Note}: This macro must be able to reload an address created by a
5139 previous invocation of this macro. If it fails to handle such addresses
5140 then the compiler may generate incorrect code or abort.
5143 The macro definition should use @code{push_reload} to indicate parts that
5144 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5145 suitable to be passed unaltered to @code{push_reload}.
5147 The code generated by this macro must not alter the substructure of
5148 @var{x}. If it transforms @var{x} into a more legitimate form, it
5149 should assign @var{x} (which will always be a C variable) a new value.
5150 This also applies to parts that you change indirectly by calling
5153 @findex strict_memory_address_p
5154 The macro definition may use @code{strict_memory_address_p} to test if
5155 the address has become legitimate.
5158 If you want to change only a part of @var{x}, one standard way of doing
5159 this is to use @code{copy_rtx}. Note, however, that is unshares only a
5160 single level of rtl. Thus, if the part to be changed is not at the
5161 top level, you'll need to replace first the top level.
5162 It is not necessary for this macro to come up with a legitimate
5163 address; but often a machine-dependent strategy can generate better code.
5166 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5167 A C statement or compound statement with a conditional @code{goto
5168 @var{label};} executed if memory address @var{x} (an RTX) can have
5169 different meanings depending on the machine mode of the memory
5170 reference it is used for or if the address is valid for some modes
5173 Autoincrement and autodecrement addresses typically have mode-dependent
5174 effects because the amount of the increment or decrement is the size
5175 of the operand being addressed. Some machines have other mode-dependent
5176 addresses. Many RISC machines have no mode-dependent addresses.
5178 You may assume that @var{addr} is a valid address for the machine.
5181 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5182 A C expression that is nonzero if @var{x} is a legitimate constant for
5183 an immediate operand on the target machine. You can assume that
5184 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5185 @samp{1} is a suitable definition for this macro on machines where
5186 anything @code{CONSTANT_P} is valid.
5189 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5190 This hook is used to undo the possibly obfuscating effects of the
5191 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5192 macros. Some backend implementations of these macros wrap symbol
5193 references inside an @code{UNSPEC} rtx to represent PIC or similar
5194 addressing modes. This target hook allows GCC's optimizers to understand
5195 the semantics of these opaque @code{UNSPEC}s by converting them back
5196 into their original form.
5199 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x})
5200 This hook should return true if @var{x} is of a form that cannot (or
5201 should not) be spilled to the constant pool. The default version of
5202 this hook returns false.
5204 The primary reason to define this hook is to prevent reload from
5205 deciding that a non-legitimate constant would be better reloaded
5206 from the constant pool instead of spilling and reloading a register
5207 holding the constant. This restriction is often true of addresses
5208 of TLS symbols for various targets.
5211 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x})
5212 This hook should return true if pool entries for constant @var{x} can
5213 be placed in an @code{object_block} structure. @var{mode} is the mode
5216 The default version returns false for all constants.
5219 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5220 This hook should return the DECL of a function @var{f} that given an
5221 address @var{addr} as an argument returns a mask @var{m} that can be
5222 used to extract from two vectors the relevant data that resides in
5223 @var{addr} in case @var{addr} is not properly aligned.
5225 The autovectrizer, when vectorizing a load operation from an address
5226 @var{addr} that may be unaligned, will generate two vector loads from
5227 the two aligned addresses around @var{addr}. It then generates a
5228 @code{REALIGN_LOAD} operation to extract the relevant data from the
5229 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5230 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5231 the third argument, @var{OFF}, defines how the data will be extracted
5232 from these two vectors: if @var{OFF} is 0, then the returned vector is
5233 @var{v2}; otherwise, the returned vector is composed from the last
5234 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5235 @var{OFF} elements of @var{v2}.
5237 If this hook is defined, the autovectorizer will generate a call
5238 to @var{f} (using the DECL tree that this hook returns) and will
5239 use the return value of @var{f} as the argument @var{OFF} to
5240 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5241 should comply with the semantics expected by @code{REALIGN_LOAD}
5243 If this hook is not defined, then @var{addr} will be used as
5244 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5245 log2(@var{VS})-1 bits of @var{addr} will be considered.
5248 @node Anchored Addresses
5249 @section Anchored Addresses
5250 @cindex anchored addresses
5251 @cindex @option{-fsection-anchors}
5253 GCC usually addresses every static object as a separate entity.
5254 For example, if we have:
5258 int foo (void) @{ return a + b + c; @}
5261 the code for @code{foo} will usually calculate three separate symbolic
5262 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5263 it would be better to calculate just one symbolic address and access
5264 the three variables relative to it. The equivalent pseudocode would
5270 register int *xr = &x;
5271 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5275 (which isn't valid C). We refer to shared addresses like @code{x} as
5276 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5278 The hooks below describe the target properties that GCC needs to know
5279 in order to make effective use of section anchors. It won't use
5280 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5281 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5283 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5284 The minimum offset that should be applied to a section anchor.
5285 On most targets, it should be the smallest offset that can be
5286 applied to a base register while still giving a legitimate address
5287 for every mode. The default value is 0.
5290 @deftypevar {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5291 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5292 offset that should be applied to section anchors. The default
5296 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5297 Write the assembly code to define section anchor @var{x}, which is a
5298 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5299 The hook is called with the assembly output position set to the beginning
5300 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5302 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5303 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5304 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5305 is @code{NULL}, which disables the use of section anchors altogether.
5308 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (rtx @var{x})
5309 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5310 @var{x}. You can assume @samp{SYMBOL_REF_IN_BLOCK_P (@var{x})} and
5311 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5313 The default version is correct for most targets, but you might need to
5314 intercept this hook to handle things like target-specific attributes
5315 or target-specific sections.
5318 @node Condition Code
5319 @section Condition Code Status
5320 @cindex condition code status
5322 @c prevent bad page break with this line
5323 This describes the condition code status.
5326 The file @file{conditions.h} defines a variable @code{cc_status} to
5327 describe how the condition code was computed (in case the interpretation of
5328 the condition code depends on the instruction that it was set by). This
5329 variable contains the RTL expressions on which the condition code is
5330 currently based, and several standard flags.
5332 Sometimes additional machine-specific flags must be defined in the machine
5333 description header file. It can also add additional machine-specific
5334 information by defining @code{CC_STATUS_MDEP}.
5336 @defmac CC_STATUS_MDEP
5337 C code for a data type which is used for declaring the @code{mdep}
5338 component of @code{cc_status}. It defaults to @code{int}.
5340 This macro is not used on machines that do not use @code{cc0}.
5343 @defmac CC_STATUS_MDEP_INIT
5344 A C expression to initialize the @code{mdep} field to ``empty''.
5345 The default definition does nothing, since most machines don't use
5346 the field anyway. If you want to use the field, you should probably
5347 define this macro to initialize it.
5349 This macro is not used on machines that do not use @code{cc0}.
5352 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5353 A C compound statement to set the components of @code{cc_status}
5354 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5355 this macro's responsibility to recognize insns that set the condition
5356 code as a byproduct of other activity as well as those that explicitly
5359 This macro is not used on machines that do not use @code{cc0}.
5361 If there are insns that do not set the condition code but do alter
5362 other machine registers, this macro must check to see whether they
5363 invalidate the expressions that the condition code is recorded as
5364 reflecting. For example, on the 68000, insns that store in address
5365 registers do not set the condition code, which means that usually
5366 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5367 insns. But suppose that the previous insn set the condition code
5368 based on location @samp{a4@@(102)} and the current insn stores a new
5369 value in @samp{a4}. Although the condition code is not changed by
5370 this, it will no longer be true that it reflects the contents of
5371 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5372 @code{cc_status} in this case to say that nothing is known about the
5373 condition code value.
5375 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5376 with the results of peephole optimization: insns whose patterns are
5377 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5378 constants which are just the operands. The RTL structure of these
5379 insns is not sufficient to indicate what the insns actually do. What
5380 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5381 @code{CC_STATUS_INIT}.
5383 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5384 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5385 @samp{cc}. This avoids having detailed information about patterns in
5386 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5389 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5390 Returns a mode from class @code{MODE_CC} to be used when comparison
5391 operation code @var{op} is applied to rtx @var{x} and @var{y}. For
5392 example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see
5393 @pxref{Jump Patterns} for a description of the reason for this
5397 #define SELECT_CC_MODE(OP,X,Y) \
5398 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5399 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5400 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5401 || GET_CODE (X) == NEG) \
5402 ? CC_NOOVmode : CCmode))
5405 You should define this macro if and only if you define extra CC modes
5406 in @file{@var{machine}-modes.def}.
5409 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
5410 On some machines not all possible comparisons are defined, but you can
5411 convert an invalid comparison into a valid one. For example, the Alpha
5412 does not have a @code{GT} comparison, but you can use an @code{LT}
5413 comparison instead and swap the order of the operands.
5415 On such machines, define this macro to be a C statement to do any
5416 required conversions. @var{code} is the initial comparison code
5417 and @var{op0} and @var{op1} are the left and right operands of the
5418 comparison, respectively. You should modify @var{code}, @var{op0}, and
5419 @var{op1} as required.
5421 GCC will not assume that the comparison resulting from this macro is
5422 valid but will see if the resulting insn matches a pattern in the
5425 You need not define this macro if it would never change the comparison
5429 @defmac REVERSIBLE_CC_MODE (@var{mode})
5430 A C expression whose value is one if it is always safe to reverse a
5431 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
5432 can ever return @var{mode} for a floating-point inequality comparison,
5433 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
5435 You need not define this macro if it would always returns zero or if the
5436 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
5437 For example, here is the definition used on the SPARC, where floating-point
5438 inequality comparisons are always given @code{CCFPEmode}:
5441 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
5445 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
5446 A C expression whose value is reversed condition code of the @var{code} for
5447 comparison done in CC_MODE @var{mode}. The macro is used only in case
5448 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
5449 machine has some non-standard way how to reverse certain conditionals. For
5450 instance in case all floating point conditions are non-trapping, compiler may
5451 freely convert unordered compares to ordered one. Then definition may look
5455 #define REVERSE_CONDITION(CODE, MODE) \
5456 ((MODE) != CCFPmode ? reverse_condition (CODE) \
5457 : reverse_condition_maybe_unordered (CODE))
5461 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
5462 A C expression that returns true if the conditional execution predicate
5463 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
5464 versa. Define this to return 0 if the target has conditional execution
5465 predicates that cannot be reversed safely. There is no need to validate
5466 that the arguments of op1 and op2 are the same, this is done separately.
5467 If no expansion is specified, this macro is defined as follows:
5470 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
5471 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
5475 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *)
5476 On targets which do not use @code{(cc0)}, and which use a hard
5477 register rather than a pseudo-register to hold condition codes, the
5478 regular CSE passes are often not able to identify cases in which the
5479 hard register is set to a common value. Use this hook to enable a
5480 small pass which optimizes such cases. This hook should return true
5481 to enable this pass, and it should set the integers to which its
5482 arguments point to the hard register numbers used for condition codes.
5483 When there is only one such register, as is true on most systems, the
5484 integer pointed to by the second argument should be set to
5485 @code{INVALID_REGNUM}.
5487 The default version of this hook returns false.
5490 @deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode)
5491 On targets which use multiple condition code modes in class
5492 @code{MODE_CC}, it is sometimes the case that a comparison can be
5493 validly done in more than one mode. On such a system, define this
5494 target hook to take two mode arguments and to return a mode in which
5495 both comparisons may be validly done. If there is no such mode,
5496 return @code{VOIDmode}.
5498 The default version of this hook checks whether the modes are the
5499 same. If they are, it returns that mode. If they are different, it
5500 returns @code{VOIDmode}.
5504 @section Describing Relative Costs of Operations
5505 @cindex costs of instructions
5506 @cindex relative costs
5507 @cindex speed of instructions
5509 These macros let you describe the relative speed of various operations
5510 on the target machine.
5512 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
5513 A C expression for the cost of moving data of mode @var{mode} from a
5514 register in class @var{from} to one in class @var{to}. The classes are
5515 expressed using the enumeration values such as @code{GENERAL_REGS}. A
5516 value of 2 is the default; other values are interpreted relative to
5519 It is not required that the cost always equal 2 when @var{from} is the
5520 same as @var{to}; on some machines it is expensive to move between
5521 registers if they are not general registers.
5523 If reload sees an insn consisting of a single @code{set} between two
5524 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
5525 classes returns a value of 2, reload does not check to ensure that the
5526 constraints of the insn are met. Setting a cost of other than 2 will
5527 allow reload to verify that the constraints are met. You should do this
5528 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
5531 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
5532 A C expression for the cost of moving data of mode @var{mode} between a
5533 register of class @var{class} and memory; @var{in} is zero if the value
5534 is to be written to memory, nonzero if it is to be read in. This cost
5535 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
5536 registers and memory is more expensive than between two registers, you
5537 should define this macro to express the relative cost.
5539 If you do not define this macro, GCC uses a default cost of 4 plus
5540 the cost of copying via a secondary reload register, if one is
5541 needed. If your machine requires a secondary reload register to copy
5542 between memory and a register of @var{class} but the reload mechanism is
5543 more complex than copying via an intermediate, define this macro to
5544 reflect the actual cost of the move.
5546 GCC defines the function @code{memory_move_secondary_cost} if
5547 secondary reloads are needed. It computes the costs due to copying via
5548 a secondary register. If your machine copies from memory using a
5549 secondary register in the conventional way but the default base value of
5550 4 is not correct for your machine, define this macro to add some other
5551 value to the result of that function. The arguments to that function
5552 are the same as to this macro.
5556 A C expression for the cost of a branch instruction. A value of 1 is
5557 the default; other values are interpreted relative to that.
5560 Here are additional macros which do not specify precise relative costs,
5561 but only that certain actions are more expensive than GCC would
5564 @defmac SLOW_BYTE_ACCESS
5565 Define this macro as a C expression which is nonzero if accessing less
5566 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
5567 faster than accessing a word of memory, i.e., if such access
5568 require more than one instruction or if there is no difference in cost
5569 between byte and (aligned) word loads.
5571 When this macro is not defined, the compiler will access a field by
5572 finding the smallest containing object; when it is defined, a fullword
5573 load will be used if alignment permits. Unless bytes accesses are
5574 faster than word accesses, using word accesses is preferable since it
5575 may eliminate subsequent memory access if subsequent accesses occur to
5576 other fields in the same word of the structure, but to different bytes.
5579 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
5580 Define this macro to be the value 1 if memory accesses described by the
5581 @var{mode} and @var{alignment} parameters have a cost many times greater
5582 than aligned accesses, for example if they are emulated in a trap
5585 When this macro is nonzero, the compiler will act as if
5586 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
5587 moves. This can cause significantly more instructions to be produced.
5588 Therefore, do not set this macro nonzero if unaligned accesses only add a
5589 cycle or two to the time for a memory access.
5591 If the value of this macro is always zero, it need not be defined. If
5592 this macro is defined, it should produce a nonzero value when
5593 @code{STRICT_ALIGNMENT} is nonzero.
5597 The threshold of number of scalar memory-to-memory move insns, @emph{below}
5598 which a sequence of insns should be generated instead of a
5599 string move insn or a library call. Increasing the value will always
5600 make code faster, but eventually incurs high cost in increased code size.
5602 Note that on machines where the corresponding move insn is a
5603 @code{define_expand} that emits a sequence of insns, this macro counts
5604 the number of such sequences.
5606 If you don't define this, a reasonable default is used.
5609 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
5610 A C expression used to determine whether @code{move_by_pieces} will be used to
5611 copy a chunk of memory, or whether some other block move mechanism
5612 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5613 than @code{MOVE_RATIO}.
5616 @defmac MOVE_MAX_PIECES
5617 A C expression used by @code{move_by_pieces} to determine the largest unit
5618 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
5622 The threshold of number of scalar move insns, @emph{below} which a sequence
5623 of insns should be generated to clear memory instead of a string clear insn
5624 or a library call. Increasing the value will always make code faster, but
5625 eventually incurs high cost in increased code size.
5627 If you don't define this, a reasonable default is used.
5630 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
5631 A C expression used to determine whether @code{clear_by_pieces} will be used
5632 to clear a chunk of memory, or whether some other block clear mechanism
5633 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5634 than @code{CLEAR_RATIO}.
5637 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
5638 A C expression used to determine whether @code{store_by_pieces} will be
5639 used to set a chunk of memory to a constant value, or whether some other
5640 mechanism will be used. Used by @code{__builtin_memset} when storing
5641 values other than constant zero and by @code{__builtin_strcpy} when
5642 when called with a constant source string.
5643 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
5644 than @code{MOVE_RATIO}.
5647 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
5648 A C expression used to determine whether a load postincrement is a good
5649 thing to use for a given mode. Defaults to the value of
5650 @code{HAVE_POST_INCREMENT}.
5653 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
5654 A C expression used to determine whether a load postdecrement is a good
5655 thing to use for a given mode. Defaults to the value of
5656 @code{HAVE_POST_DECREMENT}.
5659 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
5660 A C expression used to determine whether a load preincrement is a good
5661 thing to use for a given mode. Defaults to the value of
5662 @code{HAVE_PRE_INCREMENT}.
5665 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
5666 A C expression used to determine whether a load predecrement is a good
5667 thing to use for a given mode. Defaults to the value of
5668 @code{HAVE_PRE_DECREMENT}.
5671 @defmac USE_STORE_POST_INCREMENT (@var{mode})
5672 A C expression used to determine whether a store postincrement is a good
5673 thing to use for a given mode. Defaults to the value of
5674 @code{HAVE_POST_INCREMENT}.
5677 @defmac USE_STORE_POST_DECREMENT (@var{mode})
5678 A C expression used to determine whether a store postdecrement is a good
5679 thing to use for a given mode. Defaults to the value of
5680 @code{HAVE_POST_DECREMENT}.
5683 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
5684 This macro is used to determine whether a store preincrement is a good
5685 thing to use for a given mode. Defaults to the value of
5686 @code{HAVE_PRE_INCREMENT}.
5689 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
5690 This macro is used to determine whether a store predecrement is a good
5691 thing to use for a given mode. Defaults to the value of
5692 @code{HAVE_PRE_DECREMENT}.
5695 @defmac NO_FUNCTION_CSE
5696 Define this macro if it is as good or better to call a constant
5697 function address than to call an address kept in a register.
5700 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
5701 Define this macro if a non-short-circuit operation produced by
5702 @samp{fold_range_test ()} is optimal. This macro defaults to true if
5703 @code{BRANCH_COST} is greater than or equal to the value 2.
5706 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total})
5707 This target hook describes the relative costs of RTL expressions.
5709 The cost may depend on the precise form of the expression, which is
5710 available for examination in @var{x}, and the rtx code of the expression
5711 in which it is contained, found in @var{outer_code}. @var{code} is the
5712 expression code---redundant, since it can be obtained with
5713 @code{GET_CODE (@var{x})}.
5715 In implementing this hook, you can use the construct
5716 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
5719 On entry to the hook, @code{*@var{total}} contains a default estimate
5720 for the cost of the expression. The hook should modify this value as
5721 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
5722 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
5723 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
5725 When optimizing for code size, i.e.@: when @code{optimize_size} is
5726 nonzero, this target hook should be used to estimate the relative
5727 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
5729 The hook returns true when all subexpressions of @var{x} have been
5730 processed, and false when @code{rtx_cost} should recurse.
5733 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address})
5734 This hook computes the cost of an addressing mode that contains
5735 @var{address}. If not defined, the cost is computed from
5736 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
5738 For most CISC machines, the default cost is a good approximation of the
5739 true cost of the addressing mode. However, on RISC machines, all
5740 instructions normally have the same length and execution time. Hence
5741 all addresses will have equal costs.
5743 In cases where more than one form of an address is known, the form with
5744 the lowest cost will be used. If multiple forms have the same, lowest,
5745 cost, the one that is the most complex will be used.
5747 For example, suppose an address that is equal to the sum of a register
5748 and a constant is used twice in the same basic block. When this macro
5749 is not defined, the address will be computed in a register and memory
5750 references will be indirect through that register. On machines where
5751 the cost of the addressing mode containing the sum is no higher than
5752 that of a simple indirect reference, this will produce an additional
5753 instruction and possibly require an additional register. Proper
5754 specification of this macro eliminates this overhead for such machines.
5756 This hook is never called with an invalid address.
5758 On machines where an address involving more than one register is as
5759 cheap as an address computation involving only one register, defining
5760 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
5761 be live over a region of code where only one would have been if
5762 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
5763 should be considered in the definition of this macro. Equivalent costs
5764 should probably only be given to addresses with different numbers of
5765 registers on machines with lots of registers.
5769 @section Adjusting the Instruction Scheduler
5771 The instruction scheduler may need a fair amount of machine-specific
5772 adjustment in order to produce good code. GCC provides several target
5773 hooks for this purpose. It is usually enough to define just a few of
5774 them: try the first ones in this list first.
5776 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
5777 This hook returns the maximum number of instructions that can ever
5778 issue at the same time on the target machine. The default is one.
5779 Although the insn scheduler can define itself the possibility of issue
5780 an insn on the same cycle, the value can serve as an additional
5781 constraint to issue insns on the same simulated processor cycle (see
5782 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
5783 This value must be constant over the entire compilation. If you need
5784 it to vary depending on what the instructions are, you must use
5785 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
5788 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
5789 This hook is executed by the scheduler after it has scheduled an insn
5790 from the ready list. It should return the number of insns which can
5791 still be issued in the current cycle. The default is
5792 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
5793 @code{USE}, which normally are not counted against the issue rate.
5794 You should define this hook if some insns take more machine resources
5795 than others, so that fewer insns can follow them in the same cycle.
5796 @var{file} is either a null pointer, or a stdio stream to write any
5797 debug output to. @var{verbose} is the verbose level provided by
5798 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
5802 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
5803 This function corrects the value of @var{cost} based on the
5804 relationship between @var{insn} and @var{dep_insn} through the
5805 dependence @var{link}. It should return the new value. The default
5806 is to make no adjustment to @var{cost}. This can be used for example
5807 to specify to the scheduler using the traditional pipeline description
5808 that an output- or anti-dependence does not incur the same cost as a
5809 data-dependence. If the scheduler using the automaton based pipeline
5810 description, the cost of anti-dependence is zero and the cost of
5811 output-dependence is maximum of one and the difference of latency
5812 times of the first and the second insns. If these values are not
5813 acceptable, you could use the hook to modify them too. See also
5814 @pxref{Processor pipeline description}.
5817 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
5818 This hook adjusts the integer scheduling priority @var{priority} of
5819 @var{insn}. It should return the new priority. Reduce the priority to
5820 execute @var{insn} earlier, increase the priority to execute @var{insn}
5821 later. Do not define this hook if you do not need to adjust the
5822 scheduling priorities of insns.
5825 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
5826 This hook is executed by the scheduler after it has scheduled the ready
5827 list, to allow the machine description to reorder it (for example to
5828 combine two small instructions together on @samp{VLIW} machines).
5829 @var{file} is either a null pointer, or a stdio stream to write any
5830 debug output to. @var{verbose} is the verbose level provided by
5831 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
5832 list of instructions that are ready to be scheduled. @var{n_readyp} is
5833 a pointer to the number of elements in the ready list. The scheduler
5834 reads the ready list in reverse order, starting with
5835 @var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock}
5836 is the timer tick of the scheduler. You may modify the ready list and
5837 the number of ready insns. The return value is the number of insns that
5838 can issue this cycle; normally this is just @code{issue_rate}. See also
5839 @samp{TARGET_SCHED_REORDER2}.
5842 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock})
5843 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
5844 function is called whenever the scheduler starts a new cycle. This one
5845 is called once per iteration over a cycle, immediately after
5846 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
5847 return the number of insns to be scheduled in the same cycle. Defining
5848 this hook can be useful if there are frequent situations where
5849 scheduling one insn causes other insns to become ready in the same
5850 cycle. These other insns can then be taken into account properly.
5853 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
5854 This hook is called after evaluation forward dependencies of insns in
5855 chain given by two parameter values (@var{head} and @var{tail}
5856 correspondingly) but before insns scheduling of the insn chain. For
5857 example, it can be used for better insn classification if it requires
5858 analysis of dependencies. This hook can use backward and forward
5859 dependencies of the insn scheduler because they are already
5863 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
5864 This hook is executed by the scheduler at the beginning of each block of
5865 instructions that are to be scheduled. @var{file} is either a null
5866 pointer, or a stdio stream to write any debug output to. @var{verbose}
5867 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5868 @var{max_ready} is the maximum number of insns in the current scheduling
5869 region that can be live at the same time. This can be used to allocate
5870 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
5873 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
5874 This hook is executed by the scheduler at the end of each block of
5875 instructions that are to be scheduled. It can be used to perform
5876 cleanup of any actions done by the other scheduling hooks. @var{file}
5877 is either a null pointer, or a stdio stream to write any debug output
5878 to. @var{verbose} is the verbose level provided by
5879 @option{-fsched-verbose-@var{n}}.
5882 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
5883 This hook is executed by the scheduler after function level initializations.
5884 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5885 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5886 @var{old_max_uid} is the maximum insn uid when scheduling begins.
5889 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
5890 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
5891 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
5892 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
5895 @deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
5896 The hook returns an RTL insn. The automaton state used in the
5897 pipeline hazard recognizer is changed as if the insn were scheduled
5898 when the new simulated processor cycle starts. Usage of the hook may
5899 simplify the automaton pipeline description for some @acronym{VLIW}
5900 processors. If the hook is defined, it is used only for the automaton
5901 based pipeline description. The default is not to change the state
5902 when the new simulated processor cycle starts.
5905 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
5906 The hook can be used to initialize data used by the previous hook.
5909 @deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
5910 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
5911 to changed the state as if the insn were scheduled when the new
5912 simulated processor cycle finishes.
5915 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
5916 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
5917 used to initialize data used by the previous hook.
5920 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
5921 This hook controls better choosing an insn from the ready insn queue
5922 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
5923 chooses the first insn from the queue. If the hook returns a positive
5924 value, an additional scheduler code tries all permutations of
5925 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
5926 subsequent ready insns to choose an insn whose issue will result in
5927 maximal number of issued insns on the same cycle. For the
5928 @acronym{VLIW} processor, the code could actually solve the problem of
5929 packing simple insns into the @acronym{VLIW} insn. Of course, if the
5930 rules of @acronym{VLIW} packing are described in the automaton.
5932 This code also could be used for superscalar @acronym{RISC}
5933 processors. Let us consider a superscalar @acronym{RISC} processor
5934 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
5935 @var{B}, some insns can be executed only in pipelines @var{B} or
5936 @var{C}, and one insn can be executed in pipeline @var{B}. The
5937 processor may issue the 1st insn into @var{A} and the 2nd one into
5938 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
5939 until the next cycle. If the scheduler issues the 3rd insn the first,
5940 the processor could issue all 3 insns per cycle.
5942 Actually this code demonstrates advantages of the automaton based
5943 pipeline hazard recognizer. We try quickly and easy many insn
5944 schedules to choose the best one.
5946 The default is no multipass scheduling.
5949 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx)
5951 This hook controls what insns from the ready insn queue will be
5952 considered for the multipass insn scheduling. If the hook returns
5953 zero for insn passed as the parameter, the insn will be not chosen to
5956 The default is that any ready insns can be chosen to be issued.
5959 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *)
5961 This hook is called by the insn scheduler before issuing insn passed
5962 as the third parameter on given cycle. If the hook returns nonzero,
5963 the insn is not issued on given processors cycle. Instead of that,
5964 the processor cycle is advanced. If the value passed through the last
5965 parameter is zero, the insn ready queue is not sorted on the new cycle
5966 start as usually. The first parameter passes file for debugging
5967 output. The second one passes the scheduler verbose level of the
5968 debugging output. The forth and the fifth parameter values are
5969 correspondingly processor cycle on which the previous insn has been
5970 issued and the current processor cycle.
5973 @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})
5974 This hook is used to define which dependences are considered costly by
5975 the target, so costly that it is not advisable to schedule the insns that
5976 are involved in the dependence too close to one another. The parameters
5977 to this hook are as follows: The second parameter @var{insn2} is dependent
5978 upon the first parameter @var{insn1}. The dependence between @var{insn1}
5979 and @var{insn2} is represented by the third parameter @var{dep_link}. The
5980 fourth parameter @var{cost} is the cost of the dependence, and the fifth
5981 parameter @var{distance} is the distance in cycles between the two insns.
5982 The hook returns @code{true} if considering the distance between the two
5983 insns the dependence between them is considered costly by the target,
5984 and @code{false} otherwise.
5986 Defining this hook can be useful in multiple-issue out-of-order machines,
5987 where (a) it's practically hopeless to predict the actual data/resource
5988 delays, however: (b) there's a better chance to predict the actual grouping
5989 that will be formed, and (c) correctly emulating the grouping can be very
5990 important. In such targets one may want to allow issuing dependent insns
5991 closer to one another---i.e., closer than the dependence distance; however,
5992 not in cases of "costly dependences", which this hooks allows to define.
5996 @section Dividing the Output into Sections (Texts, Data, @dots{})
5997 @c the above section title is WAY too long. maybe cut the part between
5998 @c the (...)? --mew 10feb93
6000 An object file is divided into sections containing different types of
6001 data. In the most common case, there are three sections: the @dfn{text
6002 section}, which holds instructions and read-only data; the @dfn{data
6003 section}, which holds initialized writable data; and the @dfn{bss
6004 section}, which holds uninitialized data. Some systems have other kinds
6007 @file{varasm.c} provides several well-known sections, such as
6008 @code{text_section}, @code{data_section} and @code{bss_section}.
6009 The normal way of controlling a @code{@var{foo}_section} variable
6010 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6011 as described below. The macros are only read once, when @file{varasm.c}
6012 initializes itself, so their values must be run-time constants.
6013 They may however depend on command-line flags.
6015 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6016 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6017 to be string literals.
6019 Some assemblers require a different string to be written every time a
6020 section is selected. If your assembler falls into this category, you
6021 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6022 @code{get_unnamed_section} to set up the sections.
6024 You must always create a @code{text_section}, either by defining
6025 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6026 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6027 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6028 create a distinct @code{readonly_data_section}, the default is to
6029 reuse @code{text_section}.
6031 All the other @file{varasm.c} sections are optional, and are null
6032 if the target does not provide them.
6034 @defmac TEXT_SECTION_ASM_OP
6035 A C expression whose value is a string, including spacing, containing the
6036 assembler operation that should precede instructions and read-only data.
6037 Normally @code{"\t.text"} is right.
6040 @defmac HOT_TEXT_SECTION_NAME
6041 If defined, a C string constant for the name of the section containing most
6042 frequently executed functions of the program. If not defined, GCC will provide
6043 a default definition if the target supports named sections.
6046 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6047 If defined, a C string constant for the name of the section containing unlikely
6048 executed functions in the program.
6051 @defmac DATA_SECTION_ASM_OP
6052 A C expression whose value is a string, including spacing, containing the
6053 assembler operation to identify the following data as writable initialized
6054 data. Normally @code{"\t.data"} is right.
6057 @defmac SDATA_SECTION_ASM_OP
6058 If defined, a C expression whose value is a string, including spacing,
6059 containing the assembler operation to identify the following data as
6060 initialized, writable small data.
6063 @defmac READONLY_DATA_SECTION_ASM_OP
6064 A C expression whose value is a string, including spacing, containing the
6065 assembler operation to identify the following data as read-only initialized
6069 @defmac BSS_SECTION_ASM_OP
6070 If defined, a C expression whose value is a string, including spacing,
6071 containing the assembler operation to identify the following data as
6072 uninitialized global data. If not defined, and neither
6073 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6074 uninitialized global data will be output in the data section if
6075 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6079 @defmac SBSS_SECTION_ASM_OP
6080 If defined, a C expression whose value is a string, including spacing,
6081 containing the assembler operation to identify the following data as
6082 uninitialized, writable small data.
6085 @defmac INIT_SECTION_ASM_OP
6086 If defined, a C expression whose value is a string, including spacing,
6087 containing the assembler operation to identify the following data as
6088 initialization code. If not defined, GCC will assume such a section does
6089 not exist. This section has no corresponding @code{init_section}
6090 variable; it is used entirely in runtime code.
6093 @defmac FINI_SECTION_ASM_OP
6094 If defined, a C expression whose value is a string, including spacing,
6095 containing the assembler operation to identify the following data as
6096 finalization code. If not defined, GCC will assume such a section does
6097 not exist. This section has no corresponding @code{fini_section}
6098 variable; it is used entirely in runtime code.
6101 @defmac INIT_ARRAY_SECTION_ASM_OP
6102 If defined, a C expression whose value is a string, including spacing,
6103 containing the assembler operation to identify the following data as
6104 part of the @code{.init_array} (or equivalent) section. If not
6105 defined, GCC will assume such a section does not exist. Do not define
6106 both this macro and @code{INIT_SECTION_ASM_OP}.
6109 @defmac FINI_ARRAY_SECTION_ASM_OP
6110 If defined, a C expression whose value is a string, including spacing,
6111 containing the assembler operation to identify the following data as
6112 part of the @code{.fini_array} (or equivalent) section. If not
6113 defined, GCC will assume such a section does not exist. Do not define
6114 both this macro and @code{FINI_SECTION_ASM_OP}.
6117 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6118 If defined, an ASM statement that switches to a different section
6119 via @var{section_op}, calls @var{function}, and switches back to
6120 the text section. This is used in @file{crtstuff.c} if
6121 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6122 to initialization and finalization functions from the init and fini
6123 sections. By default, this macro uses a simple function call. Some
6124 ports need hand-crafted assembly code to avoid dependencies on
6125 registers initialized in the function prologue or to ensure that
6126 constant pools don't end up too far way in the text section.
6129 @defmac FORCE_CODE_SECTION_ALIGN
6130 If defined, an ASM statement that aligns a code section to some
6131 arbitrary boundary. This is used to force all fragments of the
6132 @code{.init} and @code{.fini} sections to have to same alignment
6133 and thus prevent the linker from having to add any padding.
6136 @defmac JUMP_TABLES_IN_TEXT_SECTION
6137 Define this macro to be an expression with a nonzero value if jump
6138 tables (for @code{tablejump} insns) should be output in the text
6139 section, along with the assembler instructions. Otherwise, the
6140 readonly data section is used.
6142 This macro is irrelevant if there is no separate readonly data section.
6145 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
6146 Define this hook if you need to do something special to set up the
6147 @file{varasm.c} sections, or if your target has some special sections
6148 of its own that you need to create.
6150 GCC calls this hook after processing the command line, but before writing
6151 any assembly code, and before calling any of the section-returning hooks
6155 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
6156 Return the section into which @var{exp} should be placed. You can
6157 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6158 some sort. @var{reloc} indicates whether the initial value of @var{exp}
6159 requires link-time relocations. Bit 0 is set when variable contains
6160 local relocations only, while bit 1 is set for global relocations.
6161 @var{align} is the constant alignment in bits.
6163 The default version of this function takes care of putting read-only
6164 variables in @code{readonly_data_section}.
6166 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
6169 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
6170 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
6171 for @code{FUNCTION_DECL}s as well as for variables and constants.
6173 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
6174 function has been determined to be likely to be called, and nonzero if
6175 it is unlikely to be called.
6178 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
6179 Build up a unique section name, expressed as a @code{STRING_CST} node,
6180 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
6181 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
6182 the initial value of @var{exp} requires link-time relocations.
6184 The default version of this function appends the symbol name to the
6185 ELF section name that would normally be used for the symbol. For
6186 example, the function @code{foo} would be placed in @code{.text.foo}.
6187 Whatever the actual target object format, this is often good enough.
6190 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
6191 Return the readonly data section associated with
6192 @samp{DECL_SECTION_NAME (@var{decl})}.
6193 The default version of this function selects @code{.gnu.linkonce.r.name} if
6194 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
6195 if function is in @code{.text.name}, and the normal readonly-data section
6199 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
6200 Return the section into which a constant @var{x}, of mode @var{mode},
6201 should be placed. You can assume that @var{x} is some kind of
6202 constant in RTL@. The argument @var{mode} is redundant except in the
6203 case of a @code{const_int} rtx. @var{align} is the constant alignment
6206 The default version of this function takes care of putting symbolic
6207 constants in @code{flag_pic} mode in @code{data_section} and everything
6208 else in @code{readonly_data_section}.
6211 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
6212 Define this hook if references to a symbol or a constant must be
6213 treated differently depending on something about the variable or
6214 function named by the symbol (such as what section it is in).
6216 The hook is executed immediately after rtl has been created for
6217 @var{decl}, which may be a variable or function declaration or
6218 an entry in the constant pool. In either case, @var{rtl} is the
6219 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
6220 in this hook; that field may not have been initialized yet.
6222 In the case of a constant, it is safe to assume that the rtl is
6223 a @code{mem} whose address is a @code{symbol_ref}. Most decls
6224 will also have this form, but that is not guaranteed. Global
6225 register variables, for instance, will have a @code{reg} for their
6226 rtl. (Normally the right thing to do with such unusual rtl is
6229 The @var{new_decl_p} argument will be true if this is the first time
6230 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
6231 be false for subsequent invocations, which will happen for duplicate
6232 declarations. Whether or not anything must be done for the duplicate
6233 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
6234 @var{new_decl_p} is always true when the hook is called for a constant.
6236 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
6237 The usual thing for this hook to do is to record flags in the
6238 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
6239 Historically, the name string was modified if it was necessary to
6240 encode more than one bit of information, but this practice is now
6241 discouraged; use @code{SYMBOL_REF_FLAGS}.
6243 The default definition of this hook, @code{default_encode_section_info}
6244 in @file{varasm.c}, sets a number of commonly-useful bits in
6245 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
6246 before overriding it.
6249 @deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name)
6250 Decode @var{name} and return the real name part, sans
6251 the characters that @code{TARGET_ENCODE_SECTION_INFO}
6255 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp})
6256 Returns true if @var{exp} should be placed into a ``small data'' section.
6257 The default version of this hook always returns false.
6260 @deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
6261 Contains the value true if the target places read-only
6262 ``small data'' into a separate section. The default value is false.
6265 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp})
6266 Returns true if @var{exp} names an object for which name resolution
6267 rules must resolve to the current ``module'' (dynamic shared library
6268 or executable image).
6270 The default version of this hook implements the name resolution rules
6271 for ELF, which has a looser model of global name binding than other
6272 currently supported object file formats.
6275 @deftypevar {Target Hook} bool TARGET_HAVE_TLS
6276 Contains the value true if the target supports thread-local storage.
6277 The default value is false.
6282 @section Position Independent Code
6283 @cindex position independent code
6286 This section describes macros that help implement generation of position
6287 independent code. Simply defining these macros is not enough to
6288 generate valid PIC; you must also add support to the macros
6289 @code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as
6290 well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of
6291 @samp{movsi} to do something appropriate when the source operand
6292 contains a symbolic address. You may also need to alter the handling of
6293 switch statements so that they use relative addresses.
6294 @c i rearranged the order of the macros above to try to force one of
6295 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
6297 @defmac PIC_OFFSET_TABLE_REGNUM
6298 The register number of the register used to address a table of static
6299 data addresses in memory. In some cases this register is defined by a
6300 processor's ``application binary interface'' (ABI)@. When this macro
6301 is defined, RTL is generated for this register once, as with the stack
6302 pointer and frame pointer registers. If this macro is not defined, it
6303 is up to the machine-dependent files to allocate such a register (if
6304 necessary). Note that this register must be fixed when in use (e.g.@:
6305 when @code{flag_pic} is true).
6308 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
6309 Define this macro if the register defined by
6310 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
6311 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
6314 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
6315 A C expression that is nonzero if @var{x} is a legitimate immediate
6316 operand on the target machine when generating position independent code.
6317 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6318 check this. You can also assume @var{flag_pic} is true, so you need not
6319 check it either. You need not define this macro if all constants
6320 (including @code{SYMBOL_REF}) can be immediate operands when generating
6321 position independent code.
6324 @node Assembler Format
6325 @section Defining the Output Assembler Language
6327 This section describes macros whose principal purpose is to describe how
6328 to write instructions in assembler language---rather than what the
6332 * File Framework:: Structural information for the assembler file.
6333 * Data Output:: Output of constants (numbers, strings, addresses).
6334 * Uninitialized Data:: Output of uninitialized variables.
6335 * Label Output:: Output and generation of labels.
6336 * Initialization:: General principles of initialization
6337 and termination routines.
6338 * Macros for Initialization::
6339 Specific macros that control the handling of
6340 initialization and termination routines.
6341 * Instruction Output:: Output of actual instructions.
6342 * Dispatch Tables:: Output of jump tables.
6343 * Exception Region Output:: Output of exception region code.
6344 * Alignment Output:: Pseudo ops for alignment and skipping data.
6347 @node File Framework
6348 @subsection The Overall Framework of an Assembler File
6349 @cindex assembler format
6350 @cindex output of assembler code
6352 @c prevent bad page break with this line
6353 This describes the overall framework of an assembly file.
6355 @deftypefn {Target Hook} void TARGET_ASM_FILE_START ()
6356 @findex default_file_start
6357 Output to @code{asm_out_file} any text which the assembler expects to
6358 find at the beginning of a file. The default behavior is controlled
6359 by two flags, documented below. Unless your target's assembler is
6360 quite unusual, if you override the default, you should call
6361 @code{default_file_start} at some point in your target hook. This
6362 lets other target files rely on these variables.
6365 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
6366 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
6367 printed as the very first line in the assembly file, unless
6368 @option{-fverbose-asm} is in effect. (If that macro has been defined
6369 to the empty string, this variable has no effect.) With the normal
6370 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
6371 assembler that it need not bother stripping comments or extra
6372 whitespace from its input. This allows it to work a bit faster.
6374 The default is false. You should not set it to true unless you have
6375 verified that your port does not generate any extra whitespace or
6376 comments that will cause GAS to issue errors in NO_APP mode.
6379 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
6380 If this flag is true, @code{output_file_directive} will be called
6381 for the primary source file, immediately after printing
6382 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
6383 this to be done. The default is false.
6386 @deftypefn {Target Hook} void TARGET_ASM_FILE_END ()
6387 Output to @code{asm_out_file} any text which the assembler expects
6388 to find at the end of a file. The default is to output nothing.
6391 @deftypefun void file_end_indicate_exec_stack ()
6392 Some systems use a common convention, the @samp{.note.GNU-stack}
6393 special section, to indicate whether or not an object file relies on
6394 the stack being executable. If your system uses this convention, you
6395 should define @code{TARGET_ASM_FILE_END} to this function. If you
6396 need to do other things in that hook, have your hook function call
6400 @defmac ASM_COMMENT_START
6401 A C string constant describing how to begin a comment in the target
6402 assembler language. The compiler assumes that the comment will end at
6403 the end of the line.
6407 A C string constant for text to be output before each @code{asm}
6408 statement or group of consecutive ones. Normally this is
6409 @code{"#APP"}, which is a comment that has no effect on most
6410 assemblers but tells the GNU assembler that it must check the lines
6411 that follow for all valid assembler constructs.
6415 A C string constant for text to be output after each @code{asm}
6416 statement or group of consecutive ones. Normally this is
6417 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
6418 time-saving assumptions that are valid for ordinary compiler output.
6421 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
6422 A C statement to output COFF information or DWARF debugging information
6423 which indicates that filename @var{name} is the current source file to
6424 the stdio stream @var{stream}.
6426 This macro need not be defined if the standard form of output
6427 for the file format in use is appropriate.
6430 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
6431 A C statement to output the string @var{string} to the stdio stream
6432 @var{stream}. If you do not call the function @code{output_quoted_string}
6433 in your config files, GCC will only call it to output filenames to
6434 the assembler source. So you can use it to canonicalize the format
6435 of the filename using this macro.
6438 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
6439 A C statement to output something to the assembler file to handle a
6440 @samp{#ident} directive containing the text @var{string}. If this
6441 macro is not defined, nothing is output for a @samp{#ident} directive.
6444 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align})
6445 Output assembly directives to switch to section @var{name}. The section
6446 should have attributes as specified by @var{flags}, which is a bit mask
6447 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align}
6448 is nonzero, it contains an alignment in bytes to be used for the section,
6449 otherwise some target default should be used. Only targets that must
6450 specify an alignment within the section directive need pay attention to
6451 @var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}.
6454 @deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
6455 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
6458 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
6459 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
6460 based on a variable or function decl, a section name, and whether or not the
6461 declaration's initializer may contain runtime relocations. @var{decl} may be
6462 null, in which case read-write data should be assumed.
6464 The default version if this function handles choosing code vs data,
6465 read-only vs read-write data, and @code{flag_pic}. You should only
6466 need to override this if your target has special flags that might be
6467 set via @code{__attribute__}.
6472 @subsection Output of Data
6475 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
6476 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
6477 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
6478 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
6479 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
6480 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
6481 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
6482 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
6483 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
6484 These hooks specify assembly directives for creating certain kinds
6485 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
6486 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
6487 aligned two-byte object, and so on. Any of the hooks may be
6488 @code{NULL}, indicating that no suitable directive is available.
6490 The compiler will print these strings at the start of a new line,
6491 followed immediately by the object's initial value. In most cases,
6492 the string should contain a tab, a pseudo-op, and then another tab.
6495 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
6496 The @code{assemble_integer} function uses this hook to output an
6497 integer object. @var{x} is the object's value, @var{size} is its size
6498 in bytes and @var{aligned_p} indicates whether it is aligned. The
6499 function should return @code{true} if it was able to output the
6500 object. If it returns false, @code{assemble_integer} will try to
6501 split the object into smaller parts.
6503 The default implementation of this hook will use the
6504 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
6505 when the relevant string is @code{NULL}.
6508 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
6509 A C statement to recognize @var{rtx} patterns that
6510 @code{output_addr_const} can't deal with, and output assembly code to
6511 @var{stream} corresponding to the pattern @var{x}. This may be used to
6512 allow machine-dependent @code{UNSPEC}s to appear within constants.
6514 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
6515 @code{goto fail}, so that a standard error message is printed. If it
6516 prints an error message itself, by calling, for example,
6517 @code{output_operand_lossage}, it may just complete normally.
6520 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
6521 A C statement to output to the stdio stream @var{stream} an assembler
6522 instruction to assemble a string constant containing the @var{len}
6523 bytes at @var{ptr}. @var{ptr} will be a C expression of type
6524 @code{char *} and @var{len} a C expression of type @code{int}.
6526 If the assembler has a @code{.ascii} pseudo-op as found in the
6527 Berkeley Unix assembler, do not define the macro
6528 @code{ASM_OUTPUT_ASCII}.
6531 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
6532 A C statement to output word @var{n} of a function descriptor for
6533 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
6534 is defined, and is otherwise unused.
6537 @defmac CONSTANT_POOL_BEFORE_FUNCTION
6538 You may define this macro as a C expression. You should define the
6539 expression to have a nonzero value if GCC should output the constant
6540 pool for a function before the code for the function, or a zero value if
6541 GCC should output the constant pool after the function. If you do
6542 not define this macro, the usual case, GCC will output the constant
6543 pool before the function.
6546 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
6547 A C statement to output assembler commands to define the start of the
6548 constant pool for a function. @var{funname} is a string giving
6549 the name of the function. Should the return type of the function
6550 be required, it can be obtained via @var{fundecl}. @var{size}
6551 is the size, in bytes, of the constant pool that will be written
6552 immediately after this call.
6554 If no constant-pool prefix is required, the usual case, this macro need
6558 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
6559 A C statement (with or without semicolon) to output a constant in the
6560 constant pool, if it needs special treatment. (This macro need not do
6561 anything for RTL expressions that can be output normally.)
6563 The argument @var{file} is the standard I/O stream to output the
6564 assembler code on. @var{x} is the RTL expression for the constant to
6565 output, and @var{mode} is the machine mode (in case @var{x} is a
6566 @samp{const_int}). @var{align} is the required alignment for the value
6567 @var{x}; you should output an assembler directive to force this much
6570 The argument @var{labelno} is a number to use in an internal label for
6571 the address of this pool entry. The definition of this macro is
6572 responsible for outputting the label definition at the proper place.
6573 Here is how to do this:
6576 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
6579 When you output a pool entry specially, you should end with a
6580 @code{goto} to the label @var{jumpto}. This will prevent the same pool
6581 entry from being output a second time in the usual manner.
6583 You need not define this macro if it would do nothing.
6586 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
6587 A C statement to output assembler commands to at the end of the constant
6588 pool for a function. @var{funname} is a string giving the name of the
6589 function. Should the return type of the function be required, you can
6590 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
6591 constant pool that GCC wrote immediately before this call.
6593 If no constant-pool epilogue is required, the usual case, you need not
6597 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C})
6598 Define this macro as a C expression which is nonzero if @var{C} is
6599 used as a logical line separator by the assembler.
6601 If you do not define this macro, the default is that only
6602 the character @samp{;} is treated as a logical line separator.
6605 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
6606 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
6607 These target hooks are C string constants, describing the syntax in the
6608 assembler for grouping arithmetic expressions. If not overridden, they
6609 default to normal parentheses, which is correct for most assemblers.
6612 These macros are provided by @file{real.h} for writing the definitions
6613 of @code{ASM_OUTPUT_DOUBLE} and the like:
6615 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
6616 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
6617 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
6618 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
6619 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
6620 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
6621 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
6622 target's floating point representation, and store its bit pattern in
6623 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
6624 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
6625 simple @code{long int}. For the others, it should be an array of
6626 @code{long int}. The number of elements in this array is determined
6627 by the size of the desired target floating point data type: 32 bits of
6628 it go in each @code{long int} array element. Each array element holds
6629 32 bits of the result, even if @code{long int} is wider than 32 bits
6630 on the host machine.
6632 The array element values are designed so that you can print them out
6633 using @code{fprintf} in the order they should appear in the target
6637 @node Uninitialized Data
6638 @subsection Output of Uninitialized Variables
6640 Each of the macros in this section is used to do the whole job of
6641 outputting a single uninitialized variable.
6643 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6644 A C statement (sans semicolon) to output to the stdio stream
6645 @var{stream} the assembler definition of a common-label named
6646 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6647 is the size rounded up to whatever alignment the caller wants.
6649 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6650 output the name itself; before and after that, output the additional
6651 assembler syntax for defining the name, and a newline.
6653 This macro controls how the assembler definitions of uninitialized
6654 common global variables are output.
6657 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
6658 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
6659 separate, explicit argument. If you define this macro, it is used in
6660 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
6661 handling the required alignment of the variable. The alignment is specified
6662 as the number of bits.
6665 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6666 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
6667 variable to be output, if there is one, or @code{NULL_TREE} if there
6668 is no corresponding variable. If you define this macro, GCC will use it
6669 in place of both @code{ASM_OUTPUT_COMMON} and
6670 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
6671 the variable's decl in order to chose what to output.
6674 @defmac ASM_OUTPUT_SHARED_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
6675 If defined, it is similar to @code{ASM_OUTPUT_COMMON}, except that it
6676 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_COMMON}
6680 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6681 A C statement (sans semicolon) to output to the stdio stream
6682 @var{stream} the assembler definition of uninitialized global @var{decl} named
6683 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6684 is the size rounded up to whatever alignment the caller wants.
6686 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
6687 defining this macro. If unable, use the expression
6688 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
6689 before and after that, output the additional assembler syntax for defining
6690 the name, and a newline.
6692 This macro controls how the assembler definitions of uninitialized global
6693 variables are output. This macro exists to properly support languages like
6694 C++ which do not have @code{common} data. However, this macro currently
6695 is not defined for all targets. If this macro and
6696 @code{ASM_OUTPUT_ALIGNED_BSS} are not defined then @code{ASM_OUTPUT_COMMON}
6697 or @code{ASM_OUTPUT_ALIGNED_COMMON} or
6698 @code{ASM_OUTPUT_ALIGNED_DECL_COMMON} is used.
6701 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6702 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
6703 separate, explicit argument. If you define this macro, it is used in
6704 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
6705 handling the required alignment of the variable. The alignment is specified
6706 as the number of bits.
6708 Try to use function @code{asm_output_aligned_bss} defined in file
6709 @file{varasm.c} when defining this macro.
6712 @defmac ASM_OUTPUT_SHARED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
6713 If defined, it is similar to @code{ASM_OUTPUT_BSS}, except that it
6714 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_BSS}
6718 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6719 A C statement (sans semicolon) to output to the stdio stream
6720 @var{stream} the assembler definition of a local-common-label named
6721 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
6722 is the size rounded up to whatever alignment the caller wants.
6724 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6725 output the name itself; before and after that, output the additional
6726 assembler syntax for defining the name, and a newline.
6728 This macro controls how the assembler definitions of uninitialized
6729 static variables are output.
6732 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
6733 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
6734 separate, explicit argument. If you define this macro, it is used in
6735 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
6736 handling the required alignment of the variable. The alignment is specified
6737 as the number of bits.
6740 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
6741 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
6742 variable to be output, if there is one, or @code{NULL_TREE} if there
6743 is no corresponding variable. If you define this macro, GCC will use it
6744 in place of both @code{ASM_OUTPUT_DECL} and
6745 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
6746 the variable's decl in order to chose what to output.
6749 @defmac ASM_OUTPUT_SHARED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
6750 If defined, it is similar to @code{ASM_OUTPUT_LOCAL}, except that it
6751 is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_LOCAL}
6756 @subsection Output and Generation of Labels
6758 @c prevent bad page break with this line
6759 This is about outputting labels.
6761 @findex assemble_name
6762 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
6763 A C statement (sans semicolon) to output to the stdio stream
6764 @var{stream} the assembler definition of a label named @var{name}.
6765 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
6766 output the name itself; before and after that, output the additional
6767 assembler syntax for defining the name, and a newline. A default
6768 definition of this macro is provided which is correct for most systems.
6771 @findex assemble_name_raw
6772 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
6773 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
6774 to refer to a compiler-generated label. The default definition uses
6775 @code{assemble_name_raw}, which is like @code{assemble_name} except
6776 that it is more efficient.
6780 A C string containing the appropriate assembler directive to specify the
6781 size of a symbol, without any arguments. On systems that use ELF, the
6782 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
6783 systems, the default is not to define this macro.
6785 Define this macro only if it is correct to use the default definitions
6786 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
6787 for your system. If you need your own custom definitions of those
6788 macros, or if you do not need explicit symbol sizes at all, do not
6792 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
6793 A C statement (sans semicolon) to output to the stdio stream
6794 @var{stream} a directive telling the assembler that the size of the
6795 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
6796 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6800 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
6801 A C statement (sans semicolon) to output to the stdio stream
6802 @var{stream} a directive telling the assembler to calculate the size of
6803 the symbol @var{name} by subtracting its address from the current
6806 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
6807 provided. The default assumes that the assembler recognizes a special
6808 @samp{.} symbol as referring to the current address, and can calculate
6809 the difference between this and another symbol. If your assembler does
6810 not recognize @samp{.} or cannot do calculations with it, you will need
6811 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
6815 A C string containing the appropriate assembler directive to specify the
6816 type of a symbol, without any arguments. On systems that use ELF, the
6817 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
6818 systems, the default is not to define this macro.
6820 Define this macro only if it is correct to use the default definition of
6821 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6822 custom definition of this macro, or if you do not need explicit symbol
6823 types at all, do not define this macro.
6826 @defmac TYPE_OPERAND_FMT
6827 A C string which specifies (using @code{printf} syntax) the format of
6828 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
6829 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
6830 the default is not to define this macro.
6832 Define this macro only if it is correct to use the default definition of
6833 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
6834 custom definition of this macro, or if you do not need explicit symbol
6835 types at all, do not define this macro.
6838 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
6839 A C statement (sans semicolon) to output to the stdio stream
6840 @var{stream} a directive telling the assembler that the type of the
6841 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
6842 that string is always either @samp{"function"} or @samp{"object"}, but
6843 you should not count on this.
6845 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
6846 definition of this macro is provided.
6849 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
6850 A C statement (sans semicolon) to output to the stdio stream
6851 @var{stream} any text necessary for declaring the name @var{name} of a
6852 function which is being defined. This macro is responsible for
6853 outputting the label definition (perhaps using
6854 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
6855 @code{FUNCTION_DECL} tree node representing the function.
6857 If this macro is not defined, then the function name is defined in the
6858 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6860 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6864 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
6865 A C statement (sans semicolon) to output to the stdio stream
6866 @var{stream} any text necessary for declaring the size of a function
6867 which is being defined. The argument @var{name} is the name of the
6868 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
6869 representing the function.
6871 If this macro is not defined, then the function size is not defined.
6873 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
6877 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
6878 A C statement (sans semicolon) to output to the stdio stream
6879 @var{stream} any text necessary for declaring the name @var{name} of an
6880 initialized variable which is being defined. This macro must output the
6881 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
6882 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
6884 If this macro is not defined, then the variable name is defined in the
6885 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6887 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
6888 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
6891 @defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size})
6892 A C statement (sans semicolon) to output to the stdio stream
6893 @var{stream} any text necessary for declaring the name @var{name} of a
6894 constant which is being defined. This macro is responsible for
6895 outputting the label definition (perhaps using
6896 @code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the
6897 value of the constant, and @var{size} is the size of the constant
6898 in bytes. @var{name} will be an internal label.
6900 If this macro is not defined, then the @var{name} is defined in the
6901 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
6903 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
6907 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
6908 A C statement (sans semicolon) to output to the stdio stream
6909 @var{stream} any text necessary for claiming a register @var{regno}
6910 for a global variable @var{decl} with name @var{name}.
6912 If you don't define this macro, that is equivalent to defining it to do
6916 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
6917 A C statement (sans semicolon) to finish up declaring a variable name
6918 once the compiler has processed its initializer fully and thus has had a
6919 chance to determine the size of an array when controlled by an
6920 initializer. This is used on systems where it's necessary to declare
6921 something about the size of the object.
6923 If you don't define this macro, that is equivalent to defining it to do
6926 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
6927 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
6930 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
6931 This target hook is a function to output to the stdio stream
6932 @var{stream} some commands that will make the label @var{name} global;
6933 that is, available for reference from other files.
6935 The default implementation relies on a proper definition of
6936 @code{GLOBAL_ASM_OP}.
6939 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
6940 A C statement (sans semicolon) to output to the stdio stream
6941 @var{stream} some commands that will make the label @var{name} weak;
6942 that is, available for reference from other files but only used if
6943 no other definition is available. Use the expression
6944 @code{assemble_name (@var{stream}, @var{name})} to output the name
6945 itself; before and after that, output the additional assembler syntax
6946 for making that name weak, and a newline.
6948 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
6949 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
6953 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
6954 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
6955 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
6956 or variable decl. If @var{value} is not @code{NULL}, this C statement
6957 should output to the stdio stream @var{stream} assembler code which
6958 defines (equates) the weak symbol @var{name} to have the value
6959 @var{value}. If @var{value} is @code{NULL}, it should output commands
6960 to make @var{name} weak.
6963 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
6964 Outputs a directive that enables @var{name} to be used to refer to
6965 symbol @var{value} with weak-symbol semantics. @code{decl} is the
6966 declaration of @code{name}.
6969 @defmac SUPPORTS_WEAK
6970 A C expression which evaluates to true if the target supports weak symbols.
6972 If you don't define this macro, @file{defaults.h} provides a default
6973 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
6974 is defined, the default definition is @samp{1}; otherwise, it is
6975 @samp{0}. Define this macro if you want to control weak symbol support
6976 with a compiler flag such as @option{-melf}.
6979 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
6980 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
6981 public symbol such that extra copies in multiple translation units will
6982 be discarded by the linker. Define this macro if your object file
6983 format provides support for this concept, such as the @samp{COMDAT}
6984 section flags in the Microsoft Windows PE/COFF format, and this support
6985 requires changes to @var{decl}, such as putting it in a separate section.
6988 @defmac SUPPORTS_ONE_ONLY
6989 A C expression which evaluates to true if the target supports one-only
6992 If you don't define this macro, @file{varasm.c} provides a default
6993 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
6994 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
6995 you want to control one-only symbol support with a compiler flag, or if
6996 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
6997 be emitted as one-only.
7000 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility})
7001 This target hook is a function to output to @var{asm_out_file} some
7002 commands that will make the symbol(s) associated with @var{decl} have
7003 hidden, protected or internal visibility as specified by @var{visibility}.
7006 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7007 A C expression that evaluates to true if the target's linker expects
7008 that weak symbols do not appear in a static archive's table of contents.
7009 The default is @code{0}.
7011 Leaving weak symbols out of an archive's table of contents means that,
7012 if a symbol will only have a definition in one translation unit and
7013 will have undefined references from other translation units, that
7014 symbol should not be weak. Defining this macro to be nonzero will
7015 thus have the effect that certain symbols that would normally be weak
7016 (explicit template instantiations, and vtables for polymorphic classes
7017 with noninline key methods) will instead be nonweak.
7019 The C++ ABI requires this macro to be zero. Define this macro for
7020 targets where full C++ ABI compliance is impossible and where linker
7021 restrictions require weak symbols to be left out of a static archive's
7025 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7026 A C statement (sans semicolon) to output to the stdio stream
7027 @var{stream} any text necessary for declaring the name of an external
7028 symbol named @var{name} which is referenced in this compilation but
7029 not defined. The value of @var{decl} is the tree node for the
7032 This macro need not be defined if it does not need to output anything.
7033 The GNU assembler and most Unix assemblers don't require anything.
7036 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
7037 This target hook is a function to output to @var{asm_out_file} an assembler
7038 pseudo-op to declare a library function name external. The name of the
7039 library function is given by @var{symref}, which is a @code{symbol_ref}.
7042 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (tree @var{decl})
7043 This target hook is a function to output to @var{asm_out_file} an assembler
7044 directive to annotate used symbol. Darwin target use .no_dead_code_strip
7048 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7049 A C statement (sans semicolon) to output to the stdio stream
7050 @var{stream} a reference in assembler syntax to a label named
7051 @var{name}. This should add @samp{_} to the front of the name, if that
7052 is customary on your operating system, as it is in most Berkeley Unix
7053 systems. This macro is used in @code{assemble_name}.
7056 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
7057 A C statement (sans semicolon) to output a reference to
7058 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
7059 will be used to output the name of the symbol. This macro may be used
7060 to modify the way a symbol is referenced depending on information
7061 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
7064 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
7065 A C statement (sans semicolon) to output a reference to @var{buf}, the
7066 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
7067 @code{assemble_name} will be used to output the name of the symbol.
7068 This macro is not used by @code{output_asm_label}, or the @code{%l}
7069 specifier that calls it; the intention is that this macro should be set
7070 when it is necessary to output a label differently when its address is
7074 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
7075 A function to output to the stdio stream @var{stream} a label whose
7076 name is made from the string @var{prefix} and the number @var{labelno}.
7078 It is absolutely essential that these labels be distinct from the labels
7079 used for user-level functions and variables. Otherwise, certain programs
7080 will have name conflicts with internal labels.
7082 It is desirable to exclude internal labels from the symbol table of the
7083 object file. Most assemblers have a naming convention for labels that
7084 should be excluded; on many systems, the letter @samp{L} at the
7085 beginning of a label has this effect. You should find out what
7086 convention your system uses, and follow it.
7088 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
7091 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
7092 A C statement to output to the stdio stream @var{stream} a debug info
7093 label whose name is made from the string @var{prefix} and the number
7094 @var{num}. This is useful for VLIW targets, where debug info labels
7095 may need to be treated differently than branch target labels. On some
7096 systems, branch target labels must be at the beginning of instruction
7097 bundles, but debug info labels can occur in the middle of instruction
7100 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
7104 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
7105 A C statement to store into the string @var{string} a label whose name
7106 is made from the string @var{prefix} and the number @var{num}.
7108 This string, when output subsequently by @code{assemble_name}, should
7109 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
7110 with the same @var{prefix} and @var{num}.
7112 If the string begins with @samp{*}, then @code{assemble_name} will
7113 output the rest of the string unchanged. It is often convenient for
7114 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
7115 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
7116 to output the string, and may change it. (Of course,
7117 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
7118 you should know what it does on your machine.)
7121 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7122 A C expression to assign to @var{outvar} (which is a variable of type
7123 @code{char *}) a newly allocated string made from the string
7124 @var{name} and the number @var{number}, with some suitable punctuation
7125 added. Use @code{alloca} to get space for the string.
7127 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
7128 produce an assembler label for an internal static variable whose name is
7129 @var{name}. Therefore, the string must be such as to result in valid
7130 assembler code. The argument @var{number} is different each time this
7131 macro is executed; it prevents conflicts between similarly-named
7132 internal static variables in different scopes.
7134 Ideally this string should not be a valid C identifier, to prevent any
7135 conflict with the user's own symbols. Most assemblers allow periods
7136 or percent signs in assembler symbols; putting at least one of these
7137 between the name and the number will suffice.
7139 If this macro is not defined, a default definition will be provided
7140 which is correct for most systems.
7143 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
7144 A C statement to output to the stdio stream @var{stream} assembler code
7145 which defines (equates) the symbol @var{name} to have the value @var{value}.
7148 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7149 correct for most systems.
7152 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
7153 A C statement to output to the stdio stream @var{stream} assembler code
7154 which defines (equates) the symbol whose tree node is @var{decl_of_name}
7155 to have the value of the tree node @var{decl_of_value}. This macro will
7156 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
7157 the tree nodes are available.
7160 If @code{SET_ASM_OP} is defined, a default definition is provided which is
7161 correct for most systems.
7164 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
7165 A C statement that evaluates to true if the assembler code which defines
7166 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
7167 of the tree node @var{decl_of_value} should be emitted near the end of the
7168 current compilation unit. The default is to not defer output of defines.
7169 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
7170 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
7173 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
7174 A C statement to output to the stdio stream @var{stream} assembler code
7175 which defines (equates) the weak symbol @var{name} to have the value
7176 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
7177 an undefined weak symbol.
7179 Define this macro if the target only supports weak aliases; define
7180 @code{ASM_OUTPUT_DEF} instead if possible.
7183 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
7184 Define this macro to override the default assembler names used for
7185 Objective-C methods.
7187 The default name is a unique method number followed by the name of the
7188 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
7189 the category is also included in the assembler name (e.g.@:
7192 These names are safe on most systems, but make debugging difficult since
7193 the method's selector is not present in the name. Therefore, particular
7194 systems define other ways of computing names.
7196 @var{buf} is an expression of type @code{char *} which gives you a
7197 buffer in which to store the name; its length is as long as
7198 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
7199 50 characters extra.
7201 The argument @var{is_inst} specifies whether the method is an instance
7202 method or a class method; @var{class_name} is the name of the class;
7203 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
7204 in a category); and @var{sel_name} is the name of the selector.
7206 On systems where the assembler can handle quoted names, you can use this
7207 macro to provide more human-readable names.
7210 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
7211 A C statement (sans semicolon) to output to the stdio stream
7212 @var{stream} commands to declare that the label @var{name} is an
7213 Objective-C class reference. This is only needed for targets whose
7214 linkers have special support for NeXT-style runtimes.
7217 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
7218 A C statement (sans semicolon) to output to the stdio stream
7219 @var{stream} commands to declare that the label @var{name} is an
7220 unresolved Objective-C class reference. This is only needed for targets
7221 whose linkers have special support for NeXT-style runtimes.
7224 @node Initialization
7225 @subsection How Initialization Functions Are Handled
7226 @cindex initialization routines
7227 @cindex termination routines
7228 @cindex constructors, output of
7229 @cindex destructors, output of
7231 The compiled code for certain languages includes @dfn{constructors}
7232 (also called @dfn{initialization routines})---functions to initialize
7233 data in the program when the program is started. These functions need
7234 to be called before the program is ``started''---that is to say, before
7235 @code{main} is called.
7237 Compiling some languages generates @dfn{destructors} (also called
7238 @dfn{termination routines}) that should be called when the program
7241 To make the initialization and termination functions work, the compiler
7242 must output something in the assembler code to cause those functions to
7243 be called at the appropriate time. When you port the compiler to a new
7244 system, you need to specify how to do this.
7246 There are two major ways that GCC currently supports the execution of
7247 initialization and termination functions. Each way has two variants.
7248 Much of the structure is common to all four variations.
7250 @findex __CTOR_LIST__
7251 @findex __DTOR_LIST__
7252 The linker must build two lists of these functions---a list of
7253 initialization functions, called @code{__CTOR_LIST__}, and a list of
7254 termination functions, called @code{__DTOR_LIST__}.
7256 Each list always begins with an ignored function pointer (which may hold
7257 0, @minus{}1, or a count of the function pointers after it, depending on
7258 the environment). This is followed by a series of zero or more function
7259 pointers to constructors (or destructors), followed by a function
7260 pointer containing zero.
7262 Depending on the operating system and its executable file format, either
7263 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
7264 time and exit time. Constructors are called in reverse order of the
7265 list; destructors in forward order.
7267 The best way to handle static constructors works only for object file
7268 formats which provide arbitrarily-named sections. A section is set
7269 aside for a list of constructors, and another for a list of destructors.
7270 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
7271 object file that defines an initialization function also puts a word in
7272 the constructor section to point to that function. The linker
7273 accumulates all these words into one contiguous @samp{.ctors} section.
7274 Termination functions are handled similarly.
7276 This method will be chosen as the default by @file{target-def.h} if
7277 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
7278 support arbitrary sections, but does support special designated
7279 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
7280 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
7282 When arbitrary sections are available, there are two variants, depending
7283 upon how the code in @file{crtstuff.c} is called. On systems that
7284 support a @dfn{.init} section which is executed at program startup,
7285 parts of @file{crtstuff.c} are compiled into that section. The
7286 program is linked by the @command{gcc} driver like this:
7289 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
7292 The prologue of a function (@code{__init}) appears in the @code{.init}
7293 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
7294 for the function @code{__fini} in the @dfn{.fini} section. Normally these
7295 files are provided by the operating system or by the GNU C library, but
7296 are provided by GCC for a few targets.
7298 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
7299 compiled from @file{crtstuff.c}. They contain, among other things, code
7300 fragments within the @code{.init} and @code{.fini} sections that branch
7301 to routines in the @code{.text} section. The linker will pull all parts
7302 of a section together, which results in a complete @code{__init} function
7303 that invokes the routines we need at startup.
7305 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
7308 If no init section is available, when GCC compiles any function called
7309 @code{main} (or more accurately, any function designated as a program
7310 entry point by the language front end calling @code{expand_main_function}),
7311 it inserts a procedure call to @code{__main} as the first executable code
7312 after the function prologue. The @code{__main} function is defined
7313 in @file{libgcc2.c} and runs the global constructors.
7315 In file formats that don't support arbitrary sections, there are again
7316 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
7317 and an `a.out' format must be used. In this case,
7318 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
7319 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
7320 and with the address of the void function containing the initialization
7321 code as its value. The GNU linker recognizes this as a request to add
7322 the value to a @dfn{set}; the values are accumulated, and are eventually
7323 placed in the executable as a vector in the format described above, with
7324 a leading (ignored) count and a trailing zero element.
7325 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
7326 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
7327 the compilation of @code{main} to call @code{__main} as above, starting
7328 the initialization process.
7330 The last variant uses neither arbitrary sections nor the GNU linker.
7331 This is preferable when you want to do dynamic linking and when using
7332 file formats which the GNU linker does not support, such as `ECOFF'@. In
7333 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
7334 termination functions are recognized simply by their names. This requires
7335 an extra program in the linkage step, called @command{collect2}. This program
7336 pretends to be the linker, for use with GCC; it does its job by running
7337 the ordinary linker, but also arranges to include the vectors of
7338 initialization and termination functions. These functions are called
7339 via @code{__main} as described above. In order to use this method,
7340 @code{use_collect2} must be defined in the target in @file{config.gcc}.
7343 The following section describes the specific macros that control and
7344 customize the handling of initialization and termination functions.
7347 @node Macros for Initialization
7348 @subsection Macros Controlling Initialization Routines
7350 Here are the macros that control how the compiler handles initialization
7351 and termination functions:
7353 @defmac INIT_SECTION_ASM_OP
7354 If defined, a C string constant, including spacing, for the assembler
7355 operation to identify the following data as initialization code. If not
7356 defined, GCC will assume such a section does not exist. When you are
7357 using special sections for initialization and termination functions, this
7358 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
7359 run the initialization functions.
7362 @defmac HAS_INIT_SECTION
7363 If defined, @code{main} will not call @code{__main} as described above.
7364 This macro should be defined for systems that control start-up code
7365 on a symbol-by-symbol basis, such as OSF/1, and should not
7366 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
7369 @defmac LD_INIT_SWITCH
7370 If defined, a C string constant for a switch that tells the linker that
7371 the following symbol is an initialization routine.
7374 @defmac LD_FINI_SWITCH
7375 If defined, a C string constant for a switch that tells the linker that
7376 the following symbol is a finalization routine.
7379 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
7380 If defined, a C statement that will write a function that can be
7381 automatically called when a shared library is loaded. The function
7382 should call @var{func}, which takes no arguments. If not defined, and
7383 the object format requires an explicit initialization function, then a
7384 function called @code{_GLOBAL__DI} will be generated.
7386 This function and the following one are used by collect2 when linking a
7387 shared library that needs constructors or destructors, or has DWARF2
7388 exception tables embedded in the code.
7391 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
7392 If defined, a C statement that will write a function that can be
7393 automatically called when a shared library is unloaded. The function
7394 should call @var{func}, which takes no arguments. If not defined, and
7395 the object format requires an explicit finalization function, then a
7396 function called @code{_GLOBAL__DD} will be generated.
7399 @defmac INVOKE__main
7400 If defined, @code{main} will call @code{__main} despite the presence of
7401 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
7402 where the init section is not actually run automatically, but is still
7403 useful for collecting the lists of constructors and destructors.
7406 @defmac SUPPORTS_INIT_PRIORITY
7407 If nonzero, the C++ @code{init_priority} attribute is supported and the
7408 compiler should emit instructions to control the order of initialization
7409 of objects. If zero, the compiler will issue an error message upon
7410 encountering an @code{init_priority} attribute.
7413 @deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS
7414 This value is true if the target supports some ``native'' method of
7415 collecting constructors and destructors to be run at startup and exit.
7416 It is false if we must use @command{collect2}.
7419 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
7420 If defined, a function that outputs assembler code to arrange to call
7421 the function referenced by @var{symbol} at initialization time.
7423 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
7424 no arguments and with no return value. If the target supports initialization
7425 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
7426 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
7428 If this macro is not defined by the target, a suitable default will
7429 be chosen if (1) the target supports arbitrary section names, (2) the
7430 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
7434 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
7435 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
7436 functions rather than initialization functions.
7439 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
7440 generated for the generated object file will have static linkage.
7442 If your system uses @command{collect2} as the means of processing
7443 constructors, then that program normally uses @command{nm} to scan
7444 an object file for constructor functions to be called.
7446 On certain kinds of systems, you can define this macro to make
7447 @command{collect2} work faster (and, in some cases, make it work at all):
7449 @defmac OBJECT_FORMAT_COFF
7450 Define this macro if the system uses COFF (Common Object File Format)
7451 object files, so that @command{collect2} can assume this format and scan
7452 object files directly for dynamic constructor/destructor functions.
7454 This macro is effective only in a native compiler; @command{collect2} as
7455 part of a cross compiler always uses @command{nm} for the target machine.
7458 @defmac REAL_NM_FILE_NAME
7459 Define this macro as a C string constant containing the file name to use
7460 to execute @command{nm}. The default is to search the path normally for
7463 If your system supports shared libraries and has a program to list the
7464 dynamic dependencies of a given library or executable, you can define
7465 these macros to enable support for running initialization and
7466 termination functions in shared libraries:
7470 Define this macro to a C string constant containing the name of the program
7471 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
7474 @defmac PARSE_LDD_OUTPUT (@var{ptr})
7475 Define this macro to be C code that extracts filenames from the output
7476 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
7477 of type @code{char *} that points to the beginning of a line of output
7478 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
7479 code must advance @var{ptr} to the beginning of the filename on that
7480 line. Otherwise, it must set @var{ptr} to @code{NULL}.
7483 @node Instruction Output
7484 @subsection Output of Assembler Instructions
7486 @c prevent bad page break with this line
7487 This describes assembler instruction output.
7489 @defmac REGISTER_NAMES
7490 A C initializer containing the assembler's names for the machine
7491 registers, each one as a C string constant. This is what translates
7492 register numbers in the compiler into assembler language.
7495 @defmac ADDITIONAL_REGISTER_NAMES
7496 If defined, a C initializer for an array of structures containing a name
7497 and a register number. This macro defines additional names for hard
7498 registers, thus allowing the @code{asm} option in declarations to refer
7499 to registers using alternate names.
7502 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
7503 Define this macro if you are using an unusual assembler that
7504 requires different names for the machine instructions.
7506 The definition is a C statement or statements which output an
7507 assembler instruction opcode to the stdio stream @var{stream}. The
7508 macro-operand @var{ptr} is a variable of type @code{char *} which
7509 points to the opcode name in its ``internal'' form---the form that is
7510 written in the machine description. The definition should output the
7511 opcode name to @var{stream}, performing any translation you desire, and
7512 increment the variable @var{ptr} to point at the end of the opcode
7513 so that it will not be output twice.
7515 In fact, your macro definition may process less than the entire opcode
7516 name, or more than the opcode name; but if you want to process text
7517 that includes @samp{%}-sequences to substitute operands, you must take
7518 care of the substitution yourself. Just be sure to increment
7519 @var{ptr} over whatever text should not be output normally.
7521 @findex recog_data.operand
7522 If you need to look at the operand values, they can be found as the
7523 elements of @code{recog_data.operand}.
7525 If the macro definition does nothing, the instruction is output
7529 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
7530 If defined, a C statement to be executed just prior to the output of
7531 assembler code for @var{insn}, to modify the extracted operands so
7532 they will be output differently.
7534 Here the argument @var{opvec} is the vector containing the operands
7535 extracted from @var{insn}, and @var{noperands} is the number of
7536 elements of the vector which contain meaningful data for this insn.
7537 The contents of this vector are what will be used to convert the insn
7538 template into assembler code, so you can change the assembler output
7539 by changing the contents of the vector.
7541 This macro is useful when various assembler syntaxes share a single
7542 file of instruction patterns; by defining this macro differently, you
7543 can cause a large class of instructions to be output differently (such
7544 as with rearranged operands). Naturally, variations in assembler
7545 syntax affecting individual insn patterns ought to be handled by
7546 writing conditional output routines in those patterns.
7548 If this macro is not defined, it is equivalent to a null statement.
7551 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
7552 A C compound statement to output to stdio stream @var{stream} the
7553 assembler syntax for an instruction operand @var{x}. @var{x} is an
7556 @var{code} is a value that can be used to specify one of several ways
7557 of printing the operand. It is used when identical operands must be
7558 printed differently depending on the context. @var{code} comes from
7559 the @samp{%} specification that was used to request printing of the
7560 operand. If the specification was just @samp{%@var{digit}} then
7561 @var{code} is 0; if the specification was @samp{%@var{ltr}
7562 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
7565 If @var{x} is a register, this macro should print the register's name.
7566 The names can be found in an array @code{reg_names} whose type is
7567 @code{char *[]}. @code{reg_names} is initialized from
7568 @code{REGISTER_NAMES}.
7570 When the machine description has a specification @samp{%@var{punct}}
7571 (a @samp{%} followed by a punctuation character), this macro is called
7572 with a null pointer for @var{x} and the punctuation character for
7576 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
7577 A C expression which evaluates to true if @var{code} is a valid
7578 punctuation character for use in the @code{PRINT_OPERAND} macro. If
7579 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
7580 punctuation characters (except for the standard one, @samp{%}) are used
7584 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
7585 A C compound statement to output to stdio stream @var{stream} the
7586 assembler syntax for an instruction operand that is a memory reference
7587 whose address is @var{x}. @var{x} is an RTL expression.
7589 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
7590 On some machines, the syntax for a symbolic address depends on the
7591 section that the address refers to. On these machines, define the hook
7592 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
7593 @code{symbol_ref}, and then check for it here. @xref{Assembler
7597 @findex dbr_sequence_length
7598 @defmac DBR_OUTPUT_SEQEND (@var{file})
7599 A C statement, to be executed after all slot-filler instructions have
7600 been output. If necessary, call @code{dbr_sequence_length} to
7601 determine the number of slots filled in a sequence (zero if not
7602 currently outputting a sequence), to decide how many no-ops to output,
7605 Don't define this macro if it has nothing to do, but it is helpful in
7606 reading assembly output if the extent of the delay sequence is made
7607 explicit (e.g.@: with white space).
7610 @findex final_sequence
7611 Note that output routines for instructions with delay slots must be
7612 prepared to deal with not being output as part of a sequence
7613 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
7614 found.) The variable @code{final_sequence} is null when not
7615 processing a sequence, otherwise it contains the @code{sequence} rtx
7619 @defmac REGISTER_PREFIX
7620 @defmacx LOCAL_LABEL_PREFIX
7621 @defmacx USER_LABEL_PREFIX
7622 @defmacx IMMEDIATE_PREFIX
7623 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
7624 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
7625 @file{final.c}). These are useful when a single @file{md} file must
7626 support multiple assembler formats. In that case, the various @file{tm.h}
7627 files can define these macros differently.
7630 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
7631 If defined this macro should expand to a series of @code{case}
7632 statements which will be parsed inside the @code{switch} statement of
7633 the @code{asm_fprintf} function. This allows targets to define extra
7634 printf formats which may useful when generating their assembler
7635 statements. Note that uppercase letters are reserved for future
7636 generic extensions to asm_fprintf, and so are not available to target
7637 specific code. The output file is given by the parameter @var{file}.
7638 The varargs input pointer is @var{argptr} and the rest of the format
7639 string, starting the character after the one that is being switched
7640 upon, is pointed to by @var{format}.
7643 @defmac ASSEMBLER_DIALECT
7644 If your target supports multiple dialects of assembler language (such as
7645 different opcodes), define this macro as a C expression that gives the
7646 numeric index of the assembler language dialect to use, with zero as the
7649 If this macro is defined, you may use constructs of the form
7651 @samp{@{option0|option1|option2@dots{}@}}
7654 in the output templates of patterns (@pxref{Output Template}) or in the
7655 first argument of @code{asm_fprintf}. This construct outputs
7656 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
7657 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
7658 within these strings retain their usual meaning. If there are fewer
7659 alternatives within the braces than the value of
7660 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
7662 If you do not define this macro, the characters @samp{@{}, @samp{|} and
7663 @samp{@}} do not have any special meaning when used in templates or
7664 operands to @code{asm_fprintf}.
7666 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
7667 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
7668 the variations in assembler language syntax with that mechanism. Define
7669 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
7670 if the syntax variant are larger and involve such things as different
7671 opcodes or operand order.
7674 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
7675 A C expression to output to @var{stream} some assembler code
7676 which will push hard register number @var{regno} onto the stack.
7677 The code need not be optimal, since this macro is used only when
7681 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
7682 A C expression to output to @var{stream} some assembler code
7683 which will pop hard register number @var{regno} off of the stack.
7684 The code need not be optimal, since this macro is used only when
7688 @node Dispatch Tables
7689 @subsection Output of Dispatch Tables
7691 @c prevent bad page break with this line
7692 This concerns dispatch tables.
7694 @cindex dispatch table
7695 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
7696 A C statement to output to the stdio stream @var{stream} an assembler
7697 pseudo-instruction to generate a difference between two labels.
7698 @var{value} and @var{rel} are the numbers of two internal labels. The
7699 definitions of these labels are output using
7700 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
7701 way here. For example,
7704 fprintf (@var{stream}, "\t.word L%d-L%d\n",
7705 @var{value}, @var{rel})
7708 You must provide this macro on machines where the addresses in a
7709 dispatch table are relative to the table's own address. If defined, GCC
7710 will also use this macro on all machines when producing PIC@.
7711 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
7712 mode and flags can be read.
7715 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
7716 This macro should be provided on machines where the addresses
7717 in a dispatch table are absolute.
7719 The definition should be a C statement to output to the stdio stream
7720 @var{stream} an assembler pseudo-instruction to generate a reference to
7721 a label. @var{value} is the number of an internal label whose
7722 definition is output using @code{(*targetm.asm_out.internal_label)}.
7726 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
7730 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
7731 Define this if the label before a jump-table needs to be output
7732 specially. The first three arguments are the same as for
7733 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
7734 jump-table which follows (a @code{jump_insn} containing an
7735 @code{addr_vec} or @code{addr_diff_vec}).
7737 This feature is used on system V to output a @code{swbeg} statement
7740 If this macro is not defined, these labels are output with
7741 @code{(*targetm.asm_out.internal_label)}.
7744 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
7745 Define this if something special must be output at the end of a
7746 jump-table. The definition should be a C statement to be executed
7747 after the assembler code for the table is written. It should write
7748 the appropriate code to stdio stream @var{stream}. The argument
7749 @var{table} is the jump-table insn, and @var{num} is the label-number
7750 of the preceding label.
7752 If this macro is not defined, nothing special is output at the end of
7756 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (@var{stream}, @var{decl}, @var{for_eh}, @var{empty})
7757 This target hook emits a label at the beginning of each FDE@. It
7758 should be defined on targets where FDEs need special labels, and it
7759 should write the appropriate label, for the FDE associated with the
7760 function declaration @var{decl}, to the stdio stream @var{stream}.
7761 The third argument, @var{for_eh}, is a boolean: true if this is for an
7762 exception table. The fourth argument, @var{empty}, is a boolean:
7763 true if this is a placeholder label for an omitted FDE@.
7765 The default is that FDEs are not given nonlocal labels.
7768 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (@var{stream})
7769 This target hook emits a label at the beginning of the exception table.
7770 It should be defined on targets where it is desirable for the table
7771 to be broken up according to function.
7773 The default is that no label is emitted.
7776 @deftypefn {Target Hook} void TARGET_UNWIND_EMIT (FILE * @var{stream}, rtx @var{insn})
7777 This target hook emits and assembly directives required to unwind the
7778 given instruction. This is only used when TARGET_UNWIND_INFO is set.
7781 @node Exception Region Output
7782 @subsection Assembler Commands for Exception Regions
7784 @c prevent bad page break with this line
7786 This describes commands marking the start and the end of an exception
7789 @defmac EH_FRAME_SECTION_NAME
7790 If defined, a C string constant for the name of the section containing
7791 exception handling frame unwind information. If not defined, GCC will
7792 provide a default definition if the target supports named sections.
7793 @file{crtstuff.c} uses this macro to switch to the appropriate section.
7795 You should define this symbol if your target supports DWARF 2 frame
7796 unwind information and the default definition does not work.
7799 @defmac EH_FRAME_IN_DATA_SECTION
7800 If defined, DWARF 2 frame unwind information will be placed in the
7801 data section even though the target supports named sections. This
7802 might be necessary, for instance, if the system linker does garbage
7803 collection and sections cannot be marked as not to be collected.
7805 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
7809 @defmac EH_TABLES_CAN_BE_READ_ONLY
7810 Define this macro to 1 if your target is such that no frame unwind
7811 information encoding used with non-PIC code will ever require a
7812 runtime relocation, but the linker may not support merging read-only
7813 and read-write sections into a single read-write section.
7816 @defmac MASK_RETURN_ADDR
7817 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
7818 that it does not contain any extraneous set bits in it.
7821 @defmac DWARF2_UNWIND_INFO
7822 Define this macro to 0 if your target supports DWARF 2 frame unwind
7823 information, but it does not yet work with exception handling.
7824 Otherwise, if your target supports this information (if it defines
7825 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
7826 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of
7829 If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder
7830 will be used in all cases. Defining this macro will enable the generation
7831 of DWARF 2 frame debugging information.
7833 If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1,
7834 the DWARF 2 unwinder will be the default exception handling mechanism;
7835 otherwise, @code{setjmp}/@code{longjmp} will be used by default.
7838 @defmac TARGET_UNWIND_INFO
7839 Define this macro if your target has ABI specified unwind tables. Usually
7840 these will be output by @code{TARGET_UNWIND_EMIT}.
7843 @deftypevar {Target Hook} bool TARGET_UNWID_TABLES_DEFAULT
7844 This variable should be set to @code{true} if the target ABI requires unwinding
7845 tables even when exceptions are not used.
7848 @defmac MUST_USE_SJLJ_EXCEPTIONS
7849 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
7850 runtime-variable. In that case, @file{except.h} cannot correctly
7851 determine the corresponding definition of
7852 @code{MUST_USE_SJLJ_EXCEPTIONS}, so the target must provide it directly.
7855 @defmac DWARF_CIE_DATA_ALIGNMENT
7856 This macro need only be defined if the target might save registers in the
7857 function prologue at an offset to the stack pointer that is not aligned to
7858 @code{UNITS_PER_WORD}. The definition should be the negative minimum
7859 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
7860 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
7861 the target supports DWARF 2 frame unwind information.
7864 @deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
7865 Contains the value true if the target should add a zero word onto the
7866 end of a Dwarf-2 frame info section when used for exception handling.
7867 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
7871 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
7872 Given a register, this hook should return a parallel of registers to
7873 represent where to find the register pieces. Define this hook if the
7874 register and its mode are represented in Dwarf in non-contiguous
7875 locations, or if the register should be represented in more than one
7876 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
7877 If not defined, the default is to return @code{NULL_RTX}.
7880 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
7881 This hook is used to output a reference from a frame unwinding table to
7882 the type_info object identified by @var{sym}. It should return @code{true}
7883 if the reference was output. Returning @code{false} will cause the
7884 reference to be output using the normal Dwarf2 routines.
7887 @deftypefn {Target Hook} bool TARGET_ARM_EABI_UNWINDER
7888 This hook should be set to @code{true} on targets that use an ARM EABI
7889 based unwinding library, and @code{false} on other targets. This effects
7890 the format of unwinding tables, and how the unwinder in entered after
7891 running a cleanup. The default is @code{false}.
7894 @node Alignment Output
7895 @subsection Assembler Commands for Alignment
7897 @c prevent bad page break with this line
7898 This describes commands for alignment.
7900 @defmac JUMP_ALIGN (@var{label})
7901 The alignment (log base 2) to put in front of @var{label}, which is
7902 a common destination of jumps and has no fallthru incoming edge.
7904 This macro need not be defined if you don't want any special alignment
7905 to be done at such a time. Most machine descriptions do not currently
7908 Unless it's necessary to inspect the @var{label} parameter, it is better
7909 to set the variable @var{align_jumps} in the target's
7910 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7911 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
7914 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
7915 The alignment (log base 2) to put in front of @var{label}, which follows
7918 This macro need not be defined if you don't want any special alignment
7919 to be done at such a time. Most machine descriptions do not currently
7923 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
7924 The maximum number of bytes to skip when applying
7925 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
7926 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7929 @defmac LOOP_ALIGN (@var{label})
7930 The alignment (log base 2) to put in front of @var{label}, which follows
7931 a @code{NOTE_INSN_LOOP_BEG} note.
7933 This macro need not be defined if you don't want any special alignment
7934 to be done at such a time. Most machine descriptions do not currently
7937 Unless it's necessary to inspect the @var{label} parameter, it is better
7938 to set the variable @code{align_loops} in the target's
7939 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7940 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
7943 @defmac LOOP_ALIGN_MAX_SKIP
7944 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
7945 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7948 @defmac LABEL_ALIGN (@var{label})
7949 The alignment (log base 2) to put in front of @var{label}.
7950 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
7951 the maximum of the specified values is used.
7953 Unless it's necessary to inspect the @var{label} parameter, it is better
7954 to set the variable @code{align_labels} in the target's
7955 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
7956 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
7959 @defmac LABEL_ALIGN_MAX_SKIP
7960 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
7961 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
7964 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
7965 A C statement to output to the stdio stream @var{stream} an assembler
7966 instruction to advance the location counter by @var{nbytes} bytes.
7967 Those bytes should be zero when loaded. @var{nbytes} will be a C
7968 expression of type @code{int}.
7971 @defmac ASM_NO_SKIP_IN_TEXT
7972 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
7973 text section because it fails to put zeros in the bytes that are skipped.
7974 This is true on many Unix systems, where the pseudo--op to skip bytes
7975 produces no-op instructions rather than zeros when used in the text
7979 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
7980 A C statement to output to the stdio stream @var{stream} an assembler
7981 command to advance the location counter to a multiple of 2 to the
7982 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
7985 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
7986 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
7987 for padding, if necessary.
7990 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
7991 A C statement to output to the stdio stream @var{stream} an assembler
7992 command to advance the location counter to a multiple of 2 to the
7993 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
7994 satisfy the alignment request. @var{power} and @var{max_skip} will be
7995 a C expression of type @code{int}.
7999 @node Debugging Info
8000 @section Controlling Debugging Information Format
8002 @c prevent bad page break with this line
8003 This describes how to specify debugging information.
8006 * All Debuggers:: Macros that affect all debugging formats uniformly.
8007 * DBX Options:: Macros enabling specific options in DBX format.
8008 * DBX Hooks:: Hook macros for varying DBX format.
8009 * File Names and DBX:: Macros controlling output of file names in DBX format.
8010 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
8011 * VMS Debug:: Macros for VMS debug format.
8015 @subsection Macros Affecting All Debugging Formats
8017 @c prevent bad page break with this line
8018 These macros affect all debugging formats.
8020 @defmac DBX_REGISTER_NUMBER (@var{regno})
8021 A C expression that returns the DBX register number for the compiler
8022 register number @var{regno}. In the default macro provided, the value
8023 of this expression will be @var{regno} itself. But sometimes there are
8024 some registers that the compiler knows about and DBX does not, or vice
8025 versa. In such cases, some register may need to have one number in the
8026 compiler and another for DBX@.
8028 If two registers have consecutive numbers inside GCC, and they can be
8029 used as a pair to hold a multiword value, then they @emph{must} have
8030 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
8031 Otherwise, debuggers will be unable to access such a pair, because they
8032 expect register pairs to be consecutive in their own numbering scheme.
8034 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
8035 does not preserve register pairs, then what you must do instead is
8036 redefine the actual register numbering scheme.
8039 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
8040 A C expression that returns the integer offset value for an automatic
8041 variable having address @var{x} (an RTL expression). The default
8042 computation assumes that @var{x} is based on the frame-pointer and
8043 gives the offset from the frame-pointer. This is required for targets
8044 that produce debugging output for DBX or COFF-style debugging output
8045 for SDB and allow the frame-pointer to be eliminated when the
8046 @option{-g} options is used.
8049 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
8050 A C expression that returns the integer offset value for an argument
8051 having address @var{x} (an RTL expression). The nominal offset is
8055 @defmac PREFERRED_DEBUGGING_TYPE
8056 A C expression that returns the type of debugging output GCC should
8057 produce when the user specifies just @option{-g}. Define
8058 this if you have arranged for GCC to support more than one format of
8059 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
8060 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
8061 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
8063 When the user specifies @option{-ggdb}, GCC normally also uses the
8064 value of this macro to select the debugging output format, but with two
8065 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
8066 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
8067 defined, GCC uses @code{DBX_DEBUG}.
8069 The value of this macro only affects the default debugging output; the
8070 user can always get a specific type of output by using @option{-gstabs},
8071 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
8075 @subsection Specific Options for DBX Output
8077 @c prevent bad page break with this line
8078 These are specific options for DBX output.
8080 @defmac DBX_DEBUGGING_INFO
8081 Define this macro if GCC should produce debugging output for DBX
8082 in response to the @option{-g} option.
8085 @defmac XCOFF_DEBUGGING_INFO
8086 Define this macro if GCC should produce XCOFF format debugging output
8087 in response to the @option{-g} option. This is a variant of DBX format.
8090 @defmac DEFAULT_GDB_EXTENSIONS
8091 Define this macro to control whether GCC should by default generate
8092 GDB's extended version of DBX debugging information (assuming DBX-format
8093 debugging information is enabled at all). If you don't define the
8094 macro, the default is 1: always generate the extended information
8095 if there is any occasion to.
8098 @defmac DEBUG_SYMS_TEXT
8099 Define this macro if all @code{.stabs} commands should be output while
8100 in the text section.
8103 @defmac ASM_STABS_OP
8104 A C string constant, including spacing, naming the assembler pseudo op to
8105 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
8106 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
8107 applies only to DBX debugging information format.
8110 @defmac ASM_STABD_OP
8111 A C string constant, including spacing, naming the assembler pseudo op to
8112 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
8113 value is the current location. If you don't define this macro,
8114 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
8118 @defmac ASM_STABN_OP
8119 A C string constant, including spacing, naming the assembler pseudo op to
8120 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
8121 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
8122 macro applies only to DBX debugging information format.
8125 @defmac DBX_NO_XREFS
8126 Define this macro if DBX on your system does not support the construct
8127 @samp{xs@var{tagname}}. On some systems, this construct is used to
8128 describe a forward reference to a structure named @var{tagname}.
8129 On other systems, this construct is not supported at all.
8132 @defmac DBX_CONTIN_LENGTH
8133 A symbol name in DBX-format debugging information is normally
8134 continued (split into two separate @code{.stabs} directives) when it
8135 exceeds a certain length (by default, 80 characters). On some
8136 operating systems, DBX requires this splitting; on others, splitting
8137 must not be done. You can inhibit splitting by defining this macro
8138 with the value zero. You can override the default splitting-length by
8139 defining this macro as an expression for the length you desire.
8142 @defmac DBX_CONTIN_CHAR
8143 Normally continuation is indicated by adding a @samp{\} character to
8144 the end of a @code{.stabs} string when a continuation follows. To use
8145 a different character instead, define this macro as a character
8146 constant for the character you want to use. Do not define this macro
8147 if backslash is correct for your system.
8150 @defmac DBX_STATIC_STAB_DATA_SECTION
8151 Define this macro if it is necessary to go to the data section before
8152 outputting the @samp{.stabs} pseudo-op for a non-global static
8156 @defmac DBX_TYPE_DECL_STABS_CODE
8157 The value to use in the ``code'' field of the @code{.stabs} directive
8158 for a typedef. The default is @code{N_LSYM}.
8161 @defmac DBX_STATIC_CONST_VAR_CODE
8162 The value to use in the ``code'' field of the @code{.stabs} directive
8163 for a static variable located in the text section. DBX format does not
8164 provide any ``right'' way to do this. The default is @code{N_FUN}.
8167 @defmac DBX_REGPARM_STABS_CODE
8168 The value to use in the ``code'' field of the @code{.stabs} directive
8169 for a parameter passed in registers. DBX format does not provide any
8170 ``right'' way to do this. The default is @code{N_RSYM}.
8173 @defmac DBX_REGPARM_STABS_LETTER
8174 The letter to use in DBX symbol data to identify a symbol as a parameter
8175 passed in registers. DBX format does not customarily provide any way to
8176 do this. The default is @code{'P'}.
8179 @defmac DBX_FUNCTION_FIRST
8180 Define this macro if the DBX information for a function and its
8181 arguments should precede the assembler code for the function. Normally,
8182 in DBX format, the debugging information entirely follows the assembler
8186 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
8187 Define this macro, with value 1, if the value of a symbol describing
8188 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
8189 relative to the start of the enclosing function. Normally, GCC uses
8190 an absolute address.
8193 @defmac DBX_LINES_FUNCTION_RELATIVE
8194 Define this macro, with value 1, if the value of a symbol indicating
8195 the current line number (@code{N_SLINE}) should be relative to the
8196 start of the enclosing function. Normally, GCC uses an absolute address.
8199 @defmac DBX_USE_BINCL
8200 Define this macro if GCC should generate @code{N_BINCL} and
8201 @code{N_EINCL} stabs for included header files, as on Sun systems. This
8202 macro also directs GCC to output a type number as a pair of a file
8203 number and a type number within the file. Normally, GCC does not
8204 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
8205 number for a type number.
8209 @subsection Open-Ended Hooks for DBX Format
8211 @c prevent bad page break with this line
8212 These are hooks for DBX format.
8214 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
8215 Define this macro to say how to output to @var{stream} the debugging
8216 information for the start of a scope level for variable names. The
8217 argument @var{name} is the name of an assembler symbol (for use with
8218 @code{assemble_name}) whose value is the address where the scope begins.
8221 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
8222 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
8225 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
8226 Define this macro if the target machine requires special handling to
8227 output an @code{N_FUN} entry for the function @var{decl}.
8230 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
8231 A C statement to output DBX debugging information before code for line
8232 number @var{line} of the current source file to the stdio stream
8233 @var{stream}. @var{counter} is the number of time the macro was
8234 invoked, including the current invocation; it is intended to generate
8235 unique labels in the assembly output.
8237 This macro should not be defined if the default output is correct, or
8238 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
8241 @defmac NO_DBX_FUNCTION_END
8242 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
8243 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
8244 On those machines, define this macro to turn this feature off without
8245 disturbing the rest of the gdb extensions.
8248 @defmac NO_DBX_BNSYM_ENSYM
8249 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
8250 extension construct. On those machines, define this macro to turn this
8251 feature off without disturbing the rest of the gdb extensions.
8254 @node File Names and DBX
8255 @subsection File Names in DBX Format
8257 @c prevent bad page break with this line
8258 This describes file names in DBX format.
8260 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
8261 A C statement to output DBX debugging information to the stdio stream
8262 @var{stream}, which indicates that file @var{name} is the main source
8263 file---the file specified as the input file for compilation.
8264 This macro is called only once, at the beginning of compilation.
8266 This macro need not be defined if the standard form of output
8267 for DBX debugging information is appropriate.
8269 It may be necessary to refer to a label equal to the beginning of the
8270 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
8271 to do so. If you do this, you must also set the variable
8272 @var{used_ltext_label_name} to @code{true}.
8275 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
8276 Define this macro, with value 1, if GCC should not emit an indication
8277 of the current directory for compilation and current source language at
8278 the beginning of the file.
8281 @defmac NO_DBX_GCC_MARKER
8282 Define this macro, with value 1, if GCC should not emit an indication
8283 that this object file was compiled by GCC@. The default is to emit
8284 an @code{N_OPT} stab at the beginning of every source file, with
8285 @samp{gcc2_compiled.} for the string and value 0.
8288 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
8289 A C statement to output DBX debugging information at the end of
8290 compilation of the main source file @var{name}. Output should be
8291 written to the stdio stream @var{stream}.
8293 If you don't define this macro, nothing special is output at the end
8294 of compilation, which is correct for most machines.
8297 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
8298 Define this macro @emph{instead of} defining
8299 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
8300 the end of compilation is a @code{N_SO} stab with an empty string,
8301 whose value is the highest absolute text address in the file.
8306 @subsection Macros for SDB and DWARF Output
8308 @c prevent bad page break with this line
8309 Here are macros for SDB and DWARF output.
8311 @defmac SDB_DEBUGGING_INFO
8312 Define this macro if GCC should produce COFF-style debugging output
8313 for SDB in response to the @option{-g} option.
8316 @defmac DWARF2_DEBUGGING_INFO
8317 Define this macro if GCC should produce dwarf version 2 format
8318 debugging output in response to the @option{-g} option.
8320 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (tree @var{function})
8321 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
8322 be emitted for each function. Instead of an integer return the enum
8323 value for the @code{DW_CC_} tag.
8326 To support optional call frame debugging information, you must also
8327 define @code{INCOMING_RETURN_ADDR_RTX} and either set
8328 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
8329 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
8330 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
8333 @defmac DWARF2_FRAME_INFO
8334 Define this macro to a nonzero value if GCC should always output
8335 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
8336 (@pxref{Exception Region Output} is nonzero, GCC will output this
8337 information not matter how you define @code{DWARF2_FRAME_INFO}.
8340 @defmac DWARF2_ASM_LINE_DEBUG_INFO
8341 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
8342 line debug info sections. This will result in much more compact line number
8343 tables, and hence is desirable if it works.
8346 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
8347 A C statement to issue assembly directives that create a difference
8348 between the two given labels, using an integer of the given size.
8351 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label})
8352 A C statement to issue assembly directives that create a
8353 section-relative reference to the given label, using an integer of the
8357 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
8358 A C statement to issue assembly directives that create a self-relative
8359 reference to the given label, using an integer of the given size.
8362 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{FILE}, int @var{size}, rtx @var{x})
8363 If defined, this target hook is a function which outputs a DTP-relative
8364 reference to the given TLS symbol of the specified size.
8367 @defmac PUT_SDB_@dots{}
8368 Define these macros to override the assembler syntax for the special
8369 SDB assembler directives. See @file{sdbout.c} for a list of these
8370 macros and their arguments. If the standard syntax is used, you need
8371 not define them yourself.
8375 Some assemblers do not support a semicolon as a delimiter, even between
8376 SDB assembler directives. In that case, define this macro to be the
8377 delimiter to use (usually @samp{\n}). It is not necessary to define
8378 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
8382 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
8383 Define this macro to allow references to unknown structure,
8384 union, or enumeration tags to be emitted. Standard COFF does not
8385 allow handling of unknown references, MIPS ECOFF has support for
8389 @defmac SDB_ALLOW_FORWARD_REFERENCES
8390 Define this macro to allow references to structure, union, or
8391 enumeration tags that have not yet been seen to be handled. Some
8392 assemblers choke if forward tags are used, while some require it.
8395 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
8396 A C statement to output SDB debugging information before code for line
8397 number @var{line} of the current source file to the stdio stream
8398 @var{stream}. The default is to emit an @code{.ln} directive.
8403 @subsection Macros for VMS Debug Format
8405 @c prevent bad page break with this line
8406 Here are macros for VMS debug format.
8408 @defmac VMS_DEBUGGING_INFO
8409 Define this macro if GCC should produce debugging output for VMS
8410 in response to the @option{-g} option. The default behavior for VMS
8411 is to generate minimal debug info for a traceback in the absence of
8412 @option{-g} unless explicitly overridden with @option{-g0}. This
8413 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
8414 @code{OVERRIDE_OPTIONS}.
8417 @node Floating Point
8418 @section Cross Compilation and Floating Point
8419 @cindex cross compilation and floating point
8420 @cindex floating point and cross compilation
8422 While all modern machines use twos-complement representation for integers,
8423 there are a variety of representations for floating point numbers. This
8424 means that in a cross-compiler the representation of floating point numbers
8425 in the compiled program may be different from that used in the machine
8426 doing the compilation.
8428 Because different representation systems may offer different amounts of
8429 range and precision, all floating point constants must be represented in
8430 the target machine's format. Therefore, the cross compiler cannot
8431 safely use the host machine's floating point arithmetic; it must emulate
8432 the target's arithmetic. To ensure consistency, GCC always uses
8433 emulation to work with floating point values, even when the host and
8434 target floating point formats are identical.
8436 The following macros are provided by @file{real.h} for the compiler to
8437 use. All parts of the compiler which generate or optimize
8438 floating-point calculations must use these macros. They may evaluate
8439 their operands more than once, so operands must not have side effects.
8441 @defmac REAL_VALUE_TYPE
8442 The C data type to be used to hold a floating point value in the target
8443 machine's format. Typically this is a @code{struct} containing an
8444 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
8448 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8449 Compares for equality the two values, @var{x} and @var{y}. If the target
8450 floating point format supports negative zeroes and/or NaNs,
8451 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
8452 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
8455 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
8456 Tests whether @var{x} is less than @var{y}.
8459 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
8460 Truncates @var{x} to a signed integer, rounding toward zero.
8463 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
8464 Truncates @var{x} to an unsigned integer, rounding toward zero. If
8465 @var{x} is negative, returns zero.
8468 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
8469 Converts @var{string} into a floating point number in the target machine's
8470 representation for mode @var{mode}. This routine can handle both
8471 decimal and hexadecimal floating point constants, using the syntax
8472 defined by the C language for both.
8475 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
8476 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
8479 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
8480 Determines whether @var{x} represents infinity (positive or negative).
8483 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
8484 Determines whether @var{x} represents a ``NaN'' (not-a-number).
8487 @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})
8488 Calculates an arithmetic operation on the two floating point values
8489 @var{x} and @var{y}, storing the result in @var{output} (which must be a
8492 The operation to be performed is specified by @var{code}. Only the
8493 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
8494 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
8496 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
8497 target's floating point format cannot represent infinity, it will call
8498 @code{abort}. Callers should check for this situation first, using
8499 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
8502 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
8503 Returns the negative of the floating point value @var{x}.
8506 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
8507 Returns the absolute value of @var{x}.
8510 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
8511 Truncates the floating point value @var{x} to fit in @var{mode}. The
8512 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
8513 appropriate bit pattern to be output asa floating constant whose
8514 precision accords with mode @var{mode}.
8517 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
8518 Converts a floating point value @var{x} into a double-precision integer
8519 which is then stored into @var{low} and @var{high}. If the value is not
8520 integral, it is truncated.
8523 @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})
8524 Converts a double-precision integer found in @var{low} and @var{high},
8525 into a floating point value which is then stored into @var{x}. The
8526 value is truncated to fit in mode @var{mode}.
8529 @node Mode Switching
8530 @section Mode Switching Instructions
8531 @cindex mode switching
8532 The following macros control mode switching optimizations:
8534 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
8535 Define this macro if the port needs extra instructions inserted for mode
8536 switching in an optimizing compilation.
8538 For an example, the SH4 can perform both single and double precision
8539 floating point operations, but to perform a single precision operation,
8540 the FPSCR PR bit has to be cleared, while for a double precision
8541 operation, this bit has to be set. Changing the PR bit requires a general
8542 purpose register as a scratch register, hence these FPSCR sets have to
8543 be inserted before reload, i.e.@: you can't put this into instruction emitting
8544 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
8546 You can have multiple entities that are mode-switched, and select at run time
8547 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
8548 return nonzero for any @var{entity} that needs mode-switching.
8549 If you define this macro, you also have to define
8550 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
8551 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
8552 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
8556 @defmac NUM_MODES_FOR_MODE_SWITCHING
8557 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
8558 initializer for an array of integers. Each initializer element
8559 N refers to an entity that needs mode switching, and specifies the number
8560 of different modes that might need to be set for this entity.
8561 The position of the initializer in the initializer---starting counting at
8562 zero---determines the integer that is used to refer to the mode-switched
8564 In macros that take mode arguments / yield a mode result, modes are
8565 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
8566 switch is needed / supplied.
8569 @defmac MODE_NEEDED (@var{entity}, @var{insn})
8570 @var{entity} is an integer specifying a mode-switched entity. If
8571 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
8572 return an integer value not larger than the corresponding element in
8573 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
8574 be switched into prior to the execution of @var{insn}.
8577 @defmac MODE_AFTER (@var{mode}, @var{insn})
8578 If this macro is defined, it is evaluated for every @var{insn} during
8579 mode switching. It determines the mode that an insn results in (if
8580 different from the incoming mode).
8583 @defmac MODE_ENTRY (@var{entity})
8584 If this macro is defined, it is evaluated for every @var{entity} that needs
8585 mode switching. It should evaluate to an integer, which is a mode that
8586 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
8587 is defined then @code{MODE_EXIT} must be defined.
8590 @defmac MODE_EXIT (@var{entity})
8591 If this macro is defined, it is evaluated for every @var{entity} that needs
8592 mode switching. It should evaluate to an integer, which is a mode that
8593 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
8594 is defined then @code{MODE_ENTRY} must be defined.
8597 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
8598 This macro specifies the order in which modes for @var{entity} are processed.
8599 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
8600 lowest. The value of the macro should be an integer designating a mode
8601 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
8602 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
8603 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
8606 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
8607 Generate one or more insns to set @var{entity} to @var{mode}.
8608 @var{hard_reg_live} is the set of hard registers live at the point where
8609 the insn(s) are to be inserted.
8612 @node Target Attributes
8613 @section Defining target-specific uses of @code{__attribute__}
8614 @cindex target attributes
8615 @cindex machine attributes
8616 @cindex attributes, target-specific
8618 Target-specific attributes may be defined for functions, data and types.
8619 These are described using the following target hooks; they also need to
8620 be documented in @file{extend.texi}.
8622 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
8623 If defined, this target hook points to an array of @samp{struct
8624 attribute_spec} (defined in @file{tree.h}) specifying the machine
8625 specific attributes for this target and some of the restrictions on the
8626 entities to which these attributes are applied and the arguments they
8630 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8631 If defined, this target hook is a function which returns zero if the attributes on
8632 @var{type1} and @var{type2} are incompatible, one if they are compatible,
8633 and two if they are nearly compatible (which causes a warning to be
8634 generated). If this is not defined, machine-specific attributes are
8635 supposed always to be compatible.
8638 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
8639 If defined, this target hook is a function which assigns default attributes to
8640 newly defined @var{type}.
8643 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
8644 Define this target hook if the merging of type attributes needs special
8645 handling. If defined, the result is a list of the combined
8646 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
8647 that @code{comptypes} has already been called and returned 1. This
8648 function may call @code{merge_attributes} to handle machine-independent
8652 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
8653 Define this target hook if the merging of decl attributes needs special
8654 handling. If defined, the result is a list of the combined
8655 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
8656 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
8657 when this is needed are when one attribute overrides another, or when an
8658 attribute is nullified by a subsequent definition. This function may
8659 call @code{merge_attributes} to handle machine-independent merging.
8661 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
8662 If the only target-specific handling you require is @samp{dllimport}
8663 for Microsoft Windows targets, you should define the macro
8664 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
8665 will then define a function called
8666 @code{merge_dllimport_decl_attributes} which can then be defined as
8667 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
8668 add @code{handle_dll_attribute} in the attribute table for your port
8669 to perform initial processing of the @samp{dllimport} and
8670 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
8671 @file{i386/i386.c}, for example.
8674 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (tree @var{decl})
8675 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
8676 specified. Use this hook if the target needs to add extra validation
8677 checks to @code{handle_dll_attribute}.
8680 @defmac TARGET_DECLSPEC
8681 Define this macro to a nonzero value if you want to treat
8682 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
8683 default, this behavior is enabled only for targets that define
8684 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
8685 of @code{__declspec} is via a built-in macro, but you should not rely
8686 on this implementation detail.
8689 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
8690 Define this target hook if you want to be able to add attributes to a decl
8691 when it is being created. This is normally useful for back ends which
8692 wish to implement a pragma by using the attributes which correspond to
8693 the pragma's effect. The @var{node} argument is the decl which is being
8694 created. The @var{attr_ptr} argument is a pointer to the attribute list
8695 for this decl. The list itself should not be modified, since it may be
8696 shared with other decls, but attributes may be chained on the head of
8697 the list and @code{*@var{attr_ptr}} modified to point to the new
8698 attributes, or a copy of the list may be made if further changes are
8702 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl})
8704 This target hook returns @code{true} if it is ok to inline @var{fndecl}
8705 into the current function, despite its having target-specific
8706 attributes, @code{false} otherwise. By default, if a function has a
8707 target specific attribute attached to it, it will not be inlined.
8710 @node MIPS Coprocessors
8711 @section Defining coprocessor specifics for MIPS targets.
8712 @cindex MIPS coprocessor-definition macros
8714 The MIPS specification allows MIPS implementations to have as many as 4
8715 coprocessors, each with as many as 32 private registers. GCC supports
8716 accessing these registers and transferring values between the registers
8717 and memory using asm-ized variables. For example:
8720 register unsigned int cp0count asm ("c0r1");
8726 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
8727 names may be added as described below, or the default names may be
8728 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
8730 Coprocessor registers are assumed to be epilogue-used; sets to them will
8731 be preserved even if it does not appear that the register is used again
8732 later in the function.
8734 Another note: according to the MIPS spec, coprocessor 1 (if present) is
8735 the FPU@. One accesses COP1 registers through standard mips
8736 floating-point support; they are not included in this mechanism.
8738 There is one macro used in defining the MIPS coprocessor interface which
8739 you may want to override in subtargets; it is described below.
8741 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
8742 A comma-separated list (with leading comma) of pairs describing the
8743 alternate names of coprocessor registers. The format of each entry should be
8745 @{ @var{alternatename}, @var{register_number}@}
8751 @section Parameters for Precompiled Header Validity Checking
8752 @cindex parameters, precompiled headers
8754 @deftypefn {Target Hook} void *TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
8755 This hook returns the data needed by @code{TARGET_PCH_VALID_P} and sets
8756 @samp{*@var{sz}} to the size of the data in bytes.
8759 @deftypefn {Target Hook} const char *TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
8760 This hook checks whether the options used to create a PCH file are
8761 compatible with the current settings. It returns @code{NULL}
8762 if so and a suitable error message if not. Error messages will
8763 be presented to the user and must be localized using @samp{_(@var{msg})}.
8765 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
8766 when the PCH file was created and @var{sz} is the size of that data in bytes.
8767 It's safe to assume that the data was created by the same version of the
8768 compiler, so no format checking is needed.
8770 The default definition of @code{default_pch_valid_p} should be
8771 suitable for most targets.
8774 @deftypefn {Target Hook} const char *TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
8775 If this hook is nonnull, the default implementation of
8776 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
8777 of @code{target_flags}. @var{pch_flags} specifies the value that
8778 @code{target_flags} had when the PCH file was created. The return
8779 value is the same as for @code{TARGET_PCH_VALID_P}.
8783 @section C++ ABI parameters
8784 @cindex parameters, c++ abi
8786 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
8787 Define this hook to override the integer type used for guard variables.
8788 These are used to implement one-time construction of static objects. The
8789 default is long_long_integer_type_node.
8792 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
8793 This hook determines how guard variables are used. It should return
8794 @code{false} (the default) if first byte should be used. A return value of
8795 @code{true} indicates the least significant bit should be used.
8798 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
8799 This hook returns the size of the cookie to use when allocating an array
8800 whose elements have the indicated @var{type}. Assumes that it is already
8801 known that a cookie is needed. The default is
8802 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
8803 IA64/Generic C++ ABI@.
8806 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
8807 This hook should return @code{true} if the element size should be stored in
8808 array cookies. The default is to return @code{false}.
8811 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
8812 If defined by a backend this hook allows the decision made to export
8813 class @var{type} to be overruled. Upon entry @var{import_export}
8814 will contain 1 if the class is going to be exported, @minus{}1 if it is going
8815 to be imported and 0 otherwise. This function should return the
8816 modified value and perform any other actions necessary to support the
8817 backend's targeted operating system.
8820 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
8821 This hook should return @code{true} if constructors and destructors return
8822 the address of the object created/destroyed. The default is to return
8826 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
8827 This hook returns true if the key method for a class (i.e., the method
8828 which, if defined in the current translation unit, causes the virtual
8829 table to be emitted) may be an inline function. Under the standard
8830 Itanium C++ ABI the key method may be an inline function so long as
8831 the function is not declared inline in the class definition. Under
8832 some variants of the ABI, an inline function can never be the key
8833 method. The default is to return @code{true}.
8836 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
8837 @var{decl} is a virtual table, virtual table table, typeinfo object,
8838 or other similar implicit class data object that will be emitted with
8839 external linkage in this translation unit. No ELF visibility has been
8840 explicitly specified. If the target needs to specify a visibility
8841 other than that of the containing class, use this hook to set
8842 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
8845 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
8846 This hook returns true (the default) if virtual tables and other
8847 similar implicit class data objects are always COMDAT if they have
8848 external linkage. If this hook returns false, then class data for
8849 classes whose virtual table will be emitted in only one translation
8850 unit will not be COMDAT.
8853 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
8854 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
8855 should be used to register static destructors when @option{-fuse-cxa-atexit}
8856 is in effect. The default is to return false to use @code{__cxa_atexit}.
8859 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
8860 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been
8861 defined. Use this hook to make adjustments to the class (eg, tweak
8862 visibility or perform any other required target modifications).
8866 @section Miscellaneous Parameters
8867 @cindex parameters, miscellaneous
8869 @c prevent bad page break with this line
8870 Here are several miscellaneous parameters.
8872 @defmac HAS_LONG_COND_BRANCH
8873 Define this boolean macro to indicate whether or not your architecture
8874 has conditional branches that can span all of memory. It is used in
8875 conjunction with an optimization that partitions hot and cold basic
8876 blocks into separate sections of the executable. If this macro is
8877 set to false, gcc will convert any conditional branches that attempt
8878 to cross between sections into unconditional branches or indirect jumps.
8881 @defmac HAS_LONG_UNCOND_BRANCH
8882 Define this boolean macro to indicate whether or not your architecture
8883 has unconditional branches that can span all of memory. It is used in
8884 conjunction with an optimization that partitions hot and cold basic
8885 blocks into separate sections of the executable. If this macro is
8886 set to false, gcc will convert any unconditional branches that attempt
8887 to cross between sections into indirect jumps.
8890 @defmac CASE_VECTOR_MODE
8891 An alias for a machine mode name. This is the machine mode that
8892 elements of a jump-table should have.
8895 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
8896 Optional: return the preferred mode for an @code{addr_diff_vec}
8897 when the minimum and maximum offset are known. If you define this,
8898 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
8899 To make this work, you also have to define @code{INSN_ALIGN} and
8900 make the alignment for @code{addr_diff_vec} explicit.
8901 The @var{body} argument is provided so that the offset_unsigned and scale
8902 flags can be updated.
8905 @defmac CASE_VECTOR_PC_RELATIVE
8906 Define this macro to be a C expression to indicate when jump-tables
8907 should contain relative addresses. You need not define this macro if
8908 jump-tables never contain relative addresses, or jump-tables should
8909 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
8913 @defmac CASE_VALUES_THRESHOLD
8914 Define this to be the smallest number of different values for which it
8915 is best to use a jump-table instead of a tree of conditional branches.
8916 The default is four for machines with a @code{casesi} instruction and
8917 five otherwise. This is best for most machines.
8920 @defmac CASE_USE_BIT_TESTS
8921 Define this macro to be a C expression to indicate whether C switch
8922 statements may be implemented by a sequence of bit tests. This is
8923 advantageous on processors that can efficiently implement left shift
8924 of 1 by the number of bits held in a register, but inappropriate on
8925 targets that would require a loop. By default, this macro returns
8926 @code{true} if the target defines an @code{ashlsi3} pattern, and
8927 @code{false} otherwise.
8930 @defmac WORD_REGISTER_OPERATIONS
8931 Define this macro if operations between registers with integral mode
8932 smaller than a word are always performed on the entire register.
8933 Most RISC machines have this property and most CISC machines do not.
8936 @defmac LOAD_EXTEND_OP (@var{mem_mode})
8937 Define this macro to be a C expression indicating when insns that read
8938 memory in @var{mem_mode}, an integral mode narrower than a word, set the
8939 bits outside of @var{mem_mode} to be either the sign-extension or the
8940 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
8941 of @var{mem_mode} for which the
8942 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
8943 @code{UNKNOWN} for other modes.
8945 This macro is not called with @var{mem_mode} non-integral or with a width
8946 greater than or equal to @code{BITS_PER_WORD}, so you may return any
8947 value in this case. Do not define this macro if it would always return
8948 @code{UNKNOWN}. On machines where this macro is defined, you will normally
8949 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
8951 You may return a non-@code{UNKNOWN} value even if for some hard registers
8952 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
8953 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
8954 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
8955 integral mode larger than this but not larger than @code{word_mode}.
8957 You must return @code{UNKNOWN} if for some hard registers that allow this
8958 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
8959 @code{word_mode}, but that they can change to another integral mode that
8960 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
8963 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
8964 Define this macro if loading short immediate values into registers sign
8968 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
8969 Define this macro if the same instructions that convert a floating
8970 point number to a signed fixed point number also convert validly to an
8974 @deftypefn {Target Hook} int TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
8975 When @option{-ffast-math} is in effect, GCC tries to optimize
8976 divisions by the same divisor, by turning them into multiplications by
8977 the reciprocal. This target hook specifies the minimum number of divisions
8978 that should be there for GCC to perform the optimization for a variable
8979 of mode @var{mode}. The default implementation returns 3 if the machine
8980 has an instruction for the division, and 2 if it does not.
8984 The maximum number of bytes that a single instruction can move quickly
8985 between memory and registers or between two memory locations.
8988 @defmac MAX_MOVE_MAX
8989 The maximum number of bytes that a single instruction can move quickly
8990 between memory and registers or between two memory locations. If this
8991 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
8992 constant value that is the largest value that @code{MOVE_MAX} can have
8996 @defmac SHIFT_COUNT_TRUNCATED
8997 A C expression that is nonzero if on this machine the number of bits
8998 actually used for the count of a shift operation is equal to the number
8999 of bits needed to represent the size of the object being shifted. When
9000 this macro is nonzero, the compiler will assume that it is safe to omit
9001 a sign-extend, zero-extend, and certain bitwise `and' instructions that
9002 truncates the count of a shift operation. On machines that have
9003 instructions that act on bit-fields at variable positions, which may
9004 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
9005 also enables deletion of truncations of the values that serve as
9006 arguments to bit-field instructions.
9008 If both types of instructions truncate the count (for shifts) and
9009 position (for bit-field operations), or if no variable-position bit-field
9010 instructions exist, you should define this macro.
9012 However, on some machines, such as the 80386 and the 680x0, truncation
9013 only applies to shift operations and not the (real or pretended)
9014 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
9015 such machines. Instead, add patterns to the @file{md} file that include
9016 the implied truncation of the shift instructions.
9018 You need not define this macro if it would always have the value of zero.
9021 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
9022 @deftypefn {Target Hook} int TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
9023 This function describes how the standard shift patterns for @var{mode}
9024 deal with shifts by negative amounts or by more than the width of the mode.
9025 @xref{shift patterns}.
9027 On many machines, the shift patterns will apply a mask @var{m} to the
9028 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
9029 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
9030 this is true for mode @var{mode}, the function should return @var{m},
9031 otherwise it should return 0. A return value of 0 indicates that no
9032 particular behavior is guaranteed.
9034 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
9035 @emph{not} apply to general shift rtxes; it applies only to instructions
9036 that are generated by the named shift patterns.
9038 The default implementation of this function returns
9039 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
9040 and 0 otherwise. This definition is always safe, but if
9041 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
9042 nevertheless truncate the shift count, you may get better code
9046 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
9047 A C expression which is nonzero if on this machine it is safe to
9048 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
9049 bits (where @var{outprec} is smaller than @var{inprec}) by merely
9050 operating on it as if it had only @var{outprec} bits.
9052 On many machines, this expression can be 1.
9054 @c rearranged this, removed the phrase "it is reported that". this was
9055 @c to fix an overfull hbox. --mew 10feb93
9056 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
9057 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
9058 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
9059 such cases may improve things.
9062 @defmac STORE_FLAG_VALUE
9063 A C expression describing the value returned by a comparison operator
9064 with an integral mode and stored by a store-flag instruction
9065 (@samp{s@var{cond}}) when the condition is true. This description must
9066 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
9067 comparison operators whose results have a @code{MODE_INT} mode.
9069 A value of 1 or @minus{}1 means that the instruction implementing the
9070 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
9071 and 0 when the comparison is false. Otherwise, the value indicates
9072 which bits of the result are guaranteed to be 1 when the comparison is
9073 true. This value is interpreted in the mode of the comparison
9074 operation, which is given by the mode of the first operand in the
9075 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
9076 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
9079 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
9080 generate code that depends only on the specified bits. It can also
9081 replace comparison operators with equivalent operations if they cause
9082 the required bits to be set, even if the remaining bits are undefined.
9083 For example, on a machine whose comparison operators return an
9084 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
9085 @samp{0x80000000}, saying that just the sign bit is relevant, the
9089 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
9096 (ashift:SI @var{x} (const_int @var{n}))
9100 where @var{n} is the appropriate shift count to move the bit being
9101 tested into the sign bit.
9103 There is no way to describe a machine that always sets the low-order bit
9104 for a true value, but does not guarantee the value of any other bits,
9105 but we do not know of any machine that has such an instruction. If you
9106 are trying to port GCC to such a machine, include an instruction to
9107 perform a logical-and of the result with 1 in the pattern for the
9108 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
9110 Often, a machine will have multiple instructions that obtain a value
9111 from a comparison (or the condition codes). Here are rules to guide the
9112 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
9117 Use the shortest sequence that yields a valid definition for
9118 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
9119 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
9120 comparison operators to do so because there may be opportunities to
9121 combine the normalization with other operations.
9124 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
9125 slightly preferred on machines with expensive jumps and 1 preferred on
9129 As a second choice, choose a value of @samp{0x80000001} if instructions
9130 exist that set both the sign and low-order bits but do not define the
9134 Otherwise, use a value of @samp{0x80000000}.
9137 Many machines can produce both the value chosen for
9138 @code{STORE_FLAG_VALUE} and its negation in the same number of
9139 instructions. On those machines, you should also define a pattern for
9140 those cases, e.g., one matching
9143 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
9146 Some machines can also perform @code{and} or @code{plus} operations on
9147 condition code values with less instructions than the corresponding
9148 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
9149 machines, define the appropriate patterns. Use the names @code{incscc}
9150 and @code{decscc}, respectively, for the patterns which perform
9151 @code{plus} or @code{minus} operations on condition code values. See
9152 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
9153 find such instruction sequences on other machines.
9155 If this macro is not defined, the default value, 1, is used. You need
9156 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
9157 instructions, or if the value generated by these instructions is 1.
9160 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
9161 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
9162 returned when comparison operators with floating-point results are true.
9163 Define this macro on machines that have comparison operations that return
9164 floating-point values. If there are no such operations, do not define
9168 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
9169 A C expression that gives a rtx representing the nonzero true element
9170 for vector comparisons. The returned rtx should be valid for the inner
9171 mode of @var{mode} which is guaranteed to be a vector mode. Define
9172 this macro on machines that have vector comparison operations that
9173 return a vector result. If there are no such operations, do not define
9174 this macro. Typically, this macro is defined as @code{const1_rtx} or
9175 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
9176 the compiler optimizing such vector comparison operations for the
9180 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9181 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
9182 A C expression that evaluates to true if the architecture defines a value
9183 for @code{clz} or @code{ctz} with a zero operand. If so, @var{value}
9184 should be set to this value. If this macro is not defined, the value of
9185 @code{clz} or @code{ctz} is assumed to be undefined.
9187 This macro must be defined if the target's expansion for @code{ffs}
9188 relies on a particular value to get correct results. Otherwise it
9189 is not necessary, though it may be used to optimize some corner cases.
9191 Note that regardless of this macro the ``definedness'' of @code{clz}
9192 and @code{ctz} at zero do @emph{not} extend to the builtin functions
9193 visible to the user. Thus one may be free to adjust the value at will
9194 to match the target expansion of these operations without fear of
9199 An alias for the machine mode for pointers. On most machines, define
9200 this to be the integer mode corresponding to the width of a hardware
9201 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
9202 On some machines you must define this to be one of the partial integer
9203 modes, such as @code{PSImode}.
9205 The width of @code{Pmode} must be at least as large as the value of
9206 @code{POINTER_SIZE}. If it is not equal, you must define the macro
9207 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
9211 @defmac FUNCTION_MODE
9212 An alias for the machine mode used for memory references to functions
9213 being called, in @code{call} RTL expressions. On most machines this
9214 should be @code{QImode}.
9217 @defmac STDC_0_IN_SYSTEM_HEADERS
9218 In normal operation, the preprocessor expands @code{__STDC__} to the
9219 constant 1, to signify that GCC conforms to ISO Standard C@. On some
9220 hosts, like Solaris, the system compiler uses a different convention,
9221 where @code{__STDC__} is normally 0, but is 1 if the user specifies
9222 strict conformance to the C Standard.
9224 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
9225 convention when processing system header files, but when processing user
9226 files @code{__STDC__} will always expand to 1.
9229 @defmac NO_IMPLICIT_EXTERN_C
9230 Define this macro if the system header files support C++ as well as C@.
9231 This macro inhibits the usual method of using system header files in
9232 C++, which is to pretend that the file's contents are enclosed in
9233 @samp{extern "C" @{@dots{}@}}.
9238 @defmac REGISTER_TARGET_PRAGMAS ()
9239 Define this macro if you want to implement any target-specific pragmas.
9240 If defined, it is a C expression which makes a series of calls to
9241 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
9242 for each pragma. The macro may also do any
9243 setup required for the pragmas.
9245 The primary reason to define this macro is to provide compatibility with
9246 other compilers for the same target. In general, we discourage
9247 definition of target-specific pragmas for GCC@.
9249 If the pragma can be implemented by attributes then you should consider
9250 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
9252 Preprocessor macros that appear on pragma lines are not expanded. All
9253 @samp{#pragma} directives that do not match any registered pragma are
9254 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
9257 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9258 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
9260 Each call to @code{c_register_pragma} or
9261 @code{c_register_pragma_with_expansion} establishes one pragma. The
9262 @var{callback} routine will be called when the preprocessor encounters a
9266 #pragma [@var{space}] @var{name} @dots{}
9269 @var{space} is the case-sensitive namespace of the pragma, or
9270 @code{NULL} to put the pragma in the global namespace. The callback
9271 routine receives @var{pfile} as its first argument, which can be passed
9272 on to cpplib's functions if necessary. You can lex tokens after the
9273 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
9274 callback will be silently ignored. The end of the line is indicated by
9275 a token of type @code{CPP_EOF}. Macro expansion occurs on the
9276 arguments of pragmas registered with
9277 @code{c_register_pragma_with_expansion} but not on the arguments of
9278 pragmas registered with @code{c_register_pragma}.
9280 For an example use of this routine, see @file{c4x.h} and the callback
9281 routines defined in @file{c4x-c.c}.
9283 Note that the use of @code{pragma_lex} is specific to the C and C++
9284 compilers. It will not work in the Java or Fortran compilers, or any
9285 other language compilers for that matter. Thus if @code{pragma_lex} is going
9286 to be called from target-specific code, it must only be done so when
9287 building the C and C++ compilers. This can be done by defining the
9288 variables @code{c_target_objs} and @code{cxx_target_objs} in the
9289 target entry in the @file{config.gcc} file. These variables should name
9290 the target-specific, language-specific object file which contains the
9291 code that uses @code{pragma_lex}. Note it will also be necessary to add a
9292 rule to the makefile fragment pointed to by @code{tmake_file} that shows
9293 how to build this object file.
9298 @defmac HANDLE_SYSV_PRAGMA
9299 Define this macro (to a value of 1) if you want the System V style
9300 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
9301 [=<value>]} to be supported by gcc.
9303 The pack pragma specifies the maximum alignment (in bytes) of fields
9304 within a structure, in much the same way as the @samp{__aligned__} and
9305 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
9306 the behavior to the default.
9308 A subtlety for Microsoft Visual C/C++ style bit-field packing
9309 (e.g.@: -mms-bitfields) for targets that support it:
9310 When a bit-field is inserted into a packed record, the whole size
9311 of the underlying type is used by one or more same-size adjacent
9312 bit-fields (that is, if its long:3, 32 bits is used in the record,
9313 and any additional adjacent long bit-fields are packed into the same
9314 chunk of 32 bits. However, if the size changes, a new field of that
9317 If both MS bit-fields and @samp{__attribute__((packed))} are used,
9318 the latter will take precedence. If @samp{__attribute__((packed))} is
9319 used on a single field when MS bit-fields are in use, it will take
9320 precedence for that field, but the alignment of the rest of the structure
9321 may affect its placement.
9323 The weak pragma only works if @code{SUPPORTS_WEAK} and
9324 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
9325 of specifically named weak labels, optionally with a value.
9330 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
9331 Define this macro (to a value of 1) if you want to support the Win32
9332 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
9333 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
9334 alignment (in bytes) of fields within a structure, in much the same way as
9335 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
9336 pack value of zero resets the behavior to the default. Successive
9337 invocations of this pragma cause the previous values to be stacked, so
9338 that invocations of @samp{#pragma pack(pop)} will return to the previous
9342 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
9343 Define this macro, as well as
9344 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
9345 arguments of @samp{#pragma pack}.
9348 @defmac TARGET_DEFAULT_PACK_STRUCT
9349 If your target requires a structure packing default other than 0 (meaning
9350 the machine default), define this macro to the necessary value (in bytes).
9351 This must be a value that would also valid to be used with
9352 @samp{#pragma pack()} (that is, a small power of two).
9355 @defmac DOLLARS_IN_IDENTIFIERS
9356 Define this macro to control use of the character @samp{$} in
9357 identifier names for the C family of languages. 0 means @samp{$} is
9358 not allowed by default; 1 means it is allowed. 1 is the default;
9359 there is no need to define this macro in that case.
9362 @defmac NO_DOLLAR_IN_LABEL
9363 Define this macro if the assembler does not accept the character
9364 @samp{$} in label names. By default constructors and destructors in
9365 G++ have @samp{$} in the identifiers. If this macro is defined,
9366 @samp{.} is used instead.
9369 @defmac NO_DOT_IN_LABEL
9370 Define this macro if the assembler does not accept the character
9371 @samp{.} in label names. By default constructors and destructors in G++
9372 have names that use @samp{.}. If this macro is defined, these names
9373 are rewritten to avoid @samp{.}.
9376 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
9377 Define this macro as a C expression that is nonzero if it is safe for the
9378 delay slot scheduler to place instructions in the delay slot of @var{insn},
9379 even if they appear to use a resource set or clobbered in @var{insn}.
9380 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
9381 every @code{call_insn} has this behavior. On machines where some @code{insn}
9382 or @code{jump_insn} is really a function call and hence has this behavior,
9383 you should define this macro.
9385 You need not define this macro if it would always return zero.
9388 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
9389 Define this macro as a C expression that is nonzero if it is safe for the
9390 delay slot scheduler to place instructions in the delay slot of @var{insn},
9391 even if they appear to set or clobber a resource referenced in @var{insn}.
9392 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
9393 some @code{insn} or @code{jump_insn} is really a function call and its operands
9394 are registers whose use is actually in the subroutine it calls, you should
9395 define this macro. Doing so allows the delay slot scheduler to move
9396 instructions which copy arguments into the argument registers into the delay
9399 You need not define this macro if it would always return zero.
9402 @defmac MULTIPLE_SYMBOL_SPACES
9403 Define this macro as a C expression that is nonzero if, in some cases,
9404 global symbols from one translation unit may not be bound to undefined
9405 symbols in another translation unit without user intervention. For
9406 instance, under Microsoft Windows symbols must be explicitly imported
9407 from shared libraries (DLLs).
9409 You need not define this macro if it would always evaluate to zero.
9412 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
9413 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
9414 any hard regs the port wishes to automatically clobber for an asm.
9415 It should return the result of the last @code{tree_cons} used to add a
9416 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
9417 corresponding parameters to the asm and may be inspected to avoid
9418 clobbering a register that is an input or output of the asm. You can use
9419 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
9420 for overlap with regards to asm-declared registers.
9423 @defmac MATH_LIBRARY
9424 Define this macro as a C string constant for the linker argument to link
9425 in the system math library, or @samp{""} if the target does not have a
9426 separate math library.
9428 You need only define this macro if the default of @samp{"-lm"} is wrong.
9431 @defmac LIBRARY_PATH_ENV
9432 Define this macro as a C string constant for the environment variable that
9433 specifies where the linker should look for libraries.
9435 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
9439 @defmac TARGET_POSIX_IO
9440 Define this macro if the target supports the following POSIX@ file
9441 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
9442 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
9443 to use file locking when exiting a program, which avoids race conditions
9444 if the program has forked. It will also create directories at run-time
9445 for cross-profiling.
9448 @defmac MAX_CONDITIONAL_EXECUTE
9450 A C expression for the maximum number of instructions to execute via
9451 conditional execution instructions instead of a branch. A value of
9452 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
9453 1 if it does use cc0.
9456 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
9457 Used if the target needs to perform machine-dependent modifications on the
9458 conditionals used for turning basic blocks into conditionally executed code.
9459 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
9460 contains information about the currently processed blocks. @var{true_expr}
9461 and @var{false_expr} are the tests that are used for converting the
9462 then-block and the else-block, respectively. Set either @var{true_expr} or
9463 @var{false_expr} to a null pointer if the tests cannot be converted.
9466 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
9467 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
9468 if-statements into conditions combined by @code{and} and @code{or} operations.
9469 @var{bb} contains the basic block that contains the test that is currently
9470 being processed and about to be turned into a condition.
9473 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
9474 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
9475 be converted to conditional execution format. @var{ce_info} points to
9476 a data structure, @code{struct ce_if_block}, which contains information
9477 about the currently processed blocks.
9480 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
9481 A C expression to perform any final machine dependent modifications in
9482 converting code to conditional execution. The involved basic blocks
9483 can be found in the @code{struct ce_if_block} structure that is pointed
9484 to by @var{ce_info}.
9487 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
9488 A C expression to cancel any machine dependent modifications in
9489 converting code to conditional execution. The involved basic blocks
9490 can be found in the @code{struct ce_if_block} structure that is pointed
9491 to by @var{ce_info}.
9494 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
9495 A C expression to initialize any extra fields in a @code{struct ce_if_block}
9496 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
9499 @defmac IFCVT_EXTRA_FIELDS
9500 If defined, it should expand to a set of field declarations that will be
9501 added to the @code{struct ce_if_block} structure. These should be initialized
9502 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
9505 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG ()
9506 If non-null, this hook performs a target-specific pass over the
9507 instruction stream. The compiler will run it at all optimization levels,
9508 just before the point at which it normally does delayed-branch scheduling.
9510 The exact purpose of the hook varies from target to target. Some use
9511 it to do transformations that are necessary for correctness, such as
9512 laying out in-function constant pools or avoiding hardware hazards.
9513 Others use it as an opportunity to do some machine-dependent optimizations.
9515 You need not implement the hook if it has nothing to do. The default
9519 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS ()
9520 Define this hook if you have any machine-specific built-in functions
9521 that need to be defined. It should be a function that performs the
9524 Machine specific built-in functions can be useful to expand special machine
9525 instructions that would otherwise not normally be generated because
9526 they have no equivalent in the source language (for example, SIMD vector
9527 instructions or prefetch instructions).
9529 To create a built-in function, call the function
9530 @code{lang_hooks.builtin_function}
9531 which is defined by the language front end. You can use any type nodes set
9532 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
9533 only language front ends that use those two functions will call
9534 @samp{TARGET_INIT_BUILTINS}.
9537 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
9539 Expand a call to a machine specific built-in function that was set up by
9540 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
9541 function call; the result should go to @var{target} if that is
9542 convenient, and have mode @var{mode} if that is convenient.
9543 @var{subtarget} may be used as the target for computing one of
9544 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
9545 ignored. This function should return the result of the call to the
9549 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (tree @var{fndecl}, tree @var{arglist})
9551 Select a replacement for a machine specific built-in function that
9552 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
9553 @emph{before} regular type checking, and so allows the target to
9554 implement a crude form of function overloading. @var{fndecl} is the
9555 declaration of the built-in function. @var{arglist} is the list of
9556 arguments passed to the built-in function. The result is a
9557 complete expression that implements the operation, usually
9558 another @code{CALL_EXPR}.
9561 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore})
9563 Fold a call to a machine specific built-in function that was set up by
9564 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
9565 built-in function. @var{arglist} is the list of arguments passed to
9566 the built-in function. The result is another tree containing a
9567 simplified expression for the call's result. If @var{ignore} is true
9568 the value will be ignored.
9571 @deftypefn {Target Hook} const char * TARGET_INVALID_WITHIN_DOLOOP (rtx @var{insn})
9573 Take an instruction in @var{insn} and return NULL if it is valid within a
9574 low-overhead loop, otherwise return a string why doloop could not be applied.
9576 Many targets use special registers for low-overhead looping. For any
9577 instruction that clobbers these this function should return a string indicating
9578 the reason why the doloop could not be applied.
9579 By default, the RTL loop optimizer does not use a present doloop pattern for
9580 loops containing function calls or branch on table instructions.
9583 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
9585 Take a branch insn in @var{branch1} and another in @var{branch2}.
9586 Return true if redirecting @var{branch1} to the destination of
9587 @var{branch2} is possible.
9589 On some targets, branches may have a limited range. Optimizing the
9590 filling of delay slots can result in branches being redirected, and this
9591 may in turn cause a branch offset to overflow.
9594 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (rtx @var{x}, @var{outer_code})
9595 This target hook returns @code{true} if @var{x} is considered to be commutative.
9596 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
9597 PLUS to be commutative inside a MEM. @var{outer_code} is the rtx code
9598 of the enclosing rtl, if known, otherwise it is UNKNOWN.
9601 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
9603 When the initial value of a hard register has been copied in a pseudo
9604 register, it is often not necessary to actually allocate another register
9605 to this pseudo register, because the original hard register or a stack slot
9606 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
9607 is called at the start of register allocation once for each hard register
9608 that had its initial value copied by using
9609 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
9610 Possible values are @code{NULL_RTX}, if you don't want
9611 to do any special allocation, a @code{REG} rtx---that would typically be
9612 the hard register itself, if it is known not to be clobbered---or a
9614 If you are returning a @code{MEM}, this is only a hint for the allocator;
9615 it might decide to use another register anyways.
9616 You may use @code{current_function_leaf_function} in the hook, functions
9617 that use @code{REG_N_SETS}, to determine if the hard
9618 register in question will not be clobbered.
9619 The default value of this hook is @code{NULL}, which disables any special
9623 @defmac TARGET_OBJECT_SUFFIX
9624 Define this macro to be a C string representing the suffix for object
9625 files on your target machine. If you do not define this macro, GCC will
9626 use @samp{.o} as the suffix for object files.
9629 @defmac TARGET_EXECUTABLE_SUFFIX
9630 Define this macro to be a C string representing the suffix to be
9631 automatically added to executable files on your target machine. If you
9632 do not define this macro, GCC will use the null string as the suffix for
9636 @defmac COLLECT_EXPORT_LIST
9637 If defined, @code{collect2} will scan the individual object files
9638 specified on its command line and create an export list for the linker.
9639 Define this macro for systems like AIX, where the linker discards
9640 object files that are not referenced from @code{main} and uses export
9644 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
9645 Define this macro to a C expression representing a variant of the
9646 method call @var{mdecl}, if Java Native Interface (JNI) methods
9647 must be invoked differently from other methods on your target.
9648 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
9649 the @code{stdcall} calling convention and this macro is then
9650 defined as this expression:
9653 build_type_attribute_variant (@var{mdecl},
9655 (get_identifier ("stdcall"),
9660 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
9661 This target hook returns @code{true} past the point in which new jump
9662 instructions could be created. On machines that require a register for
9663 every jump such as the SHmedia ISA of SH5, this point would typically be
9664 reload, so this target hook should be defined to a function such as:
9668 cannot_modify_jumps_past_reload_p ()
9670 return (reload_completed || reload_in_progress);
9675 @deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
9676 This target hook returns a register class for which branch target register
9677 optimizations should be applied. All registers in this class should be
9678 usable interchangeably. After reload, registers in this class will be
9679 re-allocated and loads will be hoisted out of loops and be subjected
9680 to inter-block scheduling.
9683 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
9684 Branch target register optimization will by default exclude callee-saved
9686 that are not already live during the current function; if this target hook
9687 returns true, they will be included. The target code must than make sure
9688 that all target registers in the class returned by
9689 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
9690 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
9691 epilogues have already been generated. Note, even if you only return
9692 true when @var{after_prologue_epilogue_gen} is false, you still are likely
9693 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
9694 to reserve space for caller-saved target registers.
9697 @defmac POWI_MAX_MULTS
9698 If defined, this macro is interpreted as a signed integer C expression
9699 that specifies the maximum number of floating point multiplications
9700 that should be emitted when expanding exponentiation by an integer
9701 constant inline. When this value is defined, exponentiation requiring
9702 more than this number of multiplications is implemented by calling the
9703 system library's @code{pow}, @code{powf} or @code{powl} routines.
9704 The default value places no upper bound on the multiplication count.
9707 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9708 This target hook should register any extra include files for the
9709 target. The parameter @var{stdinc} indicates if normal include files
9710 are present. The parameter @var{sysroot} is the system root directory.
9711 The parameter @var{iprefix} is the prefix for the gcc directory.
9714 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
9715 This target hook should register any extra include files for the
9716 target before any standard headers. The parameter @var{stdinc}
9717 indicates if normal include files are present. The parameter
9718 @var{sysroot} is the system root directory. The parameter
9719 @var{iprefix} is the prefix for the gcc directory.
9722 @deftypefn Macro void TARGET_OPTF (char *@var{path})
9723 This target hook should register special include paths for the target.
9724 The parameter @var{path} is the include to register. On Darwin
9725 systems, this is used for Framework includes, which have semantics
9726 that are different from @option{-I}.
9729 @deftypefn {Target Hook} bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
9730 This target hook returns @code{true} if it is safe to use a local alias
9731 for a virtual function @var{fndecl} when constructing thunks,
9732 @code{false} otherwise. By default, the hook returns @code{true} for all
9733 functions, if a target supports aliases (i.e.@: defines
9734 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
9737 @defmac TARGET_FORMAT_TYPES
9738 If defined, this macro is the name of a global variable containing
9739 target-specific format checking information for the @option{-Wformat}
9740 option. The default is to have no target-specific format checks.
9743 @defmac TARGET_N_FORMAT_TYPES
9744 If defined, this macro is the number of entries in
9745 @code{TARGET_FORMAT_TYPES}.
9748 @deftypefn {Target Hook} bool TARGET_RELAXED_ORDERING
9749 If set to @code{true}, means that the target's memory model does not
9750 guarantee that loads which do not depend on one another will access
9751 main memory in the order of the instruction stream; if ordering is
9752 important, an explicit memory barrier must be used. This is true of
9753 many recent processors which implement a policy of ``relaxed,''
9754 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
9755 and ia64. The default is @code{false}.
9758 @deftypefn {Target Hook} const char *TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (tree @var{typelist}, tree @var{funcdecl}, tree @var{val})
9759 If defined, this macro returns the diagnostic message when it is
9760 illegal to pass argument @var{val} to function @var{funcdecl}
9761 with prototype @var{typelist}.
9764 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (tree @var{fromtype}, tree @var{totype})
9765 If defined, this macro returns the diagnostic message when it is
9766 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
9767 if validity should be determined by the front end.
9770 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, tree @var{type})
9771 If defined, this macro returns the diagnostic message when it is
9772 invalid to apply operation @var{op} (where unary plus is denoted by
9773 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
9774 if validity should be determined by the front end.
9777 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, tree @var{type1}, tree @var{type2})
9778 If defined, this macro returns the diagnostic message when it is
9779 invalid to apply operation @var{op} to operands of types @var{type1}
9780 and @var{type2}, or @code{NULL} if validity should be determined by
9784 @defmac TARGET_USE_JCR_SECTION
9785 This macro determines whether to use the JCR section to register Java
9786 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
9787 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.