1 @c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001,
2 @c 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010
3 @c Free Software Foundation, Inc.
4 @c This is part of the GCC manual.
5 @c For copying conditions, see the file gcc.texi.
8 @chapter Target Description Macros and Functions
9 @cindex machine description macros
10 @cindex target description macros
11 @cindex macros, target description
12 @cindex @file{tm.h} macros
14 In addition to the file @file{@var{machine}.md}, a machine description
15 includes a C header file conventionally given the name
16 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
17 The header file defines numerous macros that convey the information
18 about the target machine that does not fit into the scheme of the
19 @file{.md} file. The file @file{tm.h} should be a link to
20 @file{@var{machine}.h}. The header file @file{config.h} includes
21 @file{tm.h} and most compiler source files include @file{config.h}. The
22 source file defines a variable @code{targetm}, which is a structure
23 containing pointers to functions and data relating to the target
24 machine. @file{@var{machine}.c} should also contain their definitions,
25 if they are not defined elsewhere in GCC, and other functions called
26 through the macros defined in the @file{.h} file.
29 * Target Structure:: The @code{targetm} variable.
30 * Driver:: Controlling how the driver runs the compilation passes.
31 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
32 * Per-Function Data:: Defining data structures for per-function information.
33 * Storage Layout:: Defining sizes and alignments of data.
34 * Type Layout:: Defining sizes and properties of basic user data types.
35 * Registers:: Naming and describing the hardware registers.
36 * Register Classes:: Defining the classes of hardware registers.
37 * Old Constraints:: The old way to define machine-specific constraints.
38 * Stack and Calling:: Defining which way the stack grows and by how much.
39 * Varargs:: Defining the varargs macros.
40 * Trampolines:: Code set up at run time to enter a nested function.
41 * Library Calls:: Controlling how library routines are implicitly called.
42 * Addressing Modes:: Defining addressing modes valid for memory operands.
43 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
44 * Condition Code:: Defining how insns update the condition code.
45 * Costs:: Defining relative costs of different operations.
46 * Scheduling:: Adjusting the behavior of the instruction scheduler.
47 * Sections:: Dividing storage into text, data, and other sections.
48 * PIC:: Macros for position independent code.
49 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
50 * Debugging Info:: Defining the format of debugging output.
51 * Floating Point:: Handling floating point for cross-compilers.
52 * Mode Switching:: Insertion of mode-switching instructions.
53 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
54 * Emulated TLS:: Emulated TLS support.
55 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
56 * PCH Target:: Validity checking for precompiled headers.
57 * C++ ABI:: Controlling C++ ABI changes.
58 * Named Address Spaces:: Adding support for named address spaces
59 * Misc:: Everything else.
62 @node Target Structure
63 @section The Global @code{targetm} Variable
65 @cindex target functions
67 @deftypevar {struct gcc_target} targetm
68 The target @file{.c} file must define the global @code{targetm} variable
69 which contains pointers to functions and data relating to the target
70 machine. The variable is declared in @file{target.h};
71 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
72 used to initialize the variable, and macros for the default initializers
73 for elements of the structure. The @file{.c} file should override those
74 macros for which the default definition is inappropriate. For example:
77 #include "target-def.h"
79 /* @r{Initialize the GCC target structure.} */
81 #undef TARGET_COMP_TYPE_ATTRIBUTES
82 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
84 struct gcc_target targetm = TARGET_INITIALIZER;
88 Where a macro should be defined in the @file{.c} file in this manner to
89 form part of the @code{targetm} structure, it is documented below as a
90 ``Target Hook'' with a prototype. Many macros will change in future
91 from being defined in the @file{.h} file to being part of the
92 @code{targetm} structure.
95 @section Controlling the Compilation Driver, @file{gcc}
97 @cindex controlling the compilation driver
99 @c prevent bad page break with this line
100 You can control the compilation driver.
102 @defmac SWITCH_TAKES_ARG (@var{char})
103 A C expression which determines whether the option @option{-@var{char}}
104 takes arguments. The value should be the number of arguments that
105 option takes--zero, for many options.
107 By default, this macro is defined as
108 @code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options
109 properly. You need not define @code{SWITCH_TAKES_ARG} unless you
110 wish to add additional options which take arguments. Any redefinition
111 should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for
115 @defmac WORD_SWITCH_TAKES_ARG (@var{name})
116 A C expression which determines whether the option @option{-@var{name}}
117 takes arguments. The value should be the number of arguments that
118 option takes--zero, for many options. This macro rather than
119 @code{SWITCH_TAKES_ARG} is used for multi-character option names.
121 By default, this macro is defined as
122 @code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options
123 properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you
124 wish to add additional options which take arguments. Any redefinition
125 should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for
129 @defmac SWITCH_CURTAILS_COMPILATION (@var{char})
130 A C expression which determines whether the option @option{-@var{char}}
131 stops compilation before the generation of an executable. The value is
132 boolean, nonzero if the option does stop an executable from being
133 generated, zero otherwise.
135 By default, this macro is defined as
136 @code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard
137 options properly. You need not define
138 @code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional
139 options which affect the generation of an executable. Any redefinition
140 should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check
141 for additional options.
144 @defmac SWITCHES_NEED_SPACES
145 A string-valued C expression which enumerates the options for which
146 the linker needs a space between the option and its argument.
148 If this macro is not defined, the default value is @code{""}.
151 @defmac TARGET_OPTION_TRANSLATE_TABLE
152 If defined, a list of pairs of strings, the first of which is a
153 potential command line target to the @file{gcc} driver program, and the
154 second of which is a space-separated (tabs and other whitespace are not
155 supported) list of options with which to replace the first option. The
156 target defining this list is responsible for assuring that the results
157 are valid. Replacement options may not be the @code{--opt} style, they
158 must be the @code{-opt} style. It is the intention of this macro to
159 provide a mechanism for substitution that affects the multilibs chosen,
160 such as one option that enables many options, some of which select
161 multilibs. Example nonsensical definition, where @option{-malt-abi},
162 @option{-EB}, and @option{-mspoo} cause different multilibs to be chosen:
165 #define TARGET_OPTION_TRANSLATE_TABLE \
166 @{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \
167 @{ "-compat", "-EB -malign=4 -mspoo" @}
171 @defmac DRIVER_SELF_SPECS
172 A list of specs for the driver itself. It should be a suitable
173 initializer for an array of strings, with no surrounding braces.
175 The driver applies these specs to its own command line between loading
176 default @file{specs} files (but not command-line specified ones) and
177 choosing the multilib directory or running any subcommands. It
178 applies them in the order given, so each spec can depend on the
179 options added by earlier ones. It is also possible to remove options
180 using @samp{%<@var{option}} in the usual way.
182 This macro can be useful when a port has several interdependent target
183 options. It provides a way of standardizing the command line so
184 that the other specs are easier to write.
186 Do not define this macro if it does not need to do anything.
189 @defmac OPTION_DEFAULT_SPECS
190 A list of specs used to support configure-time default options (i.e.@:
191 @option{--with} options) in the driver. It should be a suitable initializer
192 for an array of structures, each containing two strings, without the
193 outermost pair of surrounding braces.
195 The first item in the pair is the name of the default. This must match
196 the code in @file{config.gcc} for the target. The second item is a spec
197 to apply if a default with this name was specified. The string
198 @samp{%(VALUE)} in the spec will be replaced by the value of the default
199 everywhere it occurs.
201 The driver will apply these specs to its own command line between loading
202 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
203 the same mechanism as @code{DRIVER_SELF_SPECS}.
205 Do not define this macro if it does not need to do anything.
209 A C string constant that tells the GCC driver program options to
210 pass to CPP@. It can also specify how to translate options you
211 give to GCC into options for GCC to pass to the CPP@.
213 Do not define this macro if it does not need to do anything.
216 @defmac CPLUSPLUS_CPP_SPEC
217 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
218 than C@. If you do not define this macro, then the value of
219 @code{CPP_SPEC} (if any) will be used instead.
223 A C string constant that tells the GCC driver program options to
224 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
226 It can also specify how to translate options you give to GCC into options
227 for GCC to pass to front ends.
229 Do not define this macro if it does not need to do anything.
233 A C string constant that tells the GCC driver program options to
234 pass to @code{cc1plus}. It can also specify how to translate options you
235 give to GCC into options for GCC to pass to the @code{cc1plus}.
237 Do not define this macro if it does not need to do anything.
238 Note that everything defined in CC1_SPEC is already passed to
239 @code{cc1plus} so there is no need to duplicate the contents of
240 CC1_SPEC in CC1PLUS_SPEC@.
244 A C string constant that tells the GCC driver program options to
245 pass to the assembler. It can also specify how to translate options
246 you give to GCC into options for GCC to pass to the assembler.
247 See the file @file{sun3.h} for an example of this.
249 Do not define this macro if it does not need to do anything.
252 @defmac ASM_FINAL_SPEC
253 A C string constant that tells the GCC driver program how to
254 run any programs which cleanup after the normal assembler.
255 Normally, this is not needed. See the file @file{mips.h} for
258 Do not define this macro if it does not need to do anything.
261 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
262 Define this macro, with no value, if the driver should give the assembler
263 an argument consisting of a single dash, @option{-}, to instruct it to
264 read from its standard input (which will be a pipe connected to the
265 output of the compiler proper). This argument is given after any
266 @option{-o} option specifying the name of the output file.
268 If you do not define this macro, the assembler is assumed to read its
269 standard input if given no non-option arguments. If your assembler
270 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
271 see @file{mips.h} for instance.
275 A C string constant that tells the GCC driver program options to
276 pass to the linker. It can also specify how to translate options you
277 give to GCC into options for GCC to pass to the linker.
279 Do not define this macro if it does not need to do anything.
283 Another C string constant used much like @code{LINK_SPEC}. The difference
284 between the two is that @code{LIB_SPEC} is used at the end of the
285 command given to the linker.
287 If this macro is not defined, a default is provided that
288 loads the standard C library from the usual place. See @file{gcc.c}.
292 Another C string constant that tells the GCC driver program
293 how and when to place a reference to @file{libgcc.a} into the
294 linker command line. This constant is placed both before and after
295 the value of @code{LIB_SPEC}.
297 If this macro is not defined, the GCC driver provides a default that
298 passes the string @option{-lgcc} to the linker.
301 @defmac REAL_LIBGCC_SPEC
302 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
303 @code{LIBGCC_SPEC} is not directly used by the driver program but is
304 instead modified to refer to different versions of @file{libgcc.a}
305 depending on the values of the command line flags @option{-static},
306 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
307 targets where these modifications are inappropriate, define
308 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
309 driver how to place a reference to @file{libgcc} on the link command
310 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
313 @defmac USE_LD_AS_NEEDED
314 A macro that controls the modifications to @code{LIBGCC_SPEC}
315 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
316 generated that uses --as-needed and the shared libgcc in place of the
317 static exception handler library, when linking without any of
318 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
322 If defined, this C string constant is added to @code{LINK_SPEC}.
323 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
324 the modifications to @code{LIBGCC_SPEC} mentioned in
325 @code{REAL_LIBGCC_SPEC}.
328 @defmac STARTFILE_SPEC
329 Another C string constant used much like @code{LINK_SPEC}. The
330 difference between the two is that @code{STARTFILE_SPEC} is used at
331 the very beginning of the command given to the linker.
333 If this macro is not defined, a default is provided that loads the
334 standard C startup file from the usual place. See @file{gcc.c}.
338 Another C string constant used much like @code{LINK_SPEC}. The
339 difference between the two is that @code{ENDFILE_SPEC} is used at
340 the very end of the command given to the linker.
342 Do not define this macro if it does not need to do anything.
345 @defmac THREAD_MODEL_SPEC
346 GCC @code{-v} will print the thread model GCC was configured to use.
347 However, this doesn't work on platforms that are multilibbed on thread
348 models, such as AIX 4.3. On such platforms, define
349 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
350 blanks that names one of the recognized thread models. @code{%*}, the
351 default value of this macro, will expand to the value of
352 @code{thread_file} set in @file{config.gcc}.
355 @defmac SYSROOT_SUFFIX_SPEC
356 Define this macro to add a suffix to the target sysroot when GCC is
357 configured with a sysroot. This will cause GCC to search for usr/lib,
358 et al, within sysroot+suffix.
361 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
362 Define this macro to add a headers_suffix to the target sysroot when
363 GCC is configured with a sysroot. This will cause GCC to pass the
364 updated sysroot+headers_suffix to CPP, causing it to search for
365 usr/include, et al, within sysroot+headers_suffix.
369 Define this macro to provide additional specifications to put in the
370 @file{specs} file that can be used in various specifications like
373 The definition should be an initializer for an array of structures,
374 containing a string constant, that defines the specification name, and a
375 string constant that provides the specification.
377 Do not define this macro if it does not need to do anything.
379 @code{EXTRA_SPECS} is useful when an architecture contains several
380 related targets, which have various @code{@dots{}_SPECS} which are similar
381 to each other, and the maintainer would like one central place to keep
384 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
385 define either @code{_CALL_SYSV} when the System V calling sequence is
386 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
389 The @file{config/rs6000/rs6000.h} target file defines:
392 #define EXTRA_SPECS \
393 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
395 #define CPP_SYS_DEFAULT ""
398 The @file{config/rs6000/sysv.h} target file defines:
402 "%@{posix: -D_POSIX_SOURCE @} \
403 %@{mcall-sysv: -D_CALL_SYSV @} \
404 %@{!mcall-sysv: %(cpp_sysv_default) @} \
405 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
407 #undef CPP_SYSV_DEFAULT
408 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
411 while the @file{config/rs6000/eabiaix.h} target file defines
412 @code{CPP_SYSV_DEFAULT} as:
415 #undef CPP_SYSV_DEFAULT
416 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
420 @defmac LINK_LIBGCC_SPECIAL_1
421 Define this macro if the driver program should find the library
422 @file{libgcc.a}. If you do not define this macro, the driver program will pass
423 the argument @option{-lgcc} to tell the linker to do the search.
426 @defmac LINK_GCC_C_SEQUENCE_SPEC
427 The sequence in which libgcc and libc are specified to the linker.
428 By default this is @code{%G %L %G}.
431 @defmac LINK_COMMAND_SPEC
432 A C string constant giving the complete command line need to execute the
433 linker. When you do this, you will need to update your port each time a
434 change is made to the link command line within @file{gcc.c}. Therefore,
435 define this macro only if you need to completely redefine the command
436 line for invoking the linker and there is no other way to accomplish
437 the effect you need. Overriding this macro may be avoidable by overriding
438 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
441 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
442 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
443 directories from linking commands. Do not give it a nonzero value if
444 removing duplicate search directories changes the linker's semantics.
447 @defmac MULTILIB_DEFAULTS
448 Define this macro as a C expression for the initializer of an array of
449 string to tell the driver program which options are defaults for this
450 target and thus do not need to be handled specially when using
451 @code{MULTILIB_OPTIONS}.
453 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
454 the target makefile fragment or if none of the options listed in
455 @code{MULTILIB_OPTIONS} are set by default.
456 @xref{Target Fragment}.
459 @defmac RELATIVE_PREFIX_NOT_LINKDIR
460 Define this macro to tell @command{gcc} that it should only translate
461 a @option{-B} prefix into a @option{-L} linker option if the prefix
462 indicates an absolute file name.
465 @defmac MD_EXEC_PREFIX
466 If defined, this macro is an additional prefix to try after
467 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
468 when the compiler is built as a cross
469 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
470 to the list of directories used to find the assembler in @file{configure.in}.
473 @defmac STANDARD_STARTFILE_PREFIX
474 Define this macro as a C string constant if you wish to override the
475 standard choice of @code{libdir} as the default prefix to
476 try when searching for startup files such as @file{crt0.o}.
477 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
478 is built as a cross compiler.
481 @defmac STANDARD_STARTFILE_PREFIX_1
482 Define this macro as a C string constant if you wish to override the
483 standard choice of @code{/lib} as a prefix to try after the default prefix
484 when searching for startup files such as @file{crt0.o}.
485 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
486 is built as a cross compiler.
489 @defmac STANDARD_STARTFILE_PREFIX_2
490 Define this macro as a C string constant if you wish to override the
491 standard choice of @code{/lib} as yet another prefix to try after the
492 default prefix when searching for startup files such as @file{crt0.o}.
493 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
494 is built as a cross compiler.
497 @defmac MD_STARTFILE_PREFIX
498 If defined, this macro supplies an additional prefix to try after the
499 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
500 compiler is built as a cross compiler.
503 @defmac MD_STARTFILE_PREFIX_1
504 If defined, this macro supplies yet another prefix to try after the
505 standard prefixes. It is not searched when the compiler is built as a
509 @defmac INIT_ENVIRONMENT
510 Define this macro as a C string constant if you wish to set environment
511 variables for programs called by the driver, such as the assembler and
512 loader. The driver passes the value of this macro to @code{putenv} to
513 initialize the necessary environment variables.
516 @defmac LOCAL_INCLUDE_DIR
517 Define this macro as a C string constant if you wish to override the
518 standard choice of @file{/usr/local/include} as the default prefix to
519 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
520 comes before @code{SYSTEM_INCLUDE_DIR} in the search order.
522 Cross compilers do not search either @file{/usr/local/include} or its
526 @defmac MODIFY_TARGET_NAME
527 Define this macro if you wish to define command-line switches that
528 modify the default target name.
530 For each switch, you can include a string to be appended to the first
531 part of the configuration name or a string to be deleted from the
532 configuration name, if present. The definition should be an initializer
533 for an array of structures. Each array element should have three
534 elements: the switch name (a string constant, including the initial
535 dash), one of the enumeration codes @code{ADD} or @code{DELETE} to
536 indicate whether the string should be inserted or deleted, and the string
537 to be inserted or deleted (a string constant).
539 For example, on a machine where @samp{64} at the end of the
540 configuration name denotes a 64-bit target and you want the @option{-32}
541 and @option{-64} switches to select between 32- and 64-bit targets, you would
545 #define MODIFY_TARGET_NAME \
546 @{ @{ "-32", DELETE, "64"@}, \
547 @{"-64", ADD, "64"@}@}
551 @defmac SYSTEM_INCLUDE_DIR
552 Define this macro as a C string constant if you wish to specify a
553 system-specific directory to search for header files before the standard
554 directory. @code{SYSTEM_INCLUDE_DIR} comes before
555 @code{STANDARD_INCLUDE_DIR} in the search order.
557 Cross compilers do not use this macro and do not search the directory
561 @defmac STANDARD_INCLUDE_DIR
562 Define this macro as a C string constant if you wish to override the
563 standard choice of @file{/usr/include} as the default prefix to
564 try when searching for header files.
566 Cross compilers ignore this macro and do not search either
567 @file{/usr/include} or its replacement.
570 @defmac STANDARD_INCLUDE_COMPONENT
571 The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}.
572 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
573 If you do not define this macro, no component is used.
576 @defmac INCLUDE_DEFAULTS
577 Define this macro if you wish to override the entire default search path
578 for include files. For a native compiler, the default search path
579 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
580 @code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and
581 @code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
582 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
583 and specify private search areas for GCC@. The directory
584 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
586 The definition should be an initializer for an array of structures.
587 Each array element should have four elements: the directory name (a
588 string constant), the component name (also a string constant), a flag
589 for C++-only directories,
590 and a flag showing that the includes in the directory don't need to be
591 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
592 the array with a null element.
594 The component name denotes what GNU package the include file is part of,
595 if any, in all uppercase letters. For example, it might be @samp{GCC}
596 or @samp{BINUTILS}. If the package is part of a vendor-supplied
597 operating system, code the component name as @samp{0}.
599 For example, here is the definition used for VAX/VMS:
602 #define INCLUDE_DEFAULTS \
604 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
605 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
606 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
613 Here is the order of prefixes tried for exec files:
617 Any prefixes specified by the user with @option{-B}.
620 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
621 is not set and the compiler has not been installed in the configure-time
622 @var{prefix}, the location in which the compiler has actually been installed.
625 The directories specified by the environment variable @code{COMPILER_PATH}.
628 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
629 in the configured-time @var{prefix}.
632 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
635 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
638 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
642 Here is the order of prefixes tried for startfiles:
646 Any prefixes specified by the user with @option{-B}.
649 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
650 value based on the installed toolchain location.
653 The directories specified by the environment variable @code{LIBRARY_PATH}
654 (or port-specific name; native only, cross compilers do not use this).
657 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
658 in the configured @var{prefix} or this is a native compiler.
661 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
664 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
668 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
669 native compiler, or we have a target system root.
672 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
673 native compiler, or we have a target system root.
676 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
677 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
678 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
681 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
682 compiler, or we have a target system root. The default for this macro is
686 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
687 compiler, or we have a target system root. The default for this macro is
691 @node Run-time Target
692 @section Run-time Target Specification
693 @cindex run-time target specification
694 @cindex predefined macros
695 @cindex target specifications
697 @c prevent bad page break with this line
698 Here are run-time target specifications.
700 @defmac TARGET_CPU_CPP_BUILTINS ()
701 This function-like macro expands to a block of code that defines
702 built-in preprocessor macros and assertions for the target CPU, using
703 the functions @code{builtin_define}, @code{builtin_define_std} and
704 @code{builtin_assert}. When the front end
705 calls this macro it provides a trailing semicolon, and since it has
706 finished command line option processing your code can use those
709 @code{builtin_assert} takes a string in the form you pass to the
710 command-line option @option{-A}, such as @code{cpu=mips}, and creates
711 the assertion. @code{builtin_define} takes a string in the form
712 accepted by option @option{-D} and unconditionally defines the macro.
714 @code{builtin_define_std} takes a string representing the name of an
715 object-like macro. If it doesn't lie in the user's namespace,
716 @code{builtin_define_std} defines it unconditionally. Otherwise, it
717 defines a version with two leading underscores, and another version
718 with two leading and trailing underscores, and defines the original
719 only if an ISO standard was not requested on the command line. For
720 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
721 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
722 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
723 defines only @code{_ABI64}.
725 You can also test for the C dialect being compiled. The variable
726 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
727 or @code{clk_objective_c}. Note that if we are preprocessing
728 assembler, this variable will be @code{clk_c} but the function-like
729 macro @code{preprocessing_asm_p()} will return true, so you might want
730 to check for that first. If you need to check for strict ANSI, the
731 variable @code{flag_iso} can be used. The function-like macro
732 @code{preprocessing_trad_p()} can be used to check for traditional
736 @defmac TARGET_OS_CPP_BUILTINS ()
737 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
738 and is used for the target operating system instead.
741 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
742 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
743 and is used for the target object format. @file{elfos.h} uses this
744 macro to define @code{__ELF__}, so you probably do not need to define
748 @deftypevar {extern int} target_flags
749 This variable is declared in @file{options.h}, which is included before
750 any target-specific headers.
753 @deftypevr {Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
754 This variable specifies the initial value of @code{target_flags}.
755 Its default setting is 0.
758 @cindex optional hardware or system features
759 @cindex features, optional, in system conventions
761 @deftypefn {Target Hook} bool TARGET_HANDLE_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
762 This hook is called whenever the user specifies one of the
763 target-specific options described by the @file{.opt} definition files
764 (@pxref{Options}). It has the opportunity to do some option-specific
765 processing and should return true if the option is valid. The default
766 definition does nothing but return true.
768 @var{code} specifies the @code{OPT_@var{name}} enumeration value
769 associated with the selected option; @var{name} is just a rendering of
770 the option name in which non-alphanumeric characters are replaced by
771 underscores. @var{arg} specifies the string argument and is null if
772 no argument was given. If the option is flagged as a @code{UInteger}
773 (@pxref{Option properties}), @var{value} is the numeric value of the
774 argument. Otherwise @var{value} is 1 if the positive form of the
775 option was used and 0 if the ``no-'' form was.
778 @deftypefn {Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
779 This target hook is called whenever the user specifies one of the
780 target-specific C language family options described by the @file{.opt}
781 definition files(@pxref{Options}). It has the opportunity to do some
782 option-specific processing and should return true if the option is
783 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
784 default definition does nothing but return false.
786 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
787 options. However, if processing an option requires routines that are
788 only available in the C (and related language) front ends, then you
789 should use @code{TARGET_HANDLE_C_OPTION} instead.
792 @defmac TARGET_VERSION
793 This macro is a C statement to print on @code{stderr} a string
794 describing the particular machine description choice. Every machine
795 description should define @code{TARGET_VERSION}. For example:
799 #define TARGET_VERSION \
800 fprintf (stderr, " (68k, Motorola syntax)");
802 #define TARGET_VERSION \
803 fprintf (stderr, " (68k, MIT syntax)");
808 @defmac OVERRIDE_OPTIONS
809 Sometimes certain combinations of command options do not make sense on
810 a particular target machine. You can define a macro
811 @code{OVERRIDE_OPTIONS} to take account of this. This macro, if
812 defined, is executed once just after all the command options have been
815 Don't use this macro to turn on various extra optimizations for
816 @option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for.
818 If you need to do something whenever the optimization level is
819 changed via the optimize attribute or pragma, see
820 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
823 @deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)
824 This target function is similar to the macro @code{OVERRIDE_OPTIONS}
825 but is called when the optimize level is changed via an attribute or
826 pragma or when it is reset at the end of the code affected by the
827 attribute or pragma. It is not called at the beginning of compilation
828 when @code{OVERRIDE_OPTIONS} is called so if you want to perform these
829 actions then, you should have @code{OVERRIDE_OPTIONS} call
830 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
833 @defmac C_COMMON_OVERRIDE_OPTIONS
834 This is similar to @code{OVERRIDE_OPTIONS} but is only used in the C
835 language frontends (C, Objective-C, C++, Objective-C++) and so can be
836 used to alter option flag variables which only exist in those
840 @defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size})
841 Some machines may desire to change what optimizations are performed for
842 various optimization levels. This macro, if defined, is executed once
843 just after the optimization level is determined and before the remainder
844 of the command options have been parsed. Values set in this macro are
845 used as the default values for the other command line options.
847 @var{level} is the optimization level specified; 2 if @option{-O2} is
848 specified, 1 if @option{-O} is specified, and 0 if neither is specified.
850 @var{size} is nonzero if @option{-Os} is specified and zero otherwise.
852 This macro is run once at program startup and when the optimization
853 options are changed via @code{#pragma GCC optimize} or by using the
854 @code{optimize} attribute.
856 @strong{Do not examine @code{write_symbols} in
857 this macro!} The debugging options are not supposed to alter the
861 @deftypefn {Target Hook} void TARGET_HELP (void)
862 This hook is called in response to the user invoking
863 @option{--target-help} on the command line. It gives the target a
864 chance to display extra information on the target specific command
865 line options found in its @file{.opt} file.
868 @defmac CAN_DEBUG_WITHOUT_FP
869 Define this macro if debugging can be performed even without a frame
870 pointer. If this macro is defined, GCC will turn on the
871 @option{-fomit-frame-pointer} option whenever @option{-O} is specified.
874 @node Per-Function Data
875 @section Defining data structures for per-function information.
876 @cindex per-function data
877 @cindex data structures
879 If the target needs to store information on a per-function basis, GCC
880 provides a macro and a couple of variables to allow this. Note, just
881 using statics to store the information is a bad idea, since GCC supports
882 nested functions, so you can be halfway through encoding one function
883 when another one comes along.
885 GCC defines a data structure called @code{struct function} which
886 contains all of the data specific to an individual function. This
887 structure contains a field called @code{machine} whose type is
888 @code{struct machine_function *}, which can be used by targets to point
889 to their own specific data.
891 If a target needs per-function specific data it should define the type
892 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
893 This macro should be used to initialize the function pointer
894 @code{init_machine_status}. This pointer is explained below.
896 One typical use of per-function, target specific data is to create an
897 RTX to hold the register containing the function's return address. This
898 RTX can then be used to implement the @code{__builtin_return_address}
899 function, for level 0.
901 Note---earlier implementations of GCC used a single data area to hold
902 all of the per-function information. Thus when processing of a nested
903 function began the old per-function data had to be pushed onto a
904 stack, and when the processing was finished, it had to be popped off the
905 stack. GCC used to provide function pointers called
906 @code{save_machine_status} and @code{restore_machine_status} to handle
907 the saving and restoring of the target specific information. Since the
908 single data area approach is no longer used, these pointers are no
911 @defmac INIT_EXPANDERS
912 Macro called to initialize any target specific information. This macro
913 is called once per function, before generation of any RTL has begun.
914 The intention of this macro is to allow the initialization of the
915 function pointer @code{init_machine_status}.
918 @deftypevar {void (*)(struct function *)} init_machine_status
919 If this function pointer is non-@code{NULL} it will be called once per
920 function, before function compilation starts, in order to allow the
921 target to perform any target specific initialization of the
922 @code{struct function} structure. It is intended that this would be
923 used to initialize the @code{machine} of that structure.
925 @code{struct machine_function} structures are expected to be freed by GC@.
926 Generally, any memory that they reference must be allocated by using
927 @code{ggc_alloc}, including the structure itself.
931 @section Storage Layout
932 @cindex storage layout
934 Note that the definitions of the macros in this table which are sizes or
935 alignments measured in bits do not need to be constant. They can be C
936 expressions that refer to static variables, such as the @code{target_flags}.
937 @xref{Run-time Target}.
939 @defmac BITS_BIG_ENDIAN
940 Define this macro to have the value 1 if the most significant bit in a
941 byte has the lowest number; otherwise define it to have the value zero.
942 This means that bit-field instructions count from the most significant
943 bit. If the machine has no bit-field instructions, then this must still
944 be defined, but it doesn't matter which value it is defined to. This
945 macro need not be a constant.
947 This macro does not affect the way structure fields are packed into
948 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
951 @defmac BYTES_BIG_ENDIAN
952 Define this macro to have the value 1 if the most significant byte in a
953 word has the lowest number. This macro need not be a constant.
956 @defmac WORDS_BIG_ENDIAN
957 Define this macro to have the value 1 if, in a multiword object, the
958 most significant word has the lowest number. This applies to both
959 memory locations and registers; GCC fundamentally assumes that the
960 order of words in memory is the same as the order in registers. This
961 macro need not be a constant.
964 @defmac LIBGCC2_WORDS_BIG_ENDIAN
965 Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a
966 constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be
967 used only when compiling @file{libgcc2.c}. Typically the value will be set
968 based on preprocessor defines.
971 @defmac FLOAT_WORDS_BIG_ENDIAN
972 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
973 @code{TFmode} floating point numbers are stored in memory with the word
974 containing the sign bit at the lowest address; otherwise define it to
975 have the value 0. This macro need not be a constant.
977 You need not define this macro if the ordering is the same as for
981 @defmac BITS_PER_UNIT
982 Define this macro to be the number of bits in an addressable storage
983 unit (byte). If you do not define this macro the default is 8.
986 @defmac BITS_PER_WORD
987 Number of bits in a word. If you do not define this macro, the default
988 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
991 @defmac MAX_BITS_PER_WORD
992 Maximum number of bits in a word. If this is undefined, the default is
993 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
994 largest value that @code{BITS_PER_WORD} can have at run-time.
997 @defmac UNITS_PER_WORD
998 Number of storage units in a word; normally the size of a general-purpose
999 register, a power of two from 1 or 8.
1002 @defmac MIN_UNITS_PER_WORD
1003 Minimum number of units in a word. If this is undefined, the default is
1004 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
1005 smallest value that @code{UNITS_PER_WORD} can have at run-time.
1008 @defmac UNITS_PER_SIMD_WORD (@var{mode})
1009 Number of units in the vectors that the vectorizer can produce for
1010 scalar mode @var{mode}. The default is equal to @code{UNITS_PER_WORD},
1011 because the vectorizer can do some transformations even in absence of
1012 specialized @acronym{SIMD} hardware.
1015 @defmac POINTER_SIZE
1016 Width of a pointer, in bits. You must specify a value no wider than the
1017 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
1018 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
1019 a value the default is @code{BITS_PER_WORD}.
1022 @defmac POINTERS_EXTEND_UNSIGNED
1023 A C expression that determines how pointers should be extended from
1024 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
1025 greater than zero if pointers should be zero-extended, zero if they
1026 should be sign-extended, and negative if some other sort of conversion
1027 is needed. In the last case, the extension is done by the target's
1028 @code{ptr_extend} instruction.
1030 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
1031 and @code{word_mode} are all the same width.
1034 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
1035 A macro to update @var{m} and @var{unsignedp} when an object whose type
1036 is @var{type} and which has the specified mode and signedness is to be
1037 stored in a register. This macro is only called when @var{type} is a
1040 On most RISC machines, which only have operations that operate on a full
1041 register, define this macro to set @var{m} to @code{word_mode} if
1042 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
1043 cases, only integer modes should be widened because wider-precision
1044 floating-point operations are usually more expensive than their narrower
1047 For most machines, the macro definition does not change @var{unsignedp}.
1048 However, some machines, have instructions that preferentially handle
1049 either signed or unsigned quantities of certain modes. For example, on
1050 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
1051 sign-extend the result to 64 bits. On such machines, set
1052 @var{unsignedp} according to which kind of extension is more efficient.
1054 Do not define this macro if it would never modify @var{m}.
1057 @deftypefn {Target Hook} {enum machine_mode} TARGET_PROMOTE_FUNCTION_MODE (const_tree @var{type}, enum machine_mode @var{mode}, int *@var{punsignedp}, const_tree @var{funtype}, int @var{for_return})
1058 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
1059 function return values. The target hook should return the new mode
1060 and possibly change @code{*@var{punsignedp}} if the promotion should
1061 change signedness. This function is called only for scalar @emph{or
1064 @var{for_return} allows to distinguish the promotion of arguments and
1065 return values. If it is @code{1}, a return value is being promoted and
1066 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
1067 If it is @code{2}, the returned mode should be that of the register in
1068 which an incoming parameter is copied, or the outgoing result is computed;
1069 then the hook should return the same mode as @code{promote_mode}, though
1070 the signedness may be different.
1072 The default is to not promote arguments and return values. You can
1073 also define the hook to @code{default_promote_function_mode_always_promote}
1074 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
1077 @defmac PARM_BOUNDARY
1078 Normal alignment required for function parameters on the stack, in
1079 bits. All stack parameters receive at least this much alignment
1080 regardless of data type. On most machines, this is the same as the
1084 @defmac STACK_BOUNDARY
1085 Define this macro to the minimum alignment enforced by hardware for the
1086 stack pointer on this machine. The definition is a C expression for the
1087 desired alignment (measured in bits). This value is used as a default
1088 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1089 this should be the same as @code{PARM_BOUNDARY}.
1092 @defmac PREFERRED_STACK_BOUNDARY
1093 Define this macro if you wish to preserve a certain alignment for the
1094 stack pointer, greater than what the hardware enforces. The definition
1095 is a C expression for the desired alignment (measured in bits). This
1096 macro must evaluate to a value equal to or larger than
1097 @code{STACK_BOUNDARY}.
1100 @defmac INCOMING_STACK_BOUNDARY
1101 Define this macro if the incoming stack boundary may be different
1102 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
1103 to a value equal to or larger than @code{STACK_BOUNDARY}.
1106 @defmac FUNCTION_BOUNDARY
1107 Alignment required for a function entry point, in bits.
1110 @defmac BIGGEST_ALIGNMENT
1111 Biggest alignment that any data type can require on this machine, in
1112 bits. Note that this is not the biggest alignment that is supported,
1113 just the biggest alignment that, when violated, may cause a fault.
1116 @defmac MALLOC_ABI_ALIGNMENT
1117 Alignment, in bits, a C conformant malloc implementation has to
1118 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1121 @defmac ATTRIBUTE_ALIGNED_VALUE
1122 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1123 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1126 @defmac MINIMUM_ATOMIC_ALIGNMENT
1127 If defined, the smallest alignment, in bits, that can be given to an
1128 object that can be referenced in one operation, without disturbing any
1129 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1130 on machines that don't have byte or half-word store operations.
1133 @defmac BIGGEST_FIELD_ALIGNMENT
1134 Biggest alignment that any structure or union field can require on this
1135 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1136 structure and union fields only, unless the field alignment has been set
1137 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1140 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1141 An expression for the alignment of a structure field @var{field} if the
1142 alignment computed in the usual way (including applying of
1143 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1144 alignment) is @var{computed}. It overrides alignment only if the
1145 field alignment has not been set by the
1146 @code{__attribute__ ((aligned (@var{n})))} construct.
1149 @defmac MAX_STACK_ALIGNMENT
1150 Biggest stack alignment guaranteed by the backend. Use this macro
1151 to specify the maximum alignment of a variable on stack.
1153 If not defined, the default value is @code{STACK_BOUNDARY}.
1155 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1156 @c But the fix for PR 32893 indicates that we can only guarantee
1157 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1158 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1161 @defmac MAX_OFILE_ALIGNMENT
1162 Biggest alignment supported by the object file format of this machine.
1163 Use this macro to limit the alignment which can be specified using the
1164 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1165 the default value is @code{BIGGEST_ALIGNMENT}.
1167 On systems that use ELF, the default (in @file{config/elfos.h}) is
1168 the largest supported 32-bit ELF section alignment representable on
1169 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1170 On 32-bit ELF the largest supported section alignment in bits is
1171 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1174 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1175 If defined, a C expression to compute the alignment for a variable in
1176 the static store. @var{type} is the data type, and @var{basic-align} is
1177 the alignment that the object would ordinarily have. The value of this
1178 macro is used instead of that alignment to align the object.
1180 If this macro is not defined, then @var{basic-align} is used.
1183 One use of this macro is to increase alignment of medium-size data to
1184 make it all fit in fewer cache lines. Another is to cause character
1185 arrays to be word-aligned so that @code{strcpy} calls that copy
1186 constants to character arrays can be done inline.
1189 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1190 If defined, a C expression to compute the alignment given to a constant
1191 that is being placed in memory. @var{constant} is the constant and
1192 @var{basic-align} is the alignment that the object would ordinarily
1193 have. The value of this macro is used instead of that alignment to
1196 If this macro is not defined, then @var{basic-align} is used.
1198 The typical use of this macro is to increase alignment for string
1199 constants to be word aligned so that @code{strcpy} calls that copy
1200 constants can be done inline.
1203 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1204 If defined, a C expression to compute the alignment for a variable in
1205 the local store. @var{type} is the data type, and @var{basic-align} is
1206 the alignment that the object would ordinarily have. The value of this
1207 macro is used instead of that alignment to align the object.
1209 If this macro is not defined, then @var{basic-align} is used.
1211 One use of this macro is to increase alignment of medium-size data to
1212 make it all fit in fewer cache lines.
1215 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1216 If defined, a C expression to compute the alignment for stack slot.
1217 @var{type} is the data type, @var{mode} is the widest mode available,
1218 and @var{basic-align} is the alignment that the slot would ordinarily
1219 have. The value of this macro is used instead of that alignment to
1222 If this macro is not defined, then @var{basic-align} is used when
1223 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1226 This macro is to set alignment of stack slot to the maximum alignment
1227 of all possible modes which the slot may have.
1230 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1231 If defined, a C expression to compute the alignment for a local
1232 variable @var{decl}.
1234 If this macro is not defined, then
1235 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1238 One use of this macro is to increase alignment of medium-size data to
1239 make it all fit in fewer cache lines.
1242 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1243 If defined, a C expression to compute the minimum required alignment
1244 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1245 @var{mode}, assuming normal alignment @var{align}.
1247 If this macro is not defined, then @var{align} will be used.
1250 @defmac EMPTY_FIELD_BOUNDARY
1251 Alignment in bits to be given to a structure bit-field that follows an
1252 empty field such as @code{int : 0;}.
1254 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1257 @defmac STRUCTURE_SIZE_BOUNDARY
1258 Number of bits which any structure or union's size must be a multiple of.
1259 Each structure or union's size is rounded up to a multiple of this.
1261 If you do not define this macro, the default is the same as
1262 @code{BITS_PER_UNIT}.
1265 @defmac STRICT_ALIGNMENT
1266 Define this macro to be the value 1 if instructions will fail to work
1267 if given data not on the nominal alignment. If instructions will merely
1268 go slower in that case, define this macro as 0.
1271 @defmac PCC_BITFIELD_TYPE_MATTERS
1272 Define this if you wish to imitate the way many other C compilers handle
1273 alignment of bit-fields and the structures that contain them.
1275 The behavior is that the type written for a named bit-field (@code{int},
1276 @code{short}, or other integer type) imposes an alignment for the entire
1277 structure, as if the structure really did contain an ordinary field of
1278 that type. In addition, the bit-field is placed within the structure so
1279 that it would fit within such a field, not crossing a boundary for it.
1281 Thus, on most machines, a named bit-field whose type is written as
1282 @code{int} would not cross a four-byte boundary, and would force
1283 four-byte alignment for the whole structure. (The alignment used may
1284 not be four bytes; it is controlled by the other alignment parameters.)
1286 An unnamed bit-field will not affect the alignment of the containing
1289 If the macro is defined, its definition should be a C expression;
1290 a nonzero value for the expression enables this behavior.
1292 Note that if this macro is not defined, or its value is zero, some
1293 bit-fields may cross more than one alignment boundary. The compiler can
1294 support such references if there are @samp{insv}, @samp{extv}, and
1295 @samp{extzv} insns that can directly reference memory.
1297 The other known way of making bit-fields work is to define
1298 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1299 Then every structure can be accessed with fullwords.
1301 Unless the machine has bit-field instructions or you define
1302 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1303 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1305 If your aim is to make GCC use the same conventions for laying out
1306 bit-fields as are used by another compiler, here is how to investigate
1307 what the other compiler does. Compile and run this program:
1326 printf ("Size of foo1 is %d\n",
1327 sizeof (struct foo1));
1328 printf ("Size of foo2 is %d\n",
1329 sizeof (struct foo2));
1334 If this prints 2 and 5, then the compiler's behavior is what you would
1335 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1338 @defmac BITFIELD_NBYTES_LIMITED
1339 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1340 to aligning a bit-field within the structure.
1343 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void)
1344 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1345 whether unnamed bitfields affect the alignment of the containing
1346 structure. The hook should return true if the structure should inherit
1347 the alignment requirements of an unnamed bitfield's type.
1350 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void)
1351 This target hook should return @code{true} if accesses to volatile bitfields
1352 should use the narrowest mode possible. It should return @code{false} if
1353 these accesses should use the bitfield container type.
1355 The default is @code{!TARGET_STRICT_ALIGN}.
1358 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1359 Return 1 if a structure or array containing @var{field} should be accessed using
1362 If @var{field} is the only field in the structure, @var{mode} is its
1363 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1364 case where structures of one field would require the structure's mode to
1365 retain the field's mode.
1367 Normally, this is not needed.
1370 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1371 Define this macro as an expression for the alignment of a type (given
1372 by @var{type} as a tree node) if the alignment computed in the usual
1373 way is @var{computed} and the alignment explicitly specified was
1376 The default is to use @var{specified} if it is larger; otherwise, use
1377 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1380 @defmac MAX_FIXED_MODE_SIZE
1381 An integer expression for the size in bits of the largest integer
1382 machine mode that should actually be used. All integer machine modes of
1383 this size or smaller can be used for structures and unions with the
1384 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1385 (DImode)} is assumed.
1388 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1389 If defined, an expression of type @code{enum machine_mode} that
1390 specifies the mode of the save area operand of a
1391 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1392 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1393 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1394 having its mode specified.
1396 You need not define this macro if it always returns @code{Pmode}. You
1397 would most commonly define this macro if the
1398 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1402 @defmac STACK_SIZE_MODE
1403 If defined, an expression of type @code{enum machine_mode} that
1404 specifies the mode of the size increment operand of an
1405 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1407 You need not define this macro if it always returns @code{word_mode}.
1408 You would most commonly define this macro if the @code{allocate_stack}
1409 pattern needs to support both a 32- and a 64-bit mode.
1412 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_CMP_RETURN_MODE (void)
1413 This target hook should return the mode to be used for the return value
1414 of compare instructions expanded to libgcc calls. If not defined
1415 @code{word_mode} is returned which is the right choice for a majority of
1419 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_SHIFT_COUNT_MODE (void)
1420 This target hook should return the mode to be used for the shift count operand
1421 of shift instructions expanded to libgcc calls. If not defined
1422 @code{word_mode} is returned which is the right choice for a majority of
1426 @deftypefn {Target Hook} {enum machine_mode} TARGET_UNWIND_WORD_MODE (void)
1427 Return machine mode to be used for @code{_Unwind_Word} type.
1428 The default is to use @code{word_mode}.
1431 @defmac ROUND_TOWARDS_ZERO
1432 If defined, this macro should be true if the prevailing rounding
1433 mode is towards zero.
1435 Defining this macro only affects the way @file{libgcc.a} emulates
1436 floating-point arithmetic.
1438 Not defining this macro is equivalent to returning zero.
1441 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1442 This macro should return true if floats with @var{size}
1443 bits do not have a NaN or infinity representation, but use the largest
1444 exponent for normal numbers instead.
1446 Defining this macro only affects the way @file{libgcc.a} emulates
1447 floating-point arithmetic.
1449 The default definition of this macro returns false for all sizes.
1452 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type})
1453 This target hook returns @code{true} if bit-fields in the given
1454 @var{record_type} are to be laid out following the rules of Microsoft
1455 Visual C/C++, namely: (i) a bit-field won't share the same storage
1456 unit with the previous bit-field if their underlying types have
1457 different sizes, and the bit-field will be aligned to the highest
1458 alignment of the underlying types of itself and of the previous
1459 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1460 the whole enclosing structure, even if it is unnamed; except that
1461 (iii) a zero-sized bit-field will be disregarded unless it follows
1462 another bit-field of nonzero size. If this hook returns @code{true},
1463 other macros that control bit-field layout are ignored.
1465 When a bit-field is inserted into a packed record, the whole size
1466 of the underlying type is used by one or more same-size adjacent
1467 bit-fields (that is, if its long:3, 32 bits is used in the record,
1468 and any additional adjacent long bit-fields are packed into the same
1469 chunk of 32 bits. However, if the size changes, a new field of that
1470 size is allocated). In an unpacked record, this is the same as using
1471 alignment, but not equivalent when packing.
1473 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1474 the latter will take precedence. If @samp{__attribute__((packed))} is
1475 used on a single field when MS bit-fields are in use, it will take
1476 precedence for that field, but the alignment of the rest of the structure
1477 may affect its placement.
1480 @deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1481 Returns true if the target supports decimal floating point.
1484 @deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void)
1485 Returns true if the target supports fixed-point arithmetic.
1488 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1489 This hook is called just before expansion into rtl, allowing the target
1490 to perform additional initializations or analysis before the expansion.
1491 For example, the rs6000 port uses it to allocate a scratch stack slot
1492 for use in copying SDmode values between memory and floating point
1493 registers whenever the function being expanded has any SDmode
1497 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1498 This hook allows the backend to perform additional instantiations on rtl
1499 that are not actually in any insns yet, but will be later.
1502 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type})
1503 If your target defines any fundamental types, or any types your target
1504 uses should be mangled differently from the default, define this hook
1505 to return the appropriate encoding for these types as part of a C++
1506 mangled name. The @var{type} argument is the tree structure representing
1507 the type to be mangled. The hook may be applied to trees which are
1508 not target-specific fundamental types; it should return @code{NULL}
1509 for all such types, as well as arguments it does not recognize. If the
1510 return value is not @code{NULL}, it must point to a statically-allocated
1513 Target-specific fundamental types might be new fundamental types or
1514 qualified versions of ordinary fundamental types. Encode new
1515 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1516 is the name used for the type in source code, and @var{n} is the
1517 length of @var{name} in decimal. Encode qualified versions of
1518 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1519 @var{name} is the name used for the type qualifier in source code,
1520 @var{n} is the length of @var{name} as above, and @var{code} is the
1521 code used to represent the unqualified version of this type. (See
1522 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1523 codes.) In both cases the spaces are for clarity; do not include any
1524 spaces in your string.
1526 This hook is applied to types prior to typedef resolution. If the mangled
1527 name for a particular type depends only on that type's main variant, you
1528 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1531 The default version of this hook always returns @code{NULL}, which is
1532 appropriate for a target that does not define any new fundamental
1537 @section Layout of Source Language Data Types
1539 These macros define the sizes and other characteristics of the standard
1540 basic data types used in programs being compiled. Unlike the macros in
1541 the previous section, these apply to specific features of C and related
1542 languages, rather than to fundamental aspects of storage layout.
1544 @defmac INT_TYPE_SIZE
1545 A C expression for the size in bits of the type @code{int} on the
1546 target machine. If you don't define this, the default is one word.
1549 @defmac SHORT_TYPE_SIZE
1550 A C expression for the size in bits of the type @code{short} on the
1551 target machine. If you don't define this, the default is half a word.
1552 (If this would be less than one storage unit, it is rounded up to one
1556 @defmac LONG_TYPE_SIZE
1557 A C expression for the size in bits of the type @code{long} on the
1558 target machine. If you don't define this, the default is one word.
1561 @defmac ADA_LONG_TYPE_SIZE
1562 On some machines, the size used for the Ada equivalent of the type
1563 @code{long} by a native Ada compiler differs from that used by C@. In
1564 that situation, define this macro to be a C expression to be used for
1565 the size of that type. If you don't define this, the default is the
1566 value of @code{LONG_TYPE_SIZE}.
1569 @defmac LONG_LONG_TYPE_SIZE
1570 A C expression for the size in bits of the type @code{long long} on the
1571 target machine. If you don't define this, the default is two
1572 words. If you want to support GNU Ada on your machine, the value of this
1573 macro must be at least 64.
1576 @defmac CHAR_TYPE_SIZE
1577 A C expression for the size in bits of the type @code{char} on the
1578 target machine. If you don't define this, the default is
1579 @code{BITS_PER_UNIT}.
1582 @defmac BOOL_TYPE_SIZE
1583 A C expression for the size in bits of the C++ type @code{bool} and
1584 C99 type @code{_Bool} on the target machine. If you don't define
1585 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1588 @defmac FLOAT_TYPE_SIZE
1589 A C expression for the size in bits of the type @code{float} on the
1590 target machine. If you don't define this, the default is one word.
1593 @defmac DOUBLE_TYPE_SIZE
1594 A C expression for the size in bits of the type @code{double} on the
1595 target machine. If you don't define this, the default is two
1599 @defmac LONG_DOUBLE_TYPE_SIZE
1600 A C expression for the size in bits of the type @code{long double} on
1601 the target machine. If you don't define this, the default is two
1605 @defmac SHORT_FRACT_TYPE_SIZE
1606 A C expression for the size in bits of the type @code{short _Fract} on
1607 the target machine. If you don't define this, the default is
1608 @code{BITS_PER_UNIT}.
1611 @defmac FRACT_TYPE_SIZE
1612 A C expression for the size in bits of the type @code{_Fract} on
1613 the target machine. If you don't define this, the default is
1614 @code{BITS_PER_UNIT * 2}.
1617 @defmac LONG_FRACT_TYPE_SIZE
1618 A C expression for the size in bits of the type @code{long _Fract} on
1619 the target machine. If you don't define this, the default is
1620 @code{BITS_PER_UNIT * 4}.
1623 @defmac LONG_LONG_FRACT_TYPE_SIZE
1624 A C expression for the size in bits of the type @code{long long _Fract} on
1625 the target machine. If you don't define this, the default is
1626 @code{BITS_PER_UNIT * 8}.
1629 @defmac SHORT_ACCUM_TYPE_SIZE
1630 A C expression for the size in bits of the type @code{short _Accum} on
1631 the target machine. If you don't define this, the default is
1632 @code{BITS_PER_UNIT * 2}.
1635 @defmac ACCUM_TYPE_SIZE
1636 A C expression for the size in bits of the type @code{_Accum} on
1637 the target machine. If you don't define this, the default is
1638 @code{BITS_PER_UNIT * 4}.
1641 @defmac LONG_ACCUM_TYPE_SIZE
1642 A C expression for the size in bits of the type @code{long _Accum} on
1643 the target machine. If you don't define this, the default is
1644 @code{BITS_PER_UNIT * 8}.
1647 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1648 A C expression for the size in bits of the type @code{long long _Accum} on
1649 the target machine. If you don't define this, the default is
1650 @code{BITS_PER_UNIT * 16}.
1653 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1654 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1655 if you want routines in @file{libgcc2.a} for a size other than
1656 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1657 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1660 @defmac LIBGCC2_HAS_DF_MODE
1661 Define this macro if neither @code{LIBGCC2_DOUBLE_TYPE_SIZE} nor
1662 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1663 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1664 anyway. If you don't define this and either @code{LIBGCC2_DOUBLE_TYPE_SIZE}
1665 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1669 @defmac LIBGCC2_HAS_XF_MODE
1670 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1671 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1672 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1673 is 80 then the default is 1, otherwise it is 0.
1676 @defmac LIBGCC2_HAS_TF_MODE
1677 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1678 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1679 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1680 is 128 then the default is 1, otherwise it is 0.
1687 Define these macros to be the size in bits of the mantissa of
1688 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1689 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1690 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1691 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1692 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1693 @code{LIBGCC2_DOUBLE_TYPE_SIZE} or
1694 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1697 @defmac TARGET_FLT_EVAL_METHOD
1698 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1699 assuming, if applicable, that the floating-point control word is in its
1700 default state. If you do not define this macro the value of
1701 @code{FLT_EVAL_METHOD} will be zero.
1704 @defmac WIDEST_HARDWARE_FP_SIZE
1705 A C expression for the size in bits of the widest floating-point format
1706 supported by the hardware. If you define this macro, you must specify a
1707 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1708 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1712 @defmac DEFAULT_SIGNED_CHAR
1713 An expression whose value is 1 or 0, according to whether the type
1714 @code{char} should be signed or unsigned by default. The user can
1715 always override this default with the options @option{-fsigned-char}
1716 and @option{-funsigned-char}.
1719 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1720 This target hook should return true if the compiler should give an
1721 @code{enum} type only as many bytes as it takes to represent the range
1722 of possible values of that type. It should return false if all
1723 @code{enum} types should be allocated like @code{int}.
1725 The default is to return false.
1729 A C expression for a string describing the name of the data type to use
1730 for size values. The typedef name @code{size_t} is defined using the
1731 contents of the string.
1733 The string can contain more than one keyword. If so, separate them with
1734 spaces, and write first any length keyword, then @code{unsigned} if
1735 appropriate, and finally @code{int}. The string must exactly match one
1736 of the data type names defined in the function
1737 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1738 omit @code{int} or change the order---that would cause the compiler to
1741 If you don't define this macro, the default is @code{"long unsigned
1745 @defmac PTRDIFF_TYPE
1746 A C expression for a string describing the name of the data type to use
1747 for the result of subtracting two pointers. The typedef name
1748 @code{ptrdiff_t} is defined using the contents of the string. See
1749 @code{SIZE_TYPE} above for more information.
1751 If you don't define this macro, the default is @code{"long int"}.
1755 A C expression for a string describing the name of the data type to use
1756 for wide characters. The typedef name @code{wchar_t} is defined using
1757 the contents of the string. See @code{SIZE_TYPE} above for more
1760 If you don't define this macro, the default is @code{"int"}.
1763 @defmac WCHAR_TYPE_SIZE
1764 A C expression for the size in bits of the data type for wide
1765 characters. This is used in @code{cpp}, which cannot make use of
1770 A C expression for a string describing the name of the data type to
1771 use for wide characters passed to @code{printf} and returned from
1772 @code{getwc}. The typedef name @code{wint_t} is defined using the
1773 contents of the string. See @code{SIZE_TYPE} above for more
1776 If you don't define this macro, the default is @code{"unsigned int"}.
1780 A C expression for a string describing the name of the data type that
1781 can represent any value of any standard or extended signed integer type.
1782 The typedef name @code{intmax_t} is defined using the contents of the
1783 string. See @code{SIZE_TYPE} above for more information.
1785 If you don't define this macro, the default is the first of
1786 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1787 much precision as @code{long long int}.
1790 @defmac UINTMAX_TYPE
1791 A C expression for a string describing the name of the data type that
1792 can represent any value of any standard or extended unsigned integer
1793 type. The typedef name @code{uintmax_t} is defined using the contents
1794 of the string. See @code{SIZE_TYPE} above for more information.
1796 If you don't define this macro, the default is the first of
1797 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1798 unsigned int"} that has as much precision as @code{long long unsigned
1802 @defmac SIG_ATOMIC_TYPE
1808 @defmacx UINT16_TYPE
1809 @defmacx UINT32_TYPE
1810 @defmacx UINT64_TYPE
1811 @defmacx INT_LEAST8_TYPE
1812 @defmacx INT_LEAST16_TYPE
1813 @defmacx INT_LEAST32_TYPE
1814 @defmacx INT_LEAST64_TYPE
1815 @defmacx UINT_LEAST8_TYPE
1816 @defmacx UINT_LEAST16_TYPE
1817 @defmacx UINT_LEAST32_TYPE
1818 @defmacx UINT_LEAST64_TYPE
1819 @defmacx INT_FAST8_TYPE
1820 @defmacx INT_FAST16_TYPE
1821 @defmacx INT_FAST32_TYPE
1822 @defmacx INT_FAST64_TYPE
1823 @defmacx UINT_FAST8_TYPE
1824 @defmacx UINT_FAST16_TYPE
1825 @defmacx UINT_FAST32_TYPE
1826 @defmacx UINT_FAST64_TYPE
1827 @defmacx INTPTR_TYPE
1828 @defmacx UINTPTR_TYPE
1829 C expressions for the standard types @code{sig_atomic_t},
1830 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1831 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1832 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1833 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1834 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1835 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1836 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1837 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1838 @code{SIZE_TYPE} above for more information.
1840 If any of these macros evaluates to a null pointer, the corresponding
1841 type is not supported; if GCC is configured to provide
1842 @code{<stdint.h>} in such a case, the header provided may not conform
1843 to C99, depending on the type in question. The defaults for all of
1844 these macros are null pointers.
1847 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1848 The C++ compiler represents a pointer-to-member-function with a struct
1855 ptrdiff_t vtable_index;
1862 The C++ compiler must use one bit to indicate whether the function that
1863 will be called through a pointer-to-member-function is virtual.
1864 Normally, we assume that the low-order bit of a function pointer must
1865 always be zero. Then, by ensuring that the vtable_index is odd, we can
1866 distinguish which variant of the union is in use. But, on some
1867 platforms function pointers can be odd, and so this doesn't work. In
1868 that case, we use the low-order bit of the @code{delta} field, and shift
1869 the remainder of the @code{delta} field to the left.
1871 GCC will automatically make the right selection about where to store
1872 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1873 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1874 set such that functions always start at even addresses, but the lowest
1875 bit of pointers to functions indicate whether the function at that
1876 address is in ARM or Thumb mode. If this is the case of your
1877 architecture, you should define this macro to
1878 @code{ptrmemfunc_vbit_in_delta}.
1880 In general, you should not have to define this macro. On architectures
1881 in which function addresses are always even, according to
1882 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1883 @code{ptrmemfunc_vbit_in_pfn}.
1886 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1887 Normally, the C++ compiler uses function pointers in vtables. This
1888 macro allows the target to change to use ``function descriptors''
1889 instead. Function descriptors are found on targets for whom a
1890 function pointer is actually a small data structure. Normally the
1891 data structure consists of the actual code address plus a data
1892 pointer to which the function's data is relative.
1894 If vtables are used, the value of this macro should be the number
1895 of words that the function descriptor occupies.
1898 @defmac TARGET_VTABLE_ENTRY_ALIGN
1899 By default, the vtable entries are void pointers, the so the alignment
1900 is the same as pointer alignment. The value of this macro specifies
1901 the alignment of the vtable entry in bits. It should be defined only
1902 when special alignment is necessary. */
1905 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1906 There are a few non-descriptor entries in the vtable at offsets below
1907 zero. If these entries must be padded (say, to preserve the alignment
1908 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1909 of words in each data entry.
1913 @section Register Usage
1914 @cindex register usage
1916 This section explains how to describe what registers the target machine
1917 has, and how (in general) they can be used.
1919 The description of which registers a specific instruction can use is
1920 done with register classes; see @ref{Register Classes}. For information
1921 on using registers to access a stack frame, see @ref{Frame Registers}.
1922 For passing values in registers, see @ref{Register Arguments}.
1923 For returning values in registers, see @ref{Scalar Return}.
1926 * Register Basics:: Number and kinds of registers.
1927 * Allocation Order:: Order in which registers are allocated.
1928 * Values in Registers:: What kinds of values each reg can hold.
1929 * Leaf Functions:: Renumbering registers for leaf functions.
1930 * Stack Registers:: Handling a register stack such as 80387.
1933 @node Register Basics
1934 @subsection Basic Characteristics of Registers
1936 @c prevent bad page break with this line
1937 Registers have various characteristics.
1939 @defmac FIRST_PSEUDO_REGISTER
1940 Number of hardware registers known to the compiler. They receive
1941 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1942 pseudo register's number really is assigned the number
1943 @code{FIRST_PSEUDO_REGISTER}.
1946 @defmac FIXED_REGISTERS
1947 @cindex fixed register
1948 An initializer that says which registers are used for fixed purposes
1949 all throughout the compiled code and are therefore not available for
1950 general allocation. These would include the stack pointer, the frame
1951 pointer (except on machines where that can be used as a general
1952 register when no frame pointer is needed), the program counter on
1953 machines where that is considered one of the addressable registers,
1954 and any other numbered register with a standard use.
1956 This information is expressed as a sequence of numbers, separated by
1957 commas and surrounded by braces. The @var{n}th number is 1 if
1958 register @var{n} is fixed, 0 otherwise.
1960 The table initialized from this macro, and the table initialized by
1961 the following one, may be overridden at run time either automatically,
1962 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1963 the user with the command options @option{-ffixed-@var{reg}},
1964 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1967 @defmac CALL_USED_REGISTERS
1968 @cindex call-used register
1969 @cindex call-clobbered register
1970 @cindex call-saved register
1971 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1972 clobbered (in general) by function calls as well as for fixed
1973 registers. This macro therefore identifies the registers that are not
1974 available for general allocation of values that must live across
1977 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1978 automatically saves it on function entry and restores it on function
1979 exit, if the register is used within the function.
1982 @defmac CALL_REALLY_USED_REGISTERS
1983 @cindex call-used register
1984 @cindex call-clobbered register
1985 @cindex call-saved register
1986 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1987 that the entire set of @code{FIXED_REGISTERS} be included.
1988 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1989 This macro is optional. If not specified, it defaults to the value
1990 of @code{CALL_USED_REGISTERS}.
1993 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1994 @cindex call-used register
1995 @cindex call-clobbered register
1996 @cindex call-saved register
1997 A C expression that is nonzero if it is not permissible to store a
1998 value of mode @var{mode} in hard register number @var{regno} across a
1999 call without some part of it being clobbered. For most machines this
2000 macro need not be defined. It is only required for machines that do not
2001 preserve the entire contents of a register across a call.
2005 @findex call_used_regs
2008 @findex reg_class_contents
2009 @defmac CONDITIONAL_REGISTER_USAGE
2010 Zero or more C statements that may conditionally modify five variables
2011 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
2012 @code{reg_names}, and @code{reg_class_contents}, to take into account
2013 any dependence of these register sets on target flags. The first three
2014 of these are of type @code{char []} (interpreted as Boolean vectors).
2015 @code{global_regs} is a @code{const char *[]}, and
2016 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
2017 called, @code{fixed_regs}, @code{call_used_regs},
2018 @code{reg_class_contents}, and @code{reg_names} have been initialized
2019 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
2020 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
2021 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
2022 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
2023 command options have been applied.
2025 You need not define this macro if it has no work to do.
2027 @cindex disabling certain registers
2028 @cindex controlling register usage
2029 If the usage of an entire class of registers depends on the target
2030 flags, you may indicate this to GCC by using this macro to modify
2031 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
2032 registers in the classes which should not be used by GCC@. Also define
2033 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
2034 to return @code{NO_REGS} if it
2035 is called with a letter for a class that shouldn't be used.
2037 (However, if this class is not included in @code{GENERAL_REGS} and all
2038 of the insn patterns whose constraints permit this class are
2039 controlled by target switches, then GCC will automatically avoid using
2040 these registers when the target switches are opposed to them.)
2043 @defmac INCOMING_REGNO (@var{out})
2044 Define this macro if the target machine has register windows. This C
2045 expression returns the register number as seen by the called function
2046 corresponding to the register number @var{out} as seen by the calling
2047 function. Return @var{out} if register number @var{out} is not an
2051 @defmac OUTGOING_REGNO (@var{in})
2052 Define this macro if the target machine has register windows. This C
2053 expression returns the register number as seen by the calling function
2054 corresponding to the register number @var{in} as seen by the called
2055 function. Return @var{in} if register number @var{in} is not an inbound
2059 @defmac LOCAL_REGNO (@var{regno})
2060 Define this macro if the target machine has register windows. This C
2061 expression returns true if the register is call-saved but is in the
2062 register window. Unlike most call-saved registers, such registers
2063 need not be explicitly restored on function exit or during non-local
2068 If the program counter has a register number, define this as that
2069 register number. Otherwise, do not define it.
2072 @node Allocation Order
2073 @subsection Order of Allocation of Registers
2074 @cindex order of register allocation
2075 @cindex register allocation order
2077 @c prevent bad page break with this line
2078 Registers are allocated in order.
2080 @defmac REG_ALLOC_ORDER
2081 If defined, an initializer for a vector of integers, containing the
2082 numbers of hard registers in the order in which GCC should prefer
2083 to use them (from most preferred to least).
2085 If this macro is not defined, registers are used lowest numbered first
2086 (all else being equal).
2088 One use of this macro is on machines where the highest numbered
2089 registers must always be saved and the save-multiple-registers
2090 instruction supports only sequences of consecutive registers. On such
2091 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2092 the highest numbered allocable register first.
2095 @defmac ADJUST_REG_ALLOC_ORDER
2096 A C statement (sans semicolon) to choose the order in which to allocate
2097 hard registers for pseudo-registers local to a basic block.
2099 Store the desired register order in the array @code{reg_alloc_order}.
2100 Element 0 should be the register to allocate first; element 1, the next
2101 register; and so on.
2103 The macro body should not assume anything about the contents of
2104 @code{reg_alloc_order} before execution of the macro.
2106 On most machines, it is not necessary to define this macro.
2109 @defmac HONOR_REG_ALLOC_ORDER
2110 Normally, IRA tries to estimate the costs for saving a register in the
2111 prologue and restoring it in the epilogue. This discourages it from
2112 using call-saved registers. If a machine wants to ensure that IRA
2113 allocates registers in the order given by REG_ALLOC_ORDER even if some
2114 call-saved registers appear earlier than call-used ones, this macro
2118 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2119 In some case register allocation order is not enough for the
2120 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2121 If this macro is defined, it should return a floating point value
2122 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2123 be increased by approximately the pseudo's usage frequency times the
2124 value returned by this macro. Not defining this macro is equivalent
2125 to having it always return @code{0.0}.
2127 On most machines, it is not necessary to define this macro.
2130 @node Values in Registers
2131 @subsection How Values Fit in Registers
2133 This section discusses the macros that describe which kinds of values
2134 (specifically, which machine modes) each register can hold, and how many
2135 consecutive registers are needed for a given mode.
2137 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2138 A C expression for the number of consecutive hard registers, starting
2139 at register number @var{regno}, required to hold a value of mode
2140 @var{mode}. This macro must never return zero, even if a register
2141 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2142 and/or CANNOT_CHANGE_MODE_CLASS instead.
2144 On a machine where all registers are exactly one word, a suitable
2145 definition of this macro is
2148 #define HARD_REGNO_NREGS(REGNO, MODE) \
2149 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2154 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2155 A C expression that is nonzero if a value of mode @var{mode}, stored
2156 in memory, ends with padding that causes it to take up more space than
2157 in registers starting at register number @var{regno} (as determined by
2158 multiplying GCC's notion of the size of the register when containing
2159 this mode by the number of registers returned by
2160 @code{HARD_REGNO_NREGS}). By default this is zero.
2162 For example, if a floating-point value is stored in three 32-bit
2163 registers but takes up 128 bits in memory, then this would be
2166 This macros only needs to be defined if there are cases where
2167 @code{subreg_get_info}
2168 would otherwise wrongly determine that a @code{subreg} can be
2169 represented by an offset to the register number, when in fact such a
2170 @code{subreg} would contain some of the padding not stored in
2171 registers and so not be representable.
2174 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2175 For values of @var{regno} and @var{mode} for which
2176 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2177 returning the greater number of registers required to hold the value
2178 including any padding. In the example above, the value would be four.
2181 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2182 Define this macro if the natural size of registers that hold values
2183 of mode @var{mode} is not the word size. It is a C expression that
2184 should give the natural size in bytes for the specified mode. It is
2185 used by the register allocator to try to optimize its results. This
2186 happens for example on SPARC 64-bit where the natural size of
2187 floating-point registers is still 32-bit.
2190 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2191 A C expression that is nonzero if it is permissible to store a value
2192 of mode @var{mode} in hard register number @var{regno} (or in several
2193 registers starting with that one). For a machine where all registers
2194 are equivalent, a suitable definition is
2197 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2200 You need not include code to check for the numbers of fixed registers,
2201 because the allocation mechanism considers them to be always occupied.
2203 @cindex register pairs
2204 On some machines, double-precision values must be kept in even/odd
2205 register pairs. You can implement that by defining this macro to reject
2206 odd register numbers for such modes.
2208 The minimum requirement for a mode to be OK in a register is that the
2209 @samp{mov@var{mode}} instruction pattern support moves between the
2210 register and other hard register in the same class and that moving a
2211 value into the register and back out not alter it.
2213 Since the same instruction used to move @code{word_mode} will work for
2214 all narrower integer modes, it is not necessary on any machine for
2215 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2216 you define patterns @samp{movhi}, etc., to take advantage of this. This
2217 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2218 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2221 Many machines have special registers for floating point arithmetic.
2222 Often people assume that floating point machine modes are allowed only
2223 in floating point registers. This is not true. Any registers that
2224 can hold integers can safely @emph{hold} a floating point machine
2225 mode, whether or not floating arithmetic can be done on it in those
2226 registers. Integer move instructions can be used to move the values.
2228 On some machines, though, the converse is true: fixed-point machine
2229 modes may not go in floating registers. This is true if the floating
2230 registers normalize any value stored in them, because storing a
2231 non-floating value there would garble it. In this case,
2232 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2233 floating registers. But if the floating registers do not automatically
2234 normalize, if you can store any bit pattern in one and retrieve it
2235 unchanged without a trap, then any machine mode may go in a floating
2236 register, so you can define this macro to say so.
2238 The primary significance of special floating registers is rather that
2239 they are the registers acceptable in floating point arithmetic
2240 instructions. However, this is of no concern to
2241 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2242 constraints for those instructions.
2244 On some machines, the floating registers are especially slow to access,
2245 so that it is better to store a value in a stack frame than in such a
2246 register if floating point arithmetic is not being done. As long as the
2247 floating registers are not in class @code{GENERAL_REGS}, they will not
2248 be used unless some pattern's constraint asks for one.
2251 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2252 A C expression that is nonzero if it is OK to rename a hard register
2253 @var{from} to another hard register @var{to}.
2255 One common use of this macro is to prevent renaming of a register to
2256 another register that is not saved by a prologue in an interrupt
2259 The default is always nonzero.
2262 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2263 A C expression that is nonzero if a value of mode
2264 @var{mode1} is accessible in mode @var{mode2} without copying.
2266 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2267 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2268 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2269 should be nonzero. If they differ for any @var{r}, you should define
2270 this macro to return zero unless some other mechanism ensures the
2271 accessibility of the value in a narrower mode.
2273 You should define this macro to return nonzero in as many cases as
2274 possible since doing so will allow GCC to perform better register
2278 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno})
2279 This target hook should return @code{true} if it is OK to use a hard register
2280 @var{regno} as scratch reg in peephole2.
2282 One common use of this macro is to prevent using of a register that
2283 is not saved by a prologue in an interrupt handler.
2285 The default version of this hook always returns @code{true}.
2288 @defmac AVOID_CCMODE_COPIES
2289 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2290 registers. You should only define this macro if support for copying to/from
2291 @code{CCmode} is incomplete.
2294 @node Leaf Functions
2295 @subsection Handling Leaf Functions
2297 @cindex leaf functions
2298 @cindex functions, leaf
2299 On some machines, a leaf function (i.e., one which makes no calls) can run
2300 more efficiently if it does not make its own register window. Often this
2301 means it is required to receive its arguments in the registers where they
2302 are passed by the caller, instead of the registers where they would
2305 The special treatment for leaf functions generally applies only when
2306 other conditions are met; for example, often they may use only those
2307 registers for its own variables and temporaries. We use the term ``leaf
2308 function'' to mean a function that is suitable for this special
2309 handling, so that functions with no calls are not necessarily ``leaf
2312 GCC assigns register numbers before it knows whether the function is
2313 suitable for leaf function treatment. So it needs to renumber the
2314 registers in order to output a leaf function. The following macros
2317 @defmac LEAF_REGISTERS
2318 Name of a char vector, indexed by hard register number, which
2319 contains 1 for a register that is allowable in a candidate for leaf
2322 If leaf function treatment involves renumbering the registers, then the
2323 registers marked here should be the ones before renumbering---those that
2324 GCC would ordinarily allocate. The registers which will actually be
2325 used in the assembler code, after renumbering, should not be marked with 1
2328 Define this macro only if the target machine offers a way to optimize
2329 the treatment of leaf functions.
2332 @defmac LEAF_REG_REMAP (@var{regno})
2333 A C expression whose value is the register number to which @var{regno}
2334 should be renumbered, when a function is treated as a leaf function.
2336 If @var{regno} is a register number which should not appear in a leaf
2337 function before renumbering, then the expression should yield @minus{}1, which
2338 will cause the compiler to abort.
2340 Define this macro only if the target machine offers a way to optimize the
2341 treatment of leaf functions, and registers need to be renumbered to do
2345 @findex current_function_is_leaf
2346 @findex current_function_uses_only_leaf_regs
2347 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2348 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2349 specially. They can test the C variable @code{current_function_is_leaf}
2350 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2351 set prior to local register allocation and is valid for the remaining
2352 compiler passes. They can also test the C variable
2353 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2354 functions which only use leaf registers.
2355 @code{current_function_uses_only_leaf_regs} is valid after all passes
2356 that modify the instructions have been run and is only useful if
2357 @code{LEAF_REGISTERS} is defined.
2358 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2359 @c of the next paragraph?! --mew 2feb93
2361 @node Stack Registers
2362 @subsection Registers That Form a Stack
2364 There are special features to handle computers where some of the
2365 ``registers'' form a stack. Stack registers are normally written by
2366 pushing onto the stack, and are numbered relative to the top of the
2369 Currently, GCC can only handle one group of stack-like registers, and
2370 they must be consecutively numbered. Furthermore, the existing
2371 support for stack-like registers is specific to the 80387 floating
2372 point coprocessor. If you have a new architecture that uses
2373 stack-like registers, you will need to do substantial work on
2374 @file{reg-stack.c} and write your machine description to cooperate
2375 with it, as well as defining these macros.
2378 Define this if the machine has any stack-like registers.
2381 @defmac STACK_REG_COVER_CLASS
2382 This is a cover class containing the stack registers. Define this if
2383 the machine has any stack-like registers.
2386 @defmac FIRST_STACK_REG
2387 The number of the first stack-like register. This one is the top
2391 @defmac LAST_STACK_REG
2392 The number of the last stack-like register. This one is the bottom of
2396 @node Register Classes
2397 @section Register Classes
2398 @cindex register class definitions
2399 @cindex class definitions, register
2401 On many machines, the numbered registers are not all equivalent.
2402 For example, certain registers may not be allowed for indexed addressing;
2403 certain registers may not be allowed in some instructions. These machine
2404 restrictions are described to the compiler using @dfn{register classes}.
2406 You define a number of register classes, giving each one a name and saying
2407 which of the registers belong to it. Then you can specify register classes
2408 that are allowed as operands to particular instruction patterns.
2412 In general, each register will belong to several classes. In fact, one
2413 class must be named @code{ALL_REGS} and contain all the registers. Another
2414 class must be named @code{NO_REGS} and contain no registers. Often the
2415 union of two classes will be another class; however, this is not required.
2417 @findex GENERAL_REGS
2418 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2419 terribly special about the name, but the operand constraint letters
2420 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2421 the same as @code{ALL_REGS}, just define it as a macro which expands
2424 Order the classes so that if class @var{x} is contained in class @var{y}
2425 then @var{x} has a lower class number than @var{y}.
2427 The way classes other than @code{GENERAL_REGS} are specified in operand
2428 constraints is through machine-dependent operand constraint letters.
2429 You can define such letters to correspond to various classes, then use
2430 them in operand constraints.
2432 You should define a class for the union of two classes whenever some
2433 instruction allows both classes. For example, if an instruction allows
2434 either a floating point (coprocessor) register or a general register for a
2435 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2436 which includes both of them. Otherwise you will get suboptimal code.
2438 You must also specify certain redundant information about the register
2439 classes: for each class, which classes contain it and which ones are
2440 contained in it; for each pair of classes, the largest class contained
2443 When a value occupying several consecutive registers is expected in a
2444 certain class, all the registers used must belong to that class.
2445 Therefore, register classes cannot be used to enforce a requirement for
2446 a register pair to start with an even-numbered register. The way to
2447 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2449 Register classes used for input-operands of bitwise-and or shift
2450 instructions have a special requirement: each such class must have, for
2451 each fixed-point machine mode, a subclass whose registers can transfer that
2452 mode to or from memory. For example, on some machines, the operations for
2453 single-byte values (@code{QImode}) are limited to certain registers. When
2454 this is so, each register class that is used in a bitwise-and or shift
2455 instruction must have a subclass consisting of registers from which
2456 single-byte values can be loaded or stored. This is so that
2457 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2459 @deftp {Data type} {enum reg_class}
2460 An enumerated type that must be defined with all the register class names
2461 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2462 must be the last register class, followed by one more enumerated value,
2463 @code{LIM_REG_CLASSES}, which is not a register class but rather
2464 tells how many classes there are.
2466 Each register class has a number, which is the value of casting
2467 the class name to type @code{int}. The number serves as an index
2468 in many of the tables described below.
2471 @defmac N_REG_CLASSES
2472 The number of distinct register classes, defined as follows:
2475 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2479 @defmac REG_CLASS_NAMES
2480 An initializer containing the names of the register classes as C string
2481 constants. These names are used in writing some of the debugging dumps.
2484 @defmac REG_CLASS_CONTENTS
2485 An initializer containing the contents of the register classes, as integers
2486 which are bit masks. The @var{n}th integer specifies the contents of class
2487 @var{n}. The way the integer @var{mask} is interpreted is that
2488 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2490 When the machine has more than 32 registers, an integer does not suffice.
2491 Then the integers are replaced by sub-initializers, braced groupings containing
2492 several integers. Each sub-initializer must be suitable as an initializer
2493 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2494 In this situation, the first integer in each sub-initializer corresponds to
2495 registers 0 through 31, the second integer to registers 32 through 63, and
2499 @defmac REGNO_REG_CLASS (@var{regno})
2500 A C expression whose value is a register class containing hard register
2501 @var{regno}. In general there is more than one such class; choose a class
2502 which is @dfn{minimal}, meaning that no smaller class also contains the
2506 @defmac BASE_REG_CLASS
2507 A macro whose definition is the name of the class to which a valid
2508 base register must belong. A base register is one used in an address
2509 which is the register value plus a displacement.
2512 @defmac MODE_BASE_REG_CLASS (@var{mode})
2513 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2514 the selection of a base register in a mode dependent manner. If
2515 @var{mode} is VOIDmode then it should return the same value as
2516 @code{BASE_REG_CLASS}.
2519 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2520 A C expression whose value is the register class to which a valid
2521 base register must belong in order to be used in a base plus index
2522 register address. You should define this macro if base plus index
2523 addresses have different requirements than other base register uses.
2526 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{outer_code}, @var{index_code})
2527 A C expression whose value is the register class to which a valid
2528 base register must belong. @var{outer_code} and @var{index_code} define the
2529 context in which the base register occurs. @var{outer_code} is the code of
2530 the immediately enclosing expression (@code{MEM} for the top level of an
2531 address, @code{ADDRESS} for something that occurs in an
2532 @code{address_operand}). @var{index_code} is the code of the corresponding
2533 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2536 @defmac INDEX_REG_CLASS
2537 A macro whose definition is the name of the class to which a valid
2538 index register must belong. An index register is one used in an
2539 address where its value is either multiplied by a scale factor or
2540 added to another register (as well as added to a displacement).
2543 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2544 A C expression which is nonzero if register number @var{num} is
2545 suitable for use as a base register in operand addresses.
2546 Like @code{TARGET_LEGITIMATE_ADDRESS_P}, this macro should also
2547 define a strict and a non-strict variant. Both variants behave
2548 the same for hard register; for pseudos, the strict variant will
2549 pass only those that have been allocated to a valid hard registers,
2550 while the non-strict variant will pass all pseudos.
2552 @findex REG_OK_STRICT
2553 Compiler source files that want to use the strict variant of this and
2554 other macros define the macro @code{REG_OK_STRICT}. You should use an
2555 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
2556 that case and the non-strict variant otherwise.
2559 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2560 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2561 that expression may examine the mode of the memory reference in
2562 @var{mode}. You should define this macro if the mode of the memory
2563 reference affects whether a register may be used as a base register. If
2564 you define this macro, the compiler will use it instead of
2565 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2566 addresses that appear outside a @code{MEM}, i.e., as an
2567 @code{address_operand}.
2569 This macro also has strict and non-strict variants.
2572 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2573 A C expression which is nonzero if register number @var{num} is suitable for
2574 use as a base register in base plus index operand addresses, accessing
2575 memory in mode @var{mode}. It may be either a suitable hard register or a
2576 pseudo register that has been allocated such a hard register. You should
2577 define this macro if base plus index addresses have different requirements
2578 than other base register uses.
2580 Use of this macro is deprecated; please use the more general
2581 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2583 This macro also has strict and non-strict variants.
2586 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{outer_code}, @var{index_code})
2587 A C expression that is just like @code{REGNO_MODE_OK_FOR_BASE_P}, except
2588 that that expression may examine the context in which the register
2589 appears in the memory reference. @var{outer_code} is the code of the
2590 immediately enclosing expression (@code{MEM} if at the top level of the
2591 address, @code{ADDRESS} for something that occurs in an
2592 @code{address_operand}). @var{index_code} is the code of the
2593 corresponding index expression if @var{outer_code} is @code{PLUS};
2594 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2595 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2597 This macro also has strict and non-strict variants.
2600 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2601 A C expression which is nonzero if register number @var{num} is
2602 suitable for use as an index register in operand addresses. It may be
2603 either a suitable hard register or a pseudo register that has been
2604 allocated such a hard register.
2606 The difference between an index register and a base register is that
2607 the index register may be scaled. If an address involves the sum of
2608 two registers, neither one of them scaled, then either one may be
2609 labeled the ``base'' and the other the ``index''; but whichever
2610 labeling is used must fit the machine's constraints of which registers
2611 may serve in each capacity. The compiler will try both labelings,
2612 looking for one that is valid, and will reload one or both registers
2613 only if neither labeling works.
2615 This macro also has strict and non-strict variants.
2618 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2619 A C expression that places additional restrictions on the register class
2620 to use when it is necessary to copy value @var{x} into a register in class
2621 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2622 another, smaller class. On many machines, the following definition is
2626 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2629 Sometimes returning a more restrictive class makes better code. For
2630 example, on the 68000, when @var{x} is an integer constant that is in range
2631 for a @samp{moveq} instruction, the value of this macro is always
2632 @code{DATA_REGS} as long as @var{class} includes the data registers.
2633 Requiring a data register guarantees that a @samp{moveq} will be used.
2635 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2636 @var{class} is if @var{x} is a legitimate constant which cannot be
2637 loaded into some register class. By returning @code{NO_REGS} you can
2638 force @var{x} into a memory location. For example, rs6000 can load
2639 immediate values into general-purpose registers, but does not have an
2640 instruction for loading an immediate value into a floating-point
2641 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2642 @var{x} is a floating-point constant. If the constant can't be loaded
2643 into any kind of register, code generation will be better if
2644 @code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2645 of using @code{PREFERRED_RELOAD_CLASS}.
2647 If an insn has pseudos in it after register allocation, reload will go
2648 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2649 to find the best one. Returning @code{NO_REGS}, in this case, makes
2650 reload add a @code{!} in front of the constraint: the x86 back-end uses
2651 this feature to discourage usage of 387 registers when math is done in
2652 the SSE registers (and vice versa).
2655 @defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class})
2656 Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2657 input reloads. If you don't define this macro, the default is to use
2658 @var{class}, unchanged.
2660 You can also use @code{PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2661 reload from using some alternatives, like @code{PREFERRED_RELOAD_CLASS}.
2664 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2665 A C expression that places additional restrictions on the register class
2666 to use when it is necessary to be able to hold a value of mode
2667 @var{mode} in a reload register for which class @var{class} would
2670 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2671 there are certain modes that simply can't go in certain reload classes.
2673 The value is a register class; perhaps @var{class}, or perhaps another,
2676 Don't define this macro unless the target machine has limitations which
2677 require the macro to do something nontrivial.
2680 @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})
2681 Many machines have some registers that cannot be copied directly to or
2682 from memory or even from other types of registers. An example is the
2683 @samp{MQ} register, which on most machines, can only be copied to or
2684 from general registers, but not memory. Below, we shall be using the
2685 term 'intermediate register' when a move operation cannot be performed
2686 directly, but has to be done by copying the source into the intermediate
2687 register first, and then copying the intermediate register to the
2688 destination. An intermediate register always has the same mode as
2689 source and destination. Since it holds the actual value being copied,
2690 reload might apply optimizations to re-use an intermediate register
2691 and eliding the copy from the source when it can determine that the
2692 intermediate register still holds the required value.
2694 Another kind of secondary reload is required on some machines which
2695 allow copying all registers to and from memory, but require a scratch
2696 register for stores to some memory locations (e.g., those with symbolic
2697 address on the RT, and those with certain symbolic address on the SPARC
2698 when compiling PIC)@. Scratch registers need not have the same mode
2699 as the value being copied, and usually hold a different value than
2700 that being copied. Special patterns in the md file are needed to
2701 describe how the copy is performed with the help of the scratch register;
2702 these patterns also describe the number, register class(es) and mode(s)
2703 of the scratch register(s).
2705 In some cases, both an intermediate and a scratch register are required.
2707 For input reloads, this target hook is called with nonzero @var{in_p},
2708 and @var{x} is an rtx that needs to be copied to a register of class
2709 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2710 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2711 needs to be copied to rtx @var{x} in @var{reload_mode}.
2713 If copying a register of @var{reload_class} from/to @var{x} requires
2714 an intermediate register, the hook @code{secondary_reload} should
2715 return the register class required for this intermediate register.
2716 If no intermediate register is required, it should return NO_REGS.
2717 If more than one intermediate register is required, describe the one
2718 that is closest in the copy chain to the reload register.
2720 If scratch registers are needed, you also have to describe how to
2721 perform the copy from/to the reload register to/from this
2722 closest intermediate register. Or if no intermediate register is
2723 required, but still a scratch register is needed, describe the
2724 copy from/to the reload register to/from the reload operand @var{x}.
2726 You do this by setting @code{sri->icode} to the instruction code of a pattern
2727 in the md file which performs the move. Operands 0 and 1 are the output
2728 and input of this copy, respectively. Operands from operand 2 onward are
2729 for scratch operands. These scratch operands must have a mode, and a
2730 single-register-class
2731 @c [later: or memory]
2734 When an intermediate register is used, the @code{secondary_reload}
2735 hook will be called again to determine how to copy the intermediate
2736 register to/from the reload operand @var{x}, so your hook must also
2737 have code to handle the register class of the intermediate operand.
2739 @c [For later: maybe we'll allow multi-alternative reload patterns -
2740 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2741 @c and match the constraints of input and output to determine the required
2742 @c alternative. A restriction would be that constraints used to match
2743 @c against reloads registers would have to be written as register class
2744 @c constraints, or we need a new target macro / hook that tells us if an
2745 @c arbitrary constraint can match an unknown register of a given class.
2746 @c Such a macro / hook would also be useful in other places.]
2749 @var{x} might be a pseudo-register or a @code{subreg} of a
2750 pseudo-register, which could either be in a hard register or in memory.
2751 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2752 in memory and the hard register number if it is in a register.
2754 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2755 currently not supported. For the time being, you will have to continue
2756 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2758 @code{copy_cost} also uses this target hook to find out how values are
2759 copied. If you want it to include some extra cost for the need to allocate
2760 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2761 Or if two dependent moves are supposed to have a lower cost than the sum
2762 of the individual moves due to expected fortuitous scheduling and/or special
2763 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2766 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2767 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2768 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2769 These macros are obsolete, new ports should use the target hook
2770 @code{TARGET_SECONDARY_RELOAD} instead.
2772 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2773 target hook. Older ports still define these macros to indicate to the
2774 reload phase that it may
2775 need to allocate at least one register for a reload in addition to the
2776 register to contain the data. Specifically, if copying @var{x} to a
2777 register @var{class} in @var{mode} requires an intermediate register,
2778 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2779 largest register class all of whose registers can be used as
2780 intermediate registers or scratch registers.
2782 If copying a register @var{class} in @var{mode} to @var{x} requires an
2783 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2784 was supposed to be defined be defined to return the largest register
2785 class required. If the
2786 requirements for input and output reloads were the same, the macro
2787 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2790 The values returned by these macros are often @code{GENERAL_REGS}.
2791 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2792 can be directly copied to or from a register of @var{class} in
2793 @var{mode} without requiring a scratch register. Do not define this
2794 macro if it would always return @code{NO_REGS}.
2796 If a scratch register is required (either with or without an
2797 intermediate register), you were supposed to define patterns for
2798 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2799 (@pxref{Standard Names}. These patterns, which were normally
2800 implemented with a @code{define_expand}, should be similar to the
2801 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2804 These patterns need constraints for the reload register and scratch
2806 contain a single register class. If the original reload register (whose
2807 class is @var{class}) can meet the constraint given in the pattern, the
2808 value returned by these macros is used for the class of the scratch
2809 register. Otherwise, two additional reload registers are required.
2810 Their classes are obtained from the constraints in the insn pattern.
2812 @var{x} might be a pseudo-register or a @code{subreg} of a
2813 pseudo-register, which could either be in a hard register or in memory.
2814 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2815 in memory and the hard register number if it is in a register.
2817 These macros should not be used in the case where a particular class of
2818 registers can only be copied to memory and not to another class of
2819 registers. In that case, secondary reload registers are not needed and
2820 would not be helpful. Instead, a stack location must be used to perform
2821 the copy and the @code{mov@var{m}} pattern should use memory as an
2822 intermediate storage. This case often occurs between floating-point and
2826 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2827 Certain machines have the property that some registers cannot be copied
2828 to some other registers without using memory. Define this macro on
2829 those machines to be a C expression that is nonzero if objects of mode
2830 @var{m} in registers of @var{class1} can only be copied to registers of
2831 class @var{class2} by storing a register of @var{class1} into memory
2832 and loading that memory location into a register of @var{class2}.
2834 Do not define this macro if its value would always be zero.
2837 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2838 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2839 allocates a stack slot for a memory location needed for register copies.
2840 If this macro is defined, the compiler instead uses the memory location
2841 defined by this macro.
2843 Do not define this macro if you do not define
2844 @code{SECONDARY_MEMORY_NEEDED}.
2847 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2848 When the compiler needs a secondary memory location to copy between two
2849 registers of mode @var{mode}, it normally allocates sufficient memory to
2850 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2851 load operations in a mode that many bits wide and whose class is the
2852 same as that of @var{mode}.
2854 This is right thing to do on most machines because it ensures that all
2855 bits of the register are copied and prevents accesses to the registers
2856 in a narrower mode, which some machines prohibit for floating-point
2859 However, this default behavior is not correct on some machines, such as
2860 the DEC Alpha, that store short integers in floating-point registers
2861 differently than in integer registers. On those machines, the default
2862 widening will not work correctly and you must define this macro to
2863 suppress that widening in some cases. See the file @file{alpha.h} for
2866 Do not define this macro if you do not define
2867 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2868 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2871 @defmac CLASS_LIKELY_SPILLED_P (@var{class})
2872 A C expression whose value is nonzero if pseudos that have been assigned
2873 to registers of class @var{class} would likely be spilled because
2874 registers of @var{class} are needed for spill registers.
2876 The default value of this macro returns 1 if @var{class} has exactly one
2877 register and zero otherwise. On most machines, this default should be
2878 used. Only define this macro to some other expression if pseudos
2879 allocated by @file{local-alloc.c} end up in memory because their hard
2880 registers were needed for spill registers. If this macro returns nonzero
2881 for those classes, those pseudos will only be allocated by
2882 @file{global.c}, which knows how to reallocate the pseudo to another
2883 register. If there would not be another register available for
2884 reallocation, you should not change the definition of this macro since
2885 the only effect of such a definition would be to slow down register
2889 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2890 A C expression for the maximum number of consecutive registers
2891 of class @var{class} needed to hold a value of mode @var{mode}.
2893 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2894 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2895 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2896 @var{mode})} for all @var{regno} values in the class @var{class}.
2898 This macro helps control the handling of multiple-word values
2902 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2903 If defined, a C expression that returns nonzero for a @var{class} for which
2904 a change from mode @var{from} to mode @var{to} is invalid.
2906 For the example, loading 32-bit integer or floating-point objects into
2907 floating-point registers on the Alpha extends them to 64 bits.
2908 Therefore loading a 64-bit object and then storing it as a 32-bit object
2909 does not store the low-order 32 bits, as would be the case for a normal
2910 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2914 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2915 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2916 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2920 @deftypefn {Target Hook} {const enum reg_class *} TARGET_IRA_COVER_CLASSES (void)
2921 Return an array of cover classes for the Integrated Register Allocator
2922 (@acronym{IRA}). Cover classes are a set of non-intersecting register
2923 classes covering all hard registers used for register allocation
2924 purposes. If a move between two registers in the same cover class is
2925 possible, it should be cheaper than a load or store of the registers.
2926 The array is terminated by a @code{LIM_REG_CLASSES} element.
2928 The order of cover classes in the array is important. If two classes
2929 have the same cost of usage for a pseudo, the class occurred first in
2930 the array is chosen for the pseudo.
2932 This hook is called once at compiler startup, after the command-line
2933 options have been processed. It is then re-examined by every call to
2934 @code{target_reinit}.
2936 The default implementation returns @code{IRA_COVER_CLASSES}, if defined,
2937 otherwise there is no default implementation. You must define either this
2938 macro or @code{IRA_COVER_CLASSES} in order to use the integrated register
2939 allocator with Chaitin-Briggs coloring. If the macro is not defined,
2940 the only available coloring algorithm is Chow's priority coloring.
2943 @defmac IRA_COVER_CLASSES
2944 See the documentation for @code{TARGET_IRA_COVER_CLASSES}.
2947 @node Old Constraints
2948 @section Obsolete Macros for Defining Constraints
2949 @cindex defining constraints, obsolete method
2950 @cindex constraints, defining, obsolete method
2952 Machine-specific constraints can be defined with these macros instead
2953 of the machine description constructs described in @ref{Define
2954 Constraints}. This mechanism is obsolete. New ports should not use
2955 it; old ports should convert to the new mechanism.
2957 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2958 For the constraint at the start of @var{str}, which starts with the letter
2959 @var{c}, return the length. This allows you to have register class /
2960 constant / extra constraints that are longer than a single letter;
2961 you don't need to define this macro if you can do with single-letter
2962 constraints only. The definition of this macro should use
2963 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2964 to handle specially.
2965 There are some sanity checks in genoutput.c that check the constraint lengths
2966 for the md file, so you can also use this macro to help you while you are
2967 transitioning from a byzantine single-letter-constraint scheme: when you
2968 return a negative length for a constraint you want to re-use, genoutput
2969 will complain about every instance where it is used in the md file.
2972 @defmac REG_CLASS_FROM_LETTER (@var{char})
2973 A C expression which defines the machine-dependent operand constraint
2974 letters for register classes. If @var{char} is such a letter, the
2975 value should be the register class corresponding to it. Otherwise,
2976 the value should be @code{NO_REGS}. The register letter @samp{r},
2977 corresponding to class @code{GENERAL_REGS}, will not be passed
2978 to this macro; you do not need to handle it.
2981 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2982 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2983 passed in @var{str}, so that you can use suffixes to distinguish between
2987 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2988 A C expression that defines the machine-dependent operand constraint
2989 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2990 particular ranges of integer values. If @var{c} is one of those
2991 letters, the expression should check that @var{value}, an integer, is in
2992 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2993 not one of those letters, the value should be 0 regardless of
2997 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2998 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2999 string passed in @var{str}, so that you can use suffixes to distinguish
3000 between different variants.
3003 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
3004 A C expression that defines the machine-dependent operand constraint
3005 letters that specify particular ranges of @code{const_double} values
3006 (@samp{G} or @samp{H}).
3008 If @var{c} is one of those letters, the expression should check that
3009 @var{value}, an RTX of code @code{const_double}, is in the appropriate
3010 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
3011 letters, the value should be 0 regardless of @var{value}.
3013 @code{const_double} is used for all floating-point constants and for
3014 @code{DImode} fixed-point constants. A given letter can accept either
3015 or both kinds of values. It can use @code{GET_MODE} to distinguish
3016 between these kinds.
3019 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
3020 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
3021 string passed in @var{str}, so that you can use suffixes to distinguish
3022 between different variants.
3025 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
3026 A C expression that defines the optional machine-dependent constraint
3027 letters that can be used to segregate specific types of operands, usually
3028 memory references, for the target machine. Any letter that is not
3029 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
3030 @code{REG_CLASS_FROM_CONSTRAINT}
3031 may be used. Normally this macro will not be defined.
3033 If it is required for a particular target machine, it should return 1
3034 if @var{value} corresponds to the operand type represented by the
3035 constraint letter @var{c}. If @var{c} is not defined as an extra
3036 constraint, the value returned should be 0 regardless of @var{value}.
3038 For example, on the ROMP, load instructions cannot have their output
3039 in r0 if the memory reference contains a symbolic address. Constraint
3040 letter @samp{Q} is defined as representing a memory address that does
3041 @emph{not} contain a symbolic address. An alternative is specified with
3042 a @samp{Q} constraint on the input and @samp{r} on the output. The next
3043 alternative specifies @samp{m} on the input and a register class that
3044 does not include r0 on the output.
3047 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
3048 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
3049 in @var{str}, so that you can use suffixes to distinguish between different
3053 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
3054 A C expression that defines the optional machine-dependent constraint
3055 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
3056 be treated like memory constraints by the reload pass.
3058 It should return 1 if the operand type represented by the constraint
3059 at the start of @var{str}, the first letter of which is the letter @var{c},
3060 comprises a subset of all memory references including
3061 all those whose address is simply a base register. This allows the reload
3062 pass to reload an operand, if it does not directly correspond to the operand
3063 type of @var{c}, by copying its address into a base register.
3065 For example, on the S/390, some instructions do not accept arbitrary
3066 memory references, but only those that do not make use of an index
3067 register. The constraint letter @samp{Q} is defined via
3068 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
3069 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
3070 a @samp{Q} constraint can handle any memory operand, because the
3071 reload pass knows it can be reloaded by copying the memory address
3072 into a base register if required. This is analogous to the way
3073 an @samp{o} constraint can handle any memory operand.
3076 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
3077 A C expression that defines the optional machine-dependent constraint
3078 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
3079 @code{EXTRA_CONSTRAINT_STR}, that should
3080 be treated like address constraints by the reload pass.
3082 It should return 1 if the operand type represented by the constraint
3083 at the start of @var{str}, which starts with the letter @var{c}, comprises
3084 a subset of all memory addresses including
3085 all those that consist of just a base register. This allows the reload
3086 pass to reload an operand, if it does not directly correspond to the operand
3087 type of @var{str}, by copying it into a base register.
3089 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3090 be used with the @code{address_operand} predicate. It is treated
3091 analogously to the @samp{p} constraint.
3094 @node Stack and Calling
3095 @section Stack Layout and Calling Conventions
3096 @cindex calling conventions
3098 @c prevent bad page break with this line
3099 This describes the stack layout and calling conventions.
3103 * Exception Handling::
3108 * Register Arguments::
3110 * Aggregate Return::
3115 * Stack Smashing Protection::
3119 @subsection Basic Stack Layout
3120 @cindex stack frame layout
3121 @cindex frame layout
3123 @c prevent bad page break with this line
3124 Here is the basic stack layout.
3126 @defmac STACK_GROWS_DOWNWARD
3127 Define this macro if pushing a word onto the stack moves the stack
3128 pointer to a smaller address.
3130 When we say, ``define this macro if @dots{}'', it means that the
3131 compiler checks this macro only with @code{#ifdef} so the precise
3132 definition used does not matter.
3135 @defmac STACK_PUSH_CODE
3136 This macro defines the operation used when something is pushed
3137 on the stack. In RTL, a push operation will be
3138 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3140 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3141 and @code{POST_INC}. Which of these is correct depends on
3142 the stack direction and on whether the stack pointer points
3143 to the last item on the stack or whether it points to the
3144 space for the next item on the stack.
3146 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3147 defined, which is almost always right, and @code{PRE_INC} otherwise,
3148 which is often wrong.
3151 @defmac FRAME_GROWS_DOWNWARD
3152 Define this macro to nonzero value if the addresses of local variable slots
3153 are at negative offsets from the frame pointer.
3156 @defmac ARGS_GROW_DOWNWARD
3157 Define this macro if successive arguments to a function occupy decreasing
3158 addresses on the stack.
3161 @defmac STARTING_FRAME_OFFSET
3162 Offset from the frame pointer to the first local variable slot to be allocated.
3164 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3165 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3166 Otherwise, it is found by adding the length of the first slot to the
3167 value @code{STARTING_FRAME_OFFSET}.
3168 @c i'm not sure if the above is still correct.. had to change it to get
3169 @c rid of an overfull. --mew 2feb93
3172 @defmac STACK_ALIGNMENT_NEEDED
3173 Define to zero to disable final alignment of the stack during reload.
3174 The nonzero default for this macro is suitable for most ports.
3176 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3177 is a register save block following the local block that doesn't require
3178 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3179 stack alignment and do it in the backend.
3182 @defmac STACK_POINTER_OFFSET
3183 Offset from the stack pointer register to the first location at which
3184 outgoing arguments are placed. If not specified, the default value of
3185 zero is used. This is the proper value for most machines.
3187 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3188 the first location at which outgoing arguments are placed.
3191 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3192 Offset from the argument pointer register to the first argument's
3193 address. On some machines it may depend on the data type of the
3196 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3197 the first argument's address.
3200 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3201 Offset from the stack pointer register to an item dynamically allocated
3202 on the stack, e.g., by @code{alloca}.
3204 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3205 length of the outgoing arguments. The default is correct for most
3206 machines. See @file{function.c} for details.
3209 @defmac INITIAL_FRAME_ADDRESS_RTX
3210 A C expression whose value is RTL representing the address of the initial
3211 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3212 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3213 default value will be used. Define this macro in order to make frame pointer
3214 elimination work in the presence of @code{__builtin_frame_address (count)} and
3215 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3218 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3219 A C expression whose value is RTL representing the address in a stack
3220 frame where the pointer to the caller's frame is stored. Assume that
3221 @var{frameaddr} is an RTL expression for the address of the stack frame
3224 If you don't define this macro, the default is to return the value
3225 of @var{frameaddr}---that is, the stack frame address is also the
3226 address of the stack word that points to the previous frame.
3229 @defmac SETUP_FRAME_ADDRESSES
3230 If defined, a C expression that produces the machine-specific code to
3231 setup the stack so that arbitrary frames can be accessed. For example,
3232 on the SPARC, we must flush all of the register windows to the stack
3233 before we can access arbitrary stack frames. You will seldom need to
3237 @deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)
3238 This target hook should return an rtx that is used to store
3239 the address of the current frame into the built in @code{setjmp} buffer.
3240 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3241 machines. One reason you may need to define this target hook is if
3242 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3245 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3246 A C expression whose value is RTL representing the value of the frame
3247 address for the current frame. @var{frameaddr} is the frame pointer
3248 of the current frame. This is used for __builtin_frame_address.
3249 You need only define this macro if the frame address is not the same
3250 as the frame pointer. Most machines do not need to define it.
3253 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3254 A C expression whose value is RTL representing the value of the return
3255 address for the frame @var{count} steps up from the current frame, after
3256 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3257 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3258 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3260 The value of the expression must always be the correct address when
3261 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3262 determine the return address of other frames.
3265 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3266 Define this if the return address of a particular stack frame is accessed
3267 from the frame pointer of the previous stack frame.
3270 @defmac INCOMING_RETURN_ADDR_RTX
3271 A C expression whose value is RTL representing the location of the
3272 incoming return address at the beginning of any function, before the
3273 prologue. This RTL is either a @code{REG}, indicating that the return
3274 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3277 You only need to define this macro if you want to support call frame
3278 debugging information like that provided by DWARF 2.
3280 If this RTL is a @code{REG}, you should also define
3281 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3284 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3285 A C expression whose value is an integer giving a DWARF 2 column
3286 number that may be used as an alternative return column. The column
3287 must not correspond to any gcc hard register (that is, it must not
3288 be in the range of @code{DWARF_FRAME_REGNUM}).
3290 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3291 general register, but an alternative column needs to be used for signal
3292 frames. Some targets have also used different frame return columns
3296 @defmac DWARF_ZERO_REG
3297 A C expression whose value is an integer giving a DWARF 2 register
3298 number that is considered to always have the value zero. This should
3299 only be defined if the target has an architected zero register, and
3300 someone decided it was a good idea to use that register number to
3301 terminate the stack backtrace. New ports should avoid this.
3304 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3305 This target hook allows the backend to emit frame-related insns that
3306 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3307 info engine will invoke it on insns of the form
3309 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3313 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3315 to let the backend emit the call frame instructions. @var{label} is
3316 the CFI label attached to the insn, @var{pattern} is the pattern of
3317 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3320 @defmac INCOMING_FRAME_SP_OFFSET
3321 A C expression whose value is an integer giving the offset, in bytes,
3322 from the value of the stack pointer register to the top of the stack
3323 frame at the beginning of any function, before the prologue. The top of
3324 the frame is defined to be the value of the stack pointer in the
3325 previous frame, just before the call instruction.
3327 You only need to define this macro if you want to support call frame
3328 debugging information like that provided by DWARF 2.
3331 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3332 A C expression whose value is an integer giving the offset, in bytes,
3333 from the argument pointer to the canonical frame address (cfa). The
3334 final value should coincide with that calculated by
3335 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3336 during virtual register instantiation.
3338 The default value for this macro is
3339 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3340 which is correct for most machines; in general, the arguments are found
3341 immediately before the stack frame. Note that this is not the case on
3342 some targets that save registers into the caller's frame, such as SPARC
3343 and rs6000, and so such targets need to define this macro.
3345 You only need to define this macro if the default is incorrect, and you
3346 want to support call frame debugging information like that provided by
3350 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3351 If defined, a C expression whose value is an integer giving the offset
3352 in bytes from the frame pointer to the canonical frame address (cfa).
3353 The final value should coincide with that calculated by
3354 @code{INCOMING_FRAME_SP_OFFSET}.
3356 Normally the CFA is calculated as an offset from the argument pointer,
3357 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3358 variable due to the ABI, this may not be possible. If this macro is
3359 defined, it implies that the virtual register instantiation should be
3360 based on the frame pointer instead of the argument pointer. Only one
3361 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3365 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3366 If defined, a C expression whose value is an integer giving the offset
3367 in bytes from the canonical frame address (cfa) to the frame base used
3368 in DWARF 2 debug information. The default is zero. A different value
3369 may reduce the size of debug information on some ports.
3372 @node Exception Handling
3373 @subsection Exception Handling Support
3374 @cindex exception handling
3376 @defmac EH_RETURN_DATA_REGNO (@var{N})
3377 A C expression whose value is the @var{N}th register number used for
3378 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3379 @var{N} registers are usable.
3381 The exception handling library routines communicate with the exception
3382 handlers via a set of agreed upon registers. Ideally these registers
3383 should be call-clobbered; it is possible to use call-saved registers,
3384 but may negatively impact code size. The target must support at least
3385 2 data registers, but should define 4 if there are enough free registers.
3387 You must define this macro if you want to support call frame exception
3388 handling like that provided by DWARF 2.
3391 @defmac EH_RETURN_STACKADJ_RTX
3392 A C expression whose value is RTL representing a location in which
3393 to store a stack adjustment to be applied before function return.
3394 This is used to unwind the stack to an exception handler's call frame.
3395 It will be assigned zero on code paths that return normally.
3397 Typically this is a call-clobbered hard register that is otherwise
3398 untouched by the epilogue, but could also be a stack slot.
3400 Do not define this macro if the stack pointer is saved and restored
3401 by the regular prolog and epilog code in the call frame itself; in
3402 this case, the exception handling library routines will update the
3403 stack location to be restored in place. Otherwise, you must define
3404 this macro if you want to support call frame exception handling like
3405 that provided by DWARF 2.
3408 @defmac EH_RETURN_HANDLER_RTX
3409 A C expression whose value is RTL representing a location in which
3410 to store the address of an exception handler to which we should
3411 return. It will not be assigned on code paths that return normally.
3413 Typically this is the location in the call frame at which the normal
3414 return address is stored. For targets that return by popping an
3415 address off the stack, this might be a memory address just below
3416 the @emph{target} call frame rather than inside the current call
3417 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3418 been assigned, so it may be used to calculate the location of the
3421 Some targets have more complex requirements than storing to an
3422 address calculable during initial code generation. In that case
3423 the @code{eh_return} instruction pattern should be used instead.
3425 If you want to support call frame exception handling, you must
3426 define either this macro or the @code{eh_return} instruction pattern.
3429 @defmac RETURN_ADDR_OFFSET
3430 If defined, an integer-valued C expression for which rtl will be generated
3431 to add it to the exception handler address before it is searched in the
3432 exception handling tables, and to subtract it again from the address before
3433 using it to return to the exception handler.
3436 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3437 This macro chooses the encoding of pointers embedded in the exception
3438 handling sections. If at all possible, this should be defined such
3439 that the exception handling section will not require dynamic relocations,
3440 and so may be read-only.
3442 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3443 @var{global} is true if the symbol may be affected by dynamic relocations.
3444 The macro should return a combination of the @code{DW_EH_PE_*} defines
3445 as found in @file{dwarf2.h}.
3447 If this macro is not defined, pointers will not be encoded but
3448 represented directly.
3451 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3452 This macro allows the target to emit whatever special magic is required
3453 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3454 Generic code takes care of pc-relative and indirect encodings; this must
3455 be defined if the target uses text-relative or data-relative encodings.
3457 This is a C statement that branches to @var{done} if the format was
3458 handled. @var{encoding} is the format chosen, @var{size} is the number
3459 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3463 @defmac MD_UNWIND_SUPPORT
3464 A string specifying a file to be #include'd in unwind-dw2.c. The file
3465 so included typically defines @code{MD_FALLBACK_FRAME_STATE_FOR}.
3468 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3469 This macro allows the target to add CPU and operating system specific
3470 code to the call-frame unwinder for use when there is no unwind data
3471 available. The most common reason to implement this macro is to unwind
3472 through signal frames.
3474 This macro is called from @code{uw_frame_state_for} in
3475 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3476 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3477 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3478 for the address of the code being executed and @code{context->cfa} for
3479 the stack pointer value. If the frame can be decoded, the register
3480 save addresses should be updated in @var{fs} and the macro should
3481 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3482 the macro should evaluate to @code{_URC_END_OF_STACK}.
3484 For proper signal handling in Java this macro is accompanied by
3485 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3488 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3489 This macro allows the target to add operating system specific code to the
3490 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3491 usually used for signal or interrupt frames.
3493 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3494 @var{context} is an @code{_Unwind_Context};
3495 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3496 for the abi and context in the @code{.unwabi} directive. If the
3497 @code{.unwabi} directive can be handled, the register save addresses should
3498 be updated in @var{fs}.
3501 @defmac TARGET_USES_WEAK_UNWIND_INFO
3502 A C expression that evaluates to true if the target requires unwind
3503 info to be given comdat linkage. Define it to be @code{1} if comdat
3504 linkage is necessary. The default is @code{0}.
3507 @node Stack Checking
3508 @subsection Specifying How Stack Checking is Done
3510 GCC will check that stack references are within the boundaries of the
3511 stack, if the option @option{-fstack-check} is specified, in one of
3516 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3517 will assume that you have arranged for full stack checking to be done
3518 at appropriate places in the configuration files. GCC will not do
3519 other special processing.
3522 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3523 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3524 that you have arranged for static stack checking (checking of the
3525 static stack frame of functions) to be done at appropriate places
3526 in the configuration files. GCC will only emit code to do dynamic
3527 stack checking (checking on dynamic stack allocations) using the third
3531 If neither of the above are true, GCC will generate code to periodically
3532 ``probe'' the stack pointer using the values of the macros defined below.
3535 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3536 GCC will change its allocation strategy for large objects if the option
3537 @option{-fstack-check} is specified: they will always be allocated
3538 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3540 @defmac STACK_CHECK_BUILTIN
3541 A nonzero value if stack checking is done by the configuration files in a
3542 machine-dependent manner. You should define this macro if stack checking
3543 is required by the ABI of your machine or if you would like to do stack
3544 checking in some more efficient way than the generic approach. The default
3545 value of this macro is zero.
3548 @defmac STACK_CHECK_STATIC_BUILTIN
3549 A nonzero value if static stack checking is done by the configuration files
3550 in a machine-dependent manner. You should define this macro if you would
3551 like to do static stack checking in some more efficient way than the generic
3552 approach. The default value of this macro is zero.
3555 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3556 An integer specifying the interval at which GCC must generate stack probe
3557 instructions, defined as 2 raised to this integer. You will normally
3558 define this macro so that the interval be no larger than the size of
3559 the ``guard pages'' at the end of a stack area. The default value
3560 of 12 (4096-byte interval) is suitable for most systems.
3563 @defmac STACK_CHECK_MOVING_SP
3564 An integer which is nonzero if GCC should move the stack pointer page by page
3565 when doing probes. This can be necessary on systems where the stack pointer
3566 contains the bottom address of the memory area accessible to the executing
3567 thread at any point in time. In this situation an alternate signal stack
3568 is required in order to be able to recover from a stack overflow. The
3569 default value of this macro is zero.
3572 @defmac STACK_CHECK_PROTECT
3573 The number of bytes of stack needed to recover from a stack overflow, for
3574 languages where such a recovery is supported. The default value of 75 words
3575 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3576 8192 bytes with other exception handling mechanisms should be adequate for
3580 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3581 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3582 in the opposite case.
3584 @defmac STACK_CHECK_MAX_FRAME_SIZE
3585 The maximum size of a stack frame, in bytes. GCC will generate probe
3586 instructions in non-leaf functions to ensure at least this many bytes of
3587 stack are available. If a stack frame is larger than this size, stack
3588 checking will not be reliable and GCC will issue a warning. The
3589 default is chosen so that GCC only generates one instruction on most
3590 systems. You should normally not change the default value of this macro.
3593 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3594 GCC uses this value to generate the above warning message. It
3595 represents the amount of fixed frame used by a function, not including
3596 space for any callee-saved registers, temporaries and user variables.
3597 You need only specify an upper bound for this amount and will normally
3598 use the default of four words.
3601 @defmac STACK_CHECK_MAX_VAR_SIZE
3602 The maximum size, in bytes, of an object that GCC will place in the
3603 fixed area of the stack frame when the user specifies
3604 @option{-fstack-check}.
3605 GCC computed the default from the values of the above macros and you will
3606 normally not need to override that default.
3610 @node Frame Registers
3611 @subsection Registers That Address the Stack Frame
3613 @c prevent bad page break with this line
3614 This discusses registers that address the stack frame.
3616 @defmac STACK_POINTER_REGNUM
3617 The register number of the stack pointer register, which must also be a
3618 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3619 the hardware determines which register this is.
3622 @defmac FRAME_POINTER_REGNUM
3623 The register number of the frame pointer register, which is used to
3624 access automatic variables in the stack frame. On some machines, the
3625 hardware determines which register this is. On other machines, you can
3626 choose any register you wish for this purpose.
3629 @defmac HARD_FRAME_POINTER_REGNUM
3630 On some machines the offset between the frame pointer and starting
3631 offset of the automatic variables is not known until after register
3632 allocation has been done (for example, because the saved registers are
3633 between these two locations). On those machines, define
3634 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3635 be used internally until the offset is known, and define
3636 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3637 used for the frame pointer.
3639 You should define this macro only in the very rare circumstances when it
3640 is not possible to calculate the offset between the frame pointer and
3641 the automatic variables until after register allocation has been
3642 completed. When this macro is defined, you must also indicate in your
3643 definition of @code{ELIMINABLE_REGS} how to eliminate
3644 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3645 or @code{STACK_POINTER_REGNUM}.
3647 Do not define this macro if it would be the same as
3648 @code{FRAME_POINTER_REGNUM}.
3651 @defmac ARG_POINTER_REGNUM
3652 The register number of the arg pointer register, which is used to access
3653 the function's argument list. On some machines, this is the same as the
3654 frame pointer register. On some machines, the hardware determines which
3655 register this is. On other machines, you can choose any register you
3656 wish for this purpose. If this is not the same register as the frame
3657 pointer register, then you must mark it as a fixed register according to
3658 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3659 (@pxref{Elimination}).
3662 @defmac RETURN_ADDRESS_POINTER_REGNUM
3663 The register number of the return address pointer register, which is used to
3664 access the current function's return address from the stack. On some
3665 machines, the return address is not at a fixed offset from the frame
3666 pointer or stack pointer or argument pointer. This register can be defined
3667 to point to the return address on the stack, and then be converted by
3668 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3670 Do not define this macro unless there is no other way to get the return
3671 address from the stack.
3674 @defmac STATIC_CHAIN_REGNUM
3675 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3676 Register numbers used for passing a function's static chain pointer. If
3677 register windows are used, the register number as seen by the called
3678 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3679 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3680 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3683 The static chain register need not be a fixed register.
3685 If the static chain is passed in memory, these macros should not be
3686 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3689 @deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl}, bool @var{incoming_p})
3690 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3691 targets that may use different static chain locations for different
3692 nested functions. This may be required if the target has function
3693 attributes that affect the calling conventions of the function and
3694 those calling conventions use different static chain locations.
3696 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3698 If the static chain is passed in memory, this hook should be used to
3699 provide rtx giving @code{mem} expressions that denote where they are stored.
3700 Often the @code{mem} expression as seen by the caller will be at an offset
3701 from the stack pointer and the @code{mem} expression as seen by the callee
3702 will be at an offset from the frame pointer.
3703 @findex stack_pointer_rtx
3704 @findex frame_pointer_rtx
3705 @findex arg_pointer_rtx
3706 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3707 @code{arg_pointer_rtx} will have been initialized and should be used
3708 to refer to those items.
3711 @defmac DWARF_FRAME_REGISTERS
3712 This macro specifies the maximum number of hard registers that can be
3713 saved in a call frame. This is used to size data structures used in
3714 DWARF2 exception handling.
3716 Prior to GCC 3.0, this macro was needed in order to establish a stable
3717 exception handling ABI in the face of adding new hard registers for ISA
3718 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3719 in the number of hard registers. Nevertheless, this macro can still be
3720 used to reduce the runtime memory requirements of the exception handling
3721 routines, which can be substantial if the ISA contains a lot of
3722 registers that are not call-saved.
3724 If this macro is not defined, it defaults to
3725 @code{FIRST_PSEUDO_REGISTER}.
3728 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3730 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3731 for backward compatibility in pre GCC 3.0 compiled code.
3733 If this macro is not defined, it defaults to
3734 @code{DWARF_FRAME_REGISTERS}.
3737 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3739 Define this macro if the target's representation for dwarf registers
3740 is different than the internal representation for unwind column.
3741 Given a dwarf register, this macro should return the internal unwind
3742 column number to use instead.
3744 See the PowerPC's SPE target for an example.
3747 @defmac DWARF_FRAME_REGNUM (@var{regno})
3749 Define this macro if the target's representation for dwarf registers
3750 used in .eh_frame or .debug_frame is different from that used in other
3751 debug info sections. Given a GCC hard register number, this macro
3752 should return the .eh_frame register number. The default is
3753 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3757 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3759 Define this macro to map register numbers held in the call frame info
3760 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3761 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3762 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3763 return @code{@var{regno}}.
3768 @subsection Eliminating Frame Pointer and Arg Pointer
3770 @c prevent bad page break with this line
3771 This is about eliminating the frame pointer and arg pointer.
3773 @deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void)
3774 This target hook should return @code{true} if a function must have and use
3775 a frame pointer. This target hook is called in the reload pass. If its return
3776 value is @code{true} the function will have a frame pointer.
3778 This target hook can in principle examine the current function and decide
3779 according to the facts, but on most machines the constant @code{false} or the
3780 constant @code{true} suffices. Use @code{false} when the machine allows code
3781 to be generated with no frame pointer, and doing so saves some time or space.
3782 Use @code{true} when there is no possible advantage to avoiding a frame
3785 In certain cases, the compiler does not know how to produce valid code
3786 without a frame pointer. The compiler recognizes those cases and
3787 automatically gives the function a frame pointer regardless of what
3788 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3791 In a function that does not require a frame pointer, the frame pointer
3792 register can be allocated for ordinary usage, unless you mark it as a
3793 fixed register. See @code{FIXED_REGISTERS} for more information.
3795 Default return value is @code{false}.
3798 @findex get_frame_size
3799 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3800 A C statement to store in the variable @var{depth-var} the difference
3801 between the frame pointer and the stack pointer values immediately after
3802 the function prologue. The value would be computed from information
3803 such as the result of @code{get_frame_size ()} and the tables of
3804 registers @code{regs_ever_live} and @code{call_used_regs}.
3806 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3807 need not be defined. Otherwise, it must be defined even if
3808 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3809 case, you may set @var{depth-var} to anything.
3812 @defmac ELIMINABLE_REGS
3813 If defined, this macro specifies a table of register pairs used to
3814 eliminate unneeded registers that point into the stack frame. If it is not
3815 defined, the only elimination attempted by the compiler is to replace
3816 references to the frame pointer with references to the stack pointer.
3818 The definition of this macro is a list of structure initializations, each
3819 of which specifies an original and replacement register.
3821 On some machines, the position of the argument pointer is not known until
3822 the compilation is completed. In such a case, a separate hard register
3823 must be used for the argument pointer. This register can be eliminated by
3824 replacing it with either the frame pointer or the argument pointer,
3825 depending on whether or not the frame pointer has been eliminated.
3827 In this case, you might specify:
3829 #define ELIMINABLE_REGS \
3830 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3831 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3832 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3835 Note that the elimination of the argument pointer with the stack pointer is
3836 specified first since that is the preferred elimination.
3839 @deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg})
3840 This target hook should returns @code{true} if the compiler is allowed to
3841 try to replace register number @var{from_reg} with register number
3842 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3843 is defined, and will usually be @code{true}, since most of the cases
3844 preventing register elimination are things that the compiler already
3847 Default return value is @code{true}.
3850 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3851 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3852 specifies the initial difference between the specified pair of
3853 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3857 @node Stack Arguments
3858 @subsection Passing Function Arguments on the Stack
3859 @cindex arguments on stack
3860 @cindex stack arguments
3862 The macros in this section control how arguments are passed
3863 on the stack. See the following section for other macros that
3864 control passing certain arguments in registers.
3866 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype})
3867 This target hook returns @code{true} if an argument declared in a
3868 prototype as an integral type smaller than @code{int} should actually be
3869 passed as an @code{int}. In addition to avoiding errors in certain
3870 cases of mismatch, it also makes for better code on certain machines.
3871 The default is to not promote prototypes.
3875 A C expression. If nonzero, push insns will be used to pass
3877 If the target machine does not have a push instruction, set it to zero.
3878 That directs GCC to use an alternate strategy: to
3879 allocate the entire argument block and then store the arguments into
3880 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3883 @defmac PUSH_ARGS_REVERSED
3884 A C expression. If nonzero, function arguments will be evaluated from
3885 last to first, rather than from first to last. If this macro is not
3886 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3887 and args grow in opposite directions, and 0 otherwise.
3890 @defmac PUSH_ROUNDING (@var{npushed})
3891 A C expression that is the number of bytes actually pushed onto the
3892 stack when an instruction attempts to push @var{npushed} bytes.
3894 On some machines, the definition
3897 #define PUSH_ROUNDING(BYTES) (BYTES)
3901 will suffice. But on other machines, instructions that appear
3902 to push one byte actually push two bytes in an attempt to maintain
3903 alignment. Then the definition should be
3906 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3910 @findex current_function_outgoing_args_size
3911 @defmac ACCUMULATE_OUTGOING_ARGS
3912 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3913 will be computed and placed into the variable
3914 @code{current_function_outgoing_args_size}. No space will be pushed
3915 onto the stack for each call; instead, the function prologue should
3916 increase the stack frame size by this amount.
3918 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3922 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3923 Define this macro if functions should assume that stack space has been
3924 allocated for arguments even when their values are passed in
3927 The value of this macro is the size, in bytes, of the area reserved for
3928 arguments passed in registers for the function represented by @var{fndecl},
3929 which can be zero if GCC is calling a library function.
3930 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3933 This space can be allocated by the caller, or be a part of the
3934 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3937 @c above is overfull. not sure what to do. --mew 5feb93 did
3938 @c something, not sure if it looks good. --mew 10feb93
3940 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3941 Define this to a nonzero value if it is the responsibility of the
3942 caller to allocate the area reserved for arguments passed in registers
3943 when calling a function of @var{fntype}. @var{fntype} may be NULL
3944 if the function called is a library function.
3946 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3947 whether the space for these arguments counts in the value of
3948 @code{current_function_outgoing_args_size}.
3951 @defmac STACK_PARMS_IN_REG_PARM_AREA
3952 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3953 stack parameters don't skip the area specified by it.
3954 @c i changed this, makes more sens and it should have taken care of the
3955 @c overfull.. not as specific, tho. --mew 5feb93
3957 Normally, when a parameter is not passed in registers, it is placed on the
3958 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3959 suppresses this behavior and causes the parameter to be passed on the
3960 stack in its natural location.
3963 @defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size})
3964 A C expression that should indicate the number of bytes of its own
3965 arguments that a function pops on returning, or 0 if the
3966 function pops no arguments and the caller must therefore pop them all
3967 after the function returns.
3969 @var{fundecl} is a C variable whose value is a tree node that describes
3970 the function in question. Normally it is a node of type
3971 @code{FUNCTION_DECL} that describes the declaration of the function.
3972 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3974 @var{funtype} is a C variable whose value is a tree node that
3975 describes the function in question. Normally it is a node of type
3976 @code{FUNCTION_TYPE} that describes the data type of the function.
3977 From this it is possible to obtain the data types of the value and
3978 arguments (if known).
3980 When a call to a library function is being considered, @var{fundecl}
3981 will contain an identifier node for the library function. Thus, if
3982 you need to distinguish among various library functions, you can do so
3983 by their names. Note that ``library function'' in this context means
3984 a function used to perform arithmetic, whose name is known specially
3985 in the compiler and was not mentioned in the C code being compiled.
3987 @var{stack-size} is the number of bytes of arguments passed on the
3988 stack. If a variable number of bytes is passed, it is zero, and
3989 argument popping will always be the responsibility of the calling function.
3991 On the VAX, all functions always pop their arguments, so the definition
3992 of this macro is @var{stack-size}. On the 68000, using the standard
3993 calling convention, no functions pop their arguments, so the value of
3994 the macro is always 0 in this case. But an alternative calling
3995 convention is available in which functions that take a fixed number of
3996 arguments pop them but other functions (such as @code{printf}) pop
3997 nothing (the caller pops all). When this convention is in use,
3998 @var{funtype} is examined to determine whether a function takes a fixed
3999 number of arguments.
4002 @defmac CALL_POPS_ARGS (@var{cum})
4003 A C expression that should indicate the number of bytes a call sequence
4004 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
4005 when compiling a function call.
4007 @var{cum} is the variable in which all arguments to the called function
4008 have been accumulated.
4010 On certain architectures, such as the SH5, a call trampoline is used
4011 that pops certain registers off the stack, depending on the arguments
4012 that have been passed to the function. Since this is a property of the
4013 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
4017 @node Register Arguments
4018 @subsection Passing Arguments in Registers
4019 @cindex arguments in registers
4020 @cindex registers arguments
4022 This section describes the macros which let you control how various
4023 types of arguments are passed in registers or how they are arranged in
4026 @defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4027 A C expression that controls whether a function argument is passed
4028 in a register, and which register.
4030 The arguments are @var{cum}, which summarizes all the previous
4031 arguments; @var{mode}, the machine mode of the argument; @var{type},
4032 the data type of the argument as a tree node or 0 if that is not known
4033 (which happens for C support library functions); and @var{named},
4034 which is 1 for an ordinary argument and 0 for nameless arguments that
4035 correspond to @samp{@dots{}} in the called function's prototype.
4036 @var{type} can be an incomplete type if a syntax error has previously
4039 The value of the expression is usually either a @code{reg} RTX for the
4040 hard register in which to pass the argument, or zero to pass the
4041 argument on the stack.
4043 For machines like the VAX and 68000, where normally all arguments are
4044 pushed, zero suffices as a definition.
4046 The value of the expression can also be a @code{parallel} RTX@. This is
4047 used when an argument is passed in multiple locations. The mode of the
4048 @code{parallel} should be the mode of the entire argument. The
4049 @code{parallel} holds any number of @code{expr_list} pairs; each one
4050 describes where part of the argument is passed. In each
4051 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
4052 register in which to pass this part of the argument, and the mode of the
4053 register RTX indicates how large this part of the argument is. The
4054 second operand of the @code{expr_list} is a @code{const_int} which gives
4055 the offset in bytes into the entire argument of where this part starts.
4056 As a special exception the first @code{expr_list} in the @code{parallel}
4057 RTX may have a first operand of zero. This indicates that the entire
4058 argument is also stored on the stack.
4060 The last time this macro is called, it is called with @code{MODE ==
4061 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4062 pattern as operands 2 and 3 respectively.
4064 @cindex @file{stdarg.h} and register arguments
4065 The usual way to make the ISO library @file{stdarg.h} work on a machine
4066 where some arguments are usually passed in registers, is to cause
4067 nameless arguments to be passed on the stack instead. This is done
4068 by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0.
4070 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG}
4071 @cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG}
4072 You may use the hook @code{targetm.calls.must_pass_in_stack}
4073 in the definition of this macro to determine if this argument is of a
4074 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4075 is not defined and @code{FUNCTION_ARG} returns nonzero for such an
4076 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4077 defined, the argument will be computed in the stack and then loaded into
4081 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, const_tree @var{type})
4082 This target hook should return @code{true} if we should not pass @var{type}
4083 solely in registers. The file @file{expr.h} defines a
4084 definition that is usually appropriate, refer to @file{expr.h} for additional
4088 @defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named})
4089 Define this macro if the target machine has ``register windows'', so
4090 that the register in which a function sees an arguments is not
4091 necessarily the same as the one in which the caller passed the
4094 For such machines, @code{FUNCTION_ARG} computes the register in which
4095 the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should
4096 be defined in a similar fashion to tell the function being called
4097 where the arguments will arrive.
4099 If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG}
4100 serves both purposes.
4103 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4104 This target hook returns the number of bytes at the beginning of an
4105 argument that must be put in registers. The value must be zero for
4106 arguments that are passed entirely in registers or that are entirely
4107 pushed on the stack.
4109 On some machines, certain arguments must be passed partially in
4110 registers and partially in memory. On these machines, typically the
4111 first few words of arguments are passed in registers, and the rest
4112 on the stack. If a multi-word argument (a @code{double} or a
4113 structure) crosses that boundary, its first few words must be passed
4114 in registers and the rest must be pushed. This macro tells the
4115 compiler when this occurs, and how many bytes should go in registers.
4117 @code{FUNCTION_ARG} for these arguments should return the first
4118 register to be used by the caller for this argument; likewise
4119 @code{FUNCTION_INCOMING_ARG}, for the called function.
4122 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4123 This target hook should return @code{true} if an argument at the
4124 position indicated by @var{cum} should be passed by reference. This
4125 predicate is queried after target independent reasons for being
4126 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4128 If the hook returns true, a copy of that argument is made in memory and a
4129 pointer to the argument is passed instead of the argument itself.
4130 The pointer is passed in whatever way is appropriate for passing a pointer
4134 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (CUMULATIVE_ARGS *@var{cum}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4135 The function argument described by the parameters to this hook is
4136 known to be passed by reference. The hook should return true if the
4137 function argument should be copied by the callee instead of copied
4140 For any argument for which the hook returns true, if it can be
4141 determined that the argument is not modified, then a copy need
4144 The default version of this hook always returns false.
4147 @defmac CUMULATIVE_ARGS
4148 A C type for declaring a variable that is used as the first argument of
4149 @code{FUNCTION_ARG} and other related values. For some target machines,
4150 the type @code{int} suffices and can hold the number of bytes of
4153 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4154 arguments that have been passed on the stack. The compiler has other
4155 variables to keep track of that. For target machines on which all
4156 arguments are passed on the stack, there is no need to store anything in
4157 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4158 should not be empty, so use @code{int}.
4161 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4162 If defined, this macro is called before generating any code for a
4163 function, but after the @var{cfun} descriptor for the function has been
4164 created. The back end may use this macro to update @var{cfun} to
4165 reflect an ABI other than that which would normally be used by default.
4166 If the compiler is generating code for a compiler-generated function,
4167 @var{fndecl} may be @code{NULL}.
4170 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4171 A C statement (sans semicolon) for initializing the variable
4172 @var{cum} for the state at the beginning of the argument list. The
4173 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4174 is the tree node for the data type of the function which will receive
4175 the args, or 0 if the args are to a compiler support library function.
4176 For direct calls that are not libcalls, @var{fndecl} contain the
4177 declaration node of the function. @var{fndecl} is also set when
4178 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4179 being compiled. @var{n_named_args} is set to the number of named
4180 arguments, including a structure return address if it is passed as a
4181 parameter, when making a call. When processing incoming arguments,
4182 @var{n_named_args} is set to @minus{}1.
4184 When processing a call to a compiler support library function,
4185 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4186 contains the name of the function, as a string. @var{libname} is 0 when
4187 an ordinary C function call is being processed. Thus, each time this
4188 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4189 never both of them at once.
4192 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4193 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4194 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4195 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4196 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4197 0)} is used instead.
4200 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4201 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4202 finding the arguments for the function being compiled. If this macro is
4203 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4205 The value passed for @var{libname} is always 0, since library routines
4206 with special calling conventions are never compiled with GCC@. The
4207 argument @var{libname} exists for symmetry with
4208 @code{INIT_CUMULATIVE_ARGS}.
4209 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4210 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4213 @defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named})
4214 A C statement (sans semicolon) to update the summarizer variable
4215 @var{cum} to advance past an argument in the argument list. The
4216 values @var{mode}, @var{type} and @var{named} describe that argument.
4217 Once this is done, the variable @var{cum} is suitable for analyzing
4218 the @emph{following} argument with @code{FUNCTION_ARG}, etc.
4220 This macro need not do anything if the argument in question was passed
4221 on the stack. The compiler knows how to track the amount of stack space
4222 used for arguments without any special help.
4225 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4226 If defined, a C expression that is the number of bytes to add to the
4227 offset of the argument passed in memory. This is needed for the SPU,
4228 which passes @code{char} and @code{short} arguments in the preferred
4229 slot that is in the middle of the quad word instead of starting at the
4233 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4234 If defined, a C expression which determines whether, and in which direction,
4235 to pad out an argument with extra space. The value should be of type
4236 @code{enum direction}: either @code{upward} to pad above the argument,
4237 @code{downward} to pad below, or @code{none} to inhibit padding.
4239 The @emph{amount} of padding is always just enough to reach the next
4240 multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control
4243 This macro has a default definition which is right for most systems.
4244 For little-endian machines, the default is to pad upward. For
4245 big-endian machines, the default is to pad downward for an argument of
4246 constant size shorter than an @code{int}, and upward otherwise.
4249 @defmac PAD_VARARGS_DOWN
4250 If defined, a C expression which determines whether the default
4251 implementation of va_arg will attempt to pad down before reading the
4252 next argument, if that argument is smaller than its aligned space as
4253 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4254 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4257 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4258 Specify padding for the last element of a block move between registers and
4259 memory. @var{first} is nonzero if this is the only element. Defining this
4260 macro allows better control of register function parameters on big-endian
4261 machines, without using @code{PARALLEL} rtl. In particular,
4262 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4263 registers, as there is no longer a "wrong" part of a register; For example,
4264 a three byte aggregate may be passed in the high part of a register if so
4268 @defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type})
4269 If defined, a C expression that gives the alignment boundary, in bits,
4270 of an argument with the specified mode and type. If it is not defined,
4271 @code{PARM_BOUNDARY} is used for all arguments.
4274 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4275 A C expression that is nonzero if @var{regno} is the number of a hard
4276 register in which function arguments are sometimes passed. This does
4277 @emph{not} include implicit arguments such as the static chain and
4278 the structure-value address. On many machines, no registers can be
4279 used for this purpose since all function arguments are pushed on the
4283 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type})
4284 This hook should return true if parameter of type @var{type} are passed
4285 as two scalar parameters. By default, GCC will attempt to pack complex
4286 arguments into the target's word size. Some ABIs require complex arguments
4287 to be split and treated as their individual components. For example, on
4288 AIX64, complex floats should be passed in a pair of floating point
4289 registers, even though a complex float would fit in one 64-bit floating
4292 The default value of this hook is @code{NULL}, which is treated as always
4296 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4297 This hook returns a type node for @code{va_list} for the target.
4298 The default version of the hook returns @code{void*}.
4301 @deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl})
4302 This hook returns the va_list type of the calling convention specified by
4304 The default version of this hook returns @code{va_list_type_node}.
4307 @deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type})
4308 This hook returns the va_list type of the calling convention specified by the
4309 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4313 @deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, gimple_seq *@var{pre_p}, gimple_seq *@var{post_p})
4314 This hook performs target-specific gimplification of
4315 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4316 arguments to @code{va_arg}; the latter two are as in
4317 @code{gimplify.c:gimplify_expr}.
4320 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
4321 Define this to return nonzero if the port can handle pointers
4322 with machine mode @var{mode}. The default version of this
4323 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4326 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4327 Define this to return nonzero if the port is prepared to handle
4328 insns involving scalar mode @var{mode}. For a scalar mode to be
4329 considered supported, all the basic arithmetic and comparisons
4332 The default version of this hook returns true for any mode
4333 required to handle the basic C types (as defined by the port).
4334 Included here are the double-word arithmetic supported by the
4335 code in @file{optabs.c}.
4338 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4339 Define this to return nonzero if the port is prepared to handle
4340 insns involving vector mode @var{mode}. At the very least, it
4341 must have move patterns for this mode.
4344 @deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (enum machine_mode @var{mode})
4345 Define this to return nonzero for machine modes for which the port has
4346 small register classes. If this target hook returns nonzero for a given
4347 @var{mode}, the compiler will try to minimize the lifetime of registers
4348 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4349 In this case, the hook is expected to return nonzero if it returns nonzero
4352 On some machines, it is risky to let hard registers live across arbitrary
4353 insns. Typically, these machines have instructions that require values
4354 to be in specific registers (like an accumulator), and reload will fail
4355 if the required hard register is used for another purpose across such an
4358 Passes before reload do not know which hard registers will be used
4359 in an instruction, but the machine modes of the registers set or used in
4360 the instruction are already known. And for some machines, register
4361 classes are small for, say, integer registers but not for floating point
4362 registers. For example, the AMD x86-64 architecture requires specific
4363 registers for the legacy x86 integer instructions, but there are many
4364 SSE registers for floating point operations. On such targets, a good
4365 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4366 machine modes but zero for the SSE register classes.
4368 The default version of this hook retuns false for any mode. It is always
4369 safe to redefine this hook to return with a nonzero value. But if you
4370 unnecessarily define it, you will reduce the amount of optimizations
4371 that can be performed in some cases. If you do not define this hook
4372 to return a nonzero value when it is required, the compiler will run out
4373 of spill registers and print a fatal error message.
4377 @subsection How Scalar Function Values Are Returned
4378 @cindex return values in registers
4379 @cindex values, returned by functions
4380 @cindex scalars, returned as values
4382 This section discusses the macros that control returning scalars as
4383 values---values that can fit in registers.
4385 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
4387 Define this to return an RTX representing the place where a function
4388 returns or receives a value of data type @var{ret_type}, a tree node
4389 representing a data type. @var{fn_decl_or_type} is a tree node
4390 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4391 function being called. If @var{outgoing} is false, the hook should
4392 compute the register in which the caller will see the return value.
4393 Otherwise, the hook should return an RTX representing the place where
4394 a function returns a value.
4396 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4397 (Actually, on most machines, scalar values are returned in the same
4398 place regardless of mode.) The value of the expression is usually a
4399 @code{reg} RTX for the hard register where the return value is stored.
4400 The value can also be a @code{parallel} RTX, if the return value is in
4401 multiple places. See @code{FUNCTION_ARG} for an explanation of the
4402 @code{parallel} form. Note that the callee will populate every
4403 location specified in the @code{parallel}, but if the first element of
4404 the @code{parallel} contains the whole return value, callers will use
4405 that element as the canonical location and ignore the others. The m68k
4406 port uses this type of @code{parallel} to return pointers in both
4407 @samp{%a0} (the canonical location) and @samp{%d0}.
4409 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4410 the same promotion rules specified in @code{PROMOTE_MODE} if
4411 @var{valtype} is a scalar type.
4413 If the precise function being called is known, @var{func} is a tree
4414 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4415 pointer. This makes it possible to use a different value-returning
4416 convention for specific functions when all their calls are
4419 Some target machines have ``register windows'' so that the register in
4420 which a function returns its value is not the same as the one in which
4421 the caller sees the value. For such machines, you should return
4422 different RTX depending on @var{outgoing}.
4424 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4425 aggregate data types, because these are returned in another way. See
4426 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4429 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4430 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4431 a new target instead.
4434 @defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func})
4435 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4436 a new target instead.
4439 @defmac LIBCALL_VALUE (@var{mode})
4440 A C expression to create an RTX representing the place where a library
4441 function returns a value of mode @var{mode}.
4443 Note that ``library function'' in this context means a compiler
4444 support routine, used to perform arithmetic, whose name is known
4445 specially by the compiler and was not mentioned in the C code being
4449 @deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (enum machine_mode
4450 @var{mode}, const_rtx @var{fun})
4451 Define this hook if the back-end needs to know the name of the libcall
4452 function in order to determine where the result should be returned.
4454 The mode of the result is given by @var{mode} and the name of the called
4455 library function is given by @var{fun}. The hook should return an RTX
4456 representing the place where the library function result will be returned.
4458 If this hook is not defined, then LIBCALL_VALUE will be used.
4461 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4462 A C expression that is nonzero if @var{regno} is the number of a hard
4463 register in which the values of called function may come back.
4465 A register whose use for returning values is limited to serving as the
4466 second of a pair (for a value of type @code{double}, say) need not be
4467 recognized by this macro. So for most machines, this definition
4471 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4474 If the machine has register windows, so that the caller and the called
4475 function use different registers for the return value, this macro
4476 should recognize only the caller's register numbers.
4478 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4479 for a new target instead.
4482 @deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno})
4483 A target hook that return @code{true} if @var{regno} is the number of a hard
4484 register in which the values of called function may come back.
4486 A register whose use for returning values is limited to serving as the
4487 second of a pair (for a value of type @code{double}, say) need not be
4488 recognized by this target hook.
4490 If the machine has register windows, so that the caller and the called
4491 function use different registers for the return value, this target hook
4492 should recognize only the caller's register numbers.
4494 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4497 @defmac TARGET_ENUM_VA_LIST (@var{idx}, @var{pname}, @var{ptype})
4498 This target macro is used in function @code{c_common_nodes_and_builtins}
4499 to iterate through the target specific builtin types for va_list. The
4500 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4501 to a @code{const char *} and @var{ptype} a pointer to a @code{tree} typed
4503 The arguments @var{pname} and @var{ptype} are used to store the result of
4504 this macro and are set to the name of the va_list builtin type and its
4506 If the return value of this macro is zero, then there is no more element.
4507 Otherwise the @var{IDX} should be increased for the next call of this
4508 macro to iterate through all types.
4511 @defmac APPLY_RESULT_SIZE
4512 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4513 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4514 saving and restoring an arbitrary return value.
4517 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type})
4518 This hook should return true if values of type @var{type} are returned
4519 at the most significant end of a register (in other words, if they are
4520 padded at the least significant end). You can assume that @var{type}
4521 is returned in a register; the caller is required to check this.
4523 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4524 be able to hold the complete return value. For example, if a 1-, 2-
4525 or 3-byte structure is returned at the most significant end of a
4526 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4530 @node Aggregate Return
4531 @subsection How Large Values Are Returned
4532 @cindex aggregates as return values
4533 @cindex large return values
4534 @cindex returning aggregate values
4535 @cindex structure value address
4537 When a function value's mode is @code{BLKmode} (and in some other
4538 cases), the value is not returned according to
4539 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4540 caller passes the address of a block of memory in which the value
4541 should be stored. This address is called the @dfn{structure value
4544 This section describes how to control returning structure values in
4547 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype})
4548 This target hook should return a nonzero value to say to return the
4549 function value in memory, just as large structures are always returned.
4550 Here @var{type} will be the data type of the value, and @var{fntype}
4551 will be the type of the function doing the returning, or @code{NULL} for
4554 Note that values of mode @code{BLKmode} must be explicitly handled
4555 by this function. Also, the option @option{-fpcc-struct-return}
4556 takes effect regardless of this macro. On most systems, it is
4557 possible to leave the hook undefined; this causes a default
4558 definition to be used, whose value is the constant 1 for @code{BLKmode}
4559 values, and 0 otherwise.
4561 Do not use this hook to indicate that structures and unions should always
4562 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4566 @defmac DEFAULT_PCC_STRUCT_RETURN
4567 Define this macro to be 1 if all structure and union return values must be
4568 in memory. Since this results in slower code, this should be defined
4569 only if needed for compatibility with other compilers or with an ABI@.
4570 If you define this macro to be 0, then the conventions used for structure
4571 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4574 If not defined, this defaults to the value 1.
4577 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4578 This target hook should return the location of the structure value
4579 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4580 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4581 be @code{NULL}, for libcalls. You do not need to define this target
4582 hook if the address is always passed as an ``invisible'' first
4585 On some architectures the place where the structure value address
4586 is found by the called function is not the same place that the
4587 caller put it. This can be due to register windows, or it could
4588 be because the function prologue moves it to a different place.
4589 @var{incoming} is @code{1} or @code{2} when the location is needed in
4590 the context of the called function, and @code{0} in the context of
4593 If @var{incoming} is nonzero and the address is to be found on the
4594 stack, return a @code{mem} which refers to the frame pointer. If
4595 @var{incoming} is @code{2}, the result is being used to fetch the
4596 structure value address at the beginning of a function. If you need
4597 to emit adjusting code, you should do it at this point.
4600 @defmac PCC_STATIC_STRUCT_RETURN
4601 Define this macro if the usual system convention on the target machine
4602 for returning structures and unions is for the called function to return
4603 the address of a static variable containing the value.
4605 Do not define this if the usual system convention is for the caller to
4606 pass an address to the subroutine.
4608 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4609 nothing when you use @option{-freg-struct-return} mode.
4613 @subsection Caller-Saves Register Allocation
4615 If you enable it, GCC can save registers around function calls. This
4616 makes it possible to use call-clobbered registers to hold variables that
4617 must live across calls.
4619 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4620 A C expression to determine whether it is worthwhile to consider placing
4621 a pseudo-register in a call-clobbered hard register and saving and
4622 restoring it around each function call. The expression should be 1 when
4623 this is worth doing, and 0 otherwise.
4625 If you don't define this macro, a default is used which is good on most
4626 machines: @code{4 * @var{calls} < @var{refs}}.
4629 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4630 A C expression specifying which mode is required for saving @var{nregs}
4631 of a pseudo-register in call-clobbered hard register @var{regno}. If
4632 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4633 returned. For most machines this macro need not be defined since GCC
4634 will select the smallest suitable mode.
4637 @node Function Entry
4638 @subsection Function Entry and Exit
4639 @cindex function entry and exit
4643 This section describes the macros that output function entry
4644 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4646 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4647 If defined, a function that outputs the assembler code for entry to a
4648 function. The prologue is responsible for setting up the stack frame,
4649 initializing the frame pointer register, saving registers that must be
4650 saved, and allocating @var{size} additional bytes of storage for the
4651 local variables. @var{size} is an integer. @var{file} is a stdio
4652 stream to which the assembler code should be output.
4654 The label for the beginning of the function need not be output by this
4655 macro. That has already been done when the macro is run.
4657 @findex regs_ever_live
4658 To determine which registers to save, the macro can refer to the array
4659 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4660 @var{r} is used anywhere within the function. This implies the function
4661 prologue should save register @var{r}, provided it is not one of the
4662 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4663 @code{regs_ever_live}.)
4665 On machines that have ``register windows'', the function entry code does
4666 not save on the stack the registers that are in the windows, even if
4667 they are supposed to be preserved by function calls; instead it takes
4668 appropriate steps to ``push'' the register stack, if any non-call-used
4669 registers are used in the function.
4671 @findex frame_pointer_needed
4672 On machines where functions may or may not have frame-pointers, the
4673 function entry code must vary accordingly; it must set up the frame
4674 pointer if one is wanted, and not otherwise. To determine whether a
4675 frame pointer is in wanted, the macro can refer to the variable
4676 @code{frame_pointer_needed}. The variable's value will be 1 at run
4677 time in a function that needs a frame pointer. @xref{Elimination}.
4679 The function entry code is responsible for allocating any stack space
4680 required for the function. This stack space consists of the regions
4681 listed below. In most cases, these regions are allocated in the
4682 order listed, with the last listed region closest to the top of the
4683 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4684 the highest address if it is not defined). You can use a different order
4685 for a machine if doing so is more convenient or required for
4686 compatibility reasons. Except in cases where required by standard
4687 or by a debugger, there is no reason why the stack layout used by GCC
4688 need agree with that used by other compilers for a machine.
4691 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4692 If defined, a function that outputs assembler code at the end of a
4693 prologue. This should be used when the function prologue is being
4694 emitted as RTL, and you have some extra assembler that needs to be
4695 emitted. @xref{prologue instruction pattern}.
4698 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4699 If defined, a function that outputs assembler code at the start of an
4700 epilogue. This should be used when the function epilogue is being
4701 emitted as RTL, and you have some extra assembler that needs to be
4702 emitted. @xref{epilogue instruction pattern}.
4705 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4706 If defined, a function that outputs the assembler code for exit from a
4707 function. The epilogue is responsible for restoring the saved
4708 registers and stack pointer to their values when the function was
4709 called, and returning control to the caller. This macro takes the
4710 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4711 registers to restore are determined from @code{regs_ever_live} and
4712 @code{CALL_USED_REGISTERS} in the same way.
4714 On some machines, there is a single instruction that does all the work
4715 of returning from the function. On these machines, give that
4716 instruction the name @samp{return} and do not define the macro
4717 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4719 Do not define a pattern named @samp{return} if you want the
4720 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4721 switches to control whether return instructions or epilogues are used,
4722 define a @samp{return} pattern with a validity condition that tests the
4723 target switches appropriately. If the @samp{return} pattern's validity
4724 condition is false, epilogues will be used.
4726 On machines where functions may or may not have frame-pointers, the
4727 function exit code must vary accordingly. Sometimes the code for these
4728 two cases is completely different. To determine whether a frame pointer
4729 is wanted, the macro can refer to the variable
4730 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4731 a function that needs a frame pointer.
4733 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4734 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4735 The C variable @code{current_function_is_leaf} is nonzero for such a
4736 function. @xref{Leaf Functions}.
4738 On some machines, some functions pop their arguments on exit while
4739 others leave that for the caller to do. For example, the 68020 when
4740 given @option{-mrtd} pops arguments in functions that take a fixed
4741 number of arguments.
4743 @findex current_function_pops_args
4744 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4745 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4746 needs to know what was decided. The number of bytes of the current
4747 function's arguments that this function should pop is available in
4748 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4753 @findex current_function_pretend_args_size
4754 A region of @code{current_function_pretend_args_size} bytes of
4755 uninitialized space just underneath the first argument arriving on the
4756 stack. (This may not be at the very start of the allocated stack region
4757 if the calling sequence has pushed anything else since pushing the stack
4758 arguments. But usually, on such machines, nothing else has been pushed
4759 yet, because the function prologue itself does all the pushing.) This
4760 region is used on machines where an argument may be passed partly in
4761 registers and partly in memory, and, in some cases to support the
4762 features in @code{<stdarg.h>}.
4765 An area of memory used to save certain registers used by the function.
4766 The size of this area, which may also include space for such things as
4767 the return address and pointers to previous stack frames, is
4768 machine-specific and usually depends on which registers have been used
4769 in the function. Machines with register windows often do not require
4773 A region of at least @var{size} bytes, possibly rounded up to an allocation
4774 boundary, to contain the local variables of the function. On some machines,
4775 this region and the save area may occur in the opposite order, with the
4776 save area closer to the top of the stack.
4779 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4780 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4781 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4782 argument lists of the function. @xref{Stack Arguments}.
4785 @defmac EXIT_IGNORE_STACK
4786 Define this macro as a C expression that is nonzero if the return
4787 instruction or the function epilogue ignores the value of the stack
4788 pointer; in other words, if it is safe to delete an instruction to
4789 adjust the stack pointer before a return from the function. The
4792 Note that this macro's value is relevant only for functions for which
4793 frame pointers are maintained. It is never safe to delete a final
4794 stack adjustment in a function that has no frame pointer, and the
4795 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4798 @defmac EPILOGUE_USES (@var{regno})
4799 Define this macro as a C expression that is nonzero for registers that are
4800 used by the epilogue or the @samp{return} pattern. The stack and frame
4801 pointer registers are already assumed to be used as needed.
4804 @defmac EH_USES (@var{regno})
4805 Define this macro as a C expression that is nonzero for registers that are
4806 used by the exception handling mechanism, and so should be considered live
4807 on entry to an exception edge.
4810 @defmac DELAY_SLOTS_FOR_EPILOGUE
4811 Define this macro if the function epilogue contains delay slots to which
4812 instructions from the rest of the function can be ``moved''. The
4813 definition should be a C expression whose value is an integer
4814 representing the number of delay slots there.
4817 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4818 A C expression that returns 1 if @var{insn} can be placed in delay
4819 slot number @var{n} of the epilogue.
4821 The argument @var{n} is an integer which identifies the delay slot now
4822 being considered (since different slots may have different rules of
4823 eligibility). It is never negative and is always less than the number
4824 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4825 If you reject a particular insn for a given delay slot, in principle, it
4826 may be reconsidered for a subsequent delay slot. Also, other insns may
4827 (at least in principle) be considered for the so far unfilled delay
4830 @findex current_function_epilogue_delay_list
4831 @findex final_scan_insn
4832 The insns accepted to fill the epilogue delay slots are put in an RTL
4833 list made with @code{insn_list} objects, stored in the variable
4834 @code{current_function_epilogue_delay_list}. The insn for the first
4835 delay slot comes first in the list. Your definition of the macro
4836 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4837 outputting the insns in this list, usually by calling
4838 @code{final_scan_insn}.
4840 You need not define this macro if you did not define
4841 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4844 @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})
4845 A function that outputs the assembler code for a thunk
4846 function, used to implement C++ virtual function calls with multiple
4847 inheritance. The thunk acts as a wrapper around a virtual function,
4848 adjusting the implicit object parameter before handing control off to
4851 First, emit code to add the integer @var{delta} to the location that
4852 contains the incoming first argument. Assume that this argument
4853 contains a pointer, and is the one used to pass the @code{this} pointer
4854 in C++. This is the incoming argument @emph{before} the function prologue,
4855 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4856 all other incoming arguments.
4858 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4859 made after adding @code{delta}. In particular, if @var{p} is the
4860 adjusted pointer, the following adjustment should be made:
4863 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4866 After the additions, emit code to jump to @var{function}, which is a
4867 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4868 not touch the return address. Hence returning from @var{FUNCTION} will
4869 return to whoever called the current @samp{thunk}.
4871 The effect must be as if @var{function} had been called directly with
4872 the adjusted first argument. This macro is responsible for emitting all
4873 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4874 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4876 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4877 have already been extracted from it.) It might possibly be useful on
4878 some targets, but probably not.
4880 If you do not define this macro, the target-independent code in the C++
4881 front end will generate a less efficient heavyweight thunk that calls
4882 @var{function} instead of jumping to it. The generic approach does
4883 not support varargs.
4886 @deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (const_tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, const_tree @var{function})
4887 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4888 to output the assembler code for the thunk function specified by the
4889 arguments it is passed, and false otherwise. In the latter case, the
4890 generic approach will be used by the C++ front end, with the limitations
4895 @subsection Generating Code for Profiling
4896 @cindex profiling, code generation
4898 These macros will help you generate code for profiling.
4900 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4901 A C statement or compound statement to output to @var{file} some
4902 assembler code to call the profiling subroutine @code{mcount}.
4905 The details of how @code{mcount} expects to be called are determined by
4906 your operating system environment, not by GCC@. To figure them out,
4907 compile a small program for profiling using the system's installed C
4908 compiler and look at the assembler code that results.
4910 Older implementations of @code{mcount} expect the address of a counter
4911 variable to be loaded into some register. The name of this variable is
4912 @samp{LP} followed by the number @var{labelno}, so you would generate
4913 the name using @samp{LP%d} in a @code{fprintf}.
4916 @defmac PROFILE_HOOK
4917 A C statement or compound statement to output to @var{file} some assembly
4918 code to call the profiling subroutine @code{mcount} even the target does
4919 not support profiling.
4922 @defmac NO_PROFILE_COUNTERS
4923 Define this macro to be an expression with a nonzero value if the
4924 @code{mcount} subroutine on your system does not need a counter variable
4925 allocated for each function. This is true for almost all modern
4926 implementations. If you define this macro, you must not use the
4927 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4930 @defmac PROFILE_BEFORE_PROLOGUE
4931 Define this macro if the code for function profiling should come before
4932 the function prologue. Normally, the profiling code comes after.
4936 @subsection Permitting tail calls
4939 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4940 True if it is ok to do sibling call optimization for the specified
4941 call expression @var{exp}. @var{decl} will be the called function,
4942 or @code{NULL} if this is an indirect call.
4944 It is not uncommon for limitations of calling conventions to prevent
4945 tail calls to functions outside the current unit of translation, or
4946 during PIC compilation. The hook is used to enforce these restrictions,
4947 as the @code{sibcall} md pattern can not fail, or fall over to a
4948 ``normal'' call. The criteria for successful sibling call optimization
4949 may vary greatly between different architectures.
4952 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs})
4953 Add any hard registers to @var{regs} that are live on entry to the
4954 function. This hook only needs to be defined to provide registers that
4955 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4956 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4957 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4958 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4961 @node Stack Smashing Protection
4962 @subsection Stack smashing protection
4963 @cindex stack smashing protection
4965 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4966 This hook returns a @code{DECL} node for the external variable to use
4967 for the stack protection guard. This variable is initialized by the
4968 runtime to some random value and is used to initialize the guard value
4969 that is placed at the top of the local stack frame. The type of this
4970 variable must be @code{ptr_type_node}.
4972 The default version of this hook creates a variable called
4973 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4976 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4977 This hook returns a tree expression that alerts the runtime that the
4978 stack protect guard variable has been modified. This expression should
4979 involve a call to a @code{noreturn} function.
4981 The default version of this hook invokes a function called
4982 @samp{__stack_chk_fail}, taking no arguments. This function is
4983 normally defined in @file{libgcc2.c}.
4987 @section Implementing the Varargs Macros
4988 @cindex varargs implementation
4990 GCC comes with an implementation of @code{<varargs.h>} and
4991 @code{<stdarg.h>} that work without change on machines that pass arguments
4992 on the stack. Other machines require their own implementations of
4993 varargs, and the two machine independent header files must have
4994 conditionals to include it.
4996 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
4997 the calling convention for @code{va_start}. The traditional
4998 implementation takes just one argument, which is the variable in which
4999 to store the argument pointer. The ISO implementation of
5000 @code{va_start} takes an additional second argument. The user is
5001 supposed to write the last named argument of the function here.
5003 However, @code{va_start} should not use this argument. The way to find
5004 the end of the named arguments is with the built-in functions described
5007 @defmac __builtin_saveregs ()
5008 Use this built-in function to save the argument registers in memory so
5009 that the varargs mechanism can access them. Both ISO and traditional
5010 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5011 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5013 On some machines, @code{__builtin_saveregs} is open-coded under the
5014 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5015 other machines, it calls a routine written in assembler language,
5016 found in @file{libgcc2.c}.
5018 Code generated for the call to @code{__builtin_saveregs} appears at the
5019 beginning of the function, as opposed to where the call to
5020 @code{__builtin_saveregs} is written, regardless of what the code is.
5021 This is because the registers must be saved before the function starts
5022 to use them for its own purposes.
5023 @c i rewrote the first sentence above to fix an overfull hbox. --mew
5027 @defmac __builtin_args_info (@var{category})
5028 Use this built-in function to find the first anonymous arguments in
5031 In general, a machine may have several categories of registers used for
5032 arguments, each for a particular category of data types. (For example,
5033 on some machines, floating-point registers are used for floating-point
5034 arguments while other arguments are passed in the general registers.)
5035 To make non-varargs functions use the proper calling convention, you
5036 have defined the @code{CUMULATIVE_ARGS} data type to record how many
5037 registers in each category have been used so far
5039 @code{__builtin_args_info} accesses the same data structure of type
5040 @code{CUMULATIVE_ARGS} after the ordinary argument layout is finished
5041 with it, with @var{category} specifying which word to access. Thus, the
5042 value indicates the first unused register in a given category.
5044 Normally, you would use @code{__builtin_args_info} in the implementation
5045 of @code{va_start}, accessing each category just once and storing the
5046 value in the @code{va_list} object. This is because @code{va_list} will
5047 have to update the values, and there is no way to alter the
5048 values accessed by @code{__builtin_args_info}.
5051 @defmac __builtin_next_arg (@var{lastarg})
5052 This is the equivalent of @code{__builtin_args_info}, for stack
5053 arguments. It returns the address of the first anonymous stack
5054 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5055 returns the address of the location above the first anonymous stack
5056 argument. Use it in @code{va_start} to initialize the pointer for
5057 fetching arguments from the stack. Also use it in @code{va_start} to
5058 verify that the second parameter @var{lastarg} is the last named argument
5059 of the current function.
5062 @defmac __builtin_classify_type (@var{object})
5063 Since each machine has its own conventions for which data types are
5064 passed in which kind of register, your implementation of @code{va_arg}
5065 has to embody these conventions. The easiest way to categorize the
5066 specified data type is to use @code{__builtin_classify_type} together
5067 with @code{sizeof} and @code{__alignof__}.
5069 @code{__builtin_classify_type} ignores the value of @var{object},
5070 considering only its data type. It returns an integer describing what
5071 kind of type that is---integer, floating, pointer, structure, and so on.
5073 The file @file{typeclass.h} defines an enumeration that you can use to
5074 interpret the values of @code{__builtin_classify_type}.
5077 These machine description macros help implement varargs:
5079 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
5080 If defined, this hook produces the machine-specific code for a call to
5081 @code{__builtin_saveregs}. This code will be moved to the very
5082 beginning of the function, before any parameter access are made. The
5083 return value of this function should be an RTX that contains the value
5084 to use as the return of @code{__builtin_saveregs}.
5087 @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})
5088 This target hook offers an alternative to using
5089 @code{__builtin_saveregs} and defining the hook
5090 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5091 register arguments into the stack so that all the arguments appear to
5092 have been passed consecutively on the stack. Once this is done, you can
5093 use the standard implementation of varargs that works for machines that
5094 pass all their arguments on the stack.
5096 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5097 structure, containing the values that are obtained after processing the
5098 named arguments. The arguments @var{mode} and @var{type} describe the
5099 last named argument---its machine mode and its data type as a tree node.
5101 The target hook should do two things: first, push onto the stack all the
5102 argument registers @emph{not} used for the named arguments, and second,
5103 store the size of the data thus pushed into the @code{int}-valued
5104 variable pointed to by @var{pretend_args_size}. The value that you
5105 store here will serve as additional offset for setting up the stack
5108 Because you must generate code to push the anonymous arguments at
5109 compile time without knowing their data types,
5110 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5111 have just a single category of argument register and use it uniformly
5114 If the argument @var{second_time} is nonzero, it means that the
5115 arguments of the function are being analyzed for the second time. This
5116 happens for an inline function, which is not actually compiled until the
5117 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5118 not generate any instructions in this case.
5121 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca})
5122 Define this hook to return @code{true} if the location where a function
5123 argument is passed depends on whether or not it is a named argument.
5125 This hook controls how the @var{named} argument to @code{FUNCTION_ARG}
5126 is set for varargs and stdarg functions. If this hook returns
5127 @code{true}, the @var{named} argument is always true for named
5128 arguments, and false for unnamed arguments. If it returns @code{false},
5129 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5130 then all arguments are treated as named. Otherwise, all named arguments
5131 except the last are treated as named.
5133 You need not define this hook if it always returns @code{false}.
5136 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (CUMULATIVE_ARGS *@var{ca})
5137 If you need to conditionally change ABIs so that one works with
5138 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5139 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5140 defined, then define this hook to return @code{true} if
5141 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5142 Otherwise, you should not define this hook.
5146 @section Trampolines for Nested Functions
5147 @cindex trampolines for nested functions
5148 @cindex nested functions, trampolines for
5150 A @dfn{trampoline} is a small piece of code that is created at run time
5151 when the address of a nested function is taken. It normally resides on
5152 the stack, in the stack frame of the containing function. These macros
5153 tell GCC how to generate code to allocate and initialize a
5156 The instructions in the trampoline must do two things: load a constant
5157 address into the static chain register, and jump to the real address of
5158 the nested function. On CISC machines such as the m68k, this requires
5159 two instructions, a move immediate and a jump. Then the two addresses
5160 exist in the trampoline as word-long immediate operands. On RISC
5161 machines, it is often necessary to load each address into a register in
5162 two parts. Then pieces of each address form separate immediate
5165 The code generated to initialize the trampoline must store the variable
5166 parts---the static chain value and the function address---into the
5167 immediate operands of the instructions. On a CISC machine, this is
5168 simply a matter of copying each address to a memory reference at the
5169 proper offset from the start of the trampoline. On a RISC machine, it
5170 may be necessary to take out pieces of the address and store them
5173 @deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f})
5174 This hook is called by @code{assemble_trampoline_template} to output,
5175 on the stream @var{f}, assembler code for a block of data that contains
5176 the constant parts of a trampoline. This code should not include a
5177 label---the label is taken care of automatically.
5179 If you do not define this hook, it means no template is needed
5180 for the target. Do not define this hook on systems where the block move
5181 code to copy the trampoline into place would be larger than the code
5182 to generate it on the spot.
5185 @defmac TRAMPOLINE_SECTION
5186 Return the section into which the trampoline template is to be placed
5187 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5190 @defmac TRAMPOLINE_SIZE
5191 A C expression for the size in bytes of the trampoline, as an integer.
5194 @defmac TRAMPOLINE_ALIGNMENT
5195 Alignment required for trampolines, in bits.
5197 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5198 is used for aligning trampolines.
5201 @deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain})
5202 This hook is called to initialize a trampoline.
5203 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5204 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5205 RTX for the static chain value that should be passed to the function
5208 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5209 first thing this hook should do is emit a block move into @var{m_tramp}
5210 from the memory block returned by @code{assemble_trampoline_template}.
5211 Note that the block move need only cover the constant parts of the
5212 trampoline. If the target isolates the variable parts of the trampoline
5213 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5215 If the target requires any other actions, such as flushing caches or
5216 enabling stack execution, these actions should be performed after
5217 initializing the trampoline proper.
5220 @deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr})
5221 This hook should perform any machine-specific adjustment in
5222 the address of the trampoline. Its argument contains the address of the
5223 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5224 the address to be used for a function call should be different from the
5225 address at which the template was stored, the different address should
5226 be returned; otherwise @var{addr} should be returned unchanged.
5227 If this hook is not defined, @var{addr} will be used for function calls.
5230 Implementing trampolines is difficult on many machines because they have
5231 separate instruction and data caches. Writing into a stack location
5232 fails to clear the memory in the instruction cache, so when the program
5233 jumps to that location, it executes the old contents.
5235 Here are two possible solutions. One is to clear the relevant parts of
5236 the instruction cache whenever a trampoline is set up. The other is to
5237 make all trampolines identical, by having them jump to a standard
5238 subroutine. The former technique makes trampoline execution faster; the
5239 latter makes initialization faster.
5241 To clear the instruction cache when a trampoline is initialized, define
5242 the following macro.
5244 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5245 If defined, expands to a C expression clearing the @emph{instruction
5246 cache} in the specified interval. The definition of this macro would
5247 typically be a series of @code{asm} statements. Both @var{beg} and
5248 @var{end} are both pointer expressions.
5251 The operating system may also require the stack to be made executable
5252 before calling the trampoline. To implement this requirement, define
5253 the following macro.
5255 @defmac ENABLE_EXECUTE_STACK
5256 Define this macro if certain operations must be performed before executing
5257 code located on the stack. The macro should expand to a series of C
5258 file-scope constructs (e.g.@: functions) and provide a unique entry point
5259 named @code{__enable_execute_stack}. The target is responsible for
5260 emitting calls to the entry point in the code, for example from the
5261 @code{TARGET_TRAMPOLINE_INIT} hook.
5264 To use a standard subroutine, define the following macro. In addition,
5265 you must make sure that the instructions in a trampoline fill an entire
5266 cache line with identical instructions, or else ensure that the
5267 beginning of the trampoline code is always aligned at the same point in
5268 its cache line. Look in @file{m68k.h} as a guide.
5270 @defmac TRANSFER_FROM_TRAMPOLINE
5271 Define this macro if trampolines need a special subroutine to do their
5272 work. The macro should expand to a series of @code{asm} statements
5273 which will be compiled with GCC@. They go in a library function named
5274 @code{__transfer_from_trampoline}.
5276 If you need to avoid executing the ordinary prologue code of a compiled
5277 C function when you jump to the subroutine, you can do so by placing a
5278 special label of your own in the assembler code. Use one @code{asm}
5279 statement to generate an assembler label, and another to make the label
5280 global. Then trampolines can use that label to jump directly to your
5281 special assembler code.
5285 @section Implicit Calls to Library Routines
5286 @cindex library subroutine names
5287 @cindex @file{libgcc.a}
5289 @c prevent bad page break with this line
5290 Here is an explanation of implicit calls to library routines.
5292 @defmac DECLARE_LIBRARY_RENAMES
5293 This macro, if defined, should expand to a piece of C code that will get
5294 expanded when compiling functions for libgcc.a. It can be used to
5295 provide alternate names for GCC's internal library functions if there
5296 are ABI-mandated names that the compiler should provide.
5299 @findex set_optab_libfunc
5300 @findex init_one_libfunc
5301 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5302 This hook should declare additional library routines or rename
5303 existing ones, using the functions @code{set_optab_libfunc} and
5304 @code{init_one_libfunc} defined in @file{optabs.c}.
5305 @code{init_optabs} calls this macro after initializing all the normal
5308 The default is to do nothing. Most ports don't need to define this hook.
5311 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5312 This macro should return @code{true} if the library routine that
5313 implements the floating point comparison operator @var{comparison} in
5314 mode @var{mode} will return a boolean, and @var{false} if it will
5317 GCC's own floating point libraries return tristates from the
5318 comparison operators, so the default returns false always. Most ports
5319 don't need to define this macro.
5322 @defmac TARGET_LIB_INT_CMP_BIASED
5323 This macro should evaluate to @code{true} if the integer comparison
5324 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5325 operand is smaller than the second, 1 to indicate that they are equal,
5326 and 2 to indicate that the first operand is greater than the second.
5327 If this macro evaluates to @code{false} the comparison functions return
5328 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5329 in @file{libgcc.a}, you do not need to define this macro.
5332 @cindex US Software GOFAST, floating point emulation library
5333 @cindex floating point emulation library, US Software GOFAST
5334 @cindex GOFAST, floating point emulation library
5335 @findex gofast_maybe_init_libfuncs
5336 @defmac US_SOFTWARE_GOFAST
5337 Define this macro if your system C library uses the US Software GOFAST
5338 library to provide floating point emulation.
5340 In addition to defining this macro, your architecture must set
5341 @code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or
5342 else call that function from its version of that hook. It is defined
5343 in @file{config/gofast.h}, which must be included by your
5344 architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for
5347 If this macro is defined, the
5348 @code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return
5349 false for @code{SFmode} and @code{DFmode} comparisons.
5352 @cindex @code{EDOM}, implicit usage
5355 The value of @code{EDOM} on the target machine, as a C integer constant
5356 expression. If you don't define this macro, GCC does not attempt to
5357 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5358 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5361 If you do not define @code{TARGET_EDOM}, then compiled code reports
5362 domain errors by calling the library function and letting it report the
5363 error. If mathematical functions on your system use @code{matherr} when
5364 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5365 that @code{matherr} is used normally.
5368 @cindex @code{errno}, implicit usage
5369 @defmac GEN_ERRNO_RTX
5370 Define this macro as a C expression to create an rtl expression that
5371 refers to the global ``variable'' @code{errno}. (On certain systems,
5372 @code{errno} may not actually be a variable.) If you don't define this
5373 macro, a reasonable default is used.
5376 @cindex C99 math functions, implicit usage
5377 @defmac TARGET_C99_FUNCTIONS
5378 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5379 @code{sinf} and similarly for other functions defined by C99 standard. The
5380 default is zero because a number of existing systems lack support for these
5381 functions in their runtime so this macro needs to be redefined to one on
5382 systems that do support the C99 runtime.
5385 @cindex sincos math function, implicit usage
5386 @defmac TARGET_HAS_SINCOS
5387 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5388 and @code{cos} with the same argument to a call to @code{sincos}. The
5389 default is zero. The target has to provide the following functions:
5391 void sincos(double x, double *sin, double *cos);
5392 void sincosf(float x, float *sin, float *cos);
5393 void sincosl(long double x, long double *sin, long double *cos);
5397 @defmac NEXT_OBJC_RUNTIME
5398 Define this macro to generate code for Objective-C message sending using
5399 the calling convention of the NeXT system. This calling convention
5400 involves passing the object, the selector and the method arguments all
5401 at once to the method-lookup library function.
5403 The default calling convention passes just the object and the selector
5404 to the lookup function, which returns a pointer to the method.
5407 @node Addressing Modes
5408 @section Addressing Modes
5409 @cindex addressing modes
5411 @c prevent bad page break with this line
5412 This is about addressing modes.
5414 @defmac HAVE_PRE_INCREMENT
5415 @defmacx HAVE_PRE_DECREMENT
5416 @defmacx HAVE_POST_INCREMENT
5417 @defmacx HAVE_POST_DECREMENT
5418 A C expression that is nonzero if the machine supports pre-increment,
5419 pre-decrement, post-increment, or post-decrement addressing respectively.
5422 @defmac HAVE_PRE_MODIFY_DISP
5423 @defmacx HAVE_POST_MODIFY_DISP
5424 A C expression that is nonzero if the machine supports pre- or
5425 post-address side-effect generation involving constants other than
5426 the size of the memory operand.
5429 @defmac HAVE_PRE_MODIFY_REG
5430 @defmacx HAVE_POST_MODIFY_REG
5431 A C expression that is nonzero if the machine supports pre- or
5432 post-address side-effect generation involving a register displacement.
5435 @defmac CONSTANT_ADDRESS_P (@var{x})
5436 A C expression that is 1 if the RTX @var{x} is a constant which
5437 is a valid address. On most machines the default definition of
5438 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5439 is acceptable, but a few machines are more restrictive as to which
5440 constant addresses are supported.
5443 @defmac CONSTANT_P (@var{x})
5444 @code{CONSTANT_P}, which is defined by target-independent code,
5445 accepts integer-values expressions whose values are not explicitly
5446 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5447 expressions and @code{const} arithmetic expressions, in addition to
5448 @code{const_int} and @code{const_double} expressions.
5451 @defmac MAX_REGS_PER_ADDRESS
5452 A number, the maximum number of registers that can appear in a valid
5453 memory address. Note that it is up to you to specify a value equal to
5454 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5458 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (enum machine_mode @var{mode}, rtx @var{x}, bool @var{strict})
5459 A function that returns whether @var{x} (an RTX) is a legitimate memory
5460 address on the target machine for a memory operand of mode @var{mode}.
5462 Legitimate addresses are defined in two variants: a strict variant and a
5463 non-strict one. The @var{strict} parameter chooses which variant is
5464 desired by the caller.
5466 The strict variant is used in the reload pass. It must be defined so
5467 that any pseudo-register that has not been allocated a hard register is
5468 considered a memory reference. This is because in contexts where some
5469 kind of register is required, a pseudo-register with no hard register
5470 must be rejected. For non-hard registers, the strict variant should look
5471 up the @code{reg_renumber} array; it should then proceed using the hard
5472 register number in the array, or treat the pseudo as a memory reference
5473 if the array holds @code{-1}.
5475 The non-strict variant is used in other passes. It must be defined to
5476 accept all pseudo-registers in every context where some kind of
5477 register is required.
5479 Normally, constant addresses which are the sum of a @code{symbol_ref}
5480 and an integer are stored inside a @code{const} RTX to mark them as
5481 constant. Therefore, there is no need to recognize such sums
5482 specifically as legitimate addresses. Normally you would simply
5483 recognize any @code{const} as legitimate.
5485 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5486 sums that are not marked with @code{const}. It assumes that a naked
5487 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5488 naked constant sums as illegitimate addresses, so that none of them will
5489 be given to @code{PRINT_OPERAND_ADDRESS}.
5491 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5492 On some machines, whether a symbolic address is legitimate depends on
5493 the section that the address refers to. On these machines, define the
5494 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5495 into the @code{symbol_ref}, and then check for it here. When you see a
5496 @code{const}, you will have to look inside it to find the
5497 @code{symbol_ref} in order to determine the section. @xref{Assembler
5500 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5501 Some ports are still using a deprecated legacy substitute for
5502 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5506 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5510 and should @code{goto @var{label}} if the address @var{x} is a valid
5511 address on the target machine for a memory operand of mode @var{mode}.
5512 Whether the strict or non-strict variants are desired is defined by
5513 the @code{REG_OK_STRICT} macro introduced earlier in this section.
5514 Using the hook is usually simpler because it limits the number of
5515 files that are recompiled when changes are made.
5518 @defmac TARGET_MEM_CONSTRAINT
5519 A single character to be used instead of the default @code{'m'}
5520 character for general memory addresses. This defines the constraint
5521 letter which matches the memory addresses accepted by
5522 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5523 support new address formats in your back end without changing the
5524 semantics of the @code{'m'} constraint. This is necessary in order to
5525 preserve functionality of inline assembly constructs using the
5526 @code{'m'} constraint.
5529 @defmac FIND_BASE_TERM (@var{x})
5530 A C expression to determine the base term of address @var{x},
5531 or to provide a simplified version of @var{x} from which @file{alias.c}
5532 can easily find the base term. This macro is used in only two places:
5533 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5535 It is always safe for this macro to not be defined. It exists so
5536 that alias analysis can understand machine-dependent addresses.
5538 The typical use of this macro is to handle addresses containing
5539 a label_ref or symbol_ref within an UNSPEC@.
5542 @deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, enum machine_mode @var{mode})
5543 This hook is given an invalid memory address @var{x} for an
5544 operand of mode @var{mode} and should try to return a valid memory
5547 @findex break_out_memory_refs
5548 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5549 and @var{oldx} will be the operand that was given to that function to produce
5552 The code of the hook should not alter the substructure of
5553 @var{x}. If it transforms @var{x} into a more legitimate form, it
5554 should return the new @var{x}.
5556 It is not necessary for this hook to come up with a legitimate address.
5557 The compiler has standard ways of doing so in all cases. In fact, it
5558 is safe to omit this hook or make it return @var{x} if it cannot find
5559 a valid way to legitimize the address. But often a machine-dependent
5560 strategy can generate better code.
5563 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5564 A C compound statement that attempts to replace @var{x}, which is an address
5565 that needs reloading, with a valid memory address for an operand of mode
5566 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5567 It is not necessary to define this macro, but it might be useful for
5568 performance reasons.
5570 For example, on the i386, it is sometimes possible to use a single
5571 reload register instead of two by reloading a sum of two pseudo
5572 registers into a register. On the other hand, for number of RISC
5573 processors offsets are limited so that often an intermediate address
5574 needs to be generated in order to address a stack slot. By defining
5575 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5576 generated for adjacent some stack slots can be made identical, and thus
5579 @emph{Note}: This macro should be used with caution. It is necessary
5580 to know something of how reload works in order to effectively use this,
5581 and it is quite easy to produce macros that build in too much knowledge
5582 of reload internals.
5584 @emph{Note}: This macro must be able to reload an address created by a
5585 previous invocation of this macro. If it fails to handle such addresses
5586 then the compiler may generate incorrect code or abort.
5589 The macro definition should use @code{push_reload} to indicate parts that
5590 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5591 suitable to be passed unaltered to @code{push_reload}.
5593 The code generated by this macro must not alter the substructure of
5594 @var{x}. If it transforms @var{x} into a more legitimate form, it
5595 should assign @var{x} (which will always be a C variable) a new value.
5596 This also applies to parts that you change indirectly by calling
5599 @findex strict_memory_address_p
5600 The macro definition may use @code{strict_memory_address_p} to test if
5601 the address has become legitimate.
5604 If you want to change only a part of @var{x}, one standard way of doing
5605 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5606 single level of rtl. Thus, if the part to be changed is not at the
5607 top level, you'll need to replace first the top level.
5608 It is not necessary for this macro to come up with a legitimate
5609 address; but often a machine-dependent strategy can generate better code.
5612 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5613 A C statement or compound statement with a conditional @code{goto
5614 @var{label};} executed if memory address @var{x} (an RTX) can have
5615 different meanings depending on the machine mode of the memory
5616 reference it is used for or if the address is valid for some modes
5619 Autoincrement and autodecrement addresses typically have mode-dependent
5620 effects because the amount of the increment or decrement is the size
5621 of the operand being addressed. Some machines have other mode-dependent
5622 addresses. Many RISC machines have no mode-dependent addresses.
5624 You may assume that @var{addr} is a valid address for the machine.
5627 @defmac LEGITIMATE_CONSTANT_P (@var{x})
5628 A C expression that is nonzero if @var{x} is a legitimate constant for
5629 an immediate operand on the target machine. You can assume that
5630 @var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact,
5631 @samp{1} is a suitable definition for this macro on machines where
5632 anything @code{CONSTANT_P} is valid.
5635 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5636 This hook is used to undo the possibly obfuscating effects of the
5637 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5638 macros. Some backend implementations of these macros wrap symbol
5639 references inside an @code{UNSPEC} rtx to represent PIC or similar
5640 addressing modes. This target hook allows GCC's optimizers to understand
5641 the semantics of these opaque @code{UNSPEC}s by converting them back
5642 into their original form.
5645 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (rtx @var{x})
5646 This hook should return true if @var{x} is of a form that cannot (or
5647 should not) be spilled to the constant pool. The default version of
5648 this hook returns false.
5650 The primary reason to define this hook is to prevent reload from
5651 deciding that a non-legitimate constant would be better reloaded
5652 from the constant pool instead of spilling and reloading a register
5653 holding the constant. This restriction is often true of addresses
5654 of TLS symbols for various targets.
5657 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, const_rtx @var{x})
5658 This hook should return true if pool entries for constant @var{x} can
5659 be placed in an @code{object_block} structure. @var{mode} is the mode
5662 The default version returns false for all constants.
5665 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (unsigned @var{fn}, bool @var{md_fn}, bool @var{sqrt})
5666 This hook should return the DECL of a function that implements reciprocal of
5667 the builtin function with builtin function code @var{fn}, or
5668 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5669 when @var{fn} is a code of a machine-dependent builtin function. When
5670 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5671 of a square root function are performed, and only reciprocals of @code{sqrt}
5675 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5676 This hook should return the DECL of a function @var{f} that given an
5677 address @var{addr} as an argument returns a mask @var{m} that can be
5678 used to extract from two vectors the relevant data that resides in
5679 @var{addr} in case @var{addr} is not properly aligned.
5681 The autovectorizer, when vectorizing a load operation from an address
5682 @var{addr} that may be unaligned, will generate two vector loads from
5683 the two aligned addresses around @var{addr}. It then generates a
5684 @code{REALIGN_LOAD} operation to extract the relevant data from the
5685 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5686 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5687 the third argument, @var{OFF}, defines how the data will be extracted
5688 from these two vectors: if @var{OFF} is 0, then the returned vector is
5689 @var{v2}; otherwise, the returned vector is composed from the last
5690 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5691 @var{OFF} elements of @var{v2}.
5693 If this hook is defined, the autovectorizer will generate a call
5694 to @var{f} (using the DECL tree that this hook returns) and will
5695 use the return value of @var{f} as the argument @var{OFF} to
5696 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5697 should comply with the semantics expected by @code{REALIGN_LOAD}
5699 If this hook is not defined, then @var{addr} will be used as
5700 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5701 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5704 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN (tree @var{x})
5705 This hook should return the DECL of a function @var{f} that implements
5706 widening multiplication of the even elements of two input vectors of type @var{x}.
5708 If this hook is defined, the autovectorizer will use it along with the
5709 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5710 widening multiplication in cases that the order of the results does not have to be
5711 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5712 @code{widen_mult_hi/lo} idioms will be used.
5715 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD (tree @var{x})
5716 This hook should return the DECL of a function @var{f} that implements
5717 widening multiplication of the odd elements of two input vectors of type @var{x}.
5719 If this hook is defined, the autovectorizer will use it along with the
5720 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5721 widening multiplication in cases that the order of the results does not have to be
5722 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5723 @code{widen_mult_hi/lo} idioms will be used.
5726 @deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (bool @var{runtime_test})
5727 Returns the cost to be added to the overhead involved with executing
5728 the vectorized version of a loop.
5731 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed})
5732 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5735 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VEC_PERM (tree @var{type}, tree *@var{mask_element_type})
5736 Target builtin that implements vector permute.
5739 @deftypefn {Target Hook} bool TARGET_VECTORIZE_BUILTIN_VEC_PERM_OK (tree @var{vec_type}, tree @var{mask})
5740 Return true if a vector created for @code{builtin_vec_perm} is valid.
5743 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (unsigned @var{code}, tree @var{dest_type}, tree @var{src_type})
5744 This hook should return the DECL of a function that implements conversion of the
5745 input vector of type @var{src_type} to type @var{dest_type}.
5746 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5747 specifies how the conversion is to be applied
5748 (truncation, rounding, etc.).
5750 If this hook is defined, the autovectorizer will use the
5751 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5752 conversion. Otherwise, it will return @code{NULL_TREE}.
5755 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in})
5756 This hook should return the decl of a function that implements the
5757 vectorized variant of the builtin function with builtin function code
5758 @var{code} or @code{NULL_TREE} if such a function is not available.
5759 The value of @var{fndecl} is the builtin function declaration. The
5760 return type of the vectorized function shall be of vector type
5761 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5764 @deftypefn {Target Hook} bool TARGET_SUPPORT_VECTOR_MISALIGNMENT (enum machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed})
5765 This hook should return true if the target supports misaligned vector
5766 store/load of a specific factor denoted in the @var{misalignment}
5767 parameter. The vector store/load should be of machine mode @var{mode} and
5768 the elements in the vectors should be of type @var{type}. @var{is_packed}
5769 parameter is true if the memory access is defined in a packed struct.
5772 @node Anchored Addresses
5773 @section Anchored Addresses
5774 @cindex anchored addresses
5775 @cindex @option{-fsection-anchors}
5777 GCC usually addresses every static object as a separate entity.
5778 For example, if we have:
5782 int foo (void) @{ return a + b + c; @}
5785 the code for @code{foo} will usually calculate three separate symbolic
5786 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5787 it would be better to calculate just one symbolic address and access
5788 the three variables relative to it. The equivalent pseudocode would
5794 register int *xr = &x;
5795 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5799 (which isn't valid C). We refer to shared addresses like @code{x} as
5800 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5802 The hooks below describe the target properties that GCC needs to know
5803 in order to make effective use of section anchors. It won't use
5804 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5805 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5807 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5808 The minimum offset that should be applied to a section anchor.
5809 On most targets, it should be the smallest offset that can be
5810 applied to a base register while still giving a legitimate address
5811 for every mode. The default value is 0.
5814 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5815 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5816 offset that should be applied to section anchors. The default
5820 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5821 Write the assembly code to define section anchor @var{x}, which is a
5822 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5823 The hook is called with the assembly output position set to the beginning
5824 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5826 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5827 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5828 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5829 is @code{NULL}, which disables the use of section anchors altogether.
5832 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x})
5833 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5834 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5835 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5837 The default version is correct for most targets, but you might need to
5838 intercept this hook to handle things like target-specific attributes
5839 or target-specific sections.
5842 @node Condition Code
5843 @section Condition Code Status
5844 @cindex condition code status
5846 The macros in this section can be split in two families, according to the
5847 two ways of representing condition codes in GCC.
5849 The first representation is the so called @code{(cc0)} representation
5850 (@pxref{Jump Patterns}), where all instructions can have an implicit
5851 clobber of the condition codes. The second is the condition code
5852 register representation, which provides better schedulability for
5853 architectures that do have a condition code register, but on which
5854 most instructions do not affect it. The latter category includes
5857 The implicit clobbering poses a strong restriction on the placement of
5858 the definition and use of the condition code, which need to be in adjacent
5859 insns for machines using @code{(cc0)}. This can prevent important
5860 optimizations on some machines. For example, on the IBM RS/6000, there
5861 is a delay for taken branches unless the condition code register is set
5862 three instructions earlier than the conditional branch. The instruction
5863 scheduler cannot perform this optimization if it is not permitted to
5864 separate the definition and use of the condition code register.
5866 For this reason, it is possible and suggested to use a register to
5867 represent the condition code for new ports. If there is a specific
5868 condition code register in the machine, use a hard register. If the
5869 condition code or comparison result can be placed in any general register,
5870 or if there are multiple condition registers, use a pseudo register.
5871 Registers used to store the condition code value will usually have a mode
5872 that is in class @code{MODE_CC}.
5874 Alternatively, you can use @code{BImode} if the comparison operator is
5875 specified already in the compare instruction. In this case, you are not
5876 interested in most macros in this section.
5879 * CC0 Condition Codes:: Old style representation of condition codes.
5880 * MODE_CC Condition Codes:: Modern representation of condition codes.
5881 * Cond. Exec. Macros:: Macros to control conditional execution.
5884 @node CC0 Condition Codes
5885 @subsection Representation of condition codes using @code{(cc0)}
5889 The file @file{conditions.h} defines a variable @code{cc_status} to
5890 describe how the condition code was computed (in case the interpretation of
5891 the condition code depends on the instruction that it was set by). This
5892 variable contains the RTL expressions on which the condition code is
5893 currently based, and several standard flags.
5895 Sometimes additional machine-specific flags must be defined in the machine
5896 description header file. It can also add additional machine-specific
5897 information by defining @code{CC_STATUS_MDEP}.
5899 @defmac CC_STATUS_MDEP
5900 C code for a data type which is used for declaring the @code{mdep}
5901 component of @code{cc_status}. It defaults to @code{int}.
5903 This macro is not used on machines that do not use @code{cc0}.
5906 @defmac CC_STATUS_MDEP_INIT
5907 A C expression to initialize the @code{mdep} field to ``empty''.
5908 The default definition does nothing, since most machines don't use
5909 the field anyway. If you want to use the field, you should probably
5910 define this macro to initialize it.
5912 This macro is not used on machines that do not use @code{cc0}.
5915 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5916 A C compound statement to set the components of @code{cc_status}
5917 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5918 this macro's responsibility to recognize insns that set the condition
5919 code as a byproduct of other activity as well as those that explicitly
5922 This macro is not used on machines that do not use @code{cc0}.
5924 If there are insns that do not set the condition code but do alter
5925 other machine registers, this macro must check to see whether they
5926 invalidate the expressions that the condition code is recorded as
5927 reflecting. For example, on the 68000, insns that store in address
5928 registers do not set the condition code, which means that usually
5929 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5930 insns. But suppose that the previous insn set the condition code
5931 based on location @samp{a4@@(102)} and the current insn stores a new
5932 value in @samp{a4}. Although the condition code is not changed by
5933 this, it will no longer be true that it reflects the contents of
5934 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5935 @code{cc_status} in this case to say that nothing is known about the
5936 condition code value.
5938 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5939 with the results of peephole optimization: insns whose patterns are
5940 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5941 constants which are just the operands. The RTL structure of these
5942 insns is not sufficient to indicate what the insns actually do. What
5943 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5944 @code{CC_STATUS_INIT}.
5946 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5947 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5948 @samp{cc}. This avoids having detailed information about patterns in
5949 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5952 @node MODE_CC Condition Codes
5953 @subsection Representation of condition codes using registers
5957 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5958 On many machines, the condition code may be produced by other instructions
5959 than compares, for example the branch can use directly the condition
5960 code set by a subtract instruction. However, on some machines
5961 when the condition code is set this way some bits (such as the overflow
5962 bit) are not set in the same way as a test instruction, so that a different
5963 branch instruction must be used for some conditional branches. When
5964 this happens, use the machine mode of the condition code register to
5965 record different formats of the condition code register. Modes can
5966 also be used to record which compare instruction (e.g. a signed or an
5967 unsigned comparison) produced the condition codes.
5969 If other modes than @code{CCmode} are required, add them to
5970 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5971 a mode given an operand of a compare. This is needed because the modes
5972 have to be chosen not only during RTL generation but also, for example,
5973 by instruction combination. The result of @code{SELECT_CC_MODE} should
5974 be consistent with the mode used in the patterns; for example to support
5975 the case of the add on the SPARC discussed above, we have the pattern
5979 [(set (reg:CC_NOOV 0)
5981 (plus:SI (match_operand:SI 0 "register_operand" "%r")
5982 (match_operand:SI 1 "arith_operand" "rI"))
5989 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
5990 for comparisons whose argument is a @code{plus}:
5993 #define SELECT_CC_MODE(OP,X,Y) \
5994 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
5995 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
5996 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
5997 || GET_CODE (X) == NEG) \
5998 ? CC_NOOVmode : CCmode))
6001 Another reason to use modes is to retain information on which operands
6002 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6005 You should define this macro if and only if you define extra CC modes
6006 in @file{@var{machine}-modes.def}.
6009 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
6010 On some machines not all possible comparisons are defined, but you can
6011 convert an invalid comparison into a valid one. For example, the Alpha
6012 does not have a @code{GT} comparison, but you can use an @code{LT}
6013 comparison instead and swap the order of the operands.
6015 On such machines, define this macro to be a C statement to do any
6016 required conversions. @var{code} is the initial comparison code
6017 and @var{op0} and @var{op1} are the left and right operands of the
6018 comparison, respectively. You should modify @var{code}, @var{op0}, and
6019 @var{op1} as required.
6021 GCC will not assume that the comparison resulting from this macro is
6022 valid but will see if the resulting insn matches a pattern in the
6025 You need not define this macro if it would never change the comparison
6029 @defmac REVERSIBLE_CC_MODE (@var{mode})
6030 A C expression whose value is one if it is always safe to reverse a
6031 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6032 can ever return @var{mode} for a floating-point inequality comparison,
6033 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6035 You need not define this macro if it would always returns zero or if the
6036 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6037 For example, here is the definition used on the SPARC, where floating-point
6038 inequality comparisons are always given @code{CCFPEmode}:
6041 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
6045 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6046 A C expression whose value is reversed condition code of the @var{code} for
6047 comparison done in CC_MODE @var{mode}. The macro is used only in case
6048 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6049 machine has some non-standard way how to reverse certain conditionals. For
6050 instance in case all floating point conditions are non-trapping, compiler may
6051 freely convert unordered compares to ordered one. Then definition may look
6055 #define REVERSE_CONDITION(CODE, MODE) \
6056 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6057 : reverse_condition_maybe_unordered (CODE))
6061 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2})
6062 On targets which do not use @code{(cc0)}, and which use a hard
6063 register rather than a pseudo-register to hold condition codes, the
6064 regular CSE passes are often not able to identify cases in which the
6065 hard register is set to a common value. Use this hook to enable a
6066 small pass which optimizes such cases. This hook should return true
6067 to enable this pass, and it should set the integers to which its
6068 arguments point to the hard register numbers used for condition codes.
6069 When there is only one such register, as is true on most systems, the
6070 integer pointed to by @var{p2} should be set to
6071 @code{INVALID_REGNUM}.
6073 The default version of this hook returns false.
6076 @deftypefn {Target Hook} {enum machine_mode} TARGET_CC_MODES_COMPATIBLE (enum machine_mode @var{m1}, enum machine_mode @var{m2})
6077 On targets which use multiple condition code modes in class
6078 @code{MODE_CC}, it is sometimes the case that a comparison can be
6079 validly done in more than one mode. On such a system, define this
6080 target hook to take two mode arguments and to return a mode in which
6081 both comparisons may be validly done. If there is no such mode,
6082 return @code{VOIDmode}.
6084 The default version of this hook checks whether the modes are the
6085 same. If they are, it returns that mode. If they are different, it
6086 returns @code{VOIDmode}.
6089 @node Cond. Exec. Macros
6090 @subsection Macros to control conditional execution
6091 @findex conditional execution
6094 There is one macro that may need to be defined for targets
6095 supporting conditional execution, independent of how they
6096 represent conditional branches.
6098 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
6099 A C expression that returns true if the conditional execution predicate
6100 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
6101 versa. Define this to return 0 if the target has conditional execution
6102 predicates that cannot be reversed safely. There is no need to validate
6103 that the arguments of op1 and op2 are the same, this is done separately.
6104 If no expansion is specified, this macro is defined as follows:
6107 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
6108 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
6113 @section Describing Relative Costs of Operations
6114 @cindex costs of instructions
6115 @cindex relative costs
6116 @cindex speed of instructions
6118 These macros let you describe the relative speed of various operations
6119 on the target machine.
6121 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6122 A C expression for the cost of moving data of mode @var{mode} from a
6123 register in class @var{from} to one in class @var{to}. The classes are
6124 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6125 value of 2 is the default; other values are interpreted relative to
6128 It is not required that the cost always equal 2 when @var{from} is the
6129 same as @var{to}; on some machines it is expensive to move between
6130 registers if they are not general registers.
6132 If reload sees an insn consisting of a single @code{set} between two
6133 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6134 classes returns a value of 2, reload does not check to ensure that the
6135 constraints of the insn are met. Setting a cost of other than 2 will
6136 allow reload to verify that the constraints are met. You should do this
6137 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6140 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6141 A C expression for the cost of moving data of mode @var{mode} between a
6142 register of class @var{class} and memory; @var{in} is zero if the value
6143 is to be written to memory, nonzero if it is to be read in. This cost
6144 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6145 registers and memory is more expensive than between two registers, you
6146 should define this macro to express the relative cost.
6148 If you do not define this macro, GCC uses a default cost of 4 plus
6149 the cost of copying via a secondary reload register, if one is
6150 needed. If your machine requires a secondary reload register to copy
6151 between memory and a register of @var{class} but the reload mechanism is
6152 more complex than copying via an intermediate, define this macro to
6153 reflect the actual cost of the move.
6155 GCC defines the function @code{memory_move_secondary_cost} if
6156 secondary reloads are needed. It computes the costs due to copying via
6157 a secondary register. If your machine copies from memory using a
6158 secondary register in the conventional way but the default base value of
6159 4 is not correct for your machine, define this macro to add some other
6160 value to the result of that function. The arguments to that function
6161 are the same as to this macro.
6164 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6165 A C expression for the cost of a branch instruction. A value of 1 is the
6166 default; other values are interpreted relative to that. Parameter @var{speed_p}
6167 is true when the branch in question should be optimized for speed. When
6168 it is false, @code{BRANCH_COST} should be returning value optimal for code size
6169 rather then performance considerations. @var{predictable_p} is true for well
6170 predictable branches. On many architectures the @code{BRANCH_COST} can be
6174 Here are additional macros which do not specify precise relative costs,
6175 but only that certain actions are more expensive than GCC would
6178 @defmac SLOW_BYTE_ACCESS
6179 Define this macro as a C expression which is nonzero if accessing less
6180 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6181 faster than accessing a word of memory, i.e., if such access
6182 require more than one instruction or if there is no difference in cost
6183 between byte and (aligned) word loads.
6185 When this macro is not defined, the compiler will access a field by
6186 finding the smallest containing object; when it is defined, a fullword
6187 load will be used if alignment permits. Unless bytes accesses are
6188 faster than word accesses, using word accesses is preferable since it
6189 may eliminate subsequent memory access if subsequent accesses occur to
6190 other fields in the same word of the structure, but to different bytes.
6193 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6194 Define this macro to be the value 1 if memory accesses described by the
6195 @var{mode} and @var{alignment} parameters have a cost many times greater
6196 than aligned accesses, for example if they are emulated in a trap
6199 When this macro is nonzero, the compiler will act as if
6200 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6201 moves. This can cause significantly more instructions to be produced.
6202 Therefore, do not set this macro nonzero if unaligned accesses only add a
6203 cycle or two to the time for a memory access.
6205 If the value of this macro is always zero, it need not be defined. If
6206 this macro is defined, it should produce a nonzero value when
6207 @code{STRICT_ALIGNMENT} is nonzero.
6210 @defmac MOVE_RATIO (@var{speed})
6211 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6212 which a sequence of insns should be generated instead of a
6213 string move insn or a library call. Increasing the value will always
6214 make code faster, but eventually incurs high cost in increased code size.
6216 Note that on machines where the corresponding move insn is a
6217 @code{define_expand} that emits a sequence of insns, this macro counts
6218 the number of such sequences.
6220 The parameter @var{speed} is true if the code is currently being
6221 optimized for speed rather than size.
6223 If you don't define this, a reasonable default is used.
6226 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6227 A C expression used to determine whether @code{move_by_pieces} will be used to
6228 copy a chunk of memory, or whether some other block move mechanism
6229 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6230 than @code{MOVE_RATIO}.
6233 @defmac MOVE_MAX_PIECES
6234 A C expression used by @code{move_by_pieces} to determine the largest unit
6235 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6238 @defmac CLEAR_RATIO (@var{speed})
6239 The threshold of number of scalar move insns, @emph{below} which a sequence
6240 of insns should be generated to clear memory instead of a string clear insn
6241 or a library call. Increasing the value will always make code faster, but
6242 eventually incurs high cost in increased code size.
6244 The parameter @var{speed} is true if the code is currently being
6245 optimized for speed rather than size.
6247 If you don't define this, a reasonable default is used.
6250 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6251 A C expression used to determine whether @code{clear_by_pieces} will be used
6252 to clear a chunk of memory, or whether some other block clear mechanism
6253 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6254 than @code{CLEAR_RATIO}.
6257 @defmac SET_RATIO (@var{speed})
6258 The threshold of number of scalar move insns, @emph{below} which a sequence
6259 of insns should be generated to set memory to a constant value, instead of
6260 a block set insn or a library call.
6261 Increasing the value will always make code faster, but
6262 eventually incurs high cost in increased code size.
6264 The parameter @var{speed} is true if the code is currently being
6265 optimized for speed rather than size.
6267 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6270 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6271 A C expression used to determine whether @code{store_by_pieces} will be
6272 used to set a chunk of memory to a constant value, or whether some
6273 other mechanism will be used. Used by @code{__builtin_memset} when
6274 storing values other than constant zero.
6275 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6276 than @code{SET_RATIO}.
6279 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6280 A C expression used to determine whether @code{store_by_pieces} will be
6281 used to set a chunk of memory to a constant string value, or whether some
6282 other mechanism will be used. Used by @code{__builtin_strcpy} when
6283 called with a constant source string.
6284 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6285 than @code{MOVE_RATIO}.
6288 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6289 A C expression used to determine whether a load postincrement is a good
6290 thing to use for a given mode. Defaults to the value of
6291 @code{HAVE_POST_INCREMENT}.
6294 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6295 A C expression used to determine whether a load postdecrement is a good
6296 thing to use for a given mode. Defaults to the value of
6297 @code{HAVE_POST_DECREMENT}.
6300 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6301 A C expression used to determine whether a load preincrement is a good
6302 thing to use for a given mode. Defaults to the value of
6303 @code{HAVE_PRE_INCREMENT}.
6306 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6307 A C expression used to determine whether a load predecrement is a good
6308 thing to use for a given mode. Defaults to the value of
6309 @code{HAVE_PRE_DECREMENT}.
6312 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6313 A C expression used to determine whether a store postincrement is a good
6314 thing to use for a given mode. Defaults to the value of
6315 @code{HAVE_POST_INCREMENT}.
6318 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6319 A C expression used to determine whether a store postdecrement is a good
6320 thing to use for a given mode. Defaults to the value of
6321 @code{HAVE_POST_DECREMENT}.
6324 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6325 This macro is used to determine whether a store preincrement is a good
6326 thing to use for a given mode. Defaults to the value of
6327 @code{HAVE_PRE_INCREMENT}.
6330 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6331 This macro is used to determine whether a store predecrement is a good
6332 thing to use for a given mode. Defaults to the value of
6333 @code{HAVE_PRE_DECREMENT}.
6336 @defmac NO_FUNCTION_CSE
6337 Define this macro if it is as good or better to call a constant
6338 function address than to call an address kept in a register.
6341 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
6342 Define this macro if a non-short-circuit operation produced by
6343 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6344 @code{BRANCH_COST} is greater than or equal to the value 2.
6347 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total}, bool @var{speed})
6348 This target hook describes the relative costs of RTL expressions.
6350 The cost may depend on the precise form of the expression, which is
6351 available for examination in @var{x}, and the rtx code of the expression
6352 in which it is contained, found in @var{outer_code}. @var{code} is the
6353 expression code---redundant, since it can be obtained with
6354 @code{GET_CODE (@var{x})}.
6356 In implementing this hook, you can use the construct
6357 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6360 On entry to the hook, @code{*@var{total}} contains a default estimate
6361 for the cost of the expression. The hook should modify this value as
6362 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6363 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6364 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6366 When optimizing for code size, i.e.@: when @code{speed} is
6367 false, this target hook should be used to estimate the relative
6368 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6370 The hook returns true when all subexpressions of @var{x} have been
6371 processed, and false when @code{rtx_cost} should recurse.
6374 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, bool @var{speed})
6375 This hook computes the cost of an addressing mode that contains
6376 @var{address}. If not defined, the cost is computed from
6377 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6379 For most CISC machines, the default cost is a good approximation of the
6380 true cost of the addressing mode. However, on RISC machines, all
6381 instructions normally have the same length and execution time. Hence
6382 all addresses will have equal costs.
6384 In cases where more than one form of an address is known, the form with
6385 the lowest cost will be used. If multiple forms have the same, lowest,
6386 cost, the one that is the most complex will be used.
6388 For example, suppose an address that is equal to the sum of a register
6389 and a constant is used twice in the same basic block. When this macro
6390 is not defined, the address will be computed in a register and memory
6391 references will be indirect through that register. On machines where
6392 the cost of the addressing mode containing the sum is no higher than
6393 that of a simple indirect reference, this will produce an additional
6394 instruction and possibly require an additional register. Proper
6395 specification of this macro eliminates this overhead for such machines.
6397 This hook is never called with an invalid address.
6399 On machines where an address involving more than one register is as
6400 cheap as an address computation involving only one register, defining
6401 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6402 be live over a region of code where only one would have been if
6403 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6404 should be considered in the definition of this macro. Equivalent costs
6405 should probably only be given to addresses with different numbers of
6406 registers on machines with lots of registers.
6410 @section Adjusting the Instruction Scheduler
6412 The instruction scheduler may need a fair amount of machine-specific
6413 adjustment in order to produce good code. GCC provides several target
6414 hooks for this purpose. It is usually enough to define just a few of
6415 them: try the first ones in this list first.
6417 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
6418 This hook returns the maximum number of instructions that can ever
6419 issue at the same time on the target machine. The default is one.
6420 Although the insn scheduler can define itself the possibility of issue
6421 an insn on the same cycle, the value can serve as an additional
6422 constraint to issue insns on the same simulated processor cycle (see
6423 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6424 This value must be constant over the entire compilation. If you need
6425 it to vary depending on what the instructions are, you must use
6426 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6429 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
6430 This hook is executed by the scheduler after it has scheduled an insn
6431 from the ready list. It should return the number of insns which can
6432 still be issued in the current cycle. The default is
6433 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6434 @code{USE}, which normally are not counted against the issue rate.
6435 You should define this hook if some insns take more machine resources
6436 than others, so that fewer insns can follow them in the same cycle.
6437 @var{file} is either a null pointer, or a stdio stream to write any
6438 debug output to. @var{verbose} is the verbose level provided by
6439 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6443 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
6444 This function corrects the value of @var{cost} based on the
6445 relationship between @var{insn} and @var{dep_insn} through the
6446 dependence @var{link}. It should return the new value. The default
6447 is to make no adjustment to @var{cost}. This can be used for example
6448 to specify to the scheduler using the traditional pipeline description
6449 that an output- or anti-dependence does not incur the same cost as a
6450 data-dependence. If the scheduler using the automaton based pipeline
6451 description, the cost of anti-dependence is zero and the cost of
6452 output-dependence is maximum of one and the difference of latency
6453 times of the first and the second insns. If these values are not
6454 acceptable, you could use the hook to modify them too. See also
6455 @pxref{Processor pipeline description}.
6458 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
6459 This hook adjusts the integer scheduling priority @var{priority} of
6460 @var{insn}. It should return the new priority. Increase the priority to
6461 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6462 later. Do not define this hook if you do not need to adjust the
6463 scheduling priorities of insns.
6466 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6467 This hook is executed by the scheduler after it has scheduled the ready
6468 list, to allow the machine description to reorder it (for example to
6469 combine two small instructions together on @samp{VLIW} machines).
6470 @var{file} is either a null pointer, or a stdio stream to write any
6471 debug output to. @var{verbose} is the verbose level provided by
6472 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6473 list of instructions that are ready to be scheduled. @var{n_readyp} is
6474 a pointer to the number of elements in the ready list. The scheduler
6475 reads the ready list in reverse order, starting with
6476 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6477 is the timer tick of the scheduler. You may modify the ready list and
6478 the number of ready insns. The return value is the number of insns that
6479 can issue this cycle; normally this is just @code{issue_rate}. See also
6480 @samp{TARGET_SCHED_REORDER2}.
6483 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6484 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6485 function is called whenever the scheduler starts a new cycle. This one
6486 is called once per iteration over a cycle, immediately after
6487 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6488 return the number of insns to be scheduled in the same cycle. Defining
6489 this hook can be useful if there are frequent situations where
6490 scheduling one insn causes other insns to become ready in the same
6491 cycle. These other insns can then be taken into account properly.
6494 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
6495 This hook is called after evaluation forward dependencies of insns in
6496 chain given by two parameter values (@var{head} and @var{tail}
6497 correspondingly) but before insns scheduling of the insn chain. For
6498 example, it can be used for better insn classification if it requires
6499 analysis of dependencies. This hook can use backward and forward
6500 dependencies of the insn scheduler because they are already
6504 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
6505 This hook is executed by the scheduler at the beginning of each block of
6506 instructions that are to be scheduled. @var{file} is either a null
6507 pointer, or a stdio stream to write any debug output to. @var{verbose}
6508 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6509 @var{max_ready} is the maximum number of insns in the current scheduling
6510 region that can be live at the same time. This can be used to allocate
6511 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6514 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
6515 This hook is executed by the scheduler at the end of each block of
6516 instructions that are to be scheduled. It can be used to perform
6517 cleanup of any actions done by the other scheduling hooks. @var{file}
6518 is either a null pointer, or a stdio stream to write any debug output
6519 to. @var{verbose} is the verbose level provided by
6520 @option{-fsched-verbose-@var{n}}.
6523 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
6524 This hook is executed by the scheduler after function level initializations.
6525 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6526 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6527 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6530 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
6531 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6532 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6533 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6536 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
6537 The hook returns an RTL insn. The automaton state used in the
6538 pipeline hazard recognizer is changed as if the insn were scheduled
6539 when the new simulated processor cycle starts. Usage of the hook may
6540 simplify the automaton pipeline description for some @acronym{VLIW}
6541 processors. If the hook is defined, it is used only for the automaton
6542 based pipeline description. The default is not to change the state
6543 when the new simulated processor cycle starts.
6546 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
6547 The hook can be used to initialize data used by the previous hook.
6550 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
6551 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6552 to changed the state as if the insn were scheduled when the new
6553 simulated processor cycle finishes.
6556 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
6557 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6558 used to initialize data used by the previous hook.
6561 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)
6562 The hook to notify target that the current simulated cycle is about to finish.
6563 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6564 to change the state in more complicated situations - e.g., when advancing
6565 state on a single insn is not enough.
6568 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)
6569 The hook to notify target that new simulated cycle has just started.
6570 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6571 to change the state in more complicated situations - e.g., when advancing
6572 state on a single insn is not enough.
6575 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
6576 This hook controls better choosing an insn from the ready insn queue
6577 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6578 chooses the first insn from the queue. If the hook returns a positive
6579 value, an additional scheduler code tries all permutations of
6580 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6581 subsequent ready insns to choose an insn whose issue will result in
6582 maximal number of issued insns on the same cycle. For the
6583 @acronym{VLIW} processor, the code could actually solve the problem of
6584 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6585 rules of @acronym{VLIW} packing are described in the automaton.
6587 This code also could be used for superscalar @acronym{RISC}
6588 processors. Let us consider a superscalar @acronym{RISC} processor
6589 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6590 @var{B}, some insns can be executed only in pipelines @var{B} or
6591 @var{C}, and one insn can be executed in pipeline @var{B}. The
6592 processor may issue the 1st insn into @var{A} and the 2nd one into
6593 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6594 until the next cycle. If the scheduler issues the 3rd insn the first,
6595 the processor could issue all 3 insns per cycle.
6597 Actually this code demonstrates advantages of the automaton based
6598 pipeline hazard recognizer. We try quickly and easy many insn
6599 schedules to choose the best one.
6601 The default is no multipass scheduling.
6604 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx @var{insn})
6606 This hook controls what insns from the ready insn queue will be
6607 considered for the multipass insn scheduling. If the hook returns
6608 zero for @var{insn}, the insn will be not chosen to
6611 The default is that any ready insns can be chosen to be issued.
6614 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *@var{dump}, int @var{verbose}, rtx @var{insn}, int @var{last_clock}, int @var{clock}, int *@var{sort_p})
6616 This hook is called by the insn scheduler before issuing @var{insn}
6617 on cycle @var{clock}. If the hook returns nonzero,
6618 @var{insn} is not issued on this processor cycle. Instead,
6619 the processor cycle is advanced. If *@var{sort_p}
6620 is zero, the insn ready queue is not sorted on the new cycle
6621 start as usually. @var{dump} and @var{verbose} specify the file and
6622 verbosity level to use for debugging output.
6623 @var{last_clock} and @var{clock} are, respectively, the
6624 processor cycle on which the previous insn has been issued,
6625 and the current processor cycle.
6628 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance})
6629 This hook is used to define which dependences are considered costly by
6630 the target, so costly that it is not advisable to schedule the insns that
6631 are involved in the dependence too close to one another. The parameters
6632 to this hook are as follows: The first parameter @var{_dep} is the dependence
6633 being evaluated. The second parameter @var{cost} is the cost of the
6634 dependence as estimated by the scheduler, and the third
6635 parameter @var{distance} is the distance in cycles between the two insns.
6636 The hook returns @code{true} if considering the distance between the two
6637 insns the dependence between them is considered costly by the target,
6638 and @code{false} otherwise.
6640 Defining this hook can be useful in multiple-issue out-of-order machines,
6641 where (a) it's practically hopeless to predict the actual data/resource
6642 delays, however: (b) there's a better chance to predict the actual grouping
6643 that will be formed, and (c) correctly emulating the grouping can be very
6644 important. In such targets one may want to allow issuing dependent insns
6645 closer to one another---i.e., closer than the dependence distance; however,
6646 not in cases of ``costly dependences'', which this hooks allows to define.
6649 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6650 This hook is called by the insn scheduler after emitting a new instruction to
6651 the instruction stream. The hook notifies a target backend to extend its
6652 per instruction data structures.
6655 @deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
6656 Return a pointer to a store large enough to hold target scheduling context.
6659 @deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p})
6660 Initialize store pointed to by @var{tc} to hold target scheduling context.
6661 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6662 beginning of the block. Otherwise, copy the current context into @var{tc}.
6665 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc})
6666 Copy target scheduling context pointed to by @var{tc} to the current context.
6669 @deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc})
6670 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6673 @deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc})
6674 Deallocate a store for target scheduling context pointed to by @var{tc}.
6677 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat})
6678 This hook is called by the insn scheduler when @var{insn} has only
6679 speculative dependencies and therefore can be scheduled speculatively.
6680 The hook is used to check if the pattern of @var{insn} has a speculative
6681 version and, in case of successful check, to generate that speculative
6682 pattern. The hook should return 1, if the instruction has a speculative form,
6683 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6684 speculation. If the return value equals 1 then @var{new_pat} is assigned
6685 the generated speculative pattern.
6688 @deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (int @var{dep_status})
6689 This hook is called by the insn scheduler during generation of recovery code
6690 for @var{insn}. It should return @code{true}, if the corresponding check
6691 instruction should branch to recovery code, or @code{false} otherwise.
6694 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p})
6695 This hook is called by the insn scheduler to generate a pattern for recovery
6696 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6697 speculative instruction for which the check should be generated.
6698 @var{label} is either a label of a basic block, where recovery code should
6699 be emitted, or a null pointer, when requested check doesn't branch to
6700 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6701 a pattern for a branchy check corresponding to a simple check denoted by
6702 @var{insn} should be generated. In this case @var{label} can't be null.
6705 @deftypefn {Target Hook} bool TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (const_rtx @var{insn})
6706 This hook is used as a workaround for
6707 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6708 called on the first instruction of the ready list. The hook is used to
6709 discard speculative instructions that stand first in the ready list from
6710 being scheduled on the current cycle. If the hook returns @code{false},
6711 @var{insn} will not be chosen to be issued.
6712 For non-speculative instructions,
6713 the hook should always return @code{true}. For example, in the ia64 backend
6714 the hook is used to cancel data speculative insns when the ALAT table
6718 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info})
6719 This hook is used by the insn scheduler to find out what features should be
6721 The structure *@var{spec_info} should be filled in by the target.
6722 The structure describes speculation types that can be used in the scheduler.
6725 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
6726 This hook is called by the swing modulo scheduler to calculate a
6727 resource-based lower bound which is based on the resources available in
6728 the machine and the resources required by each instruction. The target
6729 backend can use @var{g} to calculate such bound. A very simple lower
6730 bound will be used in case this hook is not implemented: the total number
6731 of instructions divided by the issue rate.
6735 @section Dividing the Output into Sections (Texts, Data, @dots{})
6736 @c the above section title is WAY too long. maybe cut the part between
6737 @c the (...)? --mew 10feb93
6739 An object file is divided into sections containing different types of
6740 data. In the most common case, there are three sections: the @dfn{text
6741 section}, which holds instructions and read-only data; the @dfn{data
6742 section}, which holds initialized writable data; and the @dfn{bss
6743 section}, which holds uninitialized data. Some systems have other kinds
6746 @file{varasm.c} provides several well-known sections, such as
6747 @code{text_section}, @code{data_section} and @code{bss_section}.
6748 The normal way of controlling a @code{@var{foo}_section} variable
6749 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6750 as described below. The macros are only read once, when @file{varasm.c}
6751 initializes itself, so their values must be run-time constants.
6752 They may however depend on command-line flags.
6754 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6755 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6756 to be string literals.
6758 Some assemblers require a different string to be written every time a
6759 section is selected. If your assembler falls into this category, you
6760 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6761 @code{get_unnamed_section} to set up the sections.
6763 You must always create a @code{text_section}, either by defining
6764 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6765 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6766 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6767 create a distinct @code{readonly_data_section}, the default is to
6768 reuse @code{text_section}.
6770 All the other @file{varasm.c} sections are optional, and are null
6771 if the target does not provide them.
6773 @defmac TEXT_SECTION_ASM_OP
6774 A C expression whose value is a string, including spacing, containing the
6775 assembler operation that should precede instructions and read-only data.
6776 Normally @code{"\t.text"} is right.
6779 @defmac HOT_TEXT_SECTION_NAME
6780 If defined, a C string constant for the name of the section containing most
6781 frequently executed functions of the program. If not defined, GCC will provide
6782 a default definition if the target supports named sections.
6785 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6786 If defined, a C string constant for the name of the section containing unlikely
6787 executed functions in the program.
6790 @defmac DATA_SECTION_ASM_OP
6791 A C expression whose value is a string, including spacing, containing the
6792 assembler operation to identify the following data as writable initialized
6793 data. Normally @code{"\t.data"} is right.
6796 @defmac SDATA_SECTION_ASM_OP
6797 If defined, a C expression whose value is a string, including spacing,
6798 containing the assembler operation to identify the following data as
6799 initialized, writable small data.
6802 @defmac READONLY_DATA_SECTION_ASM_OP
6803 A C expression whose value is a string, including spacing, containing the
6804 assembler operation to identify the following data as read-only initialized
6808 @defmac BSS_SECTION_ASM_OP
6809 If defined, a C expression whose value is a string, including spacing,
6810 containing the assembler operation to identify the following data as
6811 uninitialized global data. If not defined, and neither
6812 @code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined,
6813 uninitialized global data will be output in the data section if
6814 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6818 @defmac SBSS_SECTION_ASM_OP
6819 If defined, a C expression whose value is a string, including spacing,
6820 containing the assembler operation to identify the following data as
6821 uninitialized, writable small data.
6824 @defmac TLS_COMMON_ASM_OP
6825 If defined, a C expression whose value is a string containing the
6826 assembler operation to identify the following data as thread-local
6827 common data. The default is @code{".tls_common"}.
6830 @defmac TLS_SECTION_ASM_FLAG
6831 If defined, a C expression whose value is a character constant
6832 containing the flag used to mark a section as a TLS section. The
6833 default is @code{'T'}.
6836 @defmac INIT_SECTION_ASM_OP
6837 If defined, a C expression whose value is a string, including spacing,
6838 containing the assembler operation to identify the following data as
6839 initialization code. If not defined, GCC will assume such a section does
6840 not exist. This section has no corresponding @code{init_section}
6841 variable; it is used entirely in runtime code.
6844 @defmac FINI_SECTION_ASM_OP
6845 If defined, a C expression whose value is a string, including spacing,
6846 containing the assembler operation to identify the following data as
6847 finalization code. If not defined, GCC will assume such a section does
6848 not exist. This section has no corresponding @code{fini_section}
6849 variable; it is used entirely in runtime code.
6852 @defmac INIT_ARRAY_SECTION_ASM_OP
6853 If defined, a C expression whose value is a string, including spacing,
6854 containing the assembler operation to identify the following data as
6855 part of the @code{.init_array} (or equivalent) section. If not
6856 defined, GCC will assume such a section does not exist. Do not define
6857 both this macro and @code{INIT_SECTION_ASM_OP}.
6860 @defmac FINI_ARRAY_SECTION_ASM_OP
6861 If defined, a C expression whose value is a string, including spacing,
6862 containing the assembler operation to identify the following data as
6863 part of the @code{.fini_array} (or equivalent) section. If not
6864 defined, GCC will assume such a section does not exist. Do not define
6865 both this macro and @code{FINI_SECTION_ASM_OP}.
6868 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6869 If defined, an ASM statement that switches to a different section
6870 via @var{section_op}, calls @var{function}, and switches back to
6871 the text section. This is used in @file{crtstuff.c} if
6872 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6873 to initialization and finalization functions from the init and fini
6874 sections. By default, this macro uses a simple function call. Some
6875 ports need hand-crafted assembly code to avoid dependencies on
6876 registers initialized in the function prologue or to ensure that
6877 constant pools don't end up too far way in the text section.
6880 @defmac TARGET_LIBGCC_SDATA_SECTION
6881 If defined, a string which names the section into which small
6882 variables defined in crtstuff and libgcc should go. This is useful
6883 when the target has options for optimizing access to small data, and
6884 you want the crtstuff and libgcc routines to be conservative in what
6885 they expect of your application yet liberal in what your application
6886 expects. For example, for targets with a @code{.sdata} section (like
6887 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
6888 require small data support from your application, but use this macro
6889 to put small data into @code{.sdata} so that your application can
6890 access these variables whether it uses small data or not.
6893 @defmac FORCE_CODE_SECTION_ALIGN
6894 If defined, an ASM statement that aligns a code section to some
6895 arbitrary boundary. This is used to force all fragments of the
6896 @code{.init} and @code{.fini} sections to have to same alignment
6897 and thus prevent the linker from having to add any padding.
6900 @defmac JUMP_TABLES_IN_TEXT_SECTION
6901 Define this macro to be an expression with a nonzero value if jump
6902 tables (for @code{tablejump} insns) should be output in the text
6903 section, along with the assembler instructions. Otherwise, the
6904 readonly data section is used.
6906 This macro is irrelevant if there is no separate readonly data section.
6909 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
6910 Define this hook if you need to do something special to set up the
6911 @file{varasm.c} sections, or if your target has some special sections
6912 of its own that you need to create.
6914 GCC calls this hook after processing the command line, but before writing
6915 any assembly code, and before calling any of the section-returning hooks
6919 @deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void)
6920 Return a mask describing how relocations should be treated when
6921 selecting sections. Bit 1 should be set if global relocations
6922 should be placed in a read-write section; bit 0 should be set if
6923 local relocations should be placed in a read-write section.
6925 The default version of this function returns 3 when @option{-fpic}
6926 is in effect, and 0 otherwise. The hook is typically redefined
6927 when the target cannot support (some kinds of) dynamic relocations
6928 in read-only sections even in executables.
6931 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
6932 Return the section into which @var{exp} should be placed. You can
6933 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
6934 some sort. @var{reloc} indicates whether the initial value of @var{exp}
6935 requires link-time relocations. Bit 0 is set when variable contains
6936 local relocations only, while bit 1 is set for global relocations.
6937 @var{align} is the constant alignment in bits.
6939 The default version of this function takes care of putting read-only
6940 variables in @code{readonly_data_section}.
6942 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
6945 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
6946 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
6947 for @code{FUNCTION_DECL}s as well as for variables and constants.
6949 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
6950 function has been determined to be likely to be called, and nonzero if
6951 it is unlikely to be called.
6954 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
6955 Build up a unique section name, expressed as a @code{STRING_CST} node,
6956 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
6957 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
6958 the initial value of @var{exp} requires link-time relocations.
6960 The default version of this function appends the symbol name to the
6961 ELF section name that would normally be used for the symbol. For
6962 example, the function @code{foo} would be placed in @code{.text.foo}.
6963 Whatever the actual target object format, this is often good enough.
6966 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
6967 Return the readonly data section associated with
6968 @samp{DECL_SECTION_NAME (@var{decl})}.
6969 The default version of this function selects @code{.gnu.linkonce.r.name} if
6970 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
6971 if function is in @code{.text.name}, and the normal readonly-data section
6975 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
6976 Return the section into which a constant @var{x}, of mode @var{mode},
6977 should be placed. You can assume that @var{x} is some kind of
6978 constant in RTL@. The argument @var{mode} is redundant except in the
6979 case of a @code{const_int} rtx. @var{align} is the constant alignment
6982 The default version of this function takes care of putting symbolic
6983 constants in @code{flag_pic} mode in @code{data_section} and everything
6984 else in @code{readonly_data_section}.
6987 @deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
6988 Define this hook if you need to postprocess the assembler name generated
6989 by target-independent code. The @var{id} provided to this hook will be
6990 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
6991 or the mangled name of the @var{decl} in C++). The return value of the
6992 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
6993 your target system. The default implementation of this hook just
6994 returns the @var{id} provided.
6997 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
6998 Define this hook if references to a symbol or a constant must be
6999 treated differently depending on something about the variable or
7000 function named by the symbol (such as what section it is in).
7002 The hook is executed immediately after rtl has been created for
7003 @var{decl}, which may be a variable or function declaration or
7004 an entry in the constant pool. In either case, @var{rtl} is the
7005 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7006 in this hook; that field may not have been initialized yet.
7008 In the case of a constant, it is safe to assume that the rtl is
7009 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7010 will also have this form, but that is not guaranteed. Global
7011 register variables, for instance, will have a @code{reg} for their
7012 rtl. (Normally the right thing to do with such unusual rtl is
7015 The @var{new_decl_p} argument will be true if this is the first time
7016 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7017 be false for subsequent invocations, which will happen for duplicate
7018 declarations. Whether or not anything must be done for the duplicate
7019 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7020 @var{new_decl_p} is always true when the hook is called for a constant.
7022 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7023 The usual thing for this hook to do is to record flags in the
7024 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7025 Historically, the name string was modified if it was necessary to
7026 encode more than one bit of information, but this practice is now
7027 discouraged; use @code{SYMBOL_REF_FLAGS}.
7029 The default definition of this hook, @code{default_encode_section_info}
7030 in @file{varasm.c}, sets a number of commonly-useful bits in
7031 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7032 before overriding it.
7035 @deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name})
7036 Decode @var{name} and return the real name part, sans
7037 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7041 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp})
7042 Returns true if @var{exp} should be placed into a ``small data'' section.
7043 The default version of this hook always returns false.
7046 @deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
7047 Contains the value true if the target places read-only
7048 ``small data'' into a separate section. The default value is false.
7051 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp})
7052 Returns true if @var{exp} names an object for which name resolution
7053 rules must resolve to the current ``module'' (dynamic shared library
7054 or executable image).
7056 The default version of this hook implements the name resolution rules
7057 for ELF, which has a looser model of global name binding than other
7058 currently supported object file formats.
7061 @deftypevr {Target Hook} bool TARGET_HAVE_TLS
7062 Contains the value true if the target supports thread-local storage.
7063 The default value is false.
7068 @section Position Independent Code
7069 @cindex position independent code
7072 This section describes macros that help implement generation of position
7073 independent code. Simply defining these macros is not enough to
7074 generate valid PIC; you must also add support to the hook
7075 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7076 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7077 must modify the definition of @samp{movsi} to do something appropriate
7078 when the source operand contains a symbolic address. You may also
7079 need to alter the handling of switch statements so that they use
7081 @c i rearranged the order of the macros above to try to force one of
7082 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7084 @defmac PIC_OFFSET_TABLE_REGNUM
7085 The register number of the register used to address a table of static
7086 data addresses in memory. In some cases this register is defined by a
7087 processor's ``application binary interface'' (ABI)@. When this macro
7088 is defined, RTL is generated for this register once, as with the stack
7089 pointer and frame pointer registers. If this macro is not defined, it
7090 is up to the machine-dependent files to allocate such a register (if
7091 necessary). Note that this register must be fixed when in use (e.g.@:
7092 when @code{flag_pic} is true).
7095 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7096 Define this macro if the register defined by
7097 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define
7098 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7101 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7102 A C expression that is nonzero if @var{x} is a legitimate immediate
7103 operand on the target machine when generating position independent code.
7104 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7105 check this. You can also assume @var{flag_pic} is true, so you need not
7106 check it either. You need not define this macro if all constants
7107 (including @code{SYMBOL_REF}) can be immediate operands when generating
7108 position independent code.
7111 @node Assembler Format
7112 @section Defining the Output Assembler Language
7114 This section describes macros whose principal purpose is to describe how
7115 to write instructions in assembler language---rather than what the
7119 * File Framework:: Structural information for the assembler file.
7120 * Data Output:: Output of constants (numbers, strings, addresses).
7121 * Uninitialized Data:: Output of uninitialized variables.
7122 * Label Output:: Output and generation of labels.
7123 * Initialization:: General principles of initialization
7124 and termination routines.
7125 * Macros for Initialization::
7126 Specific macros that control the handling of
7127 initialization and termination routines.
7128 * Instruction Output:: Output of actual instructions.
7129 * Dispatch Tables:: Output of jump tables.
7130 * Exception Region Output:: Output of exception region code.
7131 * Alignment Output:: Pseudo ops for alignment and skipping data.
7134 @node File Framework
7135 @subsection The Overall Framework of an Assembler File
7136 @cindex assembler format
7137 @cindex output of assembler code
7139 @c prevent bad page break with this line
7140 This describes the overall framework of an assembly file.
7142 @findex default_file_start
7143 @deftypefn {Target Hook} void TARGET_ASM_FILE_START (void)
7144 Output to @code{asm_out_file} any text which the assembler expects to
7145 find at the beginning of a file. The default behavior is controlled
7146 by two flags, documented below. Unless your target's assembler is
7147 quite unusual, if you override the default, you should call
7148 @code{default_file_start} at some point in your target hook. This
7149 lets other target files rely on these variables.
7152 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
7153 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7154 printed as the very first line in the assembly file, unless
7155 @option{-fverbose-asm} is in effect. (If that macro has been defined
7156 to the empty string, this variable has no effect.) With the normal
7157 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7158 assembler that it need not bother stripping comments or extra
7159 whitespace from its input. This allows it to work a bit faster.
7161 The default is false. You should not set it to true unless you have
7162 verified that your port does not generate any extra whitespace or
7163 comments that will cause GAS to issue errors in NO_APP mode.
7166 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
7167 If this flag is true, @code{output_file_directive} will be called
7168 for the primary source file, immediately after printing
7169 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7170 this to be done. The default is false.
7173 @deftypefn {Target Hook} void TARGET_ASM_FILE_END (void)
7174 Output to @code{asm_out_file} any text which the assembler expects
7175 to find at the end of a file. The default is to output nothing.
7178 @deftypefun void file_end_indicate_exec_stack ()
7179 Some systems use a common convention, the @samp{.note.GNU-stack}
7180 special section, to indicate whether or not an object file relies on
7181 the stack being executable. If your system uses this convention, you
7182 should define @code{TARGET_ASM_FILE_END} to this function. If you
7183 need to do other things in that hook, have your hook function call
7187 @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void)
7188 Output to @code{asm_out_file} any text which is needed before emitting
7189 unwind info and debug info at the end of a file. Some targets emit
7190 here PIC setup thunks that cannot be emitted at the end of file,
7191 because they couldn't have unwind info then. The default is to output
7195 @defmac ASM_COMMENT_START
7196 A C string constant describing how to begin a comment in the target
7197 assembler language. The compiler assumes that the comment will end at
7198 the end of the line.
7202 A C string constant for text to be output before each @code{asm}
7203 statement or group of consecutive ones. Normally this is
7204 @code{"#APP"}, which is a comment that has no effect on most
7205 assemblers but tells the GNU assembler that it must check the lines
7206 that follow for all valid assembler constructs.
7210 A C string constant for text to be output after each @code{asm}
7211 statement or group of consecutive ones. Normally this is
7212 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7213 time-saving assumptions that are valid for ordinary compiler output.
7216 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7217 A C statement to output COFF information or DWARF debugging information
7218 which indicates that filename @var{name} is the current source file to
7219 the stdio stream @var{stream}.
7221 This macro need not be defined if the standard form of output
7222 for the file format in use is appropriate.
7225 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7226 A C statement to output the string @var{string} to the stdio stream
7227 @var{stream}. If you do not call the function @code{output_quoted_string}
7228 in your config files, GCC will only call it to output filenames to
7229 the assembler source. So you can use it to canonicalize the format
7230 of the filename using this macro.
7233 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7234 A C statement to output something to the assembler file to handle a
7235 @samp{#ident} directive containing the text @var{string}. If this
7236 macro is not defined, nothing is output for a @samp{#ident} directive.
7239 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl})
7240 Output assembly directives to switch to section @var{name}. The section
7241 should have attributes as specified by @var{flags}, which is a bit mask
7242 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7243 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7244 this section is associated.
7247 @deftypevr {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
7248 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7251 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7252 @deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7253 This flag is true if we can create zeroed data by switching to a BSS
7254 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7255 This is true on most ELF targets.
7258 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
7259 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7260 based on a variable or function decl, a section name, and whether or not the
7261 declaration's initializer may contain runtime relocations. @var{decl} may be
7262 null, in which case read-write data should be assumed.
7264 The default version of this function handles choosing code vs data,
7265 read-only vs read-write data, and @code{flag_pic}. You should only
7266 need to override this if your target has special flags that might be
7267 set via @code{__attribute__}.
7270 @deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text})
7271 Provides the target with the ability to record the gcc command line
7272 switches that have been passed to the compiler, and options that are
7273 enabled. The @var{type} argument specifies what is being recorded.
7274 It can take the following values:
7277 @item SWITCH_TYPE_PASSED
7278 @var{text} is a command line switch that has been set by the user.
7280 @item SWITCH_TYPE_ENABLED
7281 @var{text} is an option which has been enabled. This might be as a
7282 direct result of a command line switch, or because it is enabled by
7283 default or because it has been enabled as a side effect of a different
7284 command line switch. For example, the @option{-O2} switch enables
7285 various different individual optimization passes.
7287 @item SWITCH_TYPE_DESCRIPTIVE
7288 @var{text} is either NULL or some descriptive text which should be
7289 ignored. If @var{text} is NULL then it is being used to warn the
7290 target hook that either recording is starting or ending. The first
7291 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7292 warning is for start up and the second time the warning is for
7293 wind down. This feature is to allow the target hook to make any
7294 necessary preparations before it starts to record switches and to
7295 perform any necessary tidying up after it has finished recording
7298 @item SWITCH_TYPE_LINE_START
7299 This option can be ignored by this target hook.
7301 @item SWITCH_TYPE_LINE_END
7302 This option can be ignored by this target hook.
7305 The hook's return value must be zero. Other return values may be
7306 supported in the future.
7308 By default this hook is set to NULL, but an example implementation is
7309 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7310 it records the switches as ASCII text inside a new, string mergeable
7311 section in the assembler output file. The name of the new section is
7312 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7316 @deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7317 This is the name of the section that will be created by the example
7318 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7324 @subsection Output of Data
7327 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
7328 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7329 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7330 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7331 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7332 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7333 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7334 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7335 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7336 These hooks specify assembly directives for creating certain kinds
7337 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7338 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7339 aligned two-byte object, and so on. Any of the hooks may be
7340 @code{NULL}, indicating that no suitable directive is available.
7342 The compiler will print these strings at the start of a new line,
7343 followed immediately by the object's initial value. In most cases,
7344 the string should contain a tab, a pseudo-op, and then another tab.
7347 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
7348 The @code{assemble_integer} function uses this hook to output an
7349 integer object. @var{x} is the object's value, @var{size} is its size
7350 in bytes and @var{aligned_p} indicates whether it is aligned. The
7351 function should return @code{true} if it was able to output the
7352 object. If it returns false, @code{assemble_integer} will try to
7353 split the object into smaller parts.
7355 The default implementation of this hook will use the
7356 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7357 when the relevant string is @code{NULL}.
7360 @defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail})
7361 A C statement to recognize @var{rtx} patterns that
7362 @code{output_addr_const} can't deal with, and output assembly code to
7363 @var{stream} corresponding to the pattern @var{x}. This may be used to
7364 allow machine-dependent @code{UNSPEC}s to appear within constants.
7366 If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must
7367 @code{goto fail}, so that a standard error message is printed. If it
7368 prints an error message itself, by calling, for example,
7369 @code{output_operand_lossage}, it may just complete normally.
7372 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7373 A C statement to output to the stdio stream @var{stream} an assembler
7374 instruction to assemble a string constant containing the @var{len}
7375 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7376 @code{char *} and @var{len} a C expression of type @code{int}.
7378 If the assembler has a @code{.ascii} pseudo-op as found in the
7379 Berkeley Unix assembler, do not define the macro
7380 @code{ASM_OUTPUT_ASCII}.
7383 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7384 A C statement to output word @var{n} of a function descriptor for
7385 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7386 is defined, and is otherwise unused.
7389 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7390 You may define this macro as a C expression. You should define the
7391 expression to have a nonzero value if GCC should output the constant
7392 pool for a function before the code for the function, or a zero value if
7393 GCC should output the constant pool after the function. If you do
7394 not define this macro, the usual case, GCC will output the constant
7395 pool before the function.
7398 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7399 A C statement to output assembler commands to define the start of the
7400 constant pool for a function. @var{funname} is a string giving
7401 the name of the function. Should the return type of the function
7402 be required, it can be obtained via @var{fundecl}. @var{size}
7403 is the size, in bytes, of the constant pool that will be written
7404 immediately after this call.
7406 If no constant-pool prefix is required, the usual case, this macro need
7410 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7411 A C statement (with or without semicolon) to output a constant in the
7412 constant pool, if it needs special treatment. (This macro need not do
7413 anything for RTL expressions that can be output normally.)
7415 The argument @var{file} is the standard I/O stream to output the
7416 assembler code on. @var{x} is the RTL expression for the constant to
7417 output, and @var{mode} is the machine mode (in case @var{x} is a
7418 @samp{const_int}). @var{align} is the required alignment for the value
7419 @var{x}; you should output an assembler directive to force this much
7422 The argument @var{labelno} is a number to use in an internal label for
7423 the address of this pool entry. The definition of this macro is
7424 responsible for outputting the label definition at the proper place.
7425 Here is how to do this:
7428 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7431 When you output a pool entry specially, you should end with a
7432 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7433 entry from being output a second time in the usual manner.
7435 You need not define this macro if it would do nothing.
7438 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7439 A C statement to output assembler commands to at the end of the constant
7440 pool for a function. @var{funname} is a string giving the name of the
7441 function. Should the return type of the function be required, you can
7442 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7443 constant pool that GCC wrote immediately before this call.
7445 If no constant-pool epilogue is required, the usual case, you need not
7449 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7450 Define this macro as a C expression which is nonzero if @var{C} is
7451 used as a logical line separator by the assembler. @var{STR} points
7452 to the position in the string where @var{C} was found; this can be used if
7453 a line separator uses multiple characters.
7455 If you do not define this macro, the default is that only
7456 the character @samp{;} is treated as a logical line separator.
7459 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
7460 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
7461 These target hooks are C string constants, describing the syntax in the
7462 assembler for grouping arithmetic expressions. If not overridden, they
7463 default to normal parentheses, which is correct for most assemblers.
7466 These macros are provided by @file{real.h} for writing the definitions
7467 of @code{ASM_OUTPUT_DOUBLE} and the like:
7469 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7470 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7471 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7472 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7473 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7474 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7475 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7476 target's floating point representation, and store its bit pattern in
7477 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7478 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7479 simple @code{long int}. For the others, it should be an array of
7480 @code{long int}. The number of elements in this array is determined
7481 by the size of the desired target floating point data type: 32 bits of
7482 it go in each @code{long int} array element. Each array element holds
7483 32 bits of the result, even if @code{long int} is wider than 32 bits
7484 on the host machine.
7486 The array element values are designed so that you can print them out
7487 using @code{fprintf} in the order they should appear in the target
7491 @node Uninitialized Data
7492 @subsection Output of Uninitialized Variables
7494 Each of the macros in this section is used to do the whole job of
7495 outputting a single uninitialized variable.
7497 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7498 A C statement (sans semicolon) to output to the stdio stream
7499 @var{stream} the assembler definition of a common-label named
7500 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7501 is the size rounded up to whatever alignment the caller wants. It is
7502 possible that @var{size} may be zero, for instance if a struct with no
7503 other member than a zero-length array is defined. In this case, the
7504 backend must output a symbol definition that allocates at least one
7505 byte, both so that the address of the resulting object does not compare
7506 equal to any other, and because some object formats cannot even express
7507 the concept of a zero-sized common symbol, as that is how they represent
7508 an ordinary undefined external.
7510 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7511 output the name itself; before and after that, output the additional
7512 assembler syntax for defining the name, and a newline.
7514 This macro controls how the assembler definitions of uninitialized
7515 common global variables are output.
7518 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7519 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7520 separate, explicit argument. If you define this macro, it is used in
7521 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7522 handling the required alignment of the variable. The alignment is specified
7523 as the number of bits.
7526 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7527 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7528 variable to be output, if there is one, or @code{NULL_TREE} if there
7529 is no corresponding variable. If you define this macro, GCC will use it
7530 in place of both @code{ASM_OUTPUT_COMMON} and
7531 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7532 the variable's decl in order to chose what to output.
7535 @defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded})
7536 A C statement (sans semicolon) to output to the stdio stream
7537 @var{stream} the assembler definition of uninitialized global @var{decl} named
7538 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7539 is the size rounded up to whatever alignment the caller wants.
7541 Try to use function @code{asm_output_bss} defined in @file{varasm.c} when
7542 defining this macro. If unable, use the expression
7543 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7544 before and after that, output the additional assembler syntax for defining
7545 the name, and a newline.
7547 There are two ways of handling global BSS@. One is to define either
7548 this macro or its aligned counterpart, @code{ASM_OUTPUT_ALIGNED_BSS}.
7549 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7550 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7551 You do not need to do both.
7553 Some languages do not have @code{common} data, and require a
7554 non-common form of global BSS in order to handle uninitialized globals
7555 efficiently. C++ is one example of this. However, if the target does
7556 not support global BSS, the front end may choose to make globals
7557 common in order to save space in the object file.
7560 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7561 Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a
7562 separate, explicit argument. If you define this macro, it is used in
7563 place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in
7564 handling the required alignment of the variable. The alignment is specified
7565 as the number of bits.
7567 Try to use function @code{asm_output_aligned_bss} defined in file
7568 @file{varasm.c} when defining this macro.
7571 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7572 A C statement (sans semicolon) to output to the stdio stream
7573 @var{stream} the assembler definition of a local-common-label named
7574 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7575 is the size rounded up to whatever alignment the caller wants.
7577 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7578 output the name itself; before and after that, output the additional
7579 assembler syntax for defining the name, and a newline.
7581 This macro controls how the assembler definitions of uninitialized
7582 static variables are output.
7585 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7586 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7587 separate, explicit argument. If you define this macro, it is used in
7588 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7589 handling the required alignment of the variable. The alignment is specified
7590 as the number of bits.
7593 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7594 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7595 variable to be output, if there is one, or @code{NULL_TREE} if there
7596 is no corresponding variable. If you define this macro, GCC will use it
7597 in place of both @code{ASM_OUTPUT_DECL} and
7598 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7599 the variable's decl in order to chose what to output.
7603 @subsection Output and Generation of Labels
7605 @c prevent bad page break with this line
7606 This is about outputting labels.
7608 @findex assemble_name
7609 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7610 A C statement (sans semicolon) to output to the stdio stream
7611 @var{stream} the assembler definition of a label named @var{name}.
7612 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7613 output the name itself; before and after that, output the additional
7614 assembler syntax for defining the name, and a newline. A default
7615 definition of this macro is provided which is correct for most systems.
7618 @findex assemble_name_raw
7619 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7620 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7621 to refer to a compiler-generated label. The default definition uses
7622 @code{assemble_name_raw}, which is like @code{assemble_name} except
7623 that it is more efficient.
7627 A C string containing the appropriate assembler directive to specify the
7628 size of a symbol, without any arguments. On systems that use ELF, the
7629 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7630 systems, the default is not to define this macro.
7632 Define this macro only if it is correct to use the default definitions
7633 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7634 for your system. If you need your own custom definitions of those
7635 macros, or if you do not need explicit symbol sizes at all, do not
7639 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7640 A C statement (sans semicolon) to output to the stdio stream
7641 @var{stream} a directive telling the assembler that the size of the
7642 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7643 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7647 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7648 A C statement (sans semicolon) to output to the stdio stream
7649 @var{stream} a directive telling the assembler to calculate the size of
7650 the symbol @var{name} by subtracting its address from the current
7653 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7654 provided. The default assumes that the assembler recognizes a special
7655 @samp{.} symbol as referring to the current address, and can calculate
7656 the difference between this and another symbol. If your assembler does
7657 not recognize @samp{.} or cannot do calculations with it, you will need
7658 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7662 A C string containing the appropriate assembler directive to specify the
7663 type of a symbol, without any arguments. On systems that use ELF, the
7664 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7665 systems, the default is not to define this macro.
7667 Define this macro only if it is correct to use the default definition of
7668 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7669 custom definition of this macro, or if you do not need explicit symbol
7670 types at all, do not define this macro.
7673 @defmac TYPE_OPERAND_FMT
7674 A C string which specifies (using @code{printf} syntax) the format of
7675 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7676 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7677 the default is not to define this macro.
7679 Define this macro only if it is correct to use the default definition of
7680 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7681 custom definition of this macro, or if you do not need explicit symbol
7682 types at all, do not define this macro.
7685 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7686 A C statement (sans semicolon) to output to the stdio stream
7687 @var{stream} a directive telling the assembler that the type of the
7688 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7689 that string is always either @samp{"function"} or @samp{"object"}, but
7690 you should not count on this.
7692 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7693 definition of this macro is provided.
7696 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7697 A C statement (sans semicolon) to output to the stdio stream
7698 @var{stream} any text necessary for declaring the name @var{name} of a
7699 function which is being defined. This macro is responsible for
7700 outputting the label definition (perhaps using
7701 @code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the
7702 @code{FUNCTION_DECL} tree node representing the function.
7704 If this macro is not defined, then the function name is defined in the
7705 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7707 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7711 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7712 A C statement (sans semicolon) to output to the stdio stream
7713 @var{stream} any text necessary for declaring the size of a function
7714 which is being defined. The argument @var{name} is the name of the
7715 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7716 representing the function.
7718 If this macro is not defined, then the function size is not defined.
7720 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7724 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7725 A C statement (sans semicolon) to output to the stdio stream
7726 @var{stream} any text necessary for declaring the name @var{name} of an
7727 initialized variable which is being defined. This macro must output the
7728 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7729 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7731 If this macro is not defined, then the variable name is defined in the
7732 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7734 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7735 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7738 @defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size})
7739 A C statement (sans semicolon) to output to the stdio stream
7740 @var{stream} any text necessary for declaring the name @var{name} of a
7741 constant which is being defined. This macro is responsible for
7742 outputting the label definition (perhaps using
7743 @code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the
7744 value of the constant, and @var{size} is the size of the constant
7745 in bytes. @var{name} will be an internal label.
7747 If this macro is not defined, then the @var{name} is defined in the
7748 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7750 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7754 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7755 A C statement (sans semicolon) to output to the stdio stream
7756 @var{stream} any text necessary for claiming a register @var{regno}
7757 for a global variable @var{decl} with name @var{name}.
7759 If you don't define this macro, that is equivalent to defining it to do
7763 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7764 A C statement (sans semicolon) to finish up declaring a variable name
7765 once the compiler has processed its initializer fully and thus has had a
7766 chance to determine the size of an array when controlled by an
7767 initializer. This is used on systems where it's necessary to declare
7768 something about the size of the object.
7770 If you don't define this macro, that is equivalent to defining it to do
7773 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7774 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7777 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
7778 This target hook is a function to output to the stdio stream
7779 @var{stream} some commands that will make the label @var{name} global;
7780 that is, available for reference from other files.
7782 The default implementation relies on a proper definition of
7783 @code{GLOBAL_ASM_OP}.
7786 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
7787 This target hook is a function to output to the stdio stream
7788 @var{stream} some commands that will make the name associated with @var{decl}
7789 global; that is, available for reference from other files.
7791 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7794 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7795 A C statement (sans semicolon) to output to the stdio stream
7796 @var{stream} some commands that will make the label @var{name} weak;
7797 that is, available for reference from other files but only used if
7798 no other definition is available. Use the expression
7799 @code{assemble_name (@var{stream}, @var{name})} to output the name
7800 itself; before and after that, output the additional assembler syntax
7801 for making that name weak, and a newline.
7803 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7804 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7808 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7809 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7810 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7811 or variable decl. If @var{value} is not @code{NULL}, this C statement
7812 should output to the stdio stream @var{stream} assembler code which
7813 defines (equates) the weak symbol @var{name} to have the value
7814 @var{value}. If @var{value} is @code{NULL}, it should output commands
7815 to make @var{name} weak.
7818 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7819 Outputs a directive that enables @var{name} to be used to refer to
7820 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7821 declaration of @code{name}.
7824 @defmac SUPPORTS_WEAK
7825 A C expression which evaluates to true if the target supports weak symbols.
7827 If you don't define this macro, @file{defaults.h} provides a default
7828 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7829 is defined, the default definition is @samp{1}; otherwise, it is
7830 @samp{0}. Define this macro if you want to control weak symbol support
7831 with a compiler flag such as @option{-melf}.
7834 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
7835 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
7836 public symbol such that extra copies in multiple translation units will
7837 be discarded by the linker. Define this macro if your object file
7838 format provides support for this concept, such as the @samp{COMDAT}
7839 section flags in the Microsoft Windows PE/COFF format, and this support
7840 requires changes to @var{decl}, such as putting it in a separate section.
7843 @defmac SUPPORTS_ONE_ONLY
7844 A C expression which evaluates to true if the target supports one-only
7847 If you don't define this macro, @file{varasm.c} provides a default
7848 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
7849 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
7850 you want to control one-only symbol support with a compiler flag, or if
7851 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
7852 be emitted as one-only.
7855 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility})
7856 This target hook is a function to output to @var{asm_out_file} some
7857 commands that will make the symbol(s) associated with @var{decl} have
7858 hidden, protected or internal visibility as specified by @var{visibility}.
7861 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
7862 A C expression that evaluates to true if the target's linker expects
7863 that weak symbols do not appear in a static archive's table of contents.
7864 The default is @code{0}.
7866 Leaving weak symbols out of an archive's table of contents means that,
7867 if a symbol will only have a definition in one translation unit and
7868 will have undefined references from other translation units, that
7869 symbol should not be weak. Defining this macro to be nonzero will
7870 thus have the effect that certain symbols that would normally be weak
7871 (explicit template instantiations, and vtables for polymorphic classes
7872 with noninline key methods) will instead be nonweak.
7874 The C++ ABI requires this macro to be zero. Define this macro for
7875 targets where full C++ ABI compliance is impossible and where linker
7876 restrictions require weak symbols to be left out of a static archive's
7880 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
7881 A C statement (sans semicolon) to output to the stdio stream
7882 @var{stream} any text necessary for declaring the name of an external
7883 symbol named @var{name} which is referenced in this compilation but
7884 not defined. The value of @var{decl} is the tree node for the
7887 This macro need not be defined if it does not need to output anything.
7888 The GNU assembler and most Unix assemblers don't require anything.
7891 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
7892 This target hook is a function to output to @var{asm_out_file} an assembler
7893 pseudo-op to declare a library function name external. The name of the
7894 library function is given by @var{symref}, which is a @code{symbol_ref}.
7897 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol})
7898 This target hook is a function to output to @var{asm_out_file} an assembler
7899 directive to annotate @var{symbol} as used. The Darwin target uses the
7900 .no_dead_code_strip directive.
7903 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
7904 A C statement (sans semicolon) to output to the stdio stream
7905 @var{stream} a reference in assembler syntax to a label named
7906 @var{name}. This should add @samp{_} to the front of the name, if that
7907 is customary on your operating system, as it is in most Berkeley Unix
7908 systems. This macro is used in @code{assemble_name}.
7911 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
7912 A C statement (sans semicolon) to output a reference to
7913 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
7914 will be used to output the name of the symbol. This macro may be used
7915 to modify the way a symbol is referenced depending on information
7916 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
7919 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
7920 A C statement (sans semicolon) to output a reference to @var{buf}, the
7921 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
7922 @code{assemble_name} will be used to output the name of the symbol.
7923 This macro is not used by @code{output_asm_label}, or the @code{%l}
7924 specifier that calls it; the intention is that this macro should be set
7925 when it is necessary to output a label differently when its address is
7929 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
7930 A function to output to the stdio stream @var{stream} a label whose
7931 name is made from the string @var{prefix} and the number @var{labelno}.
7933 It is absolutely essential that these labels be distinct from the labels
7934 used for user-level functions and variables. Otherwise, certain programs
7935 will have name conflicts with internal labels.
7937 It is desirable to exclude internal labels from the symbol table of the
7938 object file. Most assemblers have a naming convention for labels that
7939 should be excluded; on many systems, the letter @samp{L} at the
7940 beginning of a label has this effect. You should find out what
7941 convention your system uses, and follow it.
7943 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
7946 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
7947 A C statement to output to the stdio stream @var{stream} a debug info
7948 label whose name is made from the string @var{prefix} and the number
7949 @var{num}. This is useful for VLIW targets, where debug info labels
7950 may need to be treated differently than branch target labels. On some
7951 systems, branch target labels must be at the beginning of instruction
7952 bundles, but debug info labels can occur in the middle of instruction
7955 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
7959 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
7960 A C statement to store into the string @var{string} a label whose name
7961 is made from the string @var{prefix} and the number @var{num}.
7963 This string, when output subsequently by @code{assemble_name}, should
7964 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
7965 with the same @var{prefix} and @var{num}.
7967 If the string begins with @samp{*}, then @code{assemble_name} will
7968 output the rest of the string unchanged. It is often convenient for
7969 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
7970 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
7971 to output the string, and may change it. (Of course,
7972 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
7973 you should know what it does on your machine.)
7976 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
7977 A C expression to assign to @var{outvar} (which is a variable of type
7978 @code{char *}) a newly allocated string made from the string
7979 @var{name} and the number @var{number}, with some suitable punctuation
7980 added. Use @code{alloca} to get space for the string.
7982 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
7983 produce an assembler label for an internal static variable whose name is
7984 @var{name}. Therefore, the string must be such as to result in valid
7985 assembler code. The argument @var{number} is different each time this
7986 macro is executed; it prevents conflicts between similarly-named
7987 internal static variables in different scopes.
7989 Ideally this string should not be a valid C identifier, to prevent any
7990 conflict with the user's own symbols. Most assemblers allow periods
7991 or percent signs in assembler symbols; putting at least one of these
7992 between the name and the number will suffice.
7994 If this macro is not defined, a default definition will be provided
7995 which is correct for most systems.
7998 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
7999 A C statement to output to the stdio stream @var{stream} assembler code
8000 which defines (equates) the symbol @var{name} to have the value @var{value}.
8003 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8004 correct for most systems.
8007 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8008 A C statement to output to the stdio stream @var{stream} assembler code
8009 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8010 to have the value of the tree node @var{decl_of_value}. This macro will
8011 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8012 the tree nodes are available.
8015 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8016 correct for most systems.
8019 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8020 A C statement that evaluates to true if the assembler code which defines
8021 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8022 of the tree node @var{decl_of_value} should be emitted near the end of the
8023 current compilation unit. The default is to not defer output of defines.
8024 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8025 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8028 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8029 A C statement to output to the stdio stream @var{stream} assembler code
8030 which defines (equates) the weak symbol @var{name} to have the value
8031 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8032 an undefined weak symbol.
8034 Define this macro if the target only supports weak aliases; define
8035 @code{ASM_OUTPUT_DEF} instead if possible.
8038 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8039 Define this macro to override the default assembler names used for
8040 Objective-C methods.
8042 The default name is a unique method number followed by the name of the
8043 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8044 the category is also included in the assembler name (e.g.@:
8047 These names are safe on most systems, but make debugging difficult since
8048 the method's selector is not present in the name. Therefore, particular
8049 systems define other ways of computing names.
8051 @var{buf} is an expression of type @code{char *} which gives you a
8052 buffer in which to store the name; its length is as long as
8053 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8054 50 characters extra.
8056 The argument @var{is_inst} specifies whether the method is an instance
8057 method or a class method; @var{class_name} is the name of the class;
8058 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8059 in a category); and @var{sel_name} is the name of the selector.
8061 On systems where the assembler can handle quoted names, you can use this
8062 macro to provide more human-readable names.
8065 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
8066 A C statement (sans semicolon) to output to the stdio stream
8067 @var{stream} commands to declare that the label @var{name} is an
8068 Objective-C class reference. This is only needed for targets whose
8069 linkers have special support for NeXT-style runtimes.
8072 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
8073 A C statement (sans semicolon) to output to the stdio stream
8074 @var{stream} commands to declare that the label @var{name} is an
8075 unresolved Objective-C class reference. This is only needed for targets
8076 whose linkers have special support for NeXT-style runtimes.
8079 @node Initialization
8080 @subsection How Initialization Functions Are Handled
8081 @cindex initialization routines
8082 @cindex termination routines
8083 @cindex constructors, output of
8084 @cindex destructors, output of
8086 The compiled code for certain languages includes @dfn{constructors}
8087 (also called @dfn{initialization routines})---functions to initialize
8088 data in the program when the program is started. These functions need
8089 to be called before the program is ``started''---that is to say, before
8090 @code{main} is called.
8092 Compiling some languages generates @dfn{destructors} (also called
8093 @dfn{termination routines}) that should be called when the program
8096 To make the initialization and termination functions work, the compiler
8097 must output something in the assembler code to cause those functions to
8098 be called at the appropriate time. When you port the compiler to a new
8099 system, you need to specify how to do this.
8101 There are two major ways that GCC currently supports the execution of
8102 initialization and termination functions. Each way has two variants.
8103 Much of the structure is common to all four variations.
8105 @findex __CTOR_LIST__
8106 @findex __DTOR_LIST__
8107 The linker must build two lists of these functions---a list of
8108 initialization functions, called @code{__CTOR_LIST__}, and a list of
8109 termination functions, called @code{__DTOR_LIST__}.
8111 Each list always begins with an ignored function pointer (which may hold
8112 0, @minus{}1, or a count of the function pointers after it, depending on
8113 the environment). This is followed by a series of zero or more function
8114 pointers to constructors (or destructors), followed by a function
8115 pointer containing zero.
8117 Depending on the operating system and its executable file format, either
8118 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8119 time and exit time. Constructors are called in reverse order of the
8120 list; destructors in forward order.
8122 The best way to handle static constructors works only for object file
8123 formats which provide arbitrarily-named sections. A section is set
8124 aside for a list of constructors, and another for a list of destructors.
8125 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8126 object file that defines an initialization function also puts a word in
8127 the constructor section to point to that function. The linker
8128 accumulates all these words into one contiguous @samp{.ctors} section.
8129 Termination functions are handled similarly.
8131 This method will be chosen as the default by @file{target-def.h} if
8132 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8133 support arbitrary sections, but does support special designated
8134 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8135 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8137 When arbitrary sections are available, there are two variants, depending
8138 upon how the code in @file{crtstuff.c} is called. On systems that
8139 support a @dfn{.init} section which is executed at program startup,
8140 parts of @file{crtstuff.c} are compiled into that section. The
8141 program is linked by the @command{gcc} driver like this:
8144 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8147 The prologue of a function (@code{__init}) appears in the @code{.init}
8148 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8149 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8150 files are provided by the operating system or by the GNU C library, but
8151 are provided by GCC for a few targets.
8153 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8154 compiled from @file{crtstuff.c}. They contain, among other things, code
8155 fragments within the @code{.init} and @code{.fini} sections that branch
8156 to routines in the @code{.text} section. The linker will pull all parts
8157 of a section together, which results in a complete @code{__init} function
8158 that invokes the routines we need at startup.
8160 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8163 If no init section is available, when GCC compiles any function called
8164 @code{main} (or more accurately, any function designated as a program
8165 entry point by the language front end calling @code{expand_main_function}),
8166 it inserts a procedure call to @code{__main} as the first executable code
8167 after the function prologue. The @code{__main} function is defined
8168 in @file{libgcc2.c} and runs the global constructors.
8170 In file formats that don't support arbitrary sections, there are again
8171 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8172 and an `a.out' format must be used. In this case,
8173 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8174 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8175 and with the address of the void function containing the initialization
8176 code as its value. The GNU linker recognizes this as a request to add
8177 the value to a @dfn{set}; the values are accumulated, and are eventually
8178 placed in the executable as a vector in the format described above, with
8179 a leading (ignored) count and a trailing zero element.
8180 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8181 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8182 the compilation of @code{main} to call @code{__main} as above, starting
8183 the initialization process.
8185 The last variant uses neither arbitrary sections nor the GNU linker.
8186 This is preferable when you want to do dynamic linking and when using
8187 file formats which the GNU linker does not support, such as `ECOFF'@. In
8188 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8189 termination functions are recognized simply by their names. This requires
8190 an extra program in the linkage step, called @command{collect2}. This program
8191 pretends to be the linker, for use with GCC; it does its job by running
8192 the ordinary linker, but also arranges to include the vectors of
8193 initialization and termination functions. These functions are called
8194 via @code{__main} as described above. In order to use this method,
8195 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8198 The following section describes the specific macros that control and
8199 customize the handling of initialization and termination functions.
8202 @node Macros for Initialization
8203 @subsection Macros Controlling Initialization Routines
8205 Here are the macros that control how the compiler handles initialization
8206 and termination functions:
8208 @defmac INIT_SECTION_ASM_OP
8209 If defined, a C string constant, including spacing, for the assembler
8210 operation to identify the following data as initialization code. If not
8211 defined, GCC will assume such a section does not exist. When you are
8212 using special sections for initialization and termination functions, this
8213 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8214 run the initialization functions.
8217 @defmac HAS_INIT_SECTION
8218 If defined, @code{main} will not call @code{__main} as described above.
8219 This macro should be defined for systems that control start-up code
8220 on a symbol-by-symbol basis, such as OSF/1, and should not
8221 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8224 @defmac LD_INIT_SWITCH
8225 If defined, a C string constant for a switch that tells the linker that
8226 the following symbol is an initialization routine.
8229 @defmac LD_FINI_SWITCH
8230 If defined, a C string constant for a switch that tells the linker that
8231 the following symbol is a finalization routine.
8234 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8235 If defined, a C statement that will write a function that can be
8236 automatically called when a shared library is loaded. The function
8237 should call @var{func}, which takes no arguments. If not defined, and
8238 the object format requires an explicit initialization function, then a
8239 function called @code{_GLOBAL__DI} will be generated.
8241 This function and the following one are used by collect2 when linking a
8242 shared library that needs constructors or destructors, or has DWARF2
8243 exception tables embedded in the code.
8246 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8247 If defined, a C statement that will write a function that can be
8248 automatically called when a shared library is unloaded. The function
8249 should call @var{func}, which takes no arguments. If not defined, and
8250 the object format requires an explicit finalization function, then a
8251 function called @code{_GLOBAL__DD} will be generated.
8254 @defmac INVOKE__main
8255 If defined, @code{main} will call @code{__main} despite the presence of
8256 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8257 where the init section is not actually run automatically, but is still
8258 useful for collecting the lists of constructors and destructors.
8261 @defmac SUPPORTS_INIT_PRIORITY
8262 If nonzero, the C++ @code{init_priority} attribute is supported and the
8263 compiler should emit instructions to control the order of initialization
8264 of objects. If zero, the compiler will issue an error message upon
8265 encountering an @code{init_priority} attribute.
8268 @deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS
8269 This value is true if the target supports some ``native'' method of
8270 collecting constructors and destructors to be run at startup and exit.
8271 It is false if we must use @command{collect2}.
8274 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
8275 If defined, a function that outputs assembler code to arrange to call
8276 the function referenced by @var{symbol} at initialization time.
8278 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8279 no arguments and with no return value. If the target supports initialization
8280 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8281 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8283 If this macro is not defined by the target, a suitable default will
8284 be chosen if (1) the target supports arbitrary section names, (2) the
8285 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8289 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
8290 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8291 functions rather than initialization functions.
8294 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8295 generated for the generated object file will have static linkage.
8297 If your system uses @command{collect2} as the means of processing
8298 constructors, then that program normally uses @command{nm} to scan
8299 an object file for constructor functions to be called.
8301 On certain kinds of systems, you can define this macro to make
8302 @command{collect2} work faster (and, in some cases, make it work at all):
8304 @defmac OBJECT_FORMAT_COFF
8305 Define this macro if the system uses COFF (Common Object File Format)
8306 object files, so that @command{collect2} can assume this format and scan
8307 object files directly for dynamic constructor/destructor functions.
8309 This macro is effective only in a native compiler; @command{collect2} as
8310 part of a cross compiler always uses @command{nm} for the target machine.
8313 @defmac REAL_NM_FILE_NAME
8314 Define this macro as a C string constant containing the file name to use
8315 to execute @command{nm}. The default is to search the path normally for
8318 If your system supports shared libraries and has a program to list the
8319 dynamic dependencies of a given library or executable, you can define
8320 these macros to enable support for running initialization and
8321 termination functions in shared libraries:
8325 Define this macro to a C string constant containing the name of the program
8326 which lists dynamic dependencies, like @command{"ldd"} under SunOS 4.
8329 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8330 Define this macro to be C code that extracts filenames from the output
8331 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8332 of type @code{char *} that points to the beginning of a line of output
8333 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8334 code must advance @var{ptr} to the beginning of the filename on that
8335 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8338 @defmac SHLIB_SUFFIX
8339 Define this macro to a C string constant containing the default shared
8340 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8341 strips version information after this suffix when generating global
8342 constructor and destructor names. This define is only needed on targets
8343 that use @command{collect2} to process constructors and destructors.
8346 @node Instruction Output
8347 @subsection Output of Assembler Instructions
8349 @c prevent bad page break with this line
8350 This describes assembler instruction output.
8352 @defmac REGISTER_NAMES
8353 A C initializer containing the assembler's names for the machine
8354 registers, each one as a C string constant. This is what translates
8355 register numbers in the compiler into assembler language.
8358 @defmac ADDITIONAL_REGISTER_NAMES
8359 If defined, a C initializer for an array of structures containing a name
8360 and a register number. This macro defines additional names for hard
8361 registers, thus allowing the @code{asm} option in declarations to refer
8362 to registers using alternate names.
8365 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8366 Define this macro if you are using an unusual assembler that
8367 requires different names for the machine instructions.
8369 The definition is a C statement or statements which output an
8370 assembler instruction opcode to the stdio stream @var{stream}. The
8371 macro-operand @var{ptr} is a variable of type @code{char *} which
8372 points to the opcode name in its ``internal'' form---the form that is
8373 written in the machine description. The definition should output the
8374 opcode name to @var{stream}, performing any translation you desire, and
8375 increment the variable @var{ptr} to point at the end of the opcode
8376 so that it will not be output twice.
8378 In fact, your macro definition may process less than the entire opcode
8379 name, or more than the opcode name; but if you want to process text
8380 that includes @samp{%}-sequences to substitute operands, you must take
8381 care of the substitution yourself. Just be sure to increment
8382 @var{ptr} over whatever text should not be output normally.
8384 @findex recog_data.operand
8385 If you need to look at the operand values, they can be found as the
8386 elements of @code{recog_data.operand}.
8388 If the macro definition does nothing, the instruction is output
8392 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8393 If defined, a C statement to be executed just prior to the output of
8394 assembler code for @var{insn}, to modify the extracted operands so
8395 they will be output differently.
8397 Here the argument @var{opvec} is the vector containing the operands
8398 extracted from @var{insn}, and @var{noperands} is the number of
8399 elements of the vector which contain meaningful data for this insn.
8400 The contents of this vector are what will be used to convert the insn
8401 template into assembler code, so you can change the assembler output
8402 by changing the contents of the vector.
8404 This macro is useful when various assembler syntaxes share a single
8405 file of instruction patterns; by defining this macro differently, you
8406 can cause a large class of instructions to be output differently (such
8407 as with rearranged operands). Naturally, variations in assembler
8408 syntax affecting individual insn patterns ought to be handled by
8409 writing conditional output routines in those patterns.
8411 If this macro is not defined, it is equivalent to a null statement.
8414 @deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx @var{insn}, rtx *@var{opvec}, int @var{noperands})
8415 If defined, this target hook is a function which is executed just after the
8416 output of assembler code for @var{insn}, to change the mode of the assembler
8419 Here the argument @var{opvec} is the vector containing the operands
8420 extracted from @var{insn}, and @var{noperands} is the number of
8421 elements of the vector which contain meaningful data for this insn.
8422 The contents of this vector are what was used to convert the insn
8423 template into assembler code, so you can change the assembler mode
8424 by checking the contents of the vector.
8427 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8428 A C compound statement to output to stdio stream @var{stream} the
8429 assembler syntax for an instruction operand @var{x}. @var{x} is an
8432 @var{code} is a value that can be used to specify one of several ways
8433 of printing the operand. It is used when identical operands must be
8434 printed differently depending on the context. @var{code} comes from
8435 the @samp{%} specification that was used to request printing of the
8436 operand. If the specification was just @samp{%@var{digit}} then
8437 @var{code} is 0; if the specification was @samp{%@var{ltr}
8438 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8441 If @var{x} is a register, this macro should print the register's name.
8442 The names can be found in an array @code{reg_names} whose type is
8443 @code{char *[]}. @code{reg_names} is initialized from
8444 @code{REGISTER_NAMES}.
8446 When the machine description has a specification @samp{%@var{punct}}
8447 (a @samp{%} followed by a punctuation character), this macro is called
8448 with a null pointer for @var{x} and the punctuation character for
8452 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8453 A C expression which evaluates to true if @var{code} is a valid
8454 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8455 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8456 punctuation characters (except for the standard one, @samp{%}) are used
8460 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8461 A C compound statement to output to stdio stream @var{stream} the
8462 assembler syntax for an instruction operand that is a memory reference
8463 whose address is @var{x}. @var{x} is an RTL expression.
8465 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8466 On some machines, the syntax for a symbolic address depends on the
8467 section that the address refers to. On these machines, define the hook
8468 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8469 @code{symbol_ref}, and then check for it here. @xref{Assembler
8473 @findex dbr_sequence_length
8474 @defmac DBR_OUTPUT_SEQEND (@var{file})
8475 A C statement, to be executed after all slot-filler instructions have
8476 been output. If necessary, call @code{dbr_sequence_length} to
8477 determine the number of slots filled in a sequence (zero if not
8478 currently outputting a sequence), to decide how many no-ops to output,
8481 Don't define this macro if it has nothing to do, but it is helpful in
8482 reading assembly output if the extent of the delay sequence is made
8483 explicit (e.g.@: with white space).
8486 @findex final_sequence
8487 Note that output routines for instructions with delay slots must be
8488 prepared to deal with not being output as part of a sequence
8489 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8490 found.) The variable @code{final_sequence} is null when not
8491 processing a sequence, otherwise it contains the @code{sequence} rtx
8495 @defmac REGISTER_PREFIX
8496 @defmacx LOCAL_LABEL_PREFIX
8497 @defmacx USER_LABEL_PREFIX
8498 @defmacx IMMEDIATE_PREFIX
8499 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8500 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8501 @file{final.c}). These are useful when a single @file{md} file must
8502 support multiple assembler formats. In that case, the various @file{tm.h}
8503 files can define these macros differently.
8506 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8507 If defined this macro should expand to a series of @code{case}
8508 statements which will be parsed inside the @code{switch} statement of
8509 the @code{asm_fprintf} function. This allows targets to define extra
8510 printf formats which may useful when generating their assembler
8511 statements. Note that uppercase letters are reserved for future
8512 generic extensions to asm_fprintf, and so are not available to target
8513 specific code. The output file is given by the parameter @var{file}.
8514 The varargs input pointer is @var{argptr} and the rest of the format
8515 string, starting the character after the one that is being switched
8516 upon, is pointed to by @var{format}.
8519 @defmac ASSEMBLER_DIALECT
8520 If your target supports multiple dialects of assembler language (such as
8521 different opcodes), define this macro as a C expression that gives the
8522 numeric index of the assembler language dialect to use, with zero as the
8525 If this macro is defined, you may use constructs of the form
8527 @samp{@{option0|option1|option2@dots{}@}}
8530 in the output templates of patterns (@pxref{Output Template}) or in the
8531 first argument of @code{asm_fprintf}. This construct outputs
8532 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8533 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8534 within these strings retain their usual meaning. If there are fewer
8535 alternatives within the braces than the value of
8536 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8538 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8539 @samp{@}} do not have any special meaning when used in templates or
8540 operands to @code{asm_fprintf}.
8542 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8543 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8544 the variations in assembler language syntax with that mechanism. Define
8545 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8546 if the syntax variant are larger and involve such things as different
8547 opcodes or operand order.
8550 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8551 A C expression to output to @var{stream} some assembler code
8552 which will push hard register number @var{regno} onto the stack.
8553 The code need not be optimal, since this macro is used only when
8557 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8558 A C expression to output to @var{stream} some assembler code
8559 which will pop hard register number @var{regno} off of the stack.
8560 The code need not be optimal, since this macro is used only when
8564 @node Dispatch Tables
8565 @subsection Output of Dispatch Tables
8567 @c prevent bad page break with this line
8568 This concerns dispatch tables.
8570 @cindex dispatch table
8571 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8572 A C statement to output to the stdio stream @var{stream} an assembler
8573 pseudo-instruction to generate a difference between two labels.
8574 @var{value} and @var{rel} are the numbers of two internal labels. The
8575 definitions of these labels are output using
8576 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8577 way here. For example,
8580 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8581 @var{value}, @var{rel})
8584 You must provide this macro on machines where the addresses in a
8585 dispatch table are relative to the table's own address. If defined, GCC
8586 will also use this macro on all machines when producing PIC@.
8587 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8588 mode and flags can be read.
8591 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8592 This macro should be provided on machines where the addresses
8593 in a dispatch table are absolute.
8595 The definition should be a C statement to output to the stdio stream
8596 @var{stream} an assembler pseudo-instruction to generate a reference to
8597 a label. @var{value} is the number of an internal label whose
8598 definition is output using @code{(*targetm.asm_out.internal_label)}.
8602 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8606 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8607 Define this if the label before a jump-table needs to be output
8608 specially. The first three arguments are the same as for
8609 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8610 jump-table which follows (a @code{jump_insn} containing an
8611 @code{addr_vec} or @code{addr_diff_vec}).
8613 This feature is used on system V to output a @code{swbeg} statement
8616 If this macro is not defined, these labels are output with
8617 @code{(*targetm.asm_out.internal_label)}.
8620 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8621 Define this if something special must be output at the end of a
8622 jump-table. The definition should be a C statement to be executed
8623 after the assembler code for the table is written. It should write
8624 the appropriate code to stdio stream @var{stream}. The argument
8625 @var{table} is the jump-table insn, and @var{num} is the label-number
8626 of the preceding label.
8628 If this macro is not defined, nothing special is output at the end of
8632 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty})
8633 This target hook emits a label at the beginning of each FDE@. It
8634 should be defined on targets where FDEs need special labels, and it
8635 should write the appropriate label, for the FDE associated with the
8636 function declaration @var{decl}, to the stdio stream @var{stream}.
8637 The third argument, @var{for_eh}, is a boolean: true if this is for an
8638 exception table. The fourth argument, @var{empty}, is a boolean:
8639 true if this is a placeholder label for an omitted FDE@.
8641 The default is that FDEs are not given nonlocal labels.
8644 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream})
8645 This target hook emits a label at the beginning of the exception table.
8646 It should be defined on targets where it is desirable for the table
8647 to be broken up according to function.
8649 The default is that no label is emitted.
8652 @deftypefn {Target Hook} void TARGET_UNWIND_EMIT (FILE *@var{stream}, rtx @var{insn})
8653 This target hook emits assembly directives required to unwind the
8654 given instruction. This is only used when TARGET_UNWIND_INFO is set.
8657 @node Exception Region Output
8658 @subsection Assembler Commands for Exception Regions
8660 @c prevent bad page break with this line
8662 This describes commands marking the start and the end of an exception
8665 @defmac EH_FRAME_SECTION_NAME
8666 If defined, a C string constant for the name of the section containing
8667 exception handling frame unwind information. If not defined, GCC will
8668 provide a default definition if the target supports named sections.
8669 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8671 You should define this symbol if your target supports DWARF 2 frame
8672 unwind information and the default definition does not work.
8675 @defmac EH_FRAME_IN_DATA_SECTION
8676 If defined, DWARF 2 frame unwind information will be placed in the
8677 data section even though the target supports named sections. This
8678 might be necessary, for instance, if the system linker does garbage
8679 collection and sections cannot be marked as not to be collected.
8681 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8685 @defmac EH_TABLES_CAN_BE_READ_ONLY
8686 Define this macro to 1 if your target is such that no frame unwind
8687 information encoding used with non-PIC code will ever require a
8688 runtime relocation, but the linker may not support merging read-only
8689 and read-write sections into a single read-write section.
8692 @defmac MASK_RETURN_ADDR
8693 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8694 that it does not contain any extraneous set bits in it.
8697 @defmac DWARF2_UNWIND_INFO
8698 Define this macro to 0 if your target supports DWARF 2 frame unwind
8699 information, but it does not yet work with exception handling.
8700 Otherwise, if your target supports this information (if it defines
8701 @samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP}
8702 or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8704 If @code{TARGET_UNWIND_INFO} is defined, the target specific unwinder
8705 will be used in all cases. Defining this macro will enable the generation
8706 of DWARF 2 frame debugging information.
8708 If @code{TARGET_UNWIND_INFO} is not defined, and this macro is defined to 1,
8709 the DWARF 2 unwinder will be the default exception handling mechanism;
8710 otherwise, the @code{setjmp}/@code{longjmp}-based scheme will be used by
8714 @defmac TARGET_UNWIND_INFO
8715 Define this macro if your target has ABI specified unwind tables. Usually
8716 these will be output by @code{TARGET_UNWIND_EMIT}.
8719 @deftypevr {Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
8720 This variable should be set to @code{true} if the target ABI requires unwinding
8721 tables even when exceptions are not used.
8724 @defmac MUST_USE_SJLJ_EXCEPTIONS
8725 This macro need only be defined if @code{DWARF2_UNWIND_INFO} is
8726 runtime-variable. In that case, @file{except.h} cannot correctly
8727 determine the corresponding definition of @code{MUST_USE_SJLJ_EXCEPTIONS},
8728 so the target must provide it directly.
8731 @defmac DONT_USE_BUILTIN_SETJMP
8732 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8733 should use the @code{setjmp}/@code{longjmp} functions from the C library
8734 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8737 @defmac DWARF_CIE_DATA_ALIGNMENT
8738 This macro need only be defined if the target might save registers in the
8739 function prologue at an offset to the stack pointer that is not aligned to
8740 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8741 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8742 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8743 the target supports DWARF 2 frame unwind information.
8746 @deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
8747 Contains the value true if the target should add a zero word onto the
8748 end of a Dwarf-2 frame info section when used for exception handling.
8749 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8753 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
8754 Given a register, this hook should return a parallel of registers to
8755 represent where to find the register pieces. Define this hook if the
8756 register and its mode are represented in Dwarf in non-contiguous
8757 locations, or if the register should be represented in more than one
8758 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8759 If not defined, the default is to return @code{NULL_RTX}.
8762 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
8763 If some registers are represented in Dwarf-2 unwind information in
8764 multiple pieces, define this hook to fill in information about the
8765 sizes of those pieces in the table used by the unwinder at runtime.
8766 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8767 filling in a single size corresponding to each hard register;
8768 @var{address} is the address of the table.
8771 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
8772 This hook is used to output a reference from a frame unwinding table to
8773 the type_info object identified by @var{sym}. It should return @code{true}
8774 if the reference was output. Returning @code{false} will cause the
8775 reference to be output using the normal Dwarf2 routines.
8778 @deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER
8779 This flag should be set to @code{true} on targets that use an ARM EABI
8780 based unwinding library, and @code{false} on other targets. This effects
8781 the format of unwinding tables, and how the unwinder in entered after
8782 running a cleanup. The default is @code{false}.
8785 @node Alignment Output
8786 @subsection Assembler Commands for Alignment
8788 @c prevent bad page break with this line
8789 This describes commands for alignment.
8791 @defmac JUMP_ALIGN (@var{label})
8792 The alignment (log base 2) to put in front of @var{label}, which is
8793 a common destination of jumps and has no fallthru incoming edge.
8795 This macro need not be defined if you don't want any special alignment
8796 to be done at such a time. Most machine descriptions do not currently
8799 Unless it's necessary to inspect the @var{label} parameter, it is better
8800 to set the variable @var{align_jumps} in the target's
8801 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8802 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
8805 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
8806 The alignment (log base 2) to put in front of @var{label}, which follows
8809 This macro need not be defined if you don't want any special alignment
8810 to be done at such a time. Most machine descriptions do not currently
8814 @defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP
8815 The maximum number of bytes to skip when applying
8816 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
8817 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8820 @defmac LOOP_ALIGN (@var{label})
8821 The alignment (log base 2) to put in front of @var{label}, which follows
8822 a @code{NOTE_INSN_LOOP_BEG} note.
8824 This macro need not be defined if you don't want any special alignment
8825 to be done at such a time. Most machine descriptions do not currently
8828 Unless it's necessary to inspect the @var{label} parameter, it is better
8829 to set the variable @code{align_loops} in the target's
8830 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8831 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
8834 @defmac LOOP_ALIGN_MAX_SKIP
8835 The maximum number of bytes to skip when applying @code{LOOP_ALIGN}.
8836 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8839 @defmac LABEL_ALIGN (@var{label})
8840 The alignment (log base 2) to put in front of @var{label}.
8841 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
8842 the maximum of the specified values is used.
8844 Unless it's necessary to inspect the @var{label} parameter, it is better
8845 to set the variable @code{align_labels} in the target's
8846 @code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's
8847 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
8850 @defmac LABEL_ALIGN_MAX_SKIP
8851 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}.
8852 This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
8855 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
8856 A C statement to output to the stdio stream @var{stream} an assembler
8857 instruction to advance the location counter by @var{nbytes} bytes.
8858 Those bytes should be zero when loaded. @var{nbytes} will be a C
8859 expression of type @code{unsigned HOST_WIDE_INT}.
8862 @defmac ASM_NO_SKIP_IN_TEXT
8863 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
8864 text section because it fails to put zeros in the bytes that are skipped.
8865 This is true on many Unix systems, where the pseudo--op to skip bytes
8866 produces no-op instructions rather than zeros when used in the text
8870 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
8871 A C statement to output to the stdio stream @var{stream} an assembler
8872 command to advance the location counter to a multiple of 2 to the
8873 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
8876 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
8877 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
8878 for padding, if necessary.
8881 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
8882 A C statement to output to the stdio stream @var{stream} an assembler
8883 command to advance the location counter to a multiple of 2 to the
8884 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
8885 satisfy the alignment request. @var{power} and @var{max_skip} will be
8886 a C expression of type @code{int}.
8890 @node Debugging Info
8891 @section Controlling Debugging Information Format
8893 @c prevent bad page break with this line
8894 This describes how to specify debugging information.
8897 * All Debuggers:: Macros that affect all debugging formats uniformly.
8898 * DBX Options:: Macros enabling specific options in DBX format.
8899 * DBX Hooks:: Hook macros for varying DBX format.
8900 * File Names and DBX:: Macros controlling output of file names in DBX format.
8901 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
8902 * VMS Debug:: Macros for VMS debug format.
8906 @subsection Macros Affecting All Debugging Formats
8908 @c prevent bad page break with this line
8909 These macros affect all debugging formats.
8911 @defmac DBX_REGISTER_NUMBER (@var{regno})
8912 A C expression that returns the DBX register number for the compiler
8913 register number @var{regno}. In the default macro provided, the value
8914 of this expression will be @var{regno} itself. But sometimes there are
8915 some registers that the compiler knows about and DBX does not, or vice
8916 versa. In such cases, some register may need to have one number in the
8917 compiler and another for DBX@.
8919 If two registers have consecutive numbers inside GCC, and they can be
8920 used as a pair to hold a multiword value, then they @emph{must} have
8921 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
8922 Otherwise, debuggers will be unable to access such a pair, because they
8923 expect register pairs to be consecutive in their own numbering scheme.
8925 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
8926 does not preserve register pairs, then what you must do instead is
8927 redefine the actual register numbering scheme.
8930 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
8931 A C expression that returns the integer offset value for an automatic
8932 variable having address @var{x} (an RTL expression). The default
8933 computation assumes that @var{x} is based on the frame-pointer and
8934 gives the offset from the frame-pointer. This is required for targets
8935 that produce debugging output for DBX or COFF-style debugging output
8936 for SDB and allow the frame-pointer to be eliminated when the
8937 @option{-g} options is used.
8940 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
8941 A C expression that returns the integer offset value for an argument
8942 having address @var{x} (an RTL expression). The nominal offset is
8946 @defmac PREFERRED_DEBUGGING_TYPE
8947 A C expression that returns the type of debugging output GCC should
8948 produce when the user specifies just @option{-g}. Define
8949 this if you have arranged for GCC to support more than one format of
8950 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
8951 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
8952 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
8954 When the user specifies @option{-ggdb}, GCC normally also uses the
8955 value of this macro to select the debugging output format, but with two
8956 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
8957 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
8958 defined, GCC uses @code{DBX_DEBUG}.
8960 The value of this macro only affects the default debugging output; the
8961 user can always get a specific type of output by using @option{-gstabs},
8962 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
8966 @subsection Specific Options for DBX Output
8968 @c prevent bad page break with this line
8969 These are specific options for DBX output.
8971 @defmac DBX_DEBUGGING_INFO
8972 Define this macro if GCC should produce debugging output for DBX
8973 in response to the @option{-g} option.
8976 @defmac XCOFF_DEBUGGING_INFO
8977 Define this macro if GCC should produce XCOFF format debugging output
8978 in response to the @option{-g} option. This is a variant of DBX format.
8981 @defmac DEFAULT_GDB_EXTENSIONS
8982 Define this macro to control whether GCC should by default generate
8983 GDB's extended version of DBX debugging information (assuming DBX-format
8984 debugging information is enabled at all). If you don't define the
8985 macro, the default is 1: always generate the extended information
8986 if there is any occasion to.
8989 @defmac DEBUG_SYMS_TEXT
8990 Define this macro if all @code{.stabs} commands should be output while
8991 in the text section.
8994 @defmac ASM_STABS_OP
8995 A C string constant, including spacing, naming the assembler pseudo op to
8996 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
8997 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
8998 applies only to DBX debugging information format.
9001 @defmac ASM_STABD_OP
9002 A C string constant, including spacing, naming the assembler pseudo op to
9003 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9004 value is the current location. If you don't define this macro,
9005 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9009 @defmac ASM_STABN_OP
9010 A C string constant, including spacing, naming the assembler pseudo op to
9011 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9012 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9013 macro applies only to DBX debugging information format.
9016 @defmac DBX_NO_XREFS
9017 Define this macro if DBX on your system does not support the construct
9018 @samp{xs@var{tagname}}. On some systems, this construct is used to
9019 describe a forward reference to a structure named @var{tagname}.
9020 On other systems, this construct is not supported at all.
9023 @defmac DBX_CONTIN_LENGTH
9024 A symbol name in DBX-format debugging information is normally
9025 continued (split into two separate @code{.stabs} directives) when it
9026 exceeds a certain length (by default, 80 characters). On some
9027 operating systems, DBX requires this splitting; on others, splitting
9028 must not be done. You can inhibit splitting by defining this macro
9029 with the value zero. You can override the default splitting-length by
9030 defining this macro as an expression for the length you desire.
9033 @defmac DBX_CONTIN_CHAR
9034 Normally continuation is indicated by adding a @samp{\} character to
9035 the end of a @code{.stabs} string when a continuation follows. To use
9036 a different character instead, define this macro as a character
9037 constant for the character you want to use. Do not define this macro
9038 if backslash is correct for your system.
9041 @defmac DBX_STATIC_STAB_DATA_SECTION
9042 Define this macro if it is necessary to go to the data section before
9043 outputting the @samp{.stabs} pseudo-op for a non-global static
9047 @defmac DBX_TYPE_DECL_STABS_CODE
9048 The value to use in the ``code'' field of the @code{.stabs} directive
9049 for a typedef. The default is @code{N_LSYM}.
9052 @defmac DBX_STATIC_CONST_VAR_CODE
9053 The value to use in the ``code'' field of the @code{.stabs} directive
9054 for a static variable located in the text section. DBX format does not
9055 provide any ``right'' way to do this. The default is @code{N_FUN}.
9058 @defmac DBX_REGPARM_STABS_CODE
9059 The value to use in the ``code'' field of the @code{.stabs} directive
9060 for a parameter passed in registers. DBX format does not provide any
9061 ``right'' way to do this. The default is @code{N_RSYM}.
9064 @defmac DBX_REGPARM_STABS_LETTER
9065 The letter to use in DBX symbol data to identify a symbol as a parameter
9066 passed in registers. DBX format does not customarily provide any way to
9067 do this. The default is @code{'P'}.
9070 @defmac DBX_FUNCTION_FIRST
9071 Define this macro if the DBX information for a function and its
9072 arguments should precede the assembler code for the function. Normally,
9073 in DBX format, the debugging information entirely follows the assembler
9077 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9078 Define this macro, with value 1, if the value of a symbol describing
9079 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9080 relative to the start of the enclosing function. Normally, GCC uses
9081 an absolute address.
9084 @defmac DBX_LINES_FUNCTION_RELATIVE
9085 Define this macro, with value 1, if the value of a symbol indicating
9086 the current line number (@code{N_SLINE}) should be relative to the
9087 start of the enclosing function. Normally, GCC uses an absolute address.
9090 @defmac DBX_USE_BINCL
9091 Define this macro if GCC should generate @code{N_BINCL} and
9092 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9093 macro also directs GCC to output a type number as a pair of a file
9094 number and a type number within the file. Normally, GCC does not
9095 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9096 number for a type number.
9100 @subsection Open-Ended Hooks for DBX Format
9102 @c prevent bad page break with this line
9103 These are hooks for DBX format.
9105 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
9106 Define this macro to say how to output to @var{stream} the debugging
9107 information for the start of a scope level for variable names. The
9108 argument @var{name} is the name of an assembler symbol (for use with
9109 @code{assemble_name}) whose value is the address where the scope begins.
9112 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
9113 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
9116 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
9117 Define this macro if the target machine requires special handling to
9118 output an @code{N_FUN} entry for the function @var{decl}.
9121 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9122 A C statement to output DBX debugging information before code for line
9123 number @var{line} of the current source file to the stdio stream
9124 @var{stream}. @var{counter} is the number of time the macro was
9125 invoked, including the current invocation; it is intended to generate
9126 unique labels in the assembly output.
9128 This macro should not be defined if the default output is correct, or
9129 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9132 @defmac NO_DBX_FUNCTION_END
9133 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9134 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9135 On those machines, define this macro to turn this feature off without
9136 disturbing the rest of the gdb extensions.
9139 @defmac NO_DBX_BNSYM_ENSYM
9140 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9141 extension construct. On those machines, define this macro to turn this
9142 feature off without disturbing the rest of the gdb extensions.
9145 @node File Names and DBX
9146 @subsection File Names in DBX Format
9148 @c prevent bad page break with this line
9149 This describes file names in DBX format.
9151 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9152 A C statement to output DBX debugging information to the stdio stream
9153 @var{stream}, which indicates that file @var{name} is the main source
9154 file---the file specified as the input file for compilation.
9155 This macro is called only once, at the beginning of compilation.
9157 This macro need not be defined if the standard form of output
9158 for DBX debugging information is appropriate.
9160 It may be necessary to refer to a label equal to the beginning of the
9161 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9162 to do so. If you do this, you must also set the variable
9163 @var{used_ltext_label_name} to @code{true}.
9166 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9167 Define this macro, with value 1, if GCC should not emit an indication
9168 of the current directory for compilation and current source language at
9169 the beginning of the file.
9172 @defmac NO_DBX_GCC_MARKER
9173 Define this macro, with value 1, if GCC should not emit an indication
9174 that this object file was compiled by GCC@. The default is to emit
9175 an @code{N_OPT} stab at the beginning of every source file, with
9176 @samp{gcc2_compiled.} for the string and value 0.
9179 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9180 A C statement to output DBX debugging information at the end of
9181 compilation of the main source file @var{name}. Output should be
9182 written to the stdio stream @var{stream}.
9184 If you don't define this macro, nothing special is output at the end
9185 of compilation, which is correct for most machines.
9188 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9189 Define this macro @emph{instead of} defining
9190 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9191 the end of compilation is an @code{N_SO} stab with an empty string,
9192 whose value is the highest absolute text address in the file.
9197 @subsection Macros for SDB and DWARF Output
9199 @c prevent bad page break with this line
9200 Here are macros for SDB and DWARF output.
9202 @defmac SDB_DEBUGGING_INFO
9203 Define this macro if GCC should produce COFF-style debugging output
9204 for SDB in response to the @option{-g} option.
9207 @defmac DWARF2_DEBUGGING_INFO
9208 Define this macro if GCC should produce dwarf version 2 format
9209 debugging output in response to the @option{-g} option.
9211 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function})
9212 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9213 be emitted for each function. Instead of an integer return the enum
9214 value for the @code{DW_CC_} tag.
9217 To support optional call frame debugging information, you must also
9218 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9219 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9220 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9221 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9224 @defmac DWARF2_FRAME_INFO
9225 Define this macro to a nonzero value if GCC should always output
9226 Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO}
9227 (@pxref{Exception Region Output} is nonzero, GCC will output this
9228 information not matter how you define @code{DWARF2_FRAME_INFO}.
9231 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9232 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9233 line debug info sections. This will result in much more compact line number
9234 tables, and hence is desirable if it works.
9237 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9238 A C statement to issue assembly directives that create a difference
9239 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9242 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9243 A C statement to issue assembly directives that create a
9244 section-relative reference to the given @var{label}, using an integer of the
9245 given @var{size}. The label is known to be defined in the given @var{section}.
9248 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9249 A C statement to issue assembly directives that create a self-relative
9250 reference to the given @var{label}, using an integer of the given @var{size}.
9253 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9254 A C statement to issue assembly directives that create a reference to
9255 the DWARF table identifier @var{label} from the current section. This
9256 is used on some systems to avoid garbage collecting a DWARF table which
9257 is referenced by a function.
9260 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x})
9261 If defined, this target hook is a function which outputs a DTP-relative
9262 reference to the given TLS symbol of the specified size.
9265 @defmac PUT_SDB_@dots{}
9266 Define these macros to override the assembler syntax for the special
9267 SDB assembler directives. See @file{sdbout.c} for a list of these
9268 macros and their arguments. If the standard syntax is used, you need
9269 not define them yourself.
9273 Some assemblers do not support a semicolon as a delimiter, even between
9274 SDB assembler directives. In that case, define this macro to be the
9275 delimiter to use (usually @samp{\n}). It is not necessary to define
9276 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9280 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9281 Define this macro to allow references to unknown structure,
9282 union, or enumeration tags to be emitted. Standard COFF does not
9283 allow handling of unknown references, MIPS ECOFF has support for
9287 @defmac SDB_ALLOW_FORWARD_REFERENCES
9288 Define this macro to allow references to structure, union, or
9289 enumeration tags that have not yet been seen to be handled. Some
9290 assemblers choke if forward tags are used, while some require it.
9293 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9294 A C statement to output SDB debugging information before code for line
9295 number @var{line} of the current source file to the stdio stream
9296 @var{stream}. The default is to emit an @code{.ln} directive.
9301 @subsection Macros for VMS Debug Format
9303 @c prevent bad page break with this line
9304 Here are macros for VMS debug format.
9306 @defmac VMS_DEBUGGING_INFO
9307 Define this macro if GCC should produce debugging output for VMS
9308 in response to the @option{-g} option. The default behavior for VMS
9309 is to generate minimal debug info for a traceback in the absence of
9310 @option{-g} unless explicitly overridden with @option{-g0}. This
9311 behavior is controlled by @code{OPTIMIZATION_OPTIONS} and
9312 @code{OVERRIDE_OPTIONS}.
9315 @node Floating Point
9316 @section Cross Compilation and Floating Point
9317 @cindex cross compilation and floating point
9318 @cindex floating point and cross compilation
9320 While all modern machines use twos-complement representation for integers,
9321 there are a variety of representations for floating point numbers. This
9322 means that in a cross-compiler the representation of floating point numbers
9323 in the compiled program may be different from that used in the machine
9324 doing the compilation.
9326 Because different representation systems may offer different amounts of
9327 range and precision, all floating point constants must be represented in
9328 the target machine's format. Therefore, the cross compiler cannot
9329 safely use the host machine's floating point arithmetic; it must emulate
9330 the target's arithmetic. To ensure consistency, GCC always uses
9331 emulation to work with floating point values, even when the host and
9332 target floating point formats are identical.
9334 The following macros are provided by @file{real.h} for the compiler to
9335 use. All parts of the compiler which generate or optimize
9336 floating-point calculations must use these macros. They may evaluate
9337 their operands more than once, so operands must not have side effects.
9339 @defmac REAL_VALUE_TYPE
9340 The C data type to be used to hold a floating point value in the target
9341 machine's format. Typically this is a @code{struct} containing an
9342 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9346 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9347 Compares for equality the two values, @var{x} and @var{y}. If the target
9348 floating point format supports negative zeroes and/or NaNs,
9349 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9350 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9353 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9354 Tests whether @var{x} is less than @var{y}.
9357 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9358 Truncates @var{x} to a signed integer, rounding toward zero.
9361 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9362 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9363 @var{x} is negative, returns zero.
9366 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9367 Converts @var{string} into a floating point number in the target machine's
9368 representation for mode @var{mode}. This routine can handle both
9369 decimal and hexadecimal floating point constants, using the syntax
9370 defined by the C language for both.
9373 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9374 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9377 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9378 Determines whether @var{x} represents infinity (positive or negative).
9381 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9382 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9385 @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})
9386 Calculates an arithmetic operation on the two floating point values
9387 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9390 The operation to be performed is specified by @var{code}. Only the
9391 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9392 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9394 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9395 target's floating point format cannot represent infinity, it will call
9396 @code{abort}. Callers should check for this situation first, using
9397 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9400 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9401 Returns the negative of the floating point value @var{x}.
9404 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9405 Returns the absolute value of @var{x}.
9408 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
9409 Truncates the floating point value @var{x} to fit in @var{mode}. The
9410 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9411 appropriate bit pattern to be output as a floating constant whose
9412 precision accords with mode @var{mode}.
9415 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9416 Converts a floating point value @var{x} into a double-precision integer
9417 which is then stored into @var{low} and @var{high}. If the value is not
9418 integral, it is truncated.
9421 @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})
9422 Converts a double-precision integer found in @var{low} and @var{high},
9423 into a floating point value which is then stored into @var{x}. The
9424 value is truncated to fit in mode @var{mode}.
9427 @node Mode Switching
9428 @section Mode Switching Instructions
9429 @cindex mode switching
9430 The following macros control mode switching optimizations:
9432 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9433 Define this macro if the port needs extra instructions inserted for mode
9434 switching in an optimizing compilation.
9436 For an example, the SH4 can perform both single and double precision
9437 floating point operations, but to perform a single precision operation,
9438 the FPSCR PR bit has to be cleared, while for a double precision
9439 operation, this bit has to be set. Changing the PR bit requires a general
9440 purpose register as a scratch register, hence these FPSCR sets have to
9441 be inserted before reload, i.e.@: you can't put this into instruction emitting
9442 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9444 You can have multiple entities that are mode-switched, and select at run time
9445 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9446 return nonzero for any @var{entity} that needs mode-switching.
9447 If you define this macro, you also have to define
9448 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9449 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9450 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9454 @defmac NUM_MODES_FOR_MODE_SWITCHING
9455 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9456 initializer for an array of integers. Each initializer element
9457 N refers to an entity that needs mode switching, and specifies the number
9458 of different modes that might need to be set for this entity.
9459 The position of the initializer in the initializer---starting counting at
9460 zero---determines the integer that is used to refer to the mode-switched
9462 In macros that take mode arguments / yield a mode result, modes are
9463 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9464 switch is needed / supplied.
9467 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9468 @var{entity} is an integer specifying a mode-switched entity. If
9469 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9470 return an integer value not larger than the corresponding element in
9471 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9472 be switched into prior to the execution of @var{insn}.
9475 @defmac MODE_AFTER (@var{mode}, @var{insn})
9476 If this macro is defined, it is evaluated for every @var{insn} during
9477 mode switching. It determines the mode that an insn results in (if
9478 different from the incoming mode).
9481 @defmac MODE_ENTRY (@var{entity})
9482 If this macro is defined, it is evaluated for every @var{entity} that needs
9483 mode switching. It should evaluate to an integer, which is a mode that
9484 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9485 is defined then @code{MODE_EXIT} must be defined.
9488 @defmac MODE_EXIT (@var{entity})
9489 If this macro is defined, it is evaluated for every @var{entity} that needs
9490 mode switching. It should evaluate to an integer, which is a mode that
9491 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9492 is defined then @code{MODE_ENTRY} must be defined.
9495 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9496 This macro specifies the order in which modes for @var{entity} are processed.
9497 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9498 lowest. The value of the macro should be an integer designating a mode
9499 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9500 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9501 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9504 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9505 Generate one or more insns to set @var{entity} to @var{mode}.
9506 @var{hard_reg_live} is the set of hard registers live at the point where
9507 the insn(s) are to be inserted.
9510 @node Target Attributes
9511 @section Defining target-specific uses of @code{__attribute__}
9512 @cindex target attributes
9513 @cindex machine attributes
9514 @cindex attributes, target-specific
9516 Target-specific attributes may be defined for functions, data and types.
9517 These are described using the following target hooks; they also need to
9518 be documented in @file{extend.texi}.
9520 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
9521 If defined, this target hook points to an array of @samp{struct
9522 attribute_spec} (defined in @file{tree.h}) specifying the machine
9523 specific attributes for this target and some of the restrictions on the
9524 entities to which these attributes are applied and the arguments they
9528 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2})
9529 If defined, this target hook is a function which returns zero if the attributes on
9530 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9531 and two if they are nearly compatible (which causes a warning to be
9532 generated). If this is not defined, machine-specific attributes are
9533 supposed always to be compatible.
9536 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
9537 If defined, this target hook is a function which assigns default attributes to
9538 the newly defined @var{type}.
9541 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9542 Define this target hook if the merging of type attributes needs special
9543 handling. If defined, the result is a list of the combined
9544 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9545 that @code{comptypes} has already been called and returned 1. This
9546 function may call @code{merge_attributes} to handle machine-independent
9550 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
9551 Define this target hook if the merging of decl attributes needs special
9552 handling. If defined, the result is a list of the combined
9553 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9554 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9555 when this is needed are when one attribute overrides another, or when an
9556 attribute is nullified by a subsequent definition. This function may
9557 call @code{merge_attributes} to handle machine-independent merging.
9559 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9560 If the only target-specific handling you require is @samp{dllimport}
9561 for Microsoft Windows targets, you should define the macro
9562 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9563 will then define a function called
9564 @code{merge_dllimport_decl_attributes} which can then be defined as
9565 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9566 add @code{handle_dll_attribute} in the attribute table for your port
9567 to perform initial processing of the @samp{dllimport} and
9568 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9569 @file{i386/i386.c}, for example.
9572 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl})
9573 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
9574 specified. Use this hook if the target needs to add extra validation
9575 checks to @code{handle_dll_attribute}.
9578 @defmac TARGET_DECLSPEC
9579 Define this macro to a nonzero value if you want to treat
9580 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9581 default, this behavior is enabled only for targets that define
9582 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9583 of @code{__declspec} is via a built-in macro, but you should not rely
9584 on this implementation detail.
9587 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
9588 Define this target hook if you want to be able to add attributes to a decl
9589 when it is being created. This is normally useful for back ends which
9590 wish to implement a pragma by using the attributes which correspond to
9591 the pragma's effect. The @var{node} argument is the decl which is being
9592 created. The @var{attr_ptr} argument is a pointer to the attribute list
9593 for this decl. The list itself should not be modified, since it may be
9594 shared with other decls, but attributes may be chained on the head of
9595 the list and @code{*@var{attr_ptr}} modified to point to the new
9596 attributes, or a copy of the list may be made if further changes are
9600 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl})
9602 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9603 into the current function, despite its having target-specific
9604 attributes, @code{false} otherwise. By default, if a function has a
9605 target specific attribute attached to it, it will not be inlined.
9608 @deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags})
9609 This hook is called to parse the @code{attribute(option("..."))}, and
9610 it allows the function to set different target machine compile time
9611 options for the current function that might be different than the
9612 options specified on the command line. The hook should return
9613 @code{true} if the options are valid.
9615 The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9616 the function declaration to hold a pointer to a target specific
9617 @var{struct cl_target_option} structure.
9620 @deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr})
9621 This hook is called to save any additional target specific information
9622 in the @var{struct cl_target_option} structure for function specific
9624 @xref{Option file format}.
9627 @deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct cl_target_option *@var{ptr})
9628 This hook is called to restore any additional target specific
9629 information in the @var{struct cl_target_option} structure for
9630 function specific options.
9633 @deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr})
9634 This hook is called to print any additional target specific
9635 information in the @var{struct cl_target_option} structure for
9636 function specific options.
9639 @deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (target @var{args})
9640 This target hook parses the options for @code{#pragma GCC option} to
9641 set the machine specific options for functions that occur later in the
9642 input stream. The options should be the same as handled by the
9643 @code{TARGET_VALID_OPTION_ATTRIBUTE_P} hook.
9646 @deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee})
9647 This target hook returns @code{false} if the @var{caller} function
9648 cannot inline @var{callee}, based on target specific information. By
9649 default, inlining is not allowed if the callee function has function
9650 specific target options and the caller does not use the same options.
9654 @section Emulating TLS
9655 @cindex Emulated TLS
9657 For targets whose psABI does not provide Thread Local Storage via
9658 specific relocations and instruction sequences, an emulation layer is
9659 used. A set of target hooks allows this emulation layer to be
9660 configured for the requirements of a particular target. For instance
9661 the psABI may in fact specify TLS support in terms of an emulation
9664 The emulation layer works by creating a control object for every TLS
9665 object. To access the TLS object, a lookup function is provided
9666 which, when given the address of the control object, will return the
9667 address of the current thread's instance of the TLS object.
9669 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
9670 Contains the name of the helper function that uses a TLS control
9671 object to locate a TLS instance. The default causes libgcc's
9672 emulated TLS helper function to be used.
9675 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
9676 Contains the name of the helper function that should be used at
9677 program startup to register TLS objects that are implicitly
9678 initialized to zero. If this is @code{NULL}, all TLS objects will
9679 have explicit initializers. The default causes libgcc's emulated TLS
9680 registration function to be used.
9683 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
9684 Contains the name of the section in which TLS control variables should
9685 be placed. The default of @code{NULL} allows these to be placed in
9689 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
9690 Contains the name of the section in which TLS initializers should be
9691 placed. The default of @code{NULL} allows these to be placed in any
9695 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
9696 Contains the prefix to be prepended to TLS control variable names.
9697 The default of @code{NULL} uses a target-specific prefix.
9700 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
9701 Contains the prefix to be prepended to TLS initializer objects. The
9702 default of @code{NULL} uses a target-specific prefix.
9705 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
9706 Specifies a function that generates the FIELD_DECLs for a TLS control
9707 object type. @var{type} is the RECORD_TYPE the fields are for and
9708 @var{name} should be filled with the structure tag, if the default of
9709 @code{__emutls_object} is unsuitable. The default creates a type suitable
9710 for libgcc's emulated TLS function.
9713 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
9714 Specifies a function that generates the CONSTRUCTOR to initialize a
9715 TLS control object. @var{var} is the TLS control object, @var{decl}
9716 is the TLS object and @var{tmpl_addr} is the address of the
9717 initializer. The default initializes libgcc's emulated TLS control object.
9720 @deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED
9721 Specifies whether the alignment of TLS control variable objects is
9722 fixed and should not be increased as some backends may do to optimize
9723 single objects. The default is false.
9726 @deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9727 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9728 may be used to describe emulated TLS control objects.
9731 @node MIPS Coprocessors
9732 @section Defining coprocessor specifics for MIPS targets.
9733 @cindex MIPS coprocessor-definition macros
9735 The MIPS specification allows MIPS implementations to have as many as 4
9736 coprocessors, each with as many as 32 private registers. GCC supports
9737 accessing these registers and transferring values between the registers
9738 and memory using asm-ized variables. For example:
9741 register unsigned int cp0count asm ("c0r1");
9747 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
9748 names may be added as described below, or the default names may be
9749 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
9751 Coprocessor registers are assumed to be epilogue-used; sets to them will
9752 be preserved even if it does not appear that the register is used again
9753 later in the function.
9755 Another note: according to the MIPS spec, coprocessor 1 (if present) is
9756 the FPU@. One accesses COP1 registers through standard mips
9757 floating-point support; they are not included in this mechanism.
9759 There is one macro used in defining the MIPS coprocessor interface which
9760 you may want to override in subtargets; it is described below.
9762 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
9763 A comma-separated list (with leading comma) of pairs describing the
9764 alternate names of coprocessor registers. The format of each entry should be
9766 @{ @var{alternatename}, @var{register_number}@}
9772 @section Parameters for Precompiled Header Validity Checking
9773 @cindex parameters, precompiled headers
9775 @deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
9776 This hook returns a pointer to the data needed by
9777 @code{TARGET_PCH_VALID_P} and sets
9778 @samp{*@var{sz}} to the size of the data in bytes.
9781 @deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
9782 This hook checks whether the options used to create a PCH file are
9783 compatible with the current settings. It returns @code{NULL}
9784 if so and a suitable error message if not. Error messages will
9785 be presented to the user and must be localized using @samp{_(@var{msg})}.
9787 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
9788 when the PCH file was created and @var{sz} is the size of that data in bytes.
9789 It's safe to assume that the data was created by the same version of the
9790 compiler, so no format checking is needed.
9792 The default definition of @code{default_pch_valid_p} should be
9793 suitable for most targets.
9796 @deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
9797 If this hook is nonnull, the default implementation of
9798 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
9799 of @code{target_flags}. @var{pch_flags} specifies the value that
9800 @code{target_flags} had when the PCH file was created. The return
9801 value is the same as for @code{TARGET_PCH_VALID_P}.
9805 @section C++ ABI parameters
9806 @cindex parameters, c++ abi
9808 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
9809 Define this hook to override the integer type used for guard variables.
9810 These are used to implement one-time construction of static objects. The
9811 default is long_long_integer_type_node.
9814 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
9815 This hook determines how guard variables are used. It should return
9816 @code{false} (the default) if the first byte should be used. A return value of
9817 @code{true} indicates that only the least significant bit should be used.
9820 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
9821 This hook returns the size of the cookie to use when allocating an array
9822 whose elements have the indicated @var{type}. Assumes that it is already
9823 known that a cookie is needed. The default is
9824 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
9825 IA64/Generic C++ ABI@.
9828 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
9829 This hook should return @code{true} if the element size should be stored in
9830 array cookies. The default is to return @code{false}.
9833 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
9834 If defined by a backend this hook allows the decision made to export
9835 class @var{type} to be overruled. Upon entry @var{import_export}
9836 will contain 1 if the class is going to be exported, @minus{}1 if it is going
9837 to be imported and 0 otherwise. This function should return the
9838 modified value and perform any other actions necessary to support the
9839 backend's targeted operating system.
9842 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
9843 This hook should return @code{true} if constructors and destructors return
9844 the address of the object created/destroyed. The default is to return
9848 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
9849 This hook returns true if the key method for a class (i.e., the method
9850 which, if defined in the current translation unit, causes the virtual
9851 table to be emitted) may be an inline function. Under the standard
9852 Itanium C++ ABI the key method may be an inline function so long as
9853 the function is not declared inline in the class definition. Under
9854 some variants of the ABI, an inline function can never be the key
9855 method. The default is to return @code{true}.
9858 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
9859 @var{decl} is a virtual table, virtual table table, typeinfo object,
9860 or other similar implicit class data object that will be emitted with
9861 external linkage in this translation unit. No ELF visibility has been
9862 explicitly specified. If the target needs to specify a visibility
9863 other than that of the containing class, use this hook to set
9864 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
9867 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
9868 This hook returns true (the default) if virtual tables and other
9869 similar implicit class data objects are always COMDAT if they have
9870 external linkage. If this hook returns false, then class data for
9871 classes whose virtual table will be emitted in only one translation
9872 unit will not be COMDAT.
9875 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
9876 This hook returns true (the default) if the RTTI information for
9877 the basic types which is defined in the C++ runtime should always
9878 be COMDAT, false if it should not be COMDAT.
9881 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
9882 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
9883 should be used to register static destructors when @option{-fuse-cxa-atexit}
9884 is in effect. The default is to return false to use @code{__cxa_atexit}.
9887 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
9888 This hook returns true if the target @code{atexit} function can be used
9889 in the same manner as @code{__cxa_atexit} to register C++ static
9890 destructors. This requires that @code{atexit}-registered functions in
9891 shared libraries are run in the correct order when the libraries are
9892 unloaded. The default is to return false.
9895 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
9896 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been
9897 defined. Use this hook to make adjustments to the class (eg, tweak
9898 visibility or perform any other required target modifications).
9901 @node Named Address Spaces
9902 @section Adding support for named address spaces
9903 @cindex named address spaces
9905 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
9906 standards committee, @cite{Programming Languages - C - Extensions to
9907 support embedded processors}, specifies a syntax for embedded
9908 processors to specify alternate address spaces. You can configure a
9909 GCC port to support section 5.1 of the draft report to add support for
9910 address spaces other than the default address space. These address
9911 spaces are new keywords that are similar to the @code{volatile} and
9912 @code{const} type attributes.
9914 Pointers to named address spaces can have a different size than
9915 pointers to the generic address space.
9917 For example, the SPU port uses the @code{__ea} address space to refer
9918 to memory in the host processor, rather than memory local to the SPU
9919 processor. Access to memory in the @code{__ea} address space involves
9920 issuing DMA operations to move data between the host processor and the
9921 local processor memory address space. Pointers in the @code{__ea}
9922 address space are either 32 bits or 64 bits based on the
9923 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
9926 Internally, address spaces are represented as a small integer in the
9927 range 0 to 15 with address space 0 being reserved for the generic
9930 @defmac TARGET_ADDR_SPACE_KEYWORDS
9931 A list of @code{ADDR_SPACE_KEYWORD} macros to define each named
9932 address keyword. The @code{ADDR_SPACE_KEYWORD} macro takes two
9933 arguments, the keyword string and the number of the named address
9934 space. For example, the SPU port uses the following to declare
9935 @code{__ea} as the keyword for named address space #1:
9937 #define ADDR_SPACE_EA 1
9938 #define TARGET_ADDR_SPACE_KEYWORDS ADDR_SPACE_KEYWORD ("__ea", ADDR_SPACE_EA)
9942 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space})
9943 Define this to return the machine mode to use for pointers to
9944 @var{address_space} if the target supports named address spaces.
9945 The default version of this hook returns @code{ptr_mode} for the
9946 generic address space only.
9949 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space})
9950 Define this to return the machine mode to use for addresses in
9951 @var{address_space} if the target supports named address spaces.
9952 The default version of this hook returns @code{Pmode} for the
9953 generic address space only.
9956 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (enum machine_mode @var{mode}, addr_space_t @var{as})
9957 Define this to return nonzero if the port can handle pointers
9958 with machine mode @var{mode} to address space @var{as}. This target
9959 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
9960 except that it includes explicit named address space support. The default
9961 version of this hook returns true for the modes returned by either the
9962 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
9963 target hooks for the given address space.
9966 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P (enum machine_mode @var{mode}, rtx @var{exp}, bool @var{strict}, addr_space_t @var{as})
9967 Define this to return true if @var{exp} is a valid address for mode
9968 @var{mode} in the named address space @var{as}. The @var{strict}
9969 parameter says whether strict addressing is in effect after reload has
9970 finished. This target hook is the same as the
9971 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
9972 explicit named address space support.
9975 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, enum machine_mode @var{mode}, addr_space_t @var{as})
9976 Define this to modify an invalid address @var{x} to be a valid address
9977 with mode @var{mode} in the named address space @var{as}. This target
9978 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
9979 except that it includes explicit named address space support.
9982 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{superset}, addr_space_t @var{subset})
9983 Define this to return whether the @var{subset} named address space is
9984 contained within the @var{superset} named address space. Pointers to
9985 a named address space that is a subset of another named address space
9986 will be converted automatically without a cast if used together in
9987 arithmetic operations. Pointers to a superset address space can be
9988 converted to pointers to a subset address space via explicit casts.
9991 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type})
9992 Define this to convert the pointer expression represented by the RTL
9993 @var{op} with type @var{from_type} that points to a named address
9994 space to a new pointer expression with type @var{to_type} that points
9995 to a different named address space. When this hook it called, it is
9996 guaranteed that one of the two address spaces is a subset of the other,
9997 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10001 @section Miscellaneous Parameters
10002 @cindex parameters, miscellaneous
10004 @c prevent bad page break with this line
10005 Here are several miscellaneous parameters.
10007 @defmac HAS_LONG_COND_BRANCH
10008 Define this boolean macro to indicate whether or not your architecture
10009 has conditional branches that can span all of memory. It is used in
10010 conjunction with an optimization that partitions hot and cold basic
10011 blocks into separate sections of the executable. If this macro is
10012 set to false, gcc will convert any conditional branches that attempt
10013 to cross between sections into unconditional branches or indirect jumps.
10016 @defmac HAS_LONG_UNCOND_BRANCH
10017 Define this boolean macro to indicate whether or not your architecture
10018 has unconditional branches that can span all of memory. It is used in
10019 conjunction with an optimization that partitions hot and cold basic
10020 blocks into separate sections of the executable. If this macro is
10021 set to false, gcc will convert any unconditional branches that attempt
10022 to cross between sections into indirect jumps.
10025 @defmac CASE_VECTOR_MODE
10026 An alias for a machine mode name. This is the machine mode that
10027 elements of a jump-table should have.
10030 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10031 Optional: return the preferred mode for an @code{addr_diff_vec}
10032 when the minimum and maximum offset are known. If you define this,
10033 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10034 To make this work, you also have to define @code{INSN_ALIGN} and
10035 make the alignment for @code{addr_diff_vec} explicit.
10036 The @var{body} argument is provided so that the offset_unsigned and scale
10037 flags can be updated.
10040 @defmac CASE_VECTOR_PC_RELATIVE
10041 Define this macro to be a C expression to indicate when jump-tables
10042 should contain relative addresses. You need not define this macro if
10043 jump-tables never contain relative addresses, or jump-tables should
10044 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10048 @deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void)
10049 This function return the smallest number of different values for which it
10050 is best to use a jump-table instead of a tree of conditional branches.
10051 The default is four for machines with a @code{casesi} instruction and
10052 five otherwise. This is best for most machines.
10055 @defmac CASE_USE_BIT_TESTS
10056 Define this macro to be a C expression to indicate whether C switch
10057 statements may be implemented by a sequence of bit tests. This is
10058 advantageous on processors that can efficiently implement left shift
10059 of 1 by the number of bits held in a register, but inappropriate on
10060 targets that would require a loop. By default, this macro returns
10061 @code{true} if the target defines an @code{ashlsi3} pattern, and
10062 @code{false} otherwise.
10065 @defmac WORD_REGISTER_OPERATIONS
10066 Define this macro if operations between registers with integral mode
10067 smaller than a word are always performed on the entire register.
10068 Most RISC machines have this property and most CISC machines do not.
10071 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10072 Define this macro to be a C expression indicating when insns that read
10073 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10074 bits outside of @var{mem_mode} to be either the sign-extension or the
10075 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10076 of @var{mem_mode} for which the
10077 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10078 @code{UNKNOWN} for other modes.
10080 This macro is not called with @var{mem_mode} non-integral or with a width
10081 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10082 value in this case. Do not define this macro if it would always return
10083 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10084 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10086 You may return a non-@code{UNKNOWN} value even if for some hard registers
10087 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10088 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10089 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10090 integral mode larger than this but not larger than @code{word_mode}.
10092 You must return @code{UNKNOWN} if for some hard registers that allow this
10093 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10094 @code{word_mode}, but that they can change to another integral mode that
10095 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10098 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10099 Define this macro if loading short immediate values into registers sign
10103 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
10104 Define this macro if the same instructions that convert a floating
10105 point number to a signed fixed point number also convert validly to an
10109 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
10110 When @option{-ffast-math} is in effect, GCC tries to optimize
10111 divisions by the same divisor, by turning them into multiplications by
10112 the reciprocal. This target hook specifies the minimum number of divisions
10113 that should be there for GCC to perform the optimization for a variable
10114 of mode @var{mode}. The default implementation returns 3 if the machine
10115 has an instruction for the division, and 2 if it does not.
10119 The maximum number of bytes that a single instruction can move quickly
10120 between memory and registers or between two memory locations.
10123 @defmac MAX_MOVE_MAX
10124 The maximum number of bytes that a single instruction can move quickly
10125 between memory and registers or between two memory locations. If this
10126 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10127 constant value that is the largest value that @code{MOVE_MAX} can have
10131 @defmac SHIFT_COUNT_TRUNCATED
10132 A C expression that is nonzero if on this machine the number of bits
10133 actually used for the count of a shift operation is equal to the number
10134 of bits needed to represent the size of the object being shifted. When
10135 this macro is nonzero, the compiler will assume that it is safe to omit
10136 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10137 truncates the count of a shift operation. On machines that have
10138 instructions that act on bit-fields at variable positions, which may
10139 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10140 also enables deletion of truncations of the values that serve as
10141 arguments to bit-field instructions.
10143 If both types of instructions truncate the count (for shifts) and
10144 position (for bit-field operations), or if no variable-position bit-field
10145 instructions exist, you should define this macro.
10147 However, on some machines, such as the 80386 and the 680x0, truncation
10148 only applies to shift operations and not the (real or pretended)
10149 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10150 such machines. Instead, add patterns to the @file{md} file that include
10151 the implied truncation of the shift instructions.
10153 You need not define this macro if it would always have the value of zero.
10156 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10157 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
10158 This function describes how the standard shift patterns for @var{mode}
10159 deal with shifts by negative amounts or by more than the width of the mode.
10160 @xref{shift patterns}.
10162 On many machines, the shift patterns will apply a mask @var{m} to the
10163 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10164 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10165 this is true for mode @var{mode}, the function should return @var{m},
10166 otherwise it should return 0. A return value of 0 indicates that no
10167 particular behavior is guaranteed.
10169 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10170 @emph{not} apply to general shift rtxes; it applies only to instructions
10171 that are generated by the named shift patterns.
10173 The default implementation of this function returns
10174 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10175 and 0 otherwise. This definition is always safe, but if
10176 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10177 nevertheless truncate the shift count, you may get better code
10181 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10182 A C expression which is nonzero if on this machine it is safe to
10183 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10184 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10185 operating on it as if it had only @var{outprec} bits.
10187 On many machines, this expression can be 1.
10189 @c rearranged this, removed the phrase "it is reported that". this was
10190 @c to fix an overfull hbox. --mew 10feb93
10191 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10192 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10193 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10194 such cases may improve things.
10197 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode})
10198 The representation of an integral mode can be such that the values
10199 are always extended to a wider integral mode. Return
10200 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10201 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10202 otherwise. (Currently, none of the targets use zero-extended
10203 representation this way so unlike @code{LOAD_EXTEND_OP},
10204 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10205 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10206 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10207 widest integral mode and currently we take advantage of this fact.)
10209 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10210 value even if the extension is not performed on certain hard registers
10211 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10212 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10214 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10215 describe two related properties. If you define
10216 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10217 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10220 In order to enforce the representation of @code{mode},
10221 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10225 @defmac STORE_FLAG_VALUE
10226 A C expression describing the value returned by a comparison operator
10227 with an integral mode and stored by a store-flag instruction
10228 (@samp{s@var{cond}}) when the condition is true. This description must
10229 apply to @emph{all} the @samp{s@var{cond}} patterns and all the
10230 comparison operators whose results have a @code{MODE_INT} mode.
10232 A value of 1 or @minus{}1 means that the instruction implementing the
10233 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10234 and 0 when the comparison is false. Otherwise, the value indicates
10235 which bits of the result are guaranteed to be 1 when the comparison is
10236 true. This value is interpreted in the mode of the comparison
10237 operation, which is given by the mode of the first operand in the
10238 @samp{s@var{cond}} pattern. Either the low bit or the sign bit of
10239 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10242 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10243 generate code that depends only on the specified bits. It can also
10244 replace comparison operators with equivalent operations if they cause
10245 the required bits to be set, even if the remaining bits are undefined.
10246 For example, on a machine whose comparison operators return an
10247 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10248 @samp{0x80000000}, saying that just the sign bit is relevant, the
10252 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10256 can be converted to
10259 (ashift:SI @var{x} (const_int @var{n}))
10263 where @var{n} is the appropriate shift count to move the bit being
10264 tested into the sign bit.
10266 There is no way to describe a machine that always sets the low-order bit
10267 for a true value, but does not guarantee the value of any other bits,
10268 but we do not know of any machine that has such an instruction. If you
10269 are trying to port GCC to such a machine, include an instruction to
10270 perform a logical-and of the result with 1 in the pattern for the
10271 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10273 Often, a machine will have multiple instructions that obtain a value
10274 from a comparison (or the condition codes). Here are rules to guide the
10275 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10280 Use the shortest sequence that yields a valid definition for
10281 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10282 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10283 comparison operators to do so because there may be opportunities to
10284 combine the normalization with other operations.
10287 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10288 slightly preferred on machines with expensive jumps and 1 preferred on
10292 As a second choice, choose a value of @samp{0x80000001} if instructions
10293 exist that set both the sign and low-order bits but do not define the
10297 Otherwise, use a value of @samp{0x80000000}.
10300 Many machines can produce both the value chosen for
10301 @code{STORE_FLAG_VALUE} and its negation in the same number of
10302 instructions. On those machines, you should also define a pattern for
10303 those cases, e.g., one matching
10306 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10309 Some machines can also perform @code{and} or @code{plus} operations on
10310 condition code values with less instructions than the corresponding
10311 @samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those
10312 machines, define the appropriate patterns. Use the names @code{incscc}
10313 and @code{decscc}, respectively, for the patterns which perform
10314 @code{plus} or @code{minus} operations on condition code values. See
10315 @file{rs6000.md} for some examples. The GNU Superoptizer can be used to
10316 find such instruction sequences on other machines.
10318 If this macro is not defined, the default value, 1, is used. You need
10319 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10320 instructions, or if the value generated by these instructions is 1.
10323 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10324 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10325 returned when comparison operators with floating-point results are true.
10326 Define this macro on machines that have comparison operations that return
10327 floating-point values. If there are no such operations, do not define
10331 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10332 A C expression that gives a rtx representing the nonzero true element
10333 for vector comparisons. The returned rtx should be valid for the inner
10334 mode of @var{mode} which is guaranteed to be a vector mode. Define
10335 this macro on machines that have vector comparison operations that
10336 return a vector result. If there are no such operations, do not define
10337 this macro. Typically, this macro is defined as @code{const1_rtx} or
10338 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10339 the compiler optimizing such vector comparison operations for the
10343 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10344 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10345 A C expression that indicates whether the architecture defines a value
10346 for @code{clz} or @code{ctz} with a zero operand.
10347 A result of @code{0} indicates the value is undefined.
10348 If the value is defined for only the RTL expression, the macro should
10349 evaluate to @code{1}; if the value applies also to the corresponding optab
10350 entry (which is normally the case if it expands directly into
10351 the corresponding RTL), then the macro should evaluate to @code{2}.
10352 In the cases where the value is defined, @var{value} should be set to
10355 If this macro is not defined, the value of @code{clz} or
10356 @code{ctz} at zero is assumed to be undefined.
10358 This macro must be defined if the target's expansion for @code{ffs}
10359 relies on a particular value to get correct results. Otherwise it
10360 is not necessary, though it may be used to optimize some corner cases, and
10361 to provide a default expansion for the @code{ffs} optab.
10363 Note that regardless of this macro the ``definedness'' of @code{clz}
10364 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10365 visible to the user. Thus one may be free to adjust the value at will
10366 to match the target expansion of these operations without fear of
10371 An alias for the machine mode for pointers. On most machines, define
10372 this to be the integer mode corresponding to the width of a hardware
10373 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10374 On some machines you must define this to be one of the partial integer
10375 modes, such as @code{PSImode}.
10377 The width of @code{Pmode} must be at least as large as the value of
10378 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10379 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10383 @defmac FUNCTION_MODE
10384 An alias for the machine mode used for memory references to functions
10385 being called, in @code{call} RTL expressions. On most CISC machines,
10386 where an instruction can begin at any byte address, this should be
10387 @code{QImode}. On most RISC machines, where all instructions have fixed
10388 size and alignment, this should be a mode with the same size and alignment
10389 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10392 @defmac STDC_0_IN_SYSTEM_HEADERS
10393 In normal operation, the preprocessor expands @code{__STDC__} to the
10394 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10395 hosts, like Solaris, the system compiler uses a different convention,
10396 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10397 strict conformance to the C Standard.
10399 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10400 convention when processing system header files, but when processing user
10401 files @code{__STDC__} will always expand to 1.
10404 @defmac NO_IMPLICIT_EXTERN_C
10405 Define this macro if the system header files support C++ as well as C@.
10406 This macro inhibits the usual method of using system header files in
10407 C++, which is to pretend that the file's contents are enclosed in
10408 @samp{extern "C" @{@dots{}@}}.
10413 @defmac REGISTER_TARGET_PRAGMAS ()
10414 Define this macro if you want to implement any target-specific pragmas.
10415 If defined, it is a C expression which makes a series of calls to
10416 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10417 for each pragma. The macro may also do any
10418 setup required for the pragmas.
10420 The primary reason to define this macro is to provide compatibility with
10421 other compilers for the same target. In general, we discourage
10422 definition of target-specific pragmas for GCC@.
10424 If the pragma can be implemented by attributes then you should consider
10425 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10427 Preprocessor macros that appear on pragma lines are not expanded. All
10428 @samp{#pragma} directives that do not match any registered pragma are
10429 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10432 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10433 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10435 Each call to @code{c_register_pragma} or
10436 @code{c_register_pragma_with_expansion} establishes one pragma. The
10437 @var{callback} routine will be called when the preprocessor encounters a
10441 #pragma [@var{space}] @var{name} @dots{}
10444 @var{space} is the case-sensitive namespace of the pragma, or
10445 @code{NULL} to put the pragma in the global namespace. The callback
10446 routine receives @var{pfile} as its first argument, which can be passed
10447 on to cpplib's functions if necessary. You can lex tokens after the
10448 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10449 callback will be silently ignored. The end of the line is indicated by
10450 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10451 arguments of pragmas registered with
10452 @code{c_register_pragma_with_expansion} but not on the arguments of
10453 pragmas registered with @code{c_register_pragma}.
10455 Note that the use of @code{pragma_lex} is specific to the C and C++
10456 compilers. It will not work in the Java or Fortran compilers, or any
10457 other language compilers for that matter. Thus if @code{pragma_lex} is going
10458 to be called from target-specific code, it must only be done so when
10459 building the C and C++ compilers. This can be done by defining the
10460 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10461 target entry in the @file{config.gcc} file. These variables should name
10462 the target-specific, language-specific object file which contains the
10463 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10464 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10465 how to build this object file.
10470 @defmac HANDLE_SYSV_PRAGMA
10471 Define this macro (to a value of 1) if you want the System V style
10472 pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name>
10473 [=<value>]} to be supported by gcc.
10475 The pack pragma specifies the maximum alignment (in bytes) of fields
10476 within a structure, in much the same way as the @samp{__aligned__} and
10477 @samp{__packed__} @code{__attribute__}s do. A pack value of zero resets
10478 the behavior to the default.
10480 A subtlety for Microsoft Visual C/C++ style bit-field packing
10481 (e.g.@: -mms-bitfields) for targets that support it:
10482 When a bit-field is inserted into a packed record, the whole size
10483 of the underlying type is used by one or more same-size adjacent
10484 bit-fields (that is, if its long:3, 32 bits is used in the record,
10485 and any additional adjacent long bit-fields are packed into the same
10486 chunk of 32 bits. However, if the size changes, a new field of that
10487 size is allocated).
10489 If both MS bit-fields and @samp{__attribute__((packed))} are used,
10490 the latter will take precedence. If @samp{__attribute__((packed))} is
10491 used on a single field when MS bit-fields are in use, it will take
10492 precedence for that field, but the alignment of the rest of the structure
10493 may affect its placement.
10495 The weak pragma only works if @code{SUPPORTS_WEAK} and
10496 @code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation
10497 of specifically named weak labels, optionally with a value.
10502 @defmac HANDLE_PRAGMA_PACK_PUSH_POP
10503 Define this macro (to a value of 1) if you want to support the Win32
10504 style pragmas @samp{#pragma pack(push[,@var{n}])} and @samp{#pragma
10505 pack(pop)}. The @samp{pack(push,[@var{n}])} pragma specifies the maximum
10506 alignment (in bytes) of fields within a structure, in much the same way as
10507 the @samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A
10508 pack value of zero resets the behavior to the default. Successive
10509 invocations of this pragma cause the previous values to be stacked, so
10510 that invocations of @samp{#pragma pack(pop)} will return to the previous
10514 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10515 Define this macro, as well as
10516 @code{HANDLE_SYSV_PRAGMA}, if macros should be expanded in the
10517 arguments of @samp{#pragma pack}.
10520 @deftypevr {Target Hook} bool TARGET_HANDLE_PRAGMA_EXTERN_PREFIX
10521 True if @code{#pragma extern_prefix} is to be supported.
10524 @defmac TARGET_DEFAULT_PACK_STRUCT
10525 If your target requires a structure packing default other than 0 (meaning
10526 the machine default), define this macro to the necessary value (in bytes).
10527 This must be a value that would also be valid to use with
10528 @samp{#pragma pack()} (that is, a small power of two).
10531 @defmac DOLLARS_IN_IDENTIFIERS
10532 Define this macro to control use of the character @samp{$} in
10533 identifier names for the C family of languages. 0 means @samp{$} is
10534 not allowed by default; 1 means it is allowed. 1 is the default;
10535 there is no need to define this macro in that case.
10538 @defmac NO_DOLLAR_IN_LABEL
10539 Define this macro if the assembler does not accept the character
10540 @samp{$} in label names. By default constructors and destructors in
10541 G++ have @samp{$} in the identifiers. If this macro is defined,
10542 @samp{.} is used instead.
10545 @defmac NO_DOT_IN_LABEL
10546 Define this macro if the assembler does not accept the character
10547 @samp{.} in label names. By default constructors and destructors in G++
10548 have names that use @samp{.}. If this macro is defined, these names
10549 are rewritten to avoid @samp{.}.
10552 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10553 Define this macro as a C expression that is nonzero if it is safe for the
10554 delay slot scheduler to place instructions in the delay slot of @var{insn},
10555 even if they appear to use a resource set or clobbered in @var{insn}.
10556 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10557 every @code{call_insn} has this behavior. On machines where some @code{insn}
10558 or @code{jump_insn} is really a function call and hence has this behavior,
10559 you should define this macro.
10561 You need not define this macro if it would always return zero.
10564 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10565 Define this macro as a C expression that is nonzero if it is safe for the
10566 delay slot scheduler to place instructions in the delay slot of @var{insn},
10567 even if they appear to set or clobber a resource referenced in @var{insn}.
10568 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10569 some @code{insn} or @code{jump_insn} is really a function call and its operands
10570 are registers whose use is actually in the subroutine it calls, you should
10571 define this macro. Doing so allows the delay slot scheduler to move
10572 instructions which copy arguments into the argument registers into the delay
10573 slot of @var{insn}.
10575 You need not define this macro if it would always return zero.
10578 @defmac MULTIPLE_SYMBOL_SPACES
10579 Define this macro as a C expression that is nonzero if, in some cases,
10580 global symbols from one translation unit may not be bound to undefined
10581 symbols in another translation unit without user intervention. For
10582 instance, under Microsoft Windows symbols must be explicitly imported
10583 from shared libraries (DLLs).
10585 You need not define this macro if it would always evaluate to zero.
10588 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
10589 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10590 any hard regs the port wishes to automatically clobber for an asm.
10591 It should return the result of the last @code{tree_cons} used to add a
10592 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10593 corresponding parameters to the asm and may be inspected to avoid
10594 clobbering a register that is an input or output of the asm. You can use
10595 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10596 for overlap with regards to asm-declared registers.
10599 @defmac MATH_LIBRARY
10600 Define this macro as a C string constant for the linker argument to link
10601 in the system math library, or @samp{""} if the target does not have a
10602 separate math library.
10604 You need only define this macro if the default of @samp{"-lm"} is wrong.
10607 @defmac LIBRARY_PATH_ENV
10608 Define this macro as a C string constant for the environment variable that
10609 specifies where the linker should look for libraries.
10611 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10615 @defmac TARGET_POSIX_IO
10616 Define this macro if the target supports the following POSIX@ file
10617 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10618 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10619 to use file locking when exiting a program, which avoids race conditions
10620 if the program has forked. It will also create directories at run-time
10621 for cross-profiling.
10624 @defmac MAX_CONDITIONAL_EXECUTE
10626 A C expression for the maximum number of instructions to execute via
10627 conditional execution instructions instead of a branch. A value of
10628 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10629 1 if it does use cc0.
10632 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10633 Used if the target needs to perform machine-dependent modifications on the
10634 conditionals used for turning basic blocks into conditionally executed code.
10635 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10636 contains information about the currently processed blocks. @var{true_expr}
10637 and @var{false_expr} are the tests that are used for converting the
10638 then-block and the else-block, respectively. Set either @var{true_expr} or
10639 @var{false_expr} to a null pointer if the tests cannot be converted.
10642 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10643 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10644 if-statements into conditions combined by @code{and} and @code{or} operations.
10645 @var{bb} contains the basic block that contains the test that is currently
10646 being processed and about to be turned into a condition.
10649 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10650 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10651 be converted to conditional execution format. @var{ce_info} points to
10652 a data structure, @code{struct ce_if_block}, which contains information
10653 about the currently processed blocks.
10656 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10657 A C expression to perform any final machine dependent modifications in
10658 converting code to conditional execution. The involved basic blocks
10659 can be found in the @code{struct ce_if_block} structure that is pointed
10660 to by @var{ce_info}.
10663 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10664 A C expression to cancel any machine dependent modifications in
10665 converting code to conditional execution. The involved basic blocks
10666 can be found in the @code{struct ce_if_block} structure that is pointed
10667 to by @var{ce_info}.
10670 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10671 A C expression to initialize any extra fields in a @code{struct ce_if_block}
10672 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10675 @defmac IFCVT_EXTRA_FIELDS
10676 If defined, it should expand to a set of field declarations that will be
10677 added to the @code{struct ce_if_block} structure. These should be initialized
10678 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10681 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void)
10682 If non-null, this hook performs a target-specific pass over the
10683 instruction stream. The compiler will run it at all optimization levels,
10684 just before the point at which it normally does delayed-branch scheduling.
10686 The exact purpose of the hook varies from target to target. Some use
10687 it to do transformations that are necessary for correctness, such as
10688 laying out in-function constant pools or avoiding hardware hazards.
10689 Others use it as an opportunity to do some machine-dependent optimizations.
10691 You need not implement the hook if it has nothing to do. The default
10692 definition is null.
10695 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void)
10696 Define this hook if you have any machine-specific built-in functions
10697 that need to be defined. It should be a function that performs the
10700 Machine specific built-in functions can be useful to expand special machine
10701 instructions that would otherwise not normally be generated because
10702 they have no equivalent in the source language (for example, SIMD vector
10703 instructions or prefetch instructions).
10705 To create a built-in function, call the function
10706 @code{lang_hooks.builtin_function}
10707 which is defined by the language front end. You can use any type nodes set
10708 up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2};
10709 only language front ends that use those two functions will call
10710 @samp{TARGET_INIT_BUILTINS}.
10713 @deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p})
10714 Define this hook if you have any machine-specific built-in functions
10715 that need to be defined. It should be a function that returns the
10716 builtin function declaration for the builtin function code @var{code}.
10717 If there is no such builtin and it cannot be initialized at this time
10718 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10719 If @var{code} is out of range the function should return
10720 @code{error_mark_node}.
10723 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
10725 Expand a call to a machine specific built-in function that was set up by
10726 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10727 function call; the result should go to @var{target} if that is
10728 convenient, and have mode @var{mode} if that is convenient.
10729 @var{subtarget} may be used as the target for computing one of
10730 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10731 ignored. This function should return the result of the call to the
10735 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist})
10737 Select a replacement for a machine specific built-in function that
10738 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10739 @emph{before} regular type checking, and so allows the target to
10740 implement a crude form of function overloading. @var{fndecl} is the
10741 declaration of the built-in function. @var{arglist} is the list of
10742 arguments passed to the built-in function. The result is a
10743 complete expression that implements the operation, usually
10744 another @code{CALL_EXPR}.
10745 @var{arglist} really has type @samp{VEC(tree,gc)*}
10748 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, tree @var{arglist}, bool @var{ignore})
10750 Fold a call to a machine specific built-in function that was set up by
10751 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10752 built-in function. @var{arglist} is the list of arguments passed to
10753 the built-in function. The result is another tree containing a
10754 simplified expression for the call's result. If @var{ignore} is true
10755 the value will be ignored.
10758 @deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const_rtx @var{insn})
10760 Take an instruction in @var{insn} and return NULL if it is valid within a
10761 low-overhead loop, otherwise return a string explaining why doloop
10762 could not be applied.
10764 Many targets use special registers for low-overhead looping. For any
10765 instruction that clobbers these this function should return a string indicating
10766 the reason why the doloop could not be applied.
10767 By default, the RTL loop optimizer does not use a present doloop pattern for
10768 loops containing function calls or branch on table instructions.
10771 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10773 Take a branch insn in @var{branch1} and another in @var{branch2}.
10774 Return true if redirecting @var{branch1} to the destination of
10775 @var{branch2} is possible.
10777 On some targets, branches may have a limited range. Optimizing the
10778 filling of delay slots can result in branches being redirected, and this
10779 may in turn cause a branch offset to overflow.
10782 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code})
10783 This target hook returns @code{true} if @var{x} is considered to be commutative.
10784 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10785 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10786 of the enclosing rtl, if known, otherwise it is UNKNOWN.
10789 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
10791 When the initial value of a hard register has been copied in a pseudo
10792 register, it is often not necessary to actually allocate another register
10793 to this pseudo register, because the original hard register or a stack slot
10794 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10795 is called at the start of register allocation once for each hard register
10796 that had its initial value copied by using
10797 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10798 Possible values are @code{NULL_RTX}, if you don't want
10799 to do any special allocation, a @code{REG} rtx---that would typically be
10800 the hard register itself, if it is known not to be clobbered---or a
10802 If you are returning a @code{MEM}, this is only a hint for the allocator;
10803 it might decide to use another register anyways.
10804 You may use @code{current_function_leaf_function} in the hook, functions
10805 that use @code{REG_N_SETS}, to determine if the hard
10806 register in question will not be clobbered.
10807 The default value of this hook is @code{NULL}, which disables any special
10811 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
10812 This target hook returns nonzero if @var{x}, an @code{unspec} or
10813 @code{unspec_volatile} operation, might cause a trap. Targets can use
10814 this hook to enhance precision of analysis for @code{unspec} and
10815 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10816 to analyze inner elements of @var{x} in which case @var{flags} should be
10820 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
10821 The compiler invokes this hook whenever it changes its current function
10822 context (@code{cfun}). You can define this function if
10823 the back end needs to perform any initialization or reset actions on a
10824 per-function basis. For example, it may be used to implement function
10825 attributes that affect register usage or code generation patterns.
10826 The argument @var{decl} is the declaration for the new function context,
10827 and may be null to indicate that the compiler has left a function context
10828 and is returning to processing at the top level.
10829 The default hook function does nothing.
10831 GCC sets @code{cfun} to a dummy function context during initialization of
10832 some parts of the back end. The hook function is not invoked in this
10833 situation; you need not worry about the hook being invoked recursively,
10834 or when the back end is in a partially-initialized state.
10835 @code{cfun} might be @code{NULL} to indicate processing at top level,
10836 outside of any function scope.
10839 @defmac TARGET_OBJECT_SUFFIX
10840 Define this macro to be a C string representing the suffix for object
10841 files on your target machine. If you do not define this macro, GCC will
10842 use @samp{.o} as the suffix for object files.
10845 @defmac TARGET_EXECUTABLE_SUFFIX
10846 Define this macro to be a C string representing the suffix to be
10847 automatically added to executable files on your target machine. If you
10848 do not define this macro, GCC will use the null string as the suffix for
10852 @defmac COLLECT_EXPORT_LIST
10853 If defined, @code{collect2} will scan the individual object files
10854 specified on its command line and create an export list for the linker.
10855 Define this macro for systems like AIX, where the linker discards
10856 object files that are not referenced from @code{main} and uses export
10860 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
10861 Define this macro to a C expression representing a variant of the
10862 method call @var{mdecl}, if Java Native Interface (JNI) methods
10863 must be invoked differently from other methods on your target.
10864 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
10865 the @code{stdcall} calling convention and this macro is then
10866 defined as this expression:
10869 build_type_attribute_variant (@var{mdecl},
10871 (get_identifier ("stdcall"),
10876 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
10877 This target hook returns @code{true} past the point in which new jump
10878 instructions could be created. On machines that require a register for
10879 every jump such as the SHmedia ISA of SH5, this point would typically be
10880 reload, so this target hook should be defined to a function such as:
10884 cannot_modify_jumps_past_reload_p ()
10886 return (reload_completed || reload_in_progress);
10891 @deftypefn {Target Hook} {enum reg_class} TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
10892 This target hook returns a register class for which branch target register
10893 optimizations should be applied. All registers in this class should be
10894 usable interchangeably. After reload, registers in this class will be
10895 re-allocated and loads will be hoisted out of loops and be subjected
10896 to inter-block scheduling.
10899 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
10900 Branch target register optimization will by default exclude callee-saved
10902 that are not already live during the current function; if this target hook
10903 returns true, they will be included. The target code must than make sure
10904 that all target registers in the class returned by
10905 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
10906 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
10907 epilogues have already been generated. Note, even if you only return
10908 true when @var{after_prologue_epilogue_gen} is false, you still are likely
10909 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
10910 to reserve space for caller-saved target registers.
10913 @deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
10914 This target hook returns true if the target supports conditional execution.
10915 This target hook is required only when the target has several different
10916 modes and they have different conditional execution capability, such as ARM.
10919 @deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, struct loop *@var{loop})
10920 This target hook returns a new value for the number of times @var{loop}
10921 should be unrolled. The parameter @var{nunroll} is the number of times
10922 the loop is to be unrolled. The parameter @var{loop} is a pointer to
10923 the loop, which is going to be checked for unrolling. This target hook
10924 is required only when the target has special constraints like maximum
10925 number of memory accesses.
10928 @defmac POWI_MAX_MULTS
10929 If defined, this macro is interpreted as a signed integer C expression
10930 that specifies the maximum number of floating point multiplications
10931 that should be emitted when expanding exponentiation by an integer
10932 constant inline. When this value is defined, exponentiation requiring
10933 more than this number of multiplications is implemented by calling the
10934 system library's @code{pow}, @code{powf} or @code{powl} routines.
10935 The default value places no upper bound on the multiplication count.
10938 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
10939 This target hook should register any extra include files for the
10940 target. The parameter @var{stdinc} indicates if normal include files
10941 are present. The parameter @var{sysroot} is the system root directory.
10942 The parameter @var{iprefix} is the prefix for the gcc directory.
10945 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
10946 This target hook should register any extra include files for the
10947 target before any standard headers. The parameter @var{stdinc}
10948 indicates if normal include files are present. The parameter
10949 @var{sysroot} is the system root directory. The parameter
10950 @var{iprefix} is the prefix for the gcc directory.
10953 @deftypefn Macro void TARGET_OPTF (char *@var{path})
10954 This target hook should register special include paths for the target.
10955 The parameter @var{path} is the include to register. On Darwin
10956 systems, this is used for Framework includes, which have semantics
10957 that are different from @option{-I}.
10960 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
10961 This target macro returns @code{true} if it is safe to use a local alias
10962 for a virtual function @var{fndecl} when constructing thunks,
10963 @code{false} otherwise. By default, the macro returns @code{true} for all
10964 functions, if a target supports aliases (i.e.@: defines
10965 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
10968 @defmac TARGET_FORMAT_TYPES
10969 If defined, this macro is the name of a global variable containing
10970 target-specific format checking information for the @option{-Wformat}
10971 option. The default is to have no target-specific format checks.
10974 @defmac TARGET_N_FORMAT_TYPES
10975 If defined, this macro is the number of entries in
10976 @code{TARGET_FORMAT_TYPES}.
10979 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
10980 If defined, this macro is the name of a global variable containing
10981 target-specific format overrides for the @option{-Wformat} option. The
10982 default is to have no target-specific format overrides. If defined,
10983 @code{TARGET_FORMAT_TYPES} must be defined, too.
10986 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
10987 If defined, this macro specifies the number of entries in
10988 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
10991 @defmac TARGET_OVERRIDES_FORMAT_INIT
10992 If defined, this macro specifies the optional initialization
10993 routine for target specific customizations of the system printf
10994 and scanf formatter settings.
10997 @deftypevr {Target Hook} bool TARGET_RELAXED_ORDERING
10998 If set to @code{true}, means that the target's memory model does not
10999 guarantee that loads which do not depend on one another will access
11000 main memory in the order of the instruction stream; if ordering is
11001 important, an explicit memory barrier must be used. This is true of
11002 many recent processors which implement a policy of ``relaxed,''
11003 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11004 and ia64. The default is @code{false}.
11007 @deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val})
11008 If defined, this macro returns the diagnostic message when it is
11009 illegal to pass argument @var{val} to function @var{funcdecl}
11010 with prototype @var{typelist}.
11013 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype})
11014 If defined, this macro returns the diagnostic message when it is
11015 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11016 if validity should be determined by the front end.
11019 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type})
11020 If defined, this macro returns the diagnostic message when it is
11021 invalid to apply operation @var{op} (where unary plus is denoted by
11022 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11023 if validity should be determined by the front end.
11026 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2})
11027 If defined, this macro returns the diagnostic message when it is
11028 invalid to apply operation @var{op} to operands of types @var{type1}
11029 and @var{type2}, or @code{NULL} if validity should be determined by
11033 @deftypefn {Target Hook} {const char *} TARGET_INVALID_PARAMETER_TYPE (const_tree @var{type})
11034 If defined, this macro returns the diagnostic message when it is
11035 invalid for functions to include parameters of type @var{type},
11036 or @code{NULL} if validity should be determined by
11037 the front end. This is currently used only by the C and C++ front ends.
11040 @deftypefn {Target Hook} {const char *} TARGET_INVALID_RETURN_TYPE (const_tree @var{type})
11041 If defined, this macro returns the diagnostic message when it is
11042 invalid for functions to have return type @var{type},
11043 or @code{NULL} if validity should be determined by
11044 the front end. This is currently used only by the C and C++ front ends.
11047 @deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type})
11048 If defined, this target hook returns the type to which values of
11049 @var{type} should be promoted when they appear in expressions,
11050 analogous to the integer promotions, or @code{NULL_TREE} to use the
11051 front end's normal promotion rules. This hook is useful when there are
11052 target-specific types with special promotion rules.
11053 This is currently used only by the C and C++ front ends.
11056 @deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr})
11057 If defined, this hook returns the result of converting @var{expr} to
11058 @var{type}. It should return the converted expression,
11059 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11060 This hook is useful when there are target-specific types with special
11062 This is currently used only by the C and C++ front ends.
11065 @defmac TARGET_USE_JCR_SECTION
11066 This macro determines whether to use the JCR section to register Java
11067 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11068 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11072 This macro determines the size of the objective C jump buffer for the
11073 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11076 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11077 Define this macro if any target-specific attributes need to be attached
11078 to the functions in @file{libgcc} that provide low-level support for
11079 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11080 and the associated definitions of those functions.
11083 @deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void)
11084 Define this macro to update the current function stack boundary if
11088 @deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void)
11089 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11090 different argument pointer register is needed to access the function's
11091 argument list due to stack realignment. Return @code{NULL} if no DRAP
11095 @deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
11096 When optimization is disabled, this hook indicates whether or not
11097 arguments should be allocated to stack slots. Normally, GCC allocates
11098 stacks slots for arguments when not optimizing in order to make
11099 debugging easier. However, when a function is declared with
11100 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11101 cannot safely move arguments from the registers in which they are passed
11102 to the stack. Therefore, this hook should return true in general, but
11103 false for naked functions. The default implementation always returns true.
11106 @deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR
11107 On some architectures it can take multiple instructions to synthesize
11108 a constant. If there is another constant already in a register that
11109 is close enough in value then it is preferable that the new constant
11110 is computed from this register using immediate addition or
11111 subtraction. We accomplish this through CSE. Besides the value of
11112 the constant we also add a lower and an upper constant anchor to the
11113 available expressions. These are then queried when encountering new
11114 constants. The anchors are computed by rounding the constant up and
11115 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11116 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11117 accepted by immediate-add plus one. We currently assume that the
11118 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11119 MIPS, where add-immediate takes a 16-bit signed value,
11120 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11121 is zero, which disables this optimization. @end deftypevr