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, 2011
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.
94 Similarly, there is a @code{targetcm} variable for hooks that are
95 specific to front ends for C-family languages, documented as ``C
96 Target Hook''. This is declared in @file{c-family/c-target.h}, the
97 initializer @code{TARGETCM_INITIALIZER} in
98 @file{c-family/c-target-def.h}. If targets initialize @code{targetcm}
99 themselves, they should set @code{target_has_targetcm=yes} in
100 @file{config.gcc}; otherwise a default definition is used.
102 Similarly, there is a @code{targetm_common} variable for hooks that
103 are shared between the compiler driver and the compilers proper,
104 documented as ``Common Target Hook''. This is declared in
105 @file{common/common-target.h}, the initializer
106 @code{TARGETM_COMMON_INITIALIZER} in
107 @file{common/common-target-def.h}. If targets initialize
108 @code{targetm_common} themselves, they should set
109 @code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a
110 default definition is used.
113 @section Controlling the Compilation Driver, @file{gcc}
115 @cindex controlling the compilation driver
117 @c prevent bad page break with this line
118 You can control the compilation driver.
120 @defmac DRIVER_SELF_SPECS
121 A list of specs for the driver itself. It should be a suitable
122 initializer for an array of strings, with no surrounding braces.
124 The driver applies these specs to its own command line between loading
125 default @file{specs} files (but not command-line specified ones) and
126 choosing the multilib directory or running any subcommands. It
127 applies them in the order given, so each spec can depend on the
128 options added by earlier ones. It is also possible to remove options
129 using @samp{%<@var{option}} in the usual way.
131 This macro can be useful when a port has several interdependent target
132 options. It provides a way of standardizing the command line so
133 that the other specs are easier to write.
135 Do not define this macro if it does not need to do anything.
138 @defmac OPTION_DEFAULT_SPECS
139 A list of specs used to support configure-time default options (i.e.@:
140 @option{--with} options) in the driver. It should be a suitable initializer
141 for an array of structures, each containing two strings, without the
142 outermost pair of surrounding braces.
144 The first item in the pair is the name of the default. This must match
145 the code in @file{config.gcc} for the target. The second item is a spec
146 to apply if a default with this name was specified. The string
147 @samp{%(VALUE)} in the spec will be replaced by the value of the default
148 everywhere it occurs.
150 The driver will apply these specs to its own command line between loading
151 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
152 the same mechanism as @code{DRIVER_SELF_SPECS}.
154 Do not define this macro if it does not need to do anything.
158 A C string constant that tells the GCC driver program options to
159 pass to CPP@. It can also specify how to translate options you
160 give to GCC into options for GCC to pass to the CPP@.
162 Do not define this macro if it does not need to do anything.
165 @defmac CPLUSPLUS_CPP_SPEC
166 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
167 than C@. If you do not define this macro, then the value of
168 @code{CPP_SPEC} (if any) will be used instead.
172 A C string constant that tells the GCC driver program options to
173 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
175 It can also specify how to translate options you give to GCC into options
176 for GCC to pass to front ends.
178 Do not define this macro if it does not need to do anything.
182 A C string constant that tells the GCC driver program options to
183 pass to @code{cc1plus}. It can also specify how to translate options you
184 give to GCC into options for GCC to pass to the @code{cc1plus}.
186 Do not define this macro if it does not need to do anything.
187 Note that everything defined in CC1_SPEC is already passed to
188 @code{cc1plus} so there is no need to duplicate the contents of
189 CC1_SPEC in CC1PLUS_SPEC@.
193 A C string constant that tells the GCC driver program options to
194 pass to the assembler. It can also specify how to translate options
195 you give to GCC into options for GCC to pass to the assembler.
196 See the file @file{sun3.h} for an example of this.
198 Do not define this macro if it does not need to do anything.
201 @defmac ASM_FINAL_SPEC
202 A C string constant that tells the GCC driver program how to
203 run any programs which cleanup after the normal assembler.
204 Normally, this is not needed. See the file @file{mips.h} for
207 Do not define this macro if it does not need to do anything.
210 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
211 Define this macro, with no value, if the driver should give the assembler
212 an argument consisting of a single dash, @option{-}, to instruct it to
213 read from its standard input (which will be a pipe connected to the
214 output of the compiler proper). This argument is given after any
215 @option{-o} option specifying the name of the output file.
217 If you do not define this macro, the assembler is assumed to read its
218 standard input if given no non-option arguments. If your assembler
219 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
220 see @file{mips.h} for instance.
224 A C string constant that tells the GCC driver program options to
225 pass to the linker. It can also specify how to translate options you
226 give to GCC into options for GCC to pass to the linker.
228 Do not define this macro if it does not need to do anything.
232 Another C string constant used much like @code{LINK_SPEC}. The difference
233 between the two is that @code{LIB_SPEC} is used at the end of the
234 command given to the linker.
236 If this macro is not defined, a default is provided that
237 loads the standard C library from the usual place. See @file{gcc.c}.
241 Another C string constant that tells the GCC driver program
242 how and when to place a reference to @file{libgcc.a} into the
243 linker command line. This constant is placed both before and after
244 the value of @code{LIB_SPEC}.
246 If this macro is not defined, the GCC driver provides a default that
247 passes the string @option{-lgcc} to the linker.
250 @defmac REAL_LIBGCC_SPEC
251 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
252 @code{LIBGCC_SPEC} is not directly used by the driver program but is
253 instead modified to refer to different versions of @file{libgcc.a}
254 depending on the values of the command line flags @option{-static},
255 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
256 targets where these modifications are inappropriate, define
257 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
258 driver how to place a reference to @file{libgcc} on the link command
259 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
262 @defmac USE_LD_AS_NEEDED
263 A macro that controls the modifications to @code{LIBGCC_SPEC}
264 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
265 generated that uses --as-needed and the shared libgcc in place of the
266 static exception handler library, when linking without any of
267 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
271 If defined, this C string constant is added to @code{LINK_SPEC}.
272 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
273 the modifications to @code{LIBGCC_SPEC} mentioned in
274 @code{REAL_LIBGCC_SPEC}.
277 @defmac STARTFILE_SPEC
278 Another C string constant used much like @code{LINK_SPEC}. The
279 difference between the two is that @code{STARTFILE_SPEC} is used at
280 the very beginning of the command given to the linker.
282 If this macro is not defined, a default is provided that loads the
283 standard C startup file from the usual place. See @file{gcc.c}.
287 Another C string constant used much like @code{LINK_SPEC}. The
288 difference between the two is that @code{ENDFILE_SPEC} is used at
289 the very end of the command given to the linker.
291 Do not define this macro if it does not need to do anything.
294 @defmac THREAD_MODEL_SPEC
295 GCC @code{-v} will print the thread model GCC was configured to use.
296 However, this doesn't work on platforms that are multilibbed on thread
297 models, such as AIX 4.3. On such platforms, define
298 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
299 blanks that names one of the recognized thread models. @code{%*}, the
300 default value of this macro, will expand to the value of
301 @code{thread_file} set in @file{config.gcc}.
304 @defmac SYSROOT_SUFFIX_SPEC
305 Define this macro to add a suffix to the target sysroot when GCC is
306 configured with a sysroot. This will cause GCC to search for usr/lib,
307 et al, within sysroot+suffix.
310 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
311 Define this macro to add a headers_suffix to the target sysroot when
312 GCC is configured with a sysroot. This will cause GCC to pass the
313 updated sysroot+headers_suffix to CPP, causing it to search for
314 usr/include, et al, within sysroot+headers_suffix.
318 Define this macro to provide additional specifications to put in the
319 @file{specs} file that can be used in various specifications like
322 The definition should be an initializer for an array of structures,
323 containing a string constant, that defines the specification name, and a
324 string constant that provides the specification.
326 Do not define this macro if it does not need to do anything.
328 @code{EXTRA_SPECS} is useful when an architecture contains several
329 related targets, which have various @code{@dots{}_SPECS} which are similar
330 to each other, and the maintainer would like one central place to keep
333 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
334 define either @code{_CALL_SYSV} when the System V calling sequence is
335 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
338 The @file{config/rs6000/rs6000.h} target file defines:
341 #define EXTRA_SPECS \
342 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
344 #define CPP_SYS_DEFAULT ""
347 The @file{config/rs6000/sysv.h} target file defines:
351 "%@{posix: -D_POSIX_SOURCE @} \
352 %@{mcall-sysv: -D_CALL_SYSV @} \
353 %@{!mcall-sysv: %(cpp_sysv_default) @} \
354 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
356 #undef CPP_SYSV_DEFAULT
357 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
360 while the @file{config/rs6000/eabiaix.h} target file defines
361 @code{CPP_SYSV_DEFAULT} as:
364 #undef CPP_SYSV_DEFAULT
365 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
369 @defmac LINK_LIBGCC_SPECIAL_1
370 Define this macro if the driver program should find the library
371 @file{libgcc.a}. If you do not define this macro, the driver program will pass
372 the argument @option{-lgcc} to tell the linker to do the search.
375 @defmac LINK_GCC_C_SEQUENCE_SPEC
376 The sequence in which libgcc and libc are specified to the linker.
377 By default this is @code{%G %L %G}.
380 @defmac LINK_COMMAND_SPEC
381 A C string constant giving the complete command line need to execute the
382 linker. When you do this, you will need to update your port each time a
383 change is made to the link command line within @file{gcc.c}. Therefore,
384 define this macro only if you need to completely redefine the command
385 line for invoking the linker and there is no other way to accomplish
386 the effect you need. Overriding this macro may be avoidable by overriding
387 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
390 @defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES
391 A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search
392 directories from linking commands. Do not give it a nonzero value if
393 removing duplicate search directories changes the linker's semantics.
396 @deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT
397 True if @file{..} components should always be removed from directory names computed relative to GCC's internal directories, false (default) if such components should be preserved and directory names containing them passed to other tools such as the linker.
400 @defmac MULTILIB_DEFAULTS
401 Define this macro as a C expression for the initializer of an array of
402 string to tell the driver program which options are defaults for this
403 target and thus do not need to be handled specially when using
404 @code{MULTILIB_OPTIONS}.
406 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
407 the target makefile fragment or if none of the options listed in
408 @code{MULTILIB_OPTIONS} are set by default.
409 @xref{Target Fragment}.
412 @defmac RELATIVE_PREFIX_NOT_LINKDIR
413 Define this macro to tell @command{gcc} that it should only translate
414 a @option{-B} prefix into a @option{-L} linker option if the prefix
415 indicates an absolute file name.
418 @defmac MD_EXEC_PREFIX
419 If defined, this macro is an additional prefix to try after
420 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
421 when the compiler is built as a cross
422 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
423 to the list of directories used to find the assembler in @file{configure.in}.
426 @defmac STANDARD_STARTFILE_PREFIX
427 Define this macro as a C string constant if you wish to override the
428 standard choice of @code{libdir} as the default prefix to
429 try when searching for startup files such as @file{crt0.o}.
430 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
431 is built as a cross compiler.
434 @defmac STANDARD_STARTFILE_PREFIX_1
435 Define this macro as a C string constant if you wish to override the
436 standard choice of @code{/lib} as a prefix to try after the default prefix
437 when searching for startup files such as @file{crt0.o}.
438 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
439 is built as a cross compiler.
442 @defmac STANDARD_STARTFILE_PREFIX_2
443 Define this macro as a C string constant if you wish to override the
444 standard choice of @code{/lib} as yet another prefix to try after the
445 default prefix when searching for startup files such as @file{crt0.o}.
446 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
447 is built as a cross compiler.
450 @defmac MD_STARTFILE_PREFIX
451 If defined, this macro supplies an additional prefix to try after the
452 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
453 compiler is built as a cross compiler.
456 @defmac MD_STARTFILE_PREFIX_1
457 If defined, this macro supplies yet another prefix to try after the
458 standard prefixes. It is not searched when the compiler is built as a
462 @defmac INIT_ENVIRONMENT
463 Define this macro as a C string constant if you wish to set environment
464 variables for programs called by the driver, such as the assembler and
465 loader. The driver passes the value of this macro to @code{putenv} to
466 initialize the necessary environment variables.
469 @defmac LOCAL_INCLUDE_DIR
470 Define this macro as a C string constant if you wish to override the
471 standard choice of @file{/usr/local/include} as the default prefix to
472 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
473 comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in
474 @file{config.gcc}, normally @file{/usr/include}) in the search order.
476 Cross compilers do not search either @file{/usr/local/include} or its
480 @defmac NATIVE_SYSTEM_HEADER_COMPONENT
481 The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}.
482 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
483 If you do not define this macro, no component is used.
486 @defmac INCLUDE_DEFAULTS
487 Define this macro if you wish to override the entire default search path
488 for include files. For a native compiler, the default search path
489 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
490 @code{GPLUSPLUS_INCLUDE_DIR}, and
491 @code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
492 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
493 and specify private search areas for GCC@. The directory
494 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
496 The definition should be an initializer for an array of structures.
497 Each array element should have four elements: the directory name (a
498 string constant), the component name (also a string constant), a flag
499 for C++-only directories,
500 and a flag showing that the includes in the directory don't need to be
501 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
502 the array with a null element.
504 The component name denotes what GNU package the include file is part of,
505 if any, in all uppercase letters. For example, it might be @samp{GCC}
506 or @samp{BINUTILS}. If the package is part of a vendor-supplied
507 operating system, code the component name as @samp{0}.
509 For example, here is the definition used for VAX/VMS:
512 #define INCLUDE_DEFAULTS \
514 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
515 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
516 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
523 Here is the order of prefixes tried for exec files:
527 Any prefixes specified by the user with @option{-B}.
530 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
531 is not set and the compiler has not been installed in the configure-time
532 @var{prefix}, the location in which the compiler has actually been installed.
535 The directories specified by the environment variable @code{COMPILER_PATH}.
538 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
539 in the configured-time @var{prefix}.
542 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
545 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
548 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
552 Here is the order of prefixes tried for startfiles:
556 Any prefixes specified by the user with @option{-B}.
559 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
560 value based on the installed toolchain location.
563 The directories specified by the environment variable @code{LIBRARY_PATH}
564 (or port-specific name; native only, cross compilers do not use this).
567 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
568 in the configured @var{prefix} or this is a native compiler.
571 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
574 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
578 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
579 native compiler, or we have a target system root.
582 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
583 native compiler, or we have a target system root.
586 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
587 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
588 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
591 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
592 compiler, or we have a target system root. The default for this macro is
596 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
597 compiler, or we have a target system root. The default for this macro is
601 @node Run-time Target
602 @section Run-time Target Specification
603 @cindex run-time target specification
604 @cindex predefined macros
605 @cindex target specifications
607 @c prevent bad page break with this line
608 Here are run-time target specifications.
610 @defmac TARGET_CPU_CPP_BUILTINS ()
611 This function-like macro expands to a block of code that defines
612 built-in preprocessor macros and assertions for the target CPU, using
613 the functions @code{builtin_define}, @code{builtin_define_std} and
614 @code{builtin_assert}. When the front end
615 calls this macro it provides a trailing semicolon, and since it has
616 finished command line option processing your code can use those
619 @code{builtin_assert} takes a string in the form you pass to the
620 command-line option @option{-A}, such as @code{cpu=mips}, and creates
621 the assertion. @code{builtin_define} takes a string in the form
622 accepted by option @option{-D} and unconditionally defines the macro.
624 @code{builtin_define_std} takes a string representing the name of an
625 object-like macro. If it doesn't lie in the user's namespace,
626 @code{builtin_define_std} defines it unconditionally. Otherwise, it
627 defines a version with two leading underscores, and another version
628 with two leading and trailing underscores, and defines the original
629 only if an ISO standard was not requested on the command line. For
630 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
631 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
632 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
633 defines only @code{_ABI64}.
635 You can also test for the C dialect being compiled. The variable
636 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
637 or @code{clk_objective_c}. Note that if we are preprocessing
638 assembler, this variable will be @code{clk_c} but the function-like
639 macro @code{preprocessing_asm_p()} will return true, so you might want
640 to check for that first. If you need to check for strict ANSI, the
641 variable @code{flag_iso} can be used. The function-like macro
642 @code{preprocessing_trad_p()} can be used to check for traditional
646 @defmac TARGET_OS_CPP_BUILTINS ()
647 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
648 and is used for the target operating system instead.
651 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
652 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
653 and is used for the target object format. @file{elfos.h} uses this
654 macro to define @code{__ELF__}, so you probably do not need to define
658 @deftypevar {extern int} target_flags
659 This variable is declared in @file{options.h}, which is included before
660 any target-specific headers.
663 @deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
664 This variable specifies the initial value of @code{target_flags}.
665 Its default setting is 0.
668 @cindex optional hardware or system features
669 @cindex features, optional, in system conventions
671 @deftypefn {Common Target Hook} bool TARGET_HANDLE_OPTION (struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set}, const struct cl_decoded_option *@var{decoded}, location_t @var{loc})
672 This hook is called whenever the user specifies one of the
673 target-specific options described by the @file{.opt} definition files
674 (@pxref{Options}). It has the opportunity to do some option-specific
675 processing and should return true if the option is valid. The default
676 definition does nothing but return true.
678 @var{decoded} specifies the option and its arguments. @var{opts} and
679 @var{opts_set} are the @code{gcc_options} structures to be used for
680 storing option state, and @var{loc} is the location at which the
681 option was passed (@code{UNKNOWN_LOCATION} except for options passed
685 @deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
686 This target hook is called whenever the user specifies one of the
687 target-specific C language family options described by the @file{.opt}
688 definition files(@pxref{Options}). It has the opportunity to do some
689 option-specific processing and should return true if the option is
690 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
691 default definition does nothing but return false.
693 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
694 options. However, if processing an option requires routines that are
695 only available in the C (and related language) front ends, then you
696 should use @code{TARGET_HANDLE_C_OPTION} instead.
699 @deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string})
700 Targets may provide a string object type that can be used within and between C, C++ and their respective Objective-C dialects. A string object might, for example, embed encoding and length information. These objects are considered opaque to the compiler and handled as references. An ideal implementation makes the composition of the string object match that of the Objective-C @code{NSString} (@code{NXString} for GNUStep), allowing efficient interworking between C-only and Objective-C code. If a target implements string objects then this hook should return a reference to such an object constructed from the normal `C' string representation provided in @var{string}. At present, the hook is used by Objective-C only, to obtain a common-format string object when the target provides one.
703 @deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref})
704 If a target implements string objects then this hook should return @code{true} if @var{stringref} is a valid reference to such an object.
707 @deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list})
708 If a target implements string objects then this hook should should provide a facility to check the function arguments in @var{args_list} against the format specifiers in @var{format_arg} where the type of @var{format_arg} is one recognized as a valid string reference type.
711 @deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)
712 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
713 but is called when the optimize level is changed via an attribute or
714 pragma or when it is reset at the end of the code affected by the
715 attribute or pragma. It is not called at the beginning of compilation
716 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
717 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
718 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
721 @defmac C_COMMON_OVERRIDE_OPTIONS
722 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
723 but is only used in the C
724 language frontends (C, Objective-C, C++, Objective-C++) and so can be
725 used to alter option flag variables which only exist in those
729 @deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE
730 Some machines may desire to change what optimizations are performed for
731 various optimization levels. This variable, if defined, describes
732 options to enable at particular sets of optimization levels. These
733 options are processed once
734 just after the optimization level is determined and before the remainder
735 of the command options have been parsed, so may be overridden by other
736 options passed explicitly.
738 This processing is run once at program startup and when the optimization
739 options are changed via @code{#pragma GCC optimize} or by using the
740 @code{optimize} attribute.
743 @deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts})
744 Set target-dependent initial values of fields in @var{opts}.
747 @deftypefn {Common Target Hook} void TARGET_OPTION_DEFAULT_PARAMS (void)
748 Set target-dependent default values for @option{--param} settings, using calls to @code{set_default_param_value}.
751 @defmac SWITCHABLE_TARGET
752 Some targets need to switch between substantially different subtargets
753 during compilation. For example, the MIPS target has one subtarget for
754 the traditional MIPS architecture and another for MIPS16. Source code
755 can switch between these two subarchitectures using the @code{mips16}
756 and @code{nomips16} attributes.
758 Such subtargets can differ in things like the set of available
759 registers, the set of available instructions, the costs of various
760 operations, and so on. GCC caches a lot of this type of information
761 in global variables, and recomputing them for each subtarget takes a
762 significant amount of time. The compiler therefore provides a facility
763 for maintaining several versions of the global variables and quickly
764 switching between them; see @file{target-globals.h} for details.
766 Define this macro to 1 if your target needs this facility. The default
770 @node Per-Function Data
771 @section Defining data structures for per-function information.
772 @cindex per-function data
773 @cindex data structures
775 If the target needs to store information on a per-function basis, GCC
776 provides a macro and a couple of variables to allow this. Note, just
777 using statics to store the information is a bad idea, since GCC supports
778 nested functions, so you can be halfway through encoding one function
779 when another one comes along.
781 GCC defines a data structure called @code{struct function} which
782 contains all of the data specific to an individual function. This
783 structure contains a field called @code{machine} whose type is
784 @code{struct machine_function *}, which can be used by targets to point
785 to their own specific data.
787 If a target needs per-function specific data it should define the type
788 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
789 This macro should be used to initialize the function pointer
790 @code{init_machine_status}. This pointer is explained below.
792 One typical use of per-function, target specific data is to create an
793 RTX to hold the register containing the function's return address. This
794 RTX can then be used to implement the @code{__builtin_return_address}
795 function, for level 0.
797 Note---earlier implementations of GCC used a single data area to hold
798 all of the per-function information. Thus when processing of a nested
799 function began the old per-function data had to be pushed onto a
800 stack, and when the processing was finished, it had to be popped off the
801 stack. GCC used to provide function pointers called
802 @code{save_machine_status} and @code{restore_machine_status} to handle
803 the saving and restoring of the target specific information. Since the
804 single data area approach is no longer used, these pointers are no
807 @defmac INIT_EXPANDERS
808 Macro called to initialize any target specific information. This macro
809 is called once per function, before generation of any RTL has begun.
810 The intention of this macro is to allow the initialization of the
811 function pointer @code{init_machine_status}.
814 @deftypevar {void (*)(struct function *)} init_machine_status
815 If this function pointer is non-@code{NULL} it will be called once per
816 function, before function compilation starts, in order to allow the
817 target to perform any target specific initialization of the
818 @code{struct function} structure. It is intended that this would be
819 used to initialize the @code{machine} of that structure.
821 @code{struct machine_function} structures are expected to be freed by GC@.
822 Generally, any memory that they reference must be allocated by using
823 GC allocation, including the structure itself.
827 @section Storage Layout
828 @cindex storage layout
830 Note that the definitions of the macros in this table which are sizes or
831 alignments measured in bits do not need to be constant. They can be C
832 expressions that refer to static variables, such as the @code{target_flags}.
833 @xref{Run-time Target}.
835 @defmac BITS_BIG_ENDIAN
836 Define this macro to have the value 1 if the most significant bit in a
837 byte has the lowest number; otherwise define it to have the value zero.
838 This means that bit-field instructions count from the most significant
839 bit. If the machine has no bit-field instructions, then this must still
840 be defined, but it doesn't matter which value it is defined to. This
841 macro need not be a constant.
843 This macro does not affect the way structure fields are packed into
844 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
847 @defmac BYTES_BIG_ENDIAN
848 Define this macro to have the value 1 if the most significant byte in a
849 word has the lowest number. This macro need not be a constant.
852 @defmac WORDS_BIG_ENDIAN
853 Define this macro to have the value 1 if, in a multiword object, the
854 most significant word has the lowest number. This applies to both
855 memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the
856 order of words in memory is not the same as the order in registers. This
857 macro need not be a constant.
860 @defmac REG_WORDS_BIG_ENDIAN
861 On some machines, the order of words in a multiword object differs between
862 registers in memory. In such a situation, define this macro to describe
863 the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls
864 the order of words in memory.
867 @defmac FLOAT_WORDS_BIG_ENDIAN
868 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
869 @code{TFmode} floating point numbers are stored in memory with the word
870 containing the sign bit at the lowest address; otherwise define it to
871 have the value 0. This macro need not be a constant.
873 You need not define this macro if the ordering is the same as for
877 @defmac BITS_PER_UNIT
878 Define this macro to be the number of bits in an addressable storage
879 unit (byte). If you do not define this macro the default is 8.
882 @defmac BITS_PER_WORD
883 Number of bits in a word. If you do not define this macro, the default
884 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
887 @defmac MAX_BITS_PER_WORD
888 Maximum number of bits in a word. If this is undefined, the default is
889 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
890 largest value that @code{BITS_PER_WORD} can have at run-time.
893 @defmac UNITS_PER_WORD
894 Number of storage units in a word; normally the size of a general-purpose
895 register, a power of two from 1 or 8.
898 @defmac MIN_UNITS_PER_WORD
899 Minimum number of units in a word. If this is undefined, the default is
900 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
901 smallest value that @code{UNITS_PER_WORD} can have at run-time.
905 Width of a pointer, in bits. You must specify a value no wider than the
906 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
907 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
908 a value the default is @code{BITS_PER_WORD}.
911 @defmac POINTERS_EXTEND_UNSIGNED
912 A C expression that determines how pointers should be extended from
913 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
914 greater than zero if pointers should be zero-extended, zero if they
915 should be sign-extended, and negative if some other sort of conversion
916 is needed. In the last case, the extension is done by the target's
917 @code{ptr_extend} instruction.
919 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
920 and @code{word_mode} are all the same width.
923 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
924 A macro to update @var{m} and @var{unsignedp} when an object whose type
925 is @var{type} and which has the specified mode and signedness is to be
926 stored in a register. This macro is only called when @var{type} is a
929 On most RISC machines, which only have operations that operate on a full
930 register, define this macro to set @var{m} to @code{word_mode} if
931 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
932 cases, only integer modes should be widened because wider-precision
933 floating-point operations are usually more expensive than their narrower
936 For most machines, the macro definition does not change @var{unsignedp}.
937 However, some machines, have instructions that preferentially handle
938 either signed or unsigned quantities of certain modes. For example, on
939 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
940 sign-extend the result to 64 bits. On such machines, set
941 @var{unsignedp} according to which kind of extension is more efficient.
943 Do not define this macro if it would never modify @var{m}.
946 @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})
947 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
948 function return values. The target hook should return the new mode
949 and possibly change @code{*@var{punsignedp}} if the promotion should
950 change signedness. This function is called only for scalar @emph{or
953 @var{for_return} allows to distinguish the promotion of arguments and
954 return values. If it is @code{1}, a return value is being promoted and
955 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
956 If it is @code{2}, the returned mode should be that of the register in
957 which an incoming parameter is copied, or the outgoing result is computed;
958 then the hook should return the same mode as @code{promote_mode}, though
959 the signedness may be different.
961 @var{type} can be NULL when promoting function arguments of libcalls.
963 The default is to not promote arguments and return values. You can
964 also define the hook to @code{default_promote_function_mode_always_promote}
965 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
968 @defmac PARM_BOUNDARY
969 Normal alignment required for function parameters on the stack, in
970 bits. All stack parameters receive at least this much alignment
971 regardless of data type. On most machines, this is the same as the
975 @defmac STACK_BOUNDARY
976 Define this macro to the minimum alignment enforced by hardware for the
977 stack pointer on this machine. The definition is a C expression for the
978 desired alignment (measured in bits). This value is used as a default
979 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
980 this should be the same as @code{PARM_BOUNDARY}.
983 @defmac PREFERRED_STACK_BOUNDARY
984 Define this macro if you wish to preserve a certain alignment for the
985 stack pointer, greater than what the hardware enforces. The definition
986 is a C expression for the desired alignment (measured in bits). This
987 macro must evaluate to a value equal to or larger than
988 @code{STACK_BOUNDARY}.
991 @defmac INCOMING_STACK_BOUNDARY
992 Define this macro if the incoming stack boundary may be different
993 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
994 to a value equal to or larger than @code{STACK_BOUNDARY}.
997 @defmac FUNCTION_BOUNDARY
998 Alignment required for a function entry point, in bits.
1001 @defmac BIGGEST_ALIGNMENT
1002 Biggest alignment that any data type can require on this machine, in
1003 bits. Note that this is not the biggest alignment that is supported,
1004 just the biggest alignment that, when violated, may cause a fault.
1007 @defmac MALLOC_ABI_ALIGNMENT
1008 Alignment, in bits, a C conformant malloc implementation has to
1009 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1012 @defmac ATTRIBUTE_ALIGNED_VALUE
1013 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1014 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1017 @defmac MINIMUM_ATOMIC_ALIGNMENT
1018 If defined, the smallest alignment, in bits, that can be given to an
1019 object that can be referenced in one operation, without disturbing any
1020 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1021 on machines that don't have byte or half-word store operations.
1024 @defmac BIGGEST_FIELD_ALIGNMENT
1025 Biggest alignment that any structure or union field can require on this
1026 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1027 structure and union fields only, unless the field alignment has been set
1028 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1031 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1032 An expression for the alignment of a structure field @var{field} if the
1033 alignment computed in the usual way (including applying of
1034 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1035 alignment) is @var{computed}. It overrides alignment only if the
1036 field alignment has not been set by the
1037 @code{__attribute__ ((aligned (@var{n})))} construct.
1040 @defmac MAX_STACK_ALIGNMENT
1041 Biggest stack alignment guaranteed by the backend. Use this macro
1042 to specify the maximum alignment of a variable on stack.
1044 If not defined, the default value is @code{STACK_BOUNDARY}.
1046 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1047 @c But the fix for PR 32893 indicates that we can only guarantee
1048 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1049 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1052 @defmac MAX_OFILE_ALIGNMENT
1053 Biggest alignment supported by the object file format of this machine.
1054 Use this macro to limit the alignment which can be specified using the
1055 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1056 the default value is @code{BIGGEST_ALIGNMENT}.
1058 On systems that use ELF, the default (in @file{config/elfos.h}) is
1059 the largest supported 32-bit ELF section alignment representable on
1060 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1061 On 32-bit ELF the largest supported section alignment in bits is
1062 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1065 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1066 If defined, a C expression to compute the alignment for a variable in
1067 the static store. @var{type} is the data type, and @var{basic-align} is
1068 the alignment that the object would ordinarily have. The value of this
1069 macro is used instead of that alignment to align the object.
1071 If this macro is not defined, then @var{basic-align} is used.
1074 One use of this macro is to increase alignment of medium-size data to
1075 make it all fit in fewer cache lines. Another is to cause character
1076 arrays to be word-aligned so that @code{strcpy} calls that copy
1077 constants to character arrays can be done inline.
1080 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1081 If defined, a C expression to compute the alignment given to a constant
1082 that is being placed in memory. @var{constant} is the constant and
1083 @var{basic-align} is the alignment that the object would ordinarily
1084 have. The value of this macro is used instead of that alignment to
1087 If this macro is not defined, then @var{basic-align} is used.
1089 The typical use of this macro is to increase alignment for string
1090 constants to be word aligned so that @code{strcpy} calls that copy
1091 constants can be done inline.
1094 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1095 If defined, a C expression to compute the alignment for a variable in
1096 the local store. @var{type} is the data type, and @var{basic-align} is
1097 the alignment that the object would ordinarily have. The value of this
1098 macro is used instead of that alignment to align the object.
1100 If this macro is not defined, then @var{basic-align} is used.
1102 One use of this macro is to increase alignment of medium-size data to
1103 make it all fit in fewer cache lines.
1105 If the value of this macro has a type, it should be an unsigned type.
1108 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1109 If defined, a C expression to compute the alignment for stack slot.
1110 @var{type} is the data type, @var{mode} is the widest mode available,
1111 and @var{basic-align} is the alignment that the slot would ordinarily
1112 have. The value of this macro is used instead of that alignment to
1115 If this macro is not defined, then @var{basic-align} is used when
1116 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1119 This macro is to set alignment of stack slot to the maximum alignment
1120 of all possible modes which the slot may have.
1122 If the value of this macro has a type, it should be an unsigned type.
1125 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1126 If defined, a C expression to compute the alignment for a local
1127 variable @var{decl}.
1129 If this macro is not defined, then
1130 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1133 One use of this macro is to increase alignment of medium-size data to
1134 make it all fit in fewer cache lines.
1136 If the value of this macro has a type, it should be an unsigned type.
1139 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1140 If defined, a C expression to compute the minimum required alignment
1141 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1142 @var{mode}, assuming normal alignment @var{align}.
1144 If this macro is not defined, then @var{align} will be used.
1147 @defmac EMPTY_FIELD_BOUNDARY
1148 Alignment in bits to be given to a structure bit-field that follows an
1149 empty field such as @code{int : 0;}.
1151 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1154 @defmac STRUCTURE_SIZE_BOUNDARY
1155 Number of bits which any structure or union's size must be a multiple of.
1156 Each structure or union's size is rounded up to a multiple of this.
1158 If you do not define this macro, the default is the same as
1159 @code{BITS_PER_UNIT}.
1162 @defmac STRICT_ALIGNMENT
1163 Define this macro to be the value 1 if instructions will fail to work
1164 if given data not on the nominal alignment. If instructions will merely
1165 go slower in that case, define this macro as 0.
1168 @defmac PCC_BITFIELD_TYPE_MATTERS
1169 Define this if you wish to imitate the way many other C compilers handle
1170 alignment of bit-fields and the structures that contain them.
1172 The behavior is that the type written for a named bit-field (@code{int},
1173 @code{short}, or other integer type) imposes an alignment for the entire
1174 structure, as if the structure really did contain an ordinary field of
1175 that type. In addition, the bit-field is placed within the structure so
1176 that it would fit within such a field, not crossing a boundary for it.
1178 Thus, on most machines, a named bit-field whose type is written as
1179 @code{int} would not cross a four-byte boundary, and would force
1180 four-byte alignment for the whole structure. (The alignment used may
1181 not be four bytes; it is controlled by the other alignment parameters.)
1183 An unnamed bit-field will not affect the alignment of the containing
1186 If the macro is defined, its definition should be a C expression;
1187 a nonzero value for the expression enables this behavior.
1189 Note that if this macro is not defined, or its value is zero, some
1190 bit-fields may cross more than one alignment boundary. The compiler can
1191 support such references if there are @samp{insv}, @samp{extv}, and
1192 @samp{extzv} insns that can directly reference memory.
1194 The other known way of making bit-fields work is to define
1195 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1196 Then every structure can be accessed with fullwords.
1198 Unless the machine has bit-field instructions or you define
1199 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1200 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1202 If your aim is to make GCC use the same conventions for laying out
1203 bit-fields as are used by another compiler, here is how to investigate
1204 what the other compiler does. Compile and run this program:
1223 printf ("Size of foo1 is %d\n",
1224 sizeof (struct foo1));
1225 printf ("Size of foo2 is %d\n",
1226 sizeof (struct foo2));
1231 If this prints 2 and 5, then the compiler's behavior is what you would
1232 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1235 @defmac BITFIELD_NBYTES_LIMITED
1236 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1237 to aligning a bit-field within the structure.
1240 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void)
1241 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1242 whether unnamed bitfields affect the alignment of the containing
1243 structure. The hook should return true if the structure should inherit
1244 the alignment requirements of an unnamed bitfield's type.
1247 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void)
1248 This target hook should return @code{true} if accesses to volatile bitfields
1249 should use the narrowest mode possible. It should return @code{false} if
1250 these accesses should use the bitfield container type.
1252 The default is @code{!TARGET_STRICT_ALIGN}.
1255 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1256 Return 1 if a structure or array containing @var{field} should be accessed using
1259 If @var{field} is the only field in the structure, @var{mode} is its
1260 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1261 case where structures of one field would require the structure's mode to
1262 retain the field's mode.
1264 Normally, this is not needed.
1267 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1268 Define this macro as an expression for the alignment of a type (given
1269 by @var{type} as a tree node) if the alignment computed in the usual
1270 way is @var{computed} and the alignment explicitly specified was
1273 The default is to use @var{specified} if it is larger; otherwise, use
1274 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1277 @defmac MAX_FIXED_MODE_SIZE
1278 An integer expression for the size in bits of the largest integer
1279 machine mode that should actually be used. All integer machine modes of
1280 this size or smaller can be used for structures and unions with the
1281 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1282 (DImode)} is assumed.
1285 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1286 If defined, an expression of type @code{enum machine_mode} that
1287 specifies the mode of the save area operand of a
1288 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1289 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1290 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1291 having its mode specified.
1293 You need not define this macro if it always returns @code{Pmode}. You
1294 would most commonly define this macro if the
1295 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1299 @defmac STACK_SIZE_MODE
1300 If defined, an expression of type @code{enum machine_mode} that
1301 specifies the mode of the size increment operand of an
1302 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1304 You need not define this macro if it always returns @code{word_mode}.
1305 You would most commonly define this macro if the @code{allocate_stack}
1306 pattern needs to support both a 32- and a 64-bit mode.
1309 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_CMP_RETURN_MODE (void)
1310 This target hook should return the mode to be used for the return value
1311 of compare instructions expanded to libgcc calls. If not defined
1312 @code{word_mode} is returned which is the right choice for a majority of
1316 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_SHIFT_COUNT_MODE (void)
1317 This target hook should return the mode to be used for the shift count operand
1318 of shift instructions expanded to libgcc calls. If not defined
1319 @code{word_mode} is returned which is the right choice for a majority of
1323 @deftypefn {Target Hook} {enum machine_mode} TARGET_UNWIND_WORD_MODE (void)
1324 Return machine mode to be used for @code{_Unwind_Word} type.
1325 The default is to use @code{word_mode}.
1328 @defmac ROUND_TOWARDS_ZERO
1329 If defined, this macro should be true if the prevailing rounding
1330 mode is towards zero.
1332 Defining this macro only affects the way @file{libgcc.a} emulates
1333 floating-point arithmetic.
1335 Not defining this macro is equivalent to returning zero.
1338 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1339 This macro should return true if floats with @var{size}
1340 bits do not have a NaN or infinity representation, but use the largest
1341 exponent for normal numbers instead.
1343 Defining this macro only affects the way @file{libgcc.a} emulates
1344 floating-point arithmetic.
1346 The default definition of this macro returns false for all sizes.
1349 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type})
1350 This target hook returns @code{true} if bit-fields in the given
1351 @var{record_type} are to be laid out following the rules of Microsoft
1352 Visual C/C++, namely: (i) a bit-field won't share the same storage
1353 unit with the previous bit-field if their underlying types have
1354 different sizes, and the bit-field will be aligned to the highest
1355 alignment of the underlying types of itself and of the previous
1356 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1357 the whole enclosing structure, even if it is unnamed; except that
1358 (iii) a zero-sized bit-field will be disregarded unless it follows
1359 another bit-field of nonzero size. If this hook returns @code{true},
1360 other macros that control bit-field layout are ignored.
1362 When a bit-field is inserted into a packed record, the whole size
1363 of the underlying type is used by one or more same-size adjacent
1364 bit-fields (that is, if its long:3, 32 bits is used in the record,
1365 and any additional adjacent long bit-fields are packed into the same
1366 chunk of 32 bits. However, if the size changes, a new field of that
1367 size is allocated). In an unpacked record, this is the same as using
1368 alignment, but not equivalent when packing.
1370 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1371 the latter will take precedence. If @samp{__attribute__((packed))} is
1372 used on a single field when MS bit-fields are in use, it will take
1373 precedence for that field, but the alignment of the rest of the structure
1374 may affect its placement.
1377 @deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1378 Returns true if the target supports decimal floating point.
1381 @deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void)
1382 Returns true if the target supports fixed-point arithmetic.
1385 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1386 This hook is called just before expansion into rtl, allowing the target
1387 to perform additional initializations or analysis before the expansion.
1388 For example, the rs6000 port uses it to allocate a scratch stack slot
1389 for use in copying SDmode values between memory and floating point
1390 registers whenever the function being expanded has any SDmode
1394 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1395 This hook allows the backend to perform additional instantiations on rtl
1396 that are not actually in any insns yet, but will be later.
1399 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type})
1400 If your target defines any fundamental types, or any types your target
1401 uses should be mangled differently from the default, define this hook
1402 to return the appropriate encoding for these types as part of a C++
1403 mangled name. The @var{type} argument is the tree structure representing
1404 the type to be mangled. The hook may be applied to trees which are
1405 not target-specific fundamental types; it should return @code{NULL}
1406 for all such types, as well as arguments it does not recognize. If the
1407 return value is not @code{NULL}, it must point to a statically-allocated
1410 Target-specific fundamental types might be new fundamental types or
1411 qualified versions of ordinary fundamental types. Encode new
1412 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1413 is the name used for the type in source code, and @var{n} is the
1414 length of @var{name} in decimal. Encode qualified versions of
1415 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1416 @var{name} is the name used for the type qualifier in source code,
1417 @var{n} is the length of @var{name} as above, and @var{code} is the
1418 code used to represent the unqualified version of this type. (See
1419 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1420 codes.) In both cases the spaces are for clarity; do not include any
1421 spaces in your string.
1423 This hook is applied to types prior to typedef resolution. If the mangled
1424 name for a particular type depends only on that type's main variant, you
1425 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1428 The default version of this hook always returns @code{NULL}, which is
1429 appropriate for a target that does not define any new fundamental
1434 @section Layout of Source Language Data Types
1436 These macros define the sizes and other characteristics of the standard
1437 basic data types used in programs being compiled. Unlike the macros in
1438 the previous section, these apply to specific features of C and related
1439 languages, rather than to fundamental aspects of storage layout.
1441 @defmac INT_TYPE_SIZE
1442 A C expression for the size in bits of the type @code{int} on the
1443 target machine. If you don't define this, the default is one word.
1446 @defmac SHORT_TYPE_SIZE
1447 A C expression for the size in bits of the type @code{short} on the
1448 target machine. If you don't define this, the default is half a word.
1449 (If this would be less than one storage unit, it is rounded up to one
1453 @defmac LONG_TYPE_SIZE
1454 A C expression for the size in bits of the type @code{long} on the
1455 target machine. If you don't define this, the default is one word.
1458 @defmac ADA_LONG_TYPE_SIZE
1459 On some machines, the size used for the Ada equivalent of the type
1460 @code{long} by a native Ada compiler differs from that used by C@. In
1461 that situation, define this macro to be a C expression to be used for
1462 the size of that type. If you don't define this, the default is the
1463 value of @code{LONG_TYPE_SIZE}.
1466 @defmac LONG_LONG_TYPE_SIZE
1467 A C expression for the size in bits of the type @code{long long} on the
1468 target machine. If you don't define this, the default is two
1469 words. If you want to support GNU Ada on your machine, the value of this
1470 macro must be at least 64.
1473 @defmac CHAR_TYPE_SIZE
1474 A C expression for the size in bits of the type @code{char} on the
1475 target machine. If you don't define this, the default is
1476 @code{BITS_PER_UNIT}.
1479 @defmac BOOL_TYPE_SIZE
1480 A C expression for the size in bits of the C++ type @code{bool} and
1481 C99 type @code{_Bool} on the target machine. If you don't define
1482 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1485 @defmac FLOAT_TYPE_SIZE
1486 A C expression for the size in bits of the type @code{float} on the
1487 target machine. If you don't define this, the default is one word.
1490 @defmac DOUBLE_TYPE_SIZE
1491 A C expression for the size in bits of the type @code{double} on the
1492 target machine. If you don't define this, the default is two
1496 @defmac LONG_DOUBLE_TYPE_SIZE
1497 A C expression for the size in bits of the type @code{long double} on
1498 the target machine. If you don't define this, the default is two
1502 @defmac SHORT_FRACT_TYPE_SIZE
1503 A C expression for the size in bits of the type @code{short _Fract} on
1504 the target machine. If you don't define this, the default is
1505 @code{BITS_PER_UNIT}.
1508 @defmac FRACT_TYPE_SIZE
1509 A C expression for the size in bits of the type @code{_Fract} on
1510 the target machine. If you don't define this, the default is
1511 @code{BITS_PER_UNIT * 2}.
1514 @defmac LONG_FRACT_TYPE_SIZE
1515 A C expression for the size in bits of the type @code{long _Fract} on
1516 the target machine. If you don't define this, the default is
1517 @code{BITS_PER_UNIT * 4}.
1520 @defmac LONG_LONG_FRACT_TYPE_SIZE
1521 A C expression for the size in bits of the type @code{long long _Fract} on
1522 the target machine. If you don't define this, the default is
1523 @code{BITS_PER_UNIT * 8}.
1526 @defmac SHORT_ACCUM_TYPE_SIZE
1527 A C expression for the size in bits of the type @code{short _Accum} on
1528 the target machine. If you don't define this, the default is
1529 @code{BITS_PER_UNIT * 2}.
1532 @defmac ACCUM_TYPE_SIZE
1533 A C expression for the size in bits of the type @code{_Accum} on
1534 the target machine. If you don't define this, the default is
1535 @code{BITS_PER_UNIT * 4}.
1538 @defmac LONG_ACCUM_TYPE_SIZE
1539 A C expression for the size in bits of the type @code{long _Accum} on
1540 the target machine. If you don't define this, the default is
1541 @code{BITS_PER_UNIT * 8}.
1544 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1545 A C expression for the size in bits of the type @code{long long _Accum} on
1546 the target machine. If you don't define this, the default is
1547 @code{BITS_PER_UNIT * 16}.
1550 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1551 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1552 if you want routines in @file{libgcc2.a} for a size other than
1553 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1554 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1557 @defmac LIBGCC2_HAS_DF_MODE
1558 Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor
1559 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1560 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1561 anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE}
1562 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1566 @defmac LIBGCC2_HAS_XF_MODE
1567 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1568 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1569 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1570 is 80 then the default is 1, otherwise it is 0.
1573 @defmac LIBGCC2_HAS_TF_MODE
1574 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1575 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1576 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1577 is 128 then the default is 1, otherwise it is 0.
1580 @defmac LIBGCC2_GNU_PREFIX
1581 This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target
1582 hook and should be defined if that hook is overriden to be true. It
1583 causes function names in libgcc to be changed to use a @code{__gnu_}
1584 prefix for their name rather than the default @code{__}. A port which
1585 uses this macro should also arrange to use @file{t-gnu-prefix} in
1586 the libgcc @file{config.host}.
1593 Define these macros to be the size in bits of the mantissa of
1594 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1595 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1596 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1597 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1598 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1599 @code{DOUBLE_TYPE_SIZE} or
1600 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1603 @defmac TARGET_FLT_EVAL_METHOD
1604 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1605 assuming, if applicable, that the floating-point control word is in its
1606 default state. If you do not define this macro the value of
1607 @code{FLT_EVAL_METHOD} will be zero.
1610 @defmac WIDEST_HARDWARE_FP_SIZE
1611 A C expression for the size in bits of the widest floating-point format
1612 supported by the hardware. If you define this macro, you must specify a
1613 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1614 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1618 @defmac DEFAULT_SIGNED_CHAR
1619 An expression whose value is 1 or 0, according to whether the type
1620 @code{char} should be signed or unsigned by default. The user can
1621 always override this default with the options @option{-fsigned-char}
1622 and @option{-funsigned-char}.
1625 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1626 This target hook should return true if the compiler should give an
1627 @code{enum} type only as many bytes as it takes to represent the range
1628 of possible values of that type. It should return false if all
1629 @code{enum} types should be allocated like @code{int}.
1631 The default is to return false.
1635 A C expression for a string describing the name of the data type to use
1636 for size values. The typedef name @code{size_t} is defined using the
1637 contents of the string.
1639 The string can contain more than one keyword. If so, separate them with
1640 spaces, and write first any length keyword, then @code{unsigned} if
1641 appropriate, and finally @code{int}. The string must exactly match one
1642 of the data type names defined in the function
1643 @code{init_decl_processing} in the file @file{c-decl.c}. You may not
1644 omit @code{int} or change the order---that would cause the compiler to
1647 If you don't define this macro, the default is @code{"long unsigned
1651 @defmac PTRDIFF_TYPE
1652 A C expression for a string describing the name of the data type to use
1653 for the result of subtracting two pointers. The typedef name
1654 @code{ptrdiff_t} is defined using the contents of the string. See
1655 @code{SIZE_TYPE} above for more information.
1657 If you don't define this macro, the default is @code{"long int"}.
1661 A C expression for a string describing the name of the data type to use
1662 for wide characters. The typedef name @code{wchar_t} is defined using
1663 the contents of the string. See @code{SIZE_TYPE} above for more
1666 If you don't define this macro, the default is @code{"int"}.
1669 @defmac WCHAR_TYPE_SIZE
1670 A C expression for the size in bits of the data type for wide
1671 characters. This is used in @code{cpp}, which cannot make use of
1676 A C expression for a string describing the name of the data type to
1677 use for wide characters passed to @code{printf} and returned from
1678 @code{getwc}. The typedef name @code{wint_t} is defined using the
1679 contents of the string. See @code{SIZE_TYPE} above for more
1682 If you don't define this macro, the default is @code{"unsigned int"}.
1686 A C expression for a string describing the name of the data type that
1687 can represent any value of any standard or extended signed integer type.
1688 The typedef name @code{intmax_t} is defined using the contents of the
1689 string. See @code{SIZE_TYPE} above for more information.
1691 If you don't define this macro, the default is the first of
1692 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1693 much precision as @code{long long int}.
1696 @defmac UINTMAX_TYPE
1697 A C expression for a string describing the name of the data type that
1698 can represent any value of any standard or extended unsigned integer
1699 type. The typedef name @code{uintmax_t} is defined using the contents
1700 of the string. See @code{SIZE_TYPE} above for more information.
1702 If you don't define this macro, the default is the first of
1703 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1704 unsigned int"} that has as much precision as @code{long long unsigned
1708 @defmac SIG_ATOMIC_TYPE
1714 @defmacx UINT16_TYPE
1715 @defmacx UINT32_TYPE
1716 @defmacx UINT64_TYPE
1717 @defmacx INT_LEAST8_TYPE
1718 @defmacx INT_LEAST16_TYPE
1719 @defmacx INT_LEAST32_TYPE
1720 @defmacx INT_LEAST64_TYPE
1721 @defmacx UINT_LEAST8_TYPE
1722 @defmacx UINT_LEAST16_TYPE
1723 @defmacx UINT_LEAST32_TYPE
1724 @defmacx UINT_LEAST64_TYPE
1725 @defmacx INT_FAST8_TYPE
1726 @defmacx INT_FAST16_TYPE
1727 @defmacx INT_FAST32_TYPE
1728 @defmacx INT_FAST64_TYPE
1729 @defmacx UINT_FAST8_TYPE
1730 @defmacx UINT_FAST16_TYPE
1731 @defmacx UINT_FAST32_TYPE
1732 @defmacx UINT_FAST64_TYPE
1733 @defmacx INTPTR_TYPE
1734 @defmacx UINTPTR_TYPE
1735 C expressions for the standard types @code{sig_atomic_t},
1736 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1737 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1738 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1739 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1740 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1741 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1742 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1743 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1744 @code{SIZE_TYPE} above for more information.
1746 If any of these macros evaluates to a null pointer, the corresponding
1747 type is not supported; if GCC is configured to provide
1748 @code{<stdint.h>} in such a case, the header provided may not conform
1749 to C99, depending on the type in question. The defaults for all of
1750 these macros are null pointers.
1753 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1754 The C++ compiler represents a pointer-to-member-function with a struct
1761 ptrdiff_t vtable_index;
1768 The C++ compiler must use one bit to indicate whether the function that
1769 will be called through a pointer-to-member-function is virtual.
1770 Normally, we assume that the low-order bit of a function pointer must
1771 always be zero. Then, by ensuring that the vtable_index is odd, we can
1772 distinguish which variant of the union is in use. But, on some
1773 platforms function pointers can be odd, and so this doesn't work. In
1774 that case, we use the low-order bit of the @code{delta} field, and shift
1775 the remainder of the @code{delta} field to the left.
1777 GCC will automatically make the right selection about where to store
1778 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1779 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1780 set such that functions always start at even addresses, but the lowest
1781 bit of pointers to functions indicate whether the function at that
1782 address is in ARM or Thumb mode. If this is the case of your
1783 architecture, you should define this macro to
1784 @code{ptrmemfunc_vbit_in_delta}.
1786 In general, you should not have to define this macro. On architectures
1787 in which function addresses are always even, according to
1788 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1789 @code{ptrmemfunc_vbit_in_pfn}.
1792 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1793 Normally, the C++ compiler uses function pointers in vtables. This
1794 macro allows the target to change to use ``function descriptors''
1795 instead. Function descriptors are found on targets for whom a
1796 function pointer is actually a small data structure. Normally the
1797 data structure consists of the actual code address plus a data
1798 pointer to which the function's data is relative.
1800 If vtables are used, the value of this macro should be the number
1801 of words that the function descriptor occupies.
1804 @defmac TARGET_VTABLE_ENTRY_ALIGN
1805 By default, the vtable entries are void pointers, the so the alignment
1806 is the same as pointer alignment. The value of this macro specifies
1807 the alignment of the vtable entry in bits. It should be defined only
1808 when special alignment is necessary. */
1811 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1812 There are a few non-descriptor entries in the vtable at offsets below
1813 zero. If these entries must be padded (say, to preserve the alignment
1814 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1815 of words in each data entry.
1819 @section Register Usage
1820 @cindex register usage
1822 This section explains how to describe what registers the target machine
1823 has, and how (in general) they can be used.
1825 The description of which registers a specific instruction can use is
1826 done with register classes; see @ref{Register Classes}. For information
1827 on using registers to access a stack frame, see @ref{Frame Registers}.
1828 For passing values in registers, see @ref{Register Arguments}.
1829 For returning values in registers, see @ref{Scalar Return}.
1832 * Register Basics:: Number and kinds of registers.
1833 * Allocation Order:: Order in which registers are allocated.
1834 * Values in Registers:: What kinds of values each reg can hold.
1835 * Leaf Functions:: Renumbering registers for leaf functions.
1836 * Stack Registers:: Handling a register stack such as 80387.
1839 @node Register Basics
1840 @subsection Basic Characteristics of Registers
1842 @c prevent bad page break with this line
1843 Registers have various characteristics.
1845 @defmac FIRST_PSEUDO_REGISTER
1846 Number of hardware registers known to the compiler. They receive
1847 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1848 pseudo register's number really is assigned the number
1849 @code{FIRST_PSEUDO_REGISTER}.
1852 @defmac FIXED_REGISTERS
1853 @cindex fixed register
1854 An initializer that says which registers are used for fixed purposes
1855 all throughout the compiled code and are therefore not available for
1856 general allocation. These would include the stack pointer, the frame
1857 pointer (except on machines where that can be used as a general
1858 register when no frame pointer is needed), the program counter on
1859 machines where that is considered one of the addressable registers,
1860 and any other numbered register with a standard use.
1862 This information is expressed as a sequence of numbers, separated by
1863 commas and surrounded by braces. The @var{n}th number is 1 if
1864 register @var{n} is fixed, 0 otherwise.
1866 The table initialized from this macro, and the table initialized by
1867 the following one, may be overridden at run time either automatically,
1868 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1869 the user with the command options @option{-ffixed-@var{reg}},
1870 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1873 @defmac CALL_USED_REGISTERS
1874 @cindex call-used register
1875 @cindex call-clobbered register
1876 @cindex call-saved register
1877 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1878 clobbered (in general) by function calls as well as for fixed
1879 registers. This macro therefore identifies the registers that are not
1880 available for general allocation of values that must live across
1883 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1884 automatically saves it on function entry and restores it on function
1885 exit, if the register is used within the function.
1888 @defmac CALL_REALLY_USED_REGISTERS
1889 @cindex call-used register
1890 @cindex call-clobbered register
1891 @cindex call-saved register
1892 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1893 that the entire set of @code{FIXED_REGISTERS} be included.
1894 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1895 This macro is optional. If not specified, it defaults to the value
1896 of @code{CALL_USED_REGISTERS}.
1899 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1900 @cindex call-used register
1901 @cindex call-clobbered register
1902 @cindex call-saved register
1903 A C expression that is nonzero if it is not permissible to store a
1904 value of mode @var{mode} in hard register number @var{regno} across a
1905 call without some part of it being clobbered. For most machines this
1906 macro need not be defined. It is only required for machines that do not
1907 preserve the entire contents of a register across a call.
1911 @findex call_used_regs
1914 @findex reg_class_contents
1915 @deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void)
1916 This hook may conditionally modify five variables
1917 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1918 @code{reg_names}, and @code{reg_class_contents}, to take into account
1919 any dependence of these register sets on target flags. The first three
1920 of these are of type @code{char []} (interpreted as Boolean vectors).
1921 @code{global_regs} is a @code{const char *[]}, and
1922 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1923 called, @code{fixed_regs}, @code{call_used_regs},
1924 @code{reg_class_contents}, and @code{reg_names} have been initialized
1925 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1926 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1927 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1928 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1929 command options have been applied.
1931 @cindex disabling certain registers
1932 @cindex controlling register usage
1933 If the usage of an entire class of registers depends on the target
1934 flags, you may indicate this to GCC by using this macro to modify
1935 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1936 registers in the classes which should not be used by GCC@. Also define
1937 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1938 to return @code{NO_REGS} if it
1939 is called with a letter for a class that shouldn't be used.
1941 (However, if this class is not included in @code{GENERAL_REGS} and all
1942 of the insn patterns whose constraints permit this class are
1943 controlled by target switches, then GCC will automatically avoid using
1944 these registers when the target switches are opposed to them.)
1947 @defmac INCOMING_REGNO (@var{out})
1948 Define this macro if the target machine has register windows. This C
1949 expression returns the register number as seen by the called function
1950 corresponding to the register number @var{out} as seen by the calling
1951 function. Return @var{out} if register number @var{out} is not an
1955 @defmac OUTGOING_REGNO (@var{in})
1956 Define this macro if the target machine has register windows. This C
1957 expression returns the register number as seen by the calling function
1958 corresponding to the register number @var{in} as seen by the called
1959 function. Return @var{in} if register number @var{in} is not an inbound
1963 @defmac LOCAL_REGNO (@var{regno})
1964 Define this macro if the target machine has register windows. This C
1965 expression returns true if the register is call-saved but is in the
1966 register window. Unlike most call-saved registers, such registers
1967 need not be explicitly restored on function exit or during non-local
1972 If the program counter has a register number, define this as that
1973 register number. Otherwise, do not define it.
1976 @node Allocation Order
1977 @subsection Order of Allocation of Registers
1978 @cindex order of register allocation
1979 @cindex register allocation order
1981 @c prevent bad page break with this line
1982 Registers are allocated in order.
1984 @defmac REG_ALLOC_ORDER
1985 If defined, an initializer for a vector of integers, containing the
1986 numbers of hard registers in the order in which GCC should prefer
1987 to use them (from most preferred to least).
1989 If this macro is not defined, registers are used lowest numbered first
1990 (all else being equal).
1992 One use of this macro is on machines where the highest numbered
1993 registers must always be saved and the save-multiple-registers
1994 instruction supports only sequences of consecutive registers. On such
1995 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
1996 the highest numbered allocable register first.
1999 @defmac ADJUST_REG_ALLOC_ORDER
2000 A C statement (sans semicolon) to choose the order in which to allocate
2001 hard registers for pseudo-registers local to a basic block.
2003 Store the desired register order in the array @code{reg_alloc_order}.
2004 Element 0 should be the register to allocate first; element 1, the next
2005 register; and so on.
2007 The macro body should not assume anything about the contents of
2008 @code{reg_alloc_order} before execution of the macro.
2010 On most machines, it is not necessary to define this macro.
2013 @defmac HONOR_REG_ALLOC_ORDER
2014 Normally, IRA tries to estimate the costs for saving a register in the
2015 prologue and restoring it in the epilogue. This discourages it from
2016 using call-saved registers. If a machine wants to ensure that IRA
2017 allocates registers in the order given by REG_ALLOC_ORDER even if some
2018 call-saved registers appear earlier than call-used ones, this macro
2022 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2023 In some case register allocation order is not enough for the
2024 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2025 If this macro is defined, it should return a floating point value
2026 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2027 be increased by approximately the pseudo's usage frequency times the
2028 value returned by this macro. Not defining this macro is equivalent
2029 to having it always return @code{0.0}.
2031 On most machines, it is not necessary to define this macro.
2034 @node Values in Registers
2035 @subsection How Values Fit in Registers
2037 This section discusses the macros that describe which kinds of values
2038 (specifically, which machine modes) each register can hold, and how many
2039 consecutive registers are needed for a given mode.
2041 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2042 A C expression for the number of consecutive hard registers, starting
2043 at register number @var{regno}, required to hold a value of mode
2044 @var{mode}. This macro must never return zero, even if a register
2045 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2046 and/or CANNOT_CHANGE_MODE_CLASS instead.
2048 On a machine where all registers are exactly one word, a suitable
2049 definition of this macro is
2052 #define HARD_REGNO_NREGS(REGNO, MODE) \
2053 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2058 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2059 A C expression that is nonzero if a value of mode @var{mode}, stored
2060 in memory, ends with padding that causes it to take up more space than
2061 in registers starting at register number @var{regno} (as determined by
2062 multiplying GCC's notion of the size of the register when containing
2063 this mode by the number of registers returned by
2064 @code{HARD_REGNO_NREGS}). By default this is zero.
2066 For example, if a floating-point value is stored in three 32-bit
2067 registers but takes up 128 bits in memory, then this would be
2070 This macros only needs to be defined if there are cases where
2071 @code{subreg_get_info}
2072 would otherwise wrongly determine that a @code{subreg} can be
2073 represented by an offset to the register number, when in fact such a
2074 @code{subreg} would contain some of the padding not stored in
2075 registers and so not be representable.
2078 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2079 For values of @var{regno} and @var{mode} for which
2080 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2081 returning the greater number of registers required to hold the value
2082 including any padding. In the example above, the value would be four.
2085 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2086 Define this macro if the natural size of registers that hold values
2087 of mode @var{mode} is not the word size. It is a C expression that
2088 should give the natural size in bytes for the specified mode. It is
2089 used by the register allocator to try to optimize its results. This
2090 happens for example on SPARC 64-bit where the natural size of
2091 floating-point registers is still 32-bit.
2094 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2095 A C expression that is nonzero if it is permissible to store a value
2096 of mode @var{mode} in hard register number @var{regno} (or in several
2097 registers starting with that one). For a machine where all registers
2098 are equivalent, a suitable definition is
2101 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2104 You need not include code to check for the numbers of fixed registers,
2105 because the allocation mechanism considers them to be always occupied.
2107 @cindex register pairs
2108 On some machines, double-precision values must be kept in even/odd
2109 register pairs. You can implement that by defining this macro to reject
2110 odd register numbers for such modes.
2112 The minimum requirement for a mode to be OK in a register is that the
2113 @samp{mov@var{mode}} instruction pattern support moves between the
2114 register and other hard register in the same class and that moving a
2115 value into the register and back out not alter it.
2117 Since the same instruction used to move @code{word_mode} will work for
2118 all narrower integer modes, it is not necessary on any machine for
2119 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2120 you define patterns @samp{movhi}, etc., to take advantage of this. This
2121 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2122 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2125 Many machines have special registers for floating point arithmetic.
2126 Often people assume that floating point machine modes are allowed only
2127 in floating point registers. This is not true. Any registers that
2128 can hold integers can safely @emph{hold} a floating point machine
2129 mode, whether or not floating arithmetic can be done on it in those
2130 registers. Integer move instructions can be used to move the values.
2132 On some machines, though, the converse is true: fixed-point machine
2133 modes may not go in floating registers. This is true if the floating
2134 registers normalize any value stored in them, because storing a
2135 non-floating value there would garble it. In this case,
2136 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2137 floating registers. But if the floating registers do not automatically
2138 normalize, if you can store any bit pattern in one and retrieve it
2139 unchanged without a trap, then any machine mode may go in a floating
2140 register, so you can define this macro to say so.
2142 The primary significance of special floating registers is rather that
2143 they are the registers acceptable in floating point arithmetic
2144 instructions. However, this is of no concern to
2145 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2146 constraints for those instructions.
2148 On some machines, the floating registers are especially slow to access,
2149 so that it is better to store a value in a stack frame than in such a
2150 register if floating point arithmetic is not being done. As long as the
2151 floating registers are not in class @code{GENERAL_REGS}, they will not
2152 be used unless some pattern's constraint asks for one.
2155 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2156 A C expression that is nonzero if it is OK to rename a hard register
2157 @var{from} to another hard register @var{to}.
2159 One common use of this macro is to prevent renaming of a register to
2160 another register that is not saved by a prologue in an interrupt
2163 The default is always nonzero.
2166 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2167 A C expression that is nonzero if a value of mode
2168 @var{mode1} is accessible in mode @var{mode2} without copying.
2170 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2171 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2172 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2173 should be nonzero. If they differ for any @var{r}, you should define
2174 this macro to return zero unless some other mechanism ensures the
2175 accessibility of the value in a narrower mode.
2177 You should define this macro to return nonzero in as many cases as
2178 possible since doing so will allow GCC to perform better register
2182 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno})
2183 This target hook should return @code{true} if it is OK to use a hard register
2184 @var{regno} as scratch reg in peephole2.
2186 One common use of this macro is to prevent using of a register that
2187 is not saved by a prologue in an interrupt handler.
2189 The default version of this hook always returns @code{true}.
2192 @defmac AVOID_CCMODE_COPIES
2193 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2194 registers. You should only define this macro if support for copying to/from
2195 @code{CCmode} is incomplete.
2198 @node Leaf Functions
2199 @subsection Handling Leaf Functions
2201 @cindex leaf functions
2202 @cindex functions, leaf
2203 On some machines, a leaf function (i.e., one which makes no calls) can run
2204 more efficiently if it does not make its own register window. Often this
2205 means it is required to receive its arguments in the registers where they
2206 are passed by the caller, instead of the registers where they would
2209 The special treatment for leaf functions generally applies only when
2210 other conditions are met; for example, often they may use only those
2211 registers for its own variables and temporaries. We use the term ``leaf
2212 function'' to mean a function that is suitable for this special
2213 handling, so that functions with no calls are not necessarily ``leaf
2216 GCC assigns register numbers before it knows whether the function is
2217 suitable for leaf function treatment. So it needs to renumber the
2218 registers in order to output a leaf function. The following macros
2221 @defmac LEAF_REGISTERS
2222 Name of a char vector, indexed by hard register number, which
2223 contains 1 for a register that is allowable in a candidate for leaf
2226 If leaf function treatment involves renumbering the registers, then the
2227 registers marked here should be the ones before renumbering---those that
2228 GCC would ordinarily allocate. The registers which will actually be
2229 used in the assembler code, after renumbering, should not be marked with 1
2232 Define this macro only if the target machine offers a way to optimize
2233 the treatment of leaf functions.
2236 @defmac LEAF_REG_REMAP (@var{regno})
2237 A C expression whose value is the register number to which @var{regno}
2238 should be renumbered, when a function is treated as a leaf function.
2240 If @var{regno} is a register number which should not appear in a leaf
2241 function before renumbering, then the expression should yield @minus{}1, which
2242 will cause the compiler to abort.
2244 Define this macro only if the target machine offers a way to optimize the
2245 treatment of leaf functions, and registers need to be renumbered to do
2249 @findex current_function_is_leaf
2250 @findex current_function_uses_only_leaf_regs
2251 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2252 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2253 specially. They can test the C variable @code{current_function_is_leaf}
2254 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2255 set prior to local register allocation and is valid for the remaining
2256 compiler passes. They can also test the C variable
2257 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2258 functions which only use leaf registers.
2259 @code{current_function_uses_only_leaf_regs} is valid after all passes
2260 that modify the instructions have been run and is only useful if
2261 @code{LEAF_REGISTERS} is defined.
2262 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2263 @c of the next paragraph?! --mew 2feb93
2265 @node Stack Registers
2266 @subsection Registers That Form a Stack
2268 There are special features to handle computers where some of the
2269 ``registers'' form a stack. Stack registers are normally written by
2270 pushing onto the stack, and are numbered relative to the top of the
2273 Currently, GCC can only handle one group of stack-like registers, and
2274 they must be consecutively numbered. Furthermore, the existing
2275 support for stack-like registers is specific to the 80387 floating
2276 point coprocessor. If you have a new architecture that uses
2277 stack-like registers, you will need to do substantial work on
2278 @file{reg-stack.c} and write your machine description to cooperate
2279 with it, as well as defining these macros.
2282 Define this if the machine has any stack-like registers.
2285 @defmac STACK_REG_COVER_CLASS
2286 This is a cover class containing the stack registers. Define this if
2287 the machine has any stack-like registers.
2290 @defmac FIRST_STACK_REG
2291 The number of the first stack-like register. This one is the top
2295 @defmac LAST_STACK_REG
2296 The number of the last stack-like register. This one is the bottom of
2300 @node Register Classes
2301 @section Register Classes
2302 @cindex register class definitions
2303 @cindex class definitions, register
2305 On many machines, the numbered registers are not all equivalent.
2306 For example, certain registers may not be allowed for indexed addressing;
2307 certain registers may not be allowed in some instructions. These machine
2308 restrictions are described to the compiler using @dfn{register classes}.
2310 You define a number of register classes, giving each one a name and saying
2311 which of the registers belong to it. Then you can specify register classes
2312 that are allowed as operands to particular instruction patterns.
2316 In general, each register will belong to several classes. In fact, one
2317 class must be named @code{ALL_REGS} and contain all the registers. Another
2318 class must be named @code{NO_REGS} and contain no registers. Often the
2319 union of two classes will be another class; however, this is not required.
2321 @findex GENERAL_REGS
2322 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2323 terribly special about the name, but the operand constraint letters
2324 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2325 the same as @code{ALL_REGS}, just define it as a macro which expands
2328 Order the classes so that if class @var{x} is contained in class @var{y}
2329 then @var{x} has a lower class number than @var{y}.
2331 The way classes other than @code{GENERAL_REGS} are specified in operand
2332 constraints is through machine-dependent operand constraint letters.
2333 You can define such letters to correspond to various classes, then use
2334 them in operand constraints.
2336 You must define the narrowest register classes for allocatable
2337 registers, so that each class either has no subclasses, or that for
2338 some mode, the move cost between registers within the class is
2339 cheaper than moving a register in the class to or from memory
2342 You should define a class for the union of two classes whenever some
2343 instruction allows both classes. For example, if an instruction allows
2344 either a floating point (coprocessor) register or a general register for a
2345 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2346 which includes both of them. Otherwise you will get suboptimal code,
2347 or even internal compiler errors when reload cannot find a register in the
2348 class computed via @code{reg_class_subunion}.
2350 You must also specify certain redundant information about the register
2351 classes: for each class, which classes contain it and which ones are
2352 contained in it; for each pair of classes, the largest class contained
2355 When a value occupying several consecutive registers is expected in a
2356 certain class, all the registers used must belong to that class.
2357 Therefore, register classes cannot be used to enforce a requirement for
2358 a register pair to start with an even-numbered register. The way to
2359 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2361 Register classes used for input-operands of bitwise-and or shift
2362 instructions have a special requirement: each such class must have, for
2363 each fixed-point machine mode, a subclass whose registers can transfer that
2364 mode to or from memory. For example, on some machines, the operations for
2365 single-byte values (@code{QImode}) are limited to certain registers. When
2366 this is so, each register class that is used in a bitwise-and or shift
2367 instruction must have a subclass consisting of registers from which
2368 single-byte values can be loaded or stored. This is so that
2369 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2371 @deftp {Data type} {enum reg_class}
2372 An enumerated type that must be defined with all the register class names
2373 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2374 must be the last register class, followed by one more enumerated value,
2375 @code{LIM_REG_CLASSES}, which is not a register class but rather
2376 tells how many classes there are.
2378 Each register class has a number, which is the value of casting
2379 the class name to type @code{int}. The number serves as an index
2380 in many of the tables described below.
2383 @defmac N_REG_CLASSES
2384 The number of distinct register classes, defined as follows:
2387 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2391 @defmac REG_CLASS_NAMES
2392 An initializer containing the names of the register classes as C string
2393 constants. These names are used in writing some of the debugging dumps.
2396 @defmac REG_CLASS_CONTENTS
2397 An initializer containing the contents of the register classes, as integers
2398 which are bit masks. The @var{n}th integer specifies the contents of class
2399 @var{n}. The way the integer @var{mask} is interpreted is that
2400 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2402 When the machine has more than 32 registers, an integer does not suffice.
2403 Then the integers are replaced by sub-initializers, braced groupings containing
2404 several integers. Each sub-initializer must be suitable as an initializer
2405 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2406 In this situation, the first integer in each sub-initializer corresponds to
2407 registers 0 through 31, the second integer to registers 32 through 63, and
2411 @defmac REGNO_REG_CLASS (@var{regno})
2412 A C expression whose value is a register class containing hard register
2413 @var{regno}. In general there is more than one such class; choose a class
2414 which is @dfn{minimal}, meaning that no smaller class also contains the
2418 @defmac BASE_REG_CLASS
2419 A macro whose definition is the name of the class to which a valid
2420 base register must belong. A base register is one used in an address
2421 which is the register value plus a displacement.
2424 @defmac MODE_BASE_REG_CLASS (@var{mode})
2425 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2426 the selection of a base register in a mode dependent manner. If
2427 @var{mode} is VOIDmode then it should return the same value as
2428 @code{BASE_REG_CLASS}.
2431 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2432 A C expression whose value is the register class to which a valid
2433 base register must belong in order to be used in a base plus index
2434 register address. You should define this macro if base plus index
2435 addresses have different requirements than other base register uses.
2438 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2439 A C expression whose value is the register class to which a valid
2440 base register for a memory reference in mode @var{mode} to address
2441 space @var{address_space} must belong. @var{outer_code} and @var{index_code}
2442 define the context in which the base register occurs. @var{outer_code} is
2443 the code of the immediately enclosing expression (@code{MEM} for the top level
2444 of an address, @code{ADDRESS} for something that occurs in an
2445 @code{address_operand}). @var{index_code} is the code of the corresponding
2446 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2449 @defmac INDEX_REG_CLASS
2450 A macro whose definition is the name of the class to which a valid
2451 index register must belong. An index register is one used in an
2452 address where its value is either multiplied by a scale factor or
2453 added to another register (as well as added to a displacement).
2456 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2457 A C expression which is nonzero if register number @var{num} is
2458 suitable for use as a base register in operand addresses.
2461 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2462 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2463 that expression may examine the mode of the memory reference in
2464 @var{mode}. You should define this macro if the mode of the memory
2465 reference affects whether a register may be used as a base register. If
2466 you define this macro, the compiler will use it instead of
2467 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2468 addresses that appear outside a @code{MEM}, i.e., as an
2469 @code{address_operand}.
2472 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2473 A C expression which is nonzero if register number @var{num} is suitable for
2474 use as a base register in base plus index operand addresses, accessing
2475 memory in mode @var{mode}. It may be either a suitable hard register or a
2476 pseudo register that has been allocated such a hard register. You should
2477 define this macro if base plus index addresses have different requirements
2478 than other base register uses.
2480 Use of this macro is deprecated; please use the more general
2481 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2484 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2485 A C expression which is nonzero if register number @var{num} is
2486 suitable for use as a base register in operand addresses, accessing
2487 memory in mode @var{mode} in address space @var{address_space}.
2488 This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except
2489 that that expression may examine the context in which the register
2490 appears in the memory reference. @var{outer_code} is the code of the
2491 immediately enclosing expression (@code{MEM} if at the top level of the
2492 address, @code{ADDRESS} for something that occurs in an
2493 @code{address_operand}). @var{index_code} is the code of the
2494 corresponding index expression if @var{outer_code} is @code{PLUS};
2495 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2496 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2499 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2500 A C expression which is nonzero if register number @var{num} is
2501 suitable for use as an index register in operand addresses. It may be
2502 either a suitable hard register or a pseudo register that has been
2503 allocated such a hard register.
2505 The difference between an index register and a base register is that
2506 the index register may be scaled. If an address involves the sum of
2507 two registers, neither one of them scaled, then either one may be
2508 labeled the ``base'' and the other the ``index''; but whichever
2509 labeling is used must fit the machine's constraints of which registers
2510 may serve in each capacity. The compiler will try both labelings,
2511 looking for one that is valid, and will reload one or both registers
2512 only if neither labeling works.
2515 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass})
2516 A target hook that places additional preference on the register class to use when it is necessary to rename a register in class @var{rclass} to another class, or perhaps @var{NO_REGS}, if no preferred register class is found or hook @code{preferred_rename_class} is not implemented. Sometimes returning a more restrictive class makes better code. For example, on ARM, thumb-2 instructions using @code{LO_REGS} may be smaller than instructions using @code{GENERIC_REGS}. By returning @code{LO_REGS} from @code{preferred_rename_class}, code size can be reduced.
2519 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2520 A target hook that places additional restrictions on the register class
2521 to use when it is necessary to copy value @var{x} into a register in class
2522 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2523 another, smaller class.
2525 The default version of this hook always returns value of @code{rclass} argument.
2527 Sometimes returning a more restrictive class makes better code. For
2528 example, on the 68000, when @var{x} is an integer constant that is in range
2529 for a @samp{moveq} instruction, the value of this macro is always
2530 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2531 Requiring a data register guarantees that a @samp{moveq} will be used.
2533 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2534 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2535 loaded into some register class. By returning @code{NO_REGS} you can
2536 force @var{x} into a memory location. For example, rs6000 can load
2537 immediate values into general-purpose registers, but does not have an
2538 instruction for loading an immediate value into a floating-point
2539 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2540 @var{x} is a floating-point constant. If the constant can't be loaded
2541 into any kind of register, code generation will be better if
2542 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2543 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2545 If an insn has pseudos in it after register allocation, reload will go
2546 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2547 to find the best one. Returning @code{NO_REGS}, in this case, makes
2548 reload add a @code{!} in front of the constraint: the x86 back-end uses
2549 this feature to discourage usage of 387 registers when math is done in
2550 the SSE registers (and vice versa).
2553 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2554 A C expression that places additional restrictions on the register class
2555 to use when it is necessary to copy value @var{x} into a register in class
2556 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2557 another, smaller class. On many machines, the following definition is
2561 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2564 Sometimes returning a more restrictive class makes better code. For
2565 example, on the 68000, when @var{x} is an integer constant that is in range
2566 for a @samp{moveq} instruction, the value of this macro is always
2567 @code{DATA_REGS} as long as @var{class} includes the data registers.
2568 Requiring a data register guarantees that a @samp{moveq} will be used.
2570 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2571 @var{class} is if @var{x} is a legitimate constant which cannot be
2572 loaded into some register class. By returning @code{NO_REGS} you can
2573 force @var{x} into a memory location. For example, rs6000 can load
2574 immediate values into general-purpose registers, but does not have an
2575 instruction for loading an immediate value into a floating-point
2576 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2577 @var{x} is a floating-point constant. If the constant can't be loaded
2578 into any kind of register, code generation will be better if
2579 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2580 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2582 If an insn has pseudos in it after register allocation, reload will go
2583 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2584 to find the best one. Returning @code{NO_REGS}, in this case, makes
2585 reload add a @code{!} in front of the constraint: the x86 back-end uses
2586 this feature to discourage usage of 387 registers when math is done in
2587 the SSE registers (and vice versa).
2590 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2591 Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2594 The default version of this hook always returns value of @code{rclass}
2597 You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2598 reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2601 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2602 A C expression that places additional restrictions on the register class
2603 to use when it is necessary to be able to hold a value of mode
2604 @var{mode} in a reload register for which class @var{class} would
2607 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2608 there are certain modes that simply can't go in certain reload classes.
2610 The value is a register class; perhaps @var{class}, or perhaps another,
2613 Don't define this macro unless the target machine has limitations which
2614 require the macro to do something nontrivial.
2617 @deftypefn {Target Hook} reg_class_t TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, reg_class_t @var{reload_class}, enum machine_mode @var{reload_mode}, secondary_reload_info *@var{sri})
2618 Many machines have some registers that cannot be copied directly to or
2619 from memory or even from other types of registers. An example is the
2620 @samp{MQ} register, which on most machines, can only be copied to or
2621 from general registers, but not memory. Below, we shall be using the
2622 term 'intermediate register' when a move operation cannot be performed
2623 directly, but has to be done by copying the source into the intermediate
2624 register first, and then copying the intermediate register to the
2625 destination. An intermediate register always has the same mode as
2626 source and destination. Since it holds the actual value being copied,
2627 reload might apply optimizations to re-use an intermediate register
2628 and eliding the copy from the source when it can determine that the
2629 intermediate register still holds the required value.
2631 Another kind of secondary reload is required on some machines which
2632 allow copying all registers to and from memory, but require a scratch
2633 register for stores to some memory locations (e.g., those with symbolic
2634 address on the RT, and those with certain symbolic address on the SPARC
2635 when compiling PIC)@. Scratch registers need not have the same mode
2636 as the value being copied, and usually hold a different value than
2637 that being copied. Special patterns in the md file are needed to
2638 describe how the copy is performed with the help of the scratch register;
2639 these patterns also describe the number, register class(es) and mode(s)
2640 of the scratch register(s).
2642 In some cases, both an intermediate and a scratch register are required.
2644 For input reloads, this target hook is called with nonzero @var{in_p},
2645 and @var{x} is an rtx that needs to be copied to a register of class
2646 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2647 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2648 needs to be copied to rtx @var{x} in @var{reload_mode}.
2650 If copying a register of @var{reload_class} from/to @var{x} requires
2651 an intermediate register, the hook @code{secondary_reload} should
2652 return the register class required for this intermediate register.
2653 If no intermediate register is required, it should return NO_REGS.
2654 If more than one intermediate register is required, describe the one
2655 that is closest in the copy chain to the reload register.
2657 If scratch registers are needed, you also have to describe how to
2658 perform the copy from/to the reload register to/from this
2659 closest intermediate register. Or if no intermediate register is
2660 required, but still a scratch register is needed, describe the
2661 copy from/to the reload register to/from the reload operand @var{x}.
2663 You do this by setting @code{sri->icode} to the instruction code of a pattern
2664 in the md file which performs the move. Operands 0 and 1 are the output
2665 and input of this copy, respectively. Operands from operand 2 onward are
2666 for scratch operands. These scratch operands must have a mode, and a
2667 single-register-class
2668 @c [later: or memory]
2671 When an intermediate register is used, the @code{secondary_reload}
2672 hook will be called again to determine how to copy the intermediate
2673 register to/from the reload operand @var{x}, so your hook must also
2674 have code to handle the register class of the intermediate operand.
2676 @c [For later: maybe we'll allow multi-alternative reload patterns -
2677 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2678 @c and match the constraints of input and output to determine the required
2679 @c alternative. A restriction would be that constraints used to match
2680 @c against reloads registers would have to be written as register class
2681 @c constraints, or we need a new target macro / hook that tells us if an
2682 @c arbitrary constraint can match an unknown register of a given class.
2683 @c Such a macro / hook would also be useful in other places.]
2686 @var{x} might be a pseudo-register or a @code{subreg} of a
2687 pseudo-register, which could either be in a hard register or in memory.
2688 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2689 in memory and the hard register number if it is in a register.
2691 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2692 currently not supported. For the time being, you will have to continue
2693 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2695 @code{copy_cost} also uses this target hook to find out how values are
2696 copied. If you want it to include some extra cost for the need to allocate
2697 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2698 Or if two dependent moves are supposed to have a lower cost than the sum
2699 of the individual moves due to expected fortuitous scheduling and/or special
2700 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2703 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2704 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2705 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2706 These macros are obsolete, new ports should use the target hook
2707 @code{TARGET_SECONDARY_RELOAD} instead.
2709 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2710 target hook. Older ports still define these macros to indicate to the
2711 reload phase that it may
2712 need to allocate at least one register for a reload in addition to the
2713 register to contain the data. Specifically, if copying @var{x} to a
2714 register @var{class} in @var{mode} requires an intermediate register,
2715 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2716 largest register class all of whose registers can be used as
2717 intermediate registers or scratch registers.
2719 If copying a register @var{class} in @var{mode} to @var{x} requires an
2720 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2721 was supposed to be defined be defined to return the largest register
2722 class required. If the
2723 requirements for input and output reloads were the same, the macro
2724 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2727 The values returned by these macros are often @code{GENERAL_REGS}.
2728 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2729 can be directly copied to or from a register of @var{class} in
2730 @var{mode} without requiring a scratch register. Do not define this
2731 macro if it would always return @code{NO_REGS}.
2733 If a scratch register is required (either with or without an
2734 intermediate register), you were supposed to define patterns for
2735 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2736 (@pxref{Standard Names}. These patterns, which were normally
2737 implemented with a @code{define_expand}, should be similar to the
2738 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2741 These patterns need constraints for the reload register and scratch
2743 contain a single register class. If the original reload register (whose
2744 class is @var{class}) can meet the constraint given in the pattern, the
2745 value returned by these macros is used for the class of the scratch
2746 register. Otherwise, two additional reload registers are required.
2747 Their classes are obtained from the constraints in the insn pattern.
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 These macros should not be used in the case where a particular class of
2755 registers can only be copied to memory and not to another class of
2756 registers. In that case, secondary reload registers are not needed and
2757 would not be helpful. Instead, a stack location must be used to perform
2758 the copy and the @code{mov@var{m}} pattern should use memory as an
2759 intermediate storage. This case often occurs between floating-point and
2763 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2764 Certain machines have the property that some registers cannot be copied
2765 to some other registers without using memory. Define this macro on
2766 those machines to be a C expression that is nonzero if objects of mode
2767 @var{m} in registers of @var{class1} can only be copied to registers of
2768 class @var{class2} by storing a register of @var{class1} into memory
2769 and loading that memory location into a register of @var{class2}.
2771 Do not define this macro if its value would always be zero.
2774 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2775 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2776 allocates a stack slot for a memory location needed for register copies.
2777 If this macro is defined, the compiler instead uses the memory location
2778 defined by this macro.
2780 Do not define this macro if you do not define
2781 @code{SECONDARY_MEMORY_NEEDED}.
2784 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2785 When the compiler needs a secondary memory location to copy between two
2786 registers of mode @var{mode}, it normally allocates sufficient memory to
2787 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2788 load operations in a mode that many bits wide and whose class is the
2789 same as that of @var{mode}.
2791 This is right thing to do on most machines because it ensures that all
2792 bits of the register are copied and prevents accesses to the registers
2793 in a narrower mode, which some machines prohibit for floating-point
2796 However, this default behavior is not correct on some machines, such as
2797 the DEC Alpha, that store short integers in floating-point registers
2798 differently than in integer registers. On those machines, the default
2799 widening will not work correctly and you must define this macro to
2800 suppress that widening in some cases. See the file @file{alpha.h} for
2803 Do not define this macro if you do not define
2804 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2805 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2808 @deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass})
2809 A target hook which returns @code{true} if pseudos that have been assigned
2810 to registers of class @var{rclass} would likely be spilled because
2811 registers of @var{rclass} are needed for spill registers.
2813 The default version of this target hook returns @code{true} if @var{rclass}
2814 has exactly one register and @code{false} otherwise. On most machines, this
2815 default should be used. Only use this target hook to some other expression
2816 if pseudos allocated by @file{local-alloc.c} end up in memory because their
2817 hard registers were needed for spill registers. If this target hook returns
2818 @code{false} for those classes, those pseudos will only be allocated by
2819 @file{global.c}, which knows how to reallocate the pseudo to another
2820 register. If there would not be another register available for reallocation,
2821 you should not change the implementation of this target hook since
2822 the only effect of such implementation would be to slow down register
2826 @deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, enum machine_mode @var{mode})
2827 A target hook returns the maximum number of consecutive registers
2828 of class @var{rclass} needed to hold a value of mode @var{mode}.
2830 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2831 the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass},
2832 @var{mode})} target hook should be the maximum value of
2833 @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno}
2834 values in the class @var{rclass}.
2836 This target hook helps control the handling of multiple-word values
2839 The default version of this target hook returns the size of @var{mode}
2843 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2844 A C expression for the maximum number of consecutive registers
2845 of class @var{class} needed to hold a value of mode @var{mode}.
2847 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2848 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2849 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2850 @var{mode})} for all @var{regno} values in the class @var{class}.
2852 This macro helps control the handling of multiple-word values
2856 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2857 If defined, a C expression that returns nonzero for a @var{class} for which
2858 a change from mode @var{from} to mode @var{to} is invalid.
2860 For the example, loading 32-bit integer or floating-point objects into
2861 floating-point registers on the Alpha extends them to 64 bits.
2862 Therefore loading a 64-bit object and then storing it as a 32-bit object
2863 does not store the low-order 32 bits, as would be the case for a normal
2864 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2868 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2869 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2870 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2874 @node Old Constraints
2875 @section Obsolete Macros for Defining Constraints
2876 @cindex defining constraints, obsolete method
2877 @cindex constraints, defining, obsolete method
2879 Machine-specific constraints can be defined with these macros instead
2880 of the machine description constructs described in @ref{Define
2881 Constraints}. This mechanism is obsolete. New ports should not use
2882 it; old ports should convert to the new mechanism.
2884 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2885 For the constraint at the start of @var{str}, which starts with the letter
2886 @var{c}, return the length. This allows you to have register class /
2887 constant / extra constraints that are longer than a single letter;
2888 you don't need to define this macro if you can do with single-letter
2889 constraints only. The definition of this macro should use
2890 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2891 to handle specially.
2892 There are some sanity checks in genoutput.c that check the constraint lengths
2893 for the md file, so you can also use this macro to help you while you are
2894 transitioning from a byzantine single-letter-constraint scheme: when you
2895 return a negative length for a constraint you want to re-use, genoutput
2896 will complain about every instance where it is used in the md file.
2899 @defmac REG_CLASS_FROM_LETTER (@var{char})
2900 A C expression which defines the machine-dependent operand constraint
2901 letters for register classes. If @var{char} is such a letter, the
2902 value should be the register class corresponding to it. Otherwise,
2903 the value should be @code{NO_REGS}. The register letter @samp{r},
2904 corresponding to class @code{GENERAL_REGS}, will not be passed
2905 to this macro; you do not need to handle it.
2908 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2909 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2910 passed in @var{str}, so that you can use suffixes to distinguish between
2914 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2915 A C expression that defines the machine-dependent operand constraint
2916 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2917 particular ranges of integer values. If @var{c} is one of those
2918 letters, the expression should check that @var{value}, an integer, is in
2919 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2920 not one of those letters, the value should be 0 regardless of
2924 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2925 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2926 string passed in @var{str}, so that you can use suffixes to distinguish
2927 between different variants.
2930 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2931 A C expression that defines the machine-dependent operand constraint
2932 letters that specify particular ranges of @code{const_double} values
2933 (@samp{G} or @samp{H}).
2935 If @var{c} is one of those letters, the expression should check that
2936 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2937 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2938 letters, the value should be 0 regardless of @var{value}.
2940 @code{const_double} is used for all floating-point constants and for
2941 @code{DImode} fixed-point constants. A given letter can accept either
2942 or both kinds of values. It can use @code{GET_MODE} to distinguish
2943 between these kinds.
2946 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2947 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2948 string passed in @var{str}, so that you can use suffixes to distinguish
2949 between different variants.
2952 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2953 A C expression that defines the optional machine-dependent constraint
2954 letters that can be used to segregate specific types of operands, usually
2955 memory references, for the target machine. Any letter that is not
2956 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2957 @code{REG_CLASS_FROM_CONSTRAINT}
2958 may be used. Normally this macro will not be defined.
2960 If it is required for a particular target machine, it should return 1
2961 if @var{value} corresponds to the operand type represented by the
2962 constraint letter @var{c}. If @var{c} is not defined as an extra
2963 constraint, the value returned should be 0 regardless of @var{value}.
2965 For example, on the ROMP, load instructions cannot have their output
2966 in r0 if the memory reference contains a symbolic address. Constraint
2967 letter @samp{Q} is defined as representing a memory address that does
2968 @emph{not} contain a symbolic address. An alternative is specified with
2969 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2970 alternative specifies @samp{m} on the input and a register class that
2971 does not include r0 on the output.
2974 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2975 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2976 in @var{str}, so that you can use suffixes to distinguish between different
2980 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2981 A C expression that defines the optional machine-dependent constraint
2982 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2983 be treated like memory constraints by the reload pass.
2985 It should return 1 if the operand type represented by the constraint
2986 at the start of @var{str}, the first letter of which is the letter @var{c},
2987 comprises a subset of all memory references including
2988 all those whose address is simply a base register. This allows the reload
2989 pass to reload an operand, if it does not directly correspond to the operand
2990 type of @var{c}, by copying its address into a base register.
2992 For example, on the S/390, some instructions do not accept arbitrary
2993 memory references, but only those that do not make use of an index
2994 register. The constraint letter @samp{Q} is defined via
2995 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
2996 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
2997 a @samp{Q} constraint can handle any memory operand, because the
2998 reload pass knows it can be reloaded by copying the memory address
2999 into a base register if required. This is analogous to the way
3000 an @samp{o} constraint can handle any memory operand.
3003 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
3004 A C expression that defines the optional machine-dependent constraint
3005 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
3006 @code{EXTRA_CONSTRAINT_STR}, that should
3007 be treated like address constraints by the reload pass.
3009 It should return 1 if the operand type represented by the constraint
3010 at the start of @var{str}, which starts with the letter @var{c}, comprises
3011 a subset of all memory addresses including
3012 all those that consist of just a base register. This allows the reload
3013 pass to reload an operand, if it does not directly correspond to the operand
3014 type of @var{str}, by copying it into a base register.
3016 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3017 be used with the @code{address_operand} predicate. It is treated
3018 analogously to the @samp{p} constraint.
3021 @node Stack and Calling
3022 @section Stack Layout and Calling Conventions
3023 @cindex calling conventions
3025 @c prevent bad page break with this line
3026 This describes the stack layout and calling conventions.
3030 * Exception Handling::
3035 * Register Arguments::
3037 * Aggregate Return::
3042 * Stack Smashing Protection::
3046 @subsection Basic Stack Layout
3047 @cindex stack frame layout
3048 @cindex frame layout
3050 @c prevent bad page break with this line
3051 Here is the basic stack layout.
3053 @defmac STACK_GROWS_DOWNWARD
3054 Define this macro if pushing a word onto the stack moves the stack
3055 pointer to a smaller address.
3057 When we say, ``define this macro if @dots{}'', it means that the
3058 compiler checks this macro only with @code{#ifdef} so the precise
3059 definition used does not matter.
3062 @defmac STACK_PUSH_CODE
3063 This macro defines the operation used when something is pushed
3064 on the stack. In RTL, a push operation will be
3065 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3067 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3068 and @code{POST_INC}. Which of these is correct depends on
3069 the stack direction and on whether the stack pointer points
3070 to the last item on the stack or whether it points to the
3071 space for the next item on the stack.
3073 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3074 defined, which is almost always right, and @code{PRE_INC} otherwise,
3075 which is often wrong.
3078 @defmac FRAME_GROWS_DOWNWARD
3079 Define this macro to nonzero value if the addresses of local variable slots
3080 are at negative offsets from the frame pointer.
3083 @defmac ARGS_GROW_DOWNWARD
3084 Define this macro if successive arguments to a function occupy decreasing
3085 addresses on the stack.
3088 @defmac STARTING_FRAME_OFFSET
3089 Offset from the frame pointer to the first local variable slot to be allocated.
3091 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3092 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3093 Otherwise, it is found by adding the length of the first slot to the
3094 value @code{STARTING_FRAME_OFFSET}.
3095 @c i'm not sure if the above is still correct.. had to change it to get
3096 @c rid of an overfull. --mew 2feb93
3099 @defmac STACK_ALIGNMENT_NEEDED
3100 Define to zero to disable final alignment of the stack during reload.
3101 The nonzero default for this macro is suitable for most ports.
3103 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3104 is a register save block following the local block that doesn't require
3105 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3106 stack alignment and do it in the backend.
3109 @defmac STACK_POINTER_OFFSET
3110 Offset from the stack pointer register to the first location at which
3111 outgoing arguments are placed. If not specified, the default value of
3112 zero is used. This is the proper value for most machines.
3114 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3115 the first location at which outgoing arguments are placed.
3118 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3119 Offset from the argument pointer register to the first argument's
3120 address. On some machines it may depend on the data type of the
3123 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3124 the first argument's address.
3127 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3128 Offset from the stack pointer register to an item dynamically allocated
3129 on the stack, e.g., by @code{alloca}.
3131 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3132 length of the outgoing arguments. The default is correct for most
3133 machines. See @file{function.c} for details.
3136 @defmac INITIAL_FRAME_ADDRESS_RTX
3137 A C expression whose value is RTL representing the address of the initial
3138 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3139 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3140 default value will be used. Define this macro in order to make frame pointer
3141 elimination work in the presence of @code{__builtin_frame_address (count)} and
3142 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3145 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3146 A C expression whose value is RTL representing the address in a stack
3147 frame where the pointer to the caller's frame is stored. Assume that
3148 @var{frameaddr} is an RTL expression for the address of the stack frame
3151 If you don't define this macro, the default is to return the value
3152 of @var{frameaddr}---that is, the stack frame address is also the
3153 address of the stack word that points to the previous frame.
3156 @defmac SETUP_FRAME_ADDRESSES
3157 If defined, a C expression that produces the machine-specific code to
3158 setup the stack so that arbitrary frames can be accessed. For example,
3159 on the SPARC, we must flush all of the register windows to the stack
3160 before we can access arbitrary stack frames. You will seldom need to
3164 @deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)
3165 This target hook should return an rtx that is used to store
3166 the address of the current frame into the built in @code{setjmp} buffer.
3167 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3168 machines. One reason you may need to define this target hook is if
3169 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3172 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3173 A C expression whose value is RTL representing the value of the frame
3174 address for the current frame. @var{frameaddr} is the frame pointer
3175 of the current frame. This is used for __builtin_frame_address.
3176 You need only define this macro if the frame address is not the same
3177 as the frame pointer. Most machines do not need to define it.
3180 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3181 A C expression whose value is RTL representing the value of the return
3182 address for the frame @var{count} steps up from the current frame, after
3183 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3184 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3185 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3187 The value of the expression must always be the correct address when
3188 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3189 determine the return address of other frames.
3192 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3193 Define this if the return address of a particular stack frame is accessed
3194 from the frame pointer of the previous stack frame.
3197 @defmac INCOMING_RETURN_ADDR_RTX
3198 A C expression whose value is RTL representing the location of the
3199 incoming return address at the beginning of any function, before the
3200 prologue. This RTL is either a @code{REG}, indicating that the return
3201 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3204 You only need to define this macro if you want to support call frame
3205 debugging information like that provided by DWARF 2.
3207 If this RTL is a @code{REG}, you should also define
3208 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3211 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3212 A C expression whose value is an integer giving a DWARF 2 column
3213 number that may be used as an alternative return column. The column
3214 must not correspond to any gcc hard register (that is, it must not
3215 be in the range of @code{DWARF_FRAME_REGNUM}).
3217 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3218 general register, but an alternative column needs to be used for signal
3219 frames. Some targets have also used different frame return columns
3223 @defmac DWARF_ZERO_REG
3224 A C expression whose value is an integer giving a DWARF 2 register
3225 number that is considered to always have the value zero. This should
3226 only be defined if the target has an architected zero register, and
3227 someone decided it was a good idea to use that register number to
3228 terminate the stack backtrace. New ports should avoid this.
3231 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3232 This target hook allows the backend to emit frame-related insns that
3233 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3234 info engine will invoke it on insns of the form
3236 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3240 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3242 to let the backend emit the call frame instructions. @var{label} is
3243 the CFI label attached to the insn, @var{pattern} is the pattern of
3244 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3247 @defmac INCOMING_FRAME_SP_OFFSET
3248 A C expression whose value is an integer giving the offset, in bytes,
3249 from the value of the stack pointer register to the top of the stack
3250 frame at the beginning of any function, before the prologue. The top of
3251 the frame is defined to be the value of the stack pointer in the
3252 previous frame, just before the call instruction.
3254 You only need to define this macro if you want to support call frame
3255 debugging information like that provided by DWARF 2.
3258 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3259 A C expression whose value is an integer giving the offset, in bytes,
3260 from the argument pointer to the canonical frame address (cfa). The
3261 final value should coincide with that calculated by
3262 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3263 during virtual register instantiation.
3265 The default value for this macro is
3266 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3267 which is correct for most machines; in general, the arguments are found
3268 immediately before the stack frame. Note that this is not the case on
3269 some targets that save registers into the caller's frame, such as SPARC
3270 and rs6000, and so such targets need to define this macro.
3272 You only need to define this macro if the default is incorrect, and you
3273 want to support call frame debugging information like that provided by
3277 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3278 If defined, a C expression whose value is an integer giving the offset
3279 in bytes from the frame pointer to the canonical frame address (cfa).
3280 The final value should coincide with that calculated by
3281 @code{INCOMING_FRAME_SP_OFFSET}.
3283 Normally the CFA is calculated as an offset from the argument pointer,
3284 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3285 variable due to the ABI, this may not be possible. If this macro is
3286 defined, it implies that the virtual register instantiation should be
3287 based on the frame pointer instead of the argument pointer. Only one
3288 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3292 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3293 If defined, a C expression whose value is an integer giving the offset
3294 in bytes from the canonical frame address (cfa) to the frame base used
3295 in DWARF 2 debug information. The default is zero. A different value
3296 may reduce the size of debug information on some ports.
3299 @node Exception Handling
3300 @subsection Exception Handling Support
3301 @cindex exception handling
3303 @defmac EH_RETURN_DATA_REGNO (@var{N})
3304 A C expression whose value is the @var{N}th register number used for
3305 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3306 @var{N} registers are usable.
3308 The exception handling library routines communicate with the exception
3309 handlers via a set of agreed upon registers. Ideally these registers
3310 should be call-clobbered; it is possible to use call-saved registers,
3311 but may negatively impact code size. The target must support at least
3312 2 data registers, but should define 4 if there are enough free registers.
3314 You must define this macro if you want to support call frame exception
3315 handling like that provided by DWARF 2.
3318 @defmac EH_RETURN_STACKADJ_RTX
3319 A C expression whose value is RTL representing a location in which
3320 to store a stack adjustment to be applied before function return.
3321 This is used to unwind the stack to an exception handler's call frame.
3322 It will be assigned zero on code paths that return normally.
3324 Typically this is a call-clobbered hard register that is otherwise
3325 untouched by the epilogue, but could also be a stack slot.
3327 Do not define this macro if the stack pointer is saved and restored
3328 by the regular prolog and epilog code in the call frame itself; in
3329 this case, the exception handling library routines will update the
3330 stack location to be restored in place. Otherwise, you must define
3331 this macro if you want to support call frame exception handling like
3332 that provided by DWARF 2.
3335 @defmac EH_RETURN_HANDLER_RTX
3336 A C expression whose value is RTL representing a location in which
3337 to store the address of an exception handler to which we should
3338 return. It will not be assigned on code paths that return normally.
3340 Typically this is the location in the call frame at which the normal
3341 return address is stored. For targets that return by popping an
3342 address off the stack, this might be a memory address just below
3343 the @emph{target} call frame rather than inside the current call
3344 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3345 been assigned, so it may be used to calculate the location of the
3348 Some targets have more complex requirements than storing to an
3349 address calculable during initial code generation. In that case
3350 the @code{eh_return} instruction pattern should be used instead.
3352 If you want to support call frame exception handling, you must
3353 define either this macro or the @code{eh_return} instruction pattern.
3356 @defmac RETURN_ADDR_OFFSET
3357 If defined, an integer-valued C expression for which rtl will be generated
3358 to add it to the exception handler address before it is searched in the
3359 exception handling tables, and to subtract it again from the address before
3360 using it to return to the exception handler.
3363 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3364 This macro chooses the encoding of pointers embedded in the exception
3365 handling sections. If at all possible, this should be defined such
3366 that the exception handling section will not require dynamic relocations,
3367 and so may be read-only.
3369 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3370 @var{global} is true if the symbol may be affected by dynamic relocations.
3371 The macro should return a combination of the @code{DW_EH_PE_*} defines
3372 as found in @file{dwarf2.h}.
3374 If this macro is not defined, pointers will not be encoded but
3375 represented directly.
3378 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3379 This macro allows the target to emit whatever special magic is required
3380 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3381 Generic code takes care of pc-relative and indirect encodings; this must
3382 be defined if the target uses text-relative or data-relative encodings.
3384 This is a C statement that branches to @var{done} if the format was
3385 handled. @var{encoding} is the format chosen, @var{size} is the number
3386 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3390 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3391 This macro allows the target to add CPU and operating system specific
3392 code to the call-frame unwinder for use when there is no unwind data
3393 available. The most common reason to implement this macro is to unwind
3394 through signal frames.
3396 This macro is called from @code{uw_frame_state_for} in
3397 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3398 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3399 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3400 for the address of the code being executed and @code{context->cfa} for
3401 the stack pointer value. If the frame can be decoded, the register
3402 save addresses should be updated in @var{fs} and the macro should
3403 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3404 the macro should evaluate to @code{_URC_END_OF_STACK}.
3406 For proper signal handling in Java this macro is accompanied by
3407 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3410 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3411 This macro allows the target to add operating system specific code to the
3412 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3413 usually used for signal or interrupt frames.
3415 This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}.
3416 @var{context} is an @code{_Unwind_Context};
3417 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3418 for the abi and context in the @code{.unwabi} directive. If the
3419 @code{.unwabi} directive can be handled, the register save addresses should
3420 be updated in @var{fs}.
3423 @defmac TARGET_USES_WEAK_UNWIND_INFO
3424 A C expression that evaluates to true if the target requires unwind
3425 info to be given comdat linkage. Define it to be @code{1} if comdat
3426 linkage is necessary. The default is @code{0}.
3429 @node Stack Checking
3430 @subsection Specifying How Stack Checking is Done
3432 GCC will check that stack references are within the boundaries of the
3433 stack, if the option @option{-fstack-check} is specified, in one of
3438 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3439 will assume that you have arranged for full stack checking to be done
3440 at appropriate places in the configuration files. GCC will not do
3441 other special processing.
3444 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3445 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3446 that you have arranged for static stack checking (checking of the
3447 static stack frame of functions) to be done at appropriate places
3448 in the configuration files. GCC will only emit code to do dynamic
3449 stack checking (checking on dynamic stack allocations) using the third
3453 If neither of the above are true, GCC will generate code to periodically
3454 ``probe'' the stack pointer using the values of the macros defined below.
3457 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3458 GCC will change its allocation strategy for large objects if the option
3459 @option{-fstack-check} is specified: they will always be allocated
3460 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3462 @defmac STACK_CHECK_BUILTIN
3463 A nonzero value if stack checking is done by the configuration files in a
3464 machine-dependent manner. You should define this macro if stack checking
3465 is required by the ABI of your machine or if you would like to do stack
3466 checking in some more efficient way than the generic approach. The default
3467 value of this macro is zero.
3470 @defmac STACK_CHECK_STATIC_BUILTIN
3471 A nonzero value if static stack checking is done by the configuration files
3472 in a machine-dependent manner. You should define this macro if you would
3473 like to do static stack checking in some more efficient way than the generic
3474 approach. The default value of this macro is zero.
3477 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3478 An integer specifying the interval at which GCC must generate stack probe
3479 instructions, defined as 2 raised to this integer. You will normally
3480 define this macro so that the interval be no larger than the size of
3481 the ``guard pages'' at the end of a stack area. The default value
3482 of 12 (4096-byte interval) is suitable for most systems.
3485 @defmac STACK_CHECK_MOVING_SP
3486 An integer which is nonzero if GCC should move the stack pointer page by page
3487 when doing probes. This can be necessary on systems where the stack pointer
3488 contains the bottom address of the memory area accessible to the executing
3489 thread at any point in time. In this situation an alternate signal stack
3490 is required in order to be able to recover from a stack overflow. The
3491 default value of this macro is zero.
3494 @defmac STACK_CHECK_PROTECT
3495 The number of bytes of stack needed to recover from a stack overflow, for
3496 languages where such a recovery is supported. The default value of 75 words
3497 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3498 8192 bytes with other exception handling mechanisms should be adequate for
3502 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3503 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3504 in the opposite case.
3506 @defmac STACK_CHECK_MAX_FRAME_SIZE
3507 The maximum size of a stack frame, in bytes. GCC will generate probe
3508 instructions in non-leaf functions to ensure at least this many bytes of
3509 stack are available. If a stack frame is larger than this size, stack
3510 checking will not be reliable and GCC will issue a warning. The
3511 default is chosen so that GCC only generates one instruction on most
3512 systems. You should normally not change the default value of this macro.
3515 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3516 GCC uses this value to generate the above warning message. It
3517 represents the amount of fixed frame used by a function, not including
3518 space for any callee-saved registers, temporaries and user variables.
3519 You need only specify an upper bound for this amount and will normally
3520 use the default of four words.
3523 @defmac STACK_CHECK_MAX_VAR_SIZE
3524 The maximum size, in bytes, of an object that GCC will place in the
3525 fixed area of the stack frame when the user specifies
3526 @option{-fstack-check}.
3527 GCC computed the default from the values of the above macros and you will
3528 normally not need to override that default.
3532 @node Frame Registers
3533 @subsection Registers That Address the Stack Frame
3535 @c prevent bad page break with this line
3536 This discusses registers that address the stack frame.
3538 @defmac STACK_POINTER_REGNUM
3539 The register number of the stack pointer register, which must also be a
3540 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3541 the hardware determines which register this is.
3544 @defmac FRAME_POINTER_REGNUM
3545 The register number of the frame pointer register, which is used to
3546 access automatic variables in the stack frame. On some machines, the
3547 hardware determines which register this is. On other machines, you can
3548 choose any register you wish for this purpose.
3551 @defmac HARD_FRAME_POINTER_REGNUM
3552 On some machines the offset between the frame pointer and starting
3553 offset of the automatic variables is not known until after register
3554 allocation has been done (for example, because the saved registers are
3555 between these two locations). On those machines, define
3556 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3557 be used internally until the offset is known, and define
3558 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3559 used for the frame pointer.
3561 You should define this macro only in the very rare circumstances when it
3562 is not possible to calculate the offset between the frame pointer and
3563 the automatic variables until after register allocation has been
3564 completed. When this macro is defined, you must also indicate in your
3565 definition of @code{ELIMINABLE_REGS} how to eliminate
3566 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3567 or @code{STACK_POINTER_REGNUM}.
3569 Do not define this macro if it would be the same as
3570 @code{FRAME_POINTER_REGNUM}.
3573 @defmac ARG_POINTER_REGNUM
3574 The register number of the arg pointer register, which is used to access
3575 the function's argument list. On some machines, this is the same as the
3576 frame pointer register. On some machines, the hardware determines which
3577 register this is. On other machines, you can choose any register you
3578 wish for this purpose. If this is not the same register as the frame
3579 pointer register, then you must mark it as a fixed register according to
3580 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3581 (@pxref{Elimination}).
3584 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3585 Define this to a preprocessor constant that is nonzero if
3586 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3587 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3588 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3589 definition is not suitable for use in preprocessor conditionals.
3592 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3593 Define this to a preprocessor constant that is nonzero if
3594 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3595 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3596 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3597 definition is not suitable for use in preprocessor conditionals.
3600 @defmac RETURN_ADDRESS_POINTER_REGNUM
3601 The register number of the return address pointer register, which is used to
3602 access the current function's return address from the stack. On some
3603 machines, the return address is not at a fixed offset from the frame
3604 pointer or stack pointer or argument pointer. This register can be defined
3605 to point to the return address on the stack, and then be converted by
3606 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3608 Do not define this macro unless there is no other way to get the return
3609 address from the stack.
3612 @defmac STATIC_CHAIN_REGNUM
3613 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3614 Register numbers used for passing a function's static chain pointer. If
3615 register windows are used, the register number as seen by the called
3616 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3617 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3618 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3621 The static chain register need not be a fixed register.
3623 If the static chain is passed in memory, these macros should not be
3624 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3627 @deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl}, bool @var{incoming_p})
3628 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3629 targets that may use different static chain locations for different
3630 nested functions. This may be required if the target has function
3631 attributes that affect the calling conventions of the function and
3632 those calling conventions use different static chain locations.
3634 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3636 If the static chain is passed in memory, this hook should be used to
3637 provide rtx giving @code{mem} expressions that denote where they are stored.
3638 Often the @code{mem} expression as seen by the caller will be at an offset
3639 from the stack pointer and the @code{mem} expression as seen by the callee
3640 will be at an offset from the frame pointer.
3641 @findex stack_pointer_rtx
3642 @findex frame_pointer_rtx
3643 @findex arg_pointer_rtx
3644 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3645 @code{arg_pointer_rtx} will have been initialized and should be used
3646 to refer to those items.
3649 @defmac DWARF_FRAME_REGISTERS
3650 This macro specifies the maximum number of hard registers that can be
3651 saved in a call frame. This is used to size data structures used in
3652 DWARF2 exception handling.
3654 Prior to GCC 3.0, this macro was needed in order to establish a stable
3655 exception handling ABI in the face of adding new hard registers for ISA
3656 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3657 in the number of hard registers. Nevertheless, this macro can still be
3658 used to reduce the runtime memory requirements of the exception handling
3659 routines, which can be substantial if the ISA contains a lot of
3660 registers that are not call-saved.
3662 If this macro is not defined, it defaults to
3663 @code{FIRST_PSEUDO_REGISTER}.
3666 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3668 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3669 for backward compatibility in pre GCC 3.0 compiled code.
3671 If this macro is not defined, it defaults to
3672 @code{DWARF_FRAME_REGISTERS}.
3675 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3677 Define this macro if the target's representation for dwarf registers
3678 is different than the internal representation for unwind column.
3679 Given a dwarf register, this macro should return the internal unwind
3680 column number to use instead.
3682 See the PowerPC's SPE target for an example.
3685 @defmac DWARF_FRAME_REGNUM (@var{regno})
3687 Define this macro if the target's representation for dwarf registers
3688 used in .eh_frame or .debug_frame is different from that used in other
3689 debug info sections. Given a GCC hard register number, this macro
3690 should return the .eh_frame register number. The default is
3691 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3695 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3697 Define this macro to map register numbers held in the call frame info
3698 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3699 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3700 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3701 return @code{@var{regno}}.
3705 @defmac REG_VALUE_IN_UNWIND_CONTEXT
3707 Define this macro if the target stores register values as
3708 @code{_Unwind_Word} type in unwind context. It should be defined if
3709 target register size is larger than the size of @code{void *}. The
3710 default is to store register values as @code{void *} type.
3714 @defmac ASSUME_EXTENDED_UNWIND_CONTEXT
3716 Define this macro to be 1 if the target always uses extended unwind
3717 context with version, args_size and by_value fields. If it is undefined,
3718 it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is
3719 defined and 0 otherwise.
3724 @subsection Eliminating Frame Pointer and Arg Pointer
3726 @c prevent bad page break with this line
3727 This is about eliminating the frame pointer and arg pointer.
3729 @deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void)
3730 This target hook should return @code{true} if a function must have and use
3731 a frame pointer. This target hook is called in the reload pass. If its return
3732 value is @code{true} the function will have a frame pointer.
3734 This target hook can in principle examine the current function and decide
3735 according to the facts, but on most machines the constant @code{false} or the
3736 constant @code{true} suffices. Use @code{false} when the machine allows code
3737 to be generated with no frame pointer, and doing so saves some time or space.
3738 Use @code{true} when there is no possible advantage to avoiding a frame
3741 In certain cases, the compiler does not know how to produce valid code
3742 without a frame pointer. The compiler recognizes those cases and
3743 automatically gives the function a frame pointer regardless of what
3744 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3747 In a function that does not require a frame pointer, the frame pointer
3748 register can be allocated for ordinary usage, unless you mark it as a
3749 fixed register. See @code{FIXED_REGISTERS} for more information.
3751 Default return value is @code{false}.
3754 @findex get_frame_size
3755 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3756 A C statement to store in the variable @var{depth-var} the difference
3757 between the frame pointer and the stack pointer values immediately after
3758 the function prologue. The value would be computed from information
3759 such as the result of @code{get_frame_size ()} and the tables of
3760 registers @code{regs_ever_live} and @code{call_used_regs}.
3762 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3763 need not be defined. Otherwise, it must be defined even if
3764 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3765 case, you may set @var{depth-var} to anything.
3768 @defmac ELIMINABLE_REGS
3769 If defined, this macro specifies a table of register pairs used to
3770 eliminate unneeded registers that point into the stack frame. If it is not
3771 defined, the only elimination attempted by the compiler is to replace
3772 references to the frame pointer with references to the stack pointer.
3774 The definition of this macro is a list of structure initializations, each
3775 of which specifies an original and replacement register.
3777 On some machines, the position of the argument pointer is not known until
3778 the compilation is completed. In such a case, a separate hard register
3779 must be used for the argument pointer. This register can be eliminated by
3780 replacing it with either the frame pointer or the argument pointer,
3781 depending on whether or not the frame pointer has been eliminated.
3783 In this case, you might specify:
3785 #define ELIMINABLE_REGS \
3786 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3787 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3788 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3791 Note that the elimination of the argument pointer with the stack pointer is
3792 specified first since that is the preferred elimination.
3795 @deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg})
3796 This target hook should returns @code{true} if the compiler is allowed to
3797 try to replace register number @var{from_reg} with register number
3798 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3799 is defined, and will usually be @code{true}, since most of the cases
3800 preventing register elimination are things that the compiler already
3803 Default return value is @code{true}.
3806 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3807 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3808 specifies the initial difference between the specified pair of
3809 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3813 @node Stack Arguments
3814 @subsection Passing Function Arguments on the Stack
3815 @cindex arguments on stack
3816 @cindex stack arguments
3818 The macros in this section control how arguments are passed
3819 on the stack. See the following section for other macros that
3820 control passing certain arguments in registers.
3822 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype})
3823 This target hook returns @code{true} if an argument declared in a
3824 prototype as an integral type smaller than @code{int} should actually be
3825 passed as an @code{int}. In addition to avoiding errors in certain
3826 cases of mismatch, it also makes for better code on certain machines.
3827 The default is to not promote prototypes.
3831 A C expression. If nonzero, push insns will be used to pass
3833 If the target machine does not have a push instruction, set it to zero.
3834 That directs GCC to use an alternate strategy: to
3835 allocate the entire argument block and then store the arguments into
3836 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3839 @defmac PUSH_ARGS_REVERSED
3840 A C expression. If nonzero, function arguments will be evaluated from
3841 last to first, rather than from first to last. If this macro is not
3842 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3843 and args grow in opposite directions, and 0 otherwise.
3846 @defmac PUSH_ROUNDING (@var{npushed})
3847 A C expression that is the number of bytes actually pushed onto the
3848 stack when an instruction attempts to push @var{npushed} bytes.
3850 On some machines, the definition
3853 #define PUSH_ROUNDING(BYTES) (BYTES)
3857 will suffice. But on other machines, instructions that appear
3858 to push one byte actually push two bytes in an attempt to maintain
3859 alignment. Then the definition should be
3862 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3865 If the value of this macro has a type, it should be an unsigned type.
3868 @findex current_function_outgoing_args_size
3869 @defmac ACCUMULATE_OUTGOING_ARGS
3870 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3871 will be computed and placed into the variable
3872 @code{current_function_outgoing_args_size}. No space will be pushed
3873 onto the stack for each call; instead, the function prologue should
3874 increase the stack frame size by this amount.
3876 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3880 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3881 Define this macro if functions should assume that stack space has been
3882 allocated for arguments even when their values are passed in
3885 The value of this macro is the size, in bytes, of the area reserved for
3886 arguments passed in registers for the function represented by @var{fndecl},
3887 which can be zero if GCC is calling a library function.
3888 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3891 This space can be allocated by the caller, or be a part of the
3892 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3895 @c above is overfull. not sure what to do. --mew 5feb93 did
3896 @c something, not sure if it looks good. --mew 10feb93
3898 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3899 Define this to a nonzero value if it is the responsibility of the
3900 caller to allocate the area reserved for arguments passed in registers
3901 when calling a function of @var{fntype}. @var{fntype} may be NULL
3902 if the function called is a library function.
3904 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3905 whether the space for these arguments counts in the value of
3906 @code{current_function_outgoing_args_size}.
3909 @defmac STACK_PARMS_IN_REG_PARM_AREA
3910 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3911 stack parameters don't skip the area specified by it.
3912 @c i changed this, makes more sens and it should have taken care of the
3913 @c overfull.. not as specific, tho. --mew 5feb93
3915 Normally, when a parameter is not passed in registers, it is placed on the
3916 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3917 suppresses this behavior and causes the parameter to be passed on the
3918 stack in its natural location.
3921 @deftypefn {Target Hook} int TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, int @var{size})
3922 This target hook returns the number of bytes of its own arguments that
3923 a function pops on returning, or 0 if the function pops no arguments
3924 and the caller must therefore pop them all after the function returns.
3926 @var{fundecl} is a C variable whose value is a tree node that describes
3927 the function in question. Normally it is a node of type
3928 @code{FUNCTION_DECL} that describes the declaration of the function.
3929 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3931 @var{funtype} is a C variable whose value is a tree node that
3932 describes the function in question. Normally it is a node of type
3933 @code{FUNCTION_TYPE} that describes the data type of the function.
3934 From this it is possible to obtain the data types of the value and
3935 arguments (if known).
3937 When a call to a library function is being considered, @var{fundecl}
3938 will contain an identifier node for the library function. Thus, if
3939 you need to distinguish among various library functions, you can do so
3940 by their names. Note that ``library function'' in this context means
3941 a function used to perform arithmetic, whose name is known specially
3942 in the compiler and was not mentioned in the C code being compiled.
3944 @var{size} is the number of bytes of arguments passed on the
3945 stack. If a variable number of bytes is passed, it is zero, and
3946 argument popping will always be the responsibility of the calling function.
3948 On the VAX, all functions always pop their arguments, so the definition
3949 of this macro is @var{size}. On the 68000, using the standard
3950 calling convention, no functions pop their arguments, so the value of
3951 the macro is always 0 in this case. But an alternative calling
3952 convention is available in which functions that take a fixed number of
3953 arguments pop them but other functions (such as @code{printf}) pop
3954 nothing (the caller pops all). When this convention is in use,
3955 @var{funtype} is examined to determine whether a function takes a fixed
3956 number of arguments.
3959 @defmac CALL_POPS_ARGS (@var{cum})
3960 A C expression that should indicate the number of bytes a call sequence
3961 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3962 when compiling a function call.
3964 @var{cum} is the variable in which all arguments to the called function
3965 have been accumulated.
3967 On certain architectures, such as the SH5, a call trampoline is used
3968 that pops certain registers off the stack, depending on the arguments
3969 that have been passed to the function. Since this is a property of the
3970 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3974 @node Register Arguments
3975 @subsection Passing Arguments in Registers
3976 @cindex arguments in registers
3977 @cindex registers arguments
3979 This section describes the macros which let you control how various
3980 types of arguments are passed in registers or how they are arranged in
3983 @deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3984 Return an RTX indicating whether a function argument is passed in a
3985 register and if so, which register.
3987 The arguments are @var{ca}, which summarizes all the previous
3988 arguments; @var{mode}, the machine mode of the argument; @var{type},
3989 the data type of the argument as a tree node or 0 if that is not known
3990 (which happens for C support library functions); and @var{named},
3991 which is @code{true} for an ordinary argument and @code{false} for
3992 nameless arguments that correspond to @samp{@dots{}} in the called
3993 function's prototype. @var{type} can be an incomplete type if a
3994 syntax error has previously occurred.
3996 The return value is usually either a @code{reg} RTX for the hard
3997 register in which to pass the argument, or zero to pass the argument
4000 The value of the expression can also be a @code{parallel} RTX@. This is
4001 used when an argument is passed in multiple locations. The mode of the
4002 @code{parallel} should be the mode of the entire argument. The
4003 @code{parallel} holds any number of @code{expr_list} pairs; each one
4004 describes where part of the argument is passed. In each
4005 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
4006 register in which to pass this part of the argument, and the mode of the
4007 register RTX indicates how large this part of the argument is. The
4008 second operand of the @code{expr_list} is a @code{const_int} which gives
4009 the offset in bytes into the entire argument of where this part starts.
4010 As a special exception the first @code{expr_list} in the @code{parallel}
4011 RTX may have a first operand of zero. This indicates that the entire
4012 argument is also stored on the stack.
4014 The last time this hook is called, it is called with @code{MODE ==
4015 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4016 pattern as operands 2 and 3 respectively.
4018 @cindex @file{stdarg.h} and register arguments
4019 The usual way to make the ISO library @file{stdarg.h} work on a
4020 machine where some arguments are usually passed in registers, is to
4021 cause nameless arguments to be passed on the stack instead. This is
4022 done by making @code{TARGET_FUNCTION_ARG} return 0 whenever
4023 @var{named} is @code{false}.
4025 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG}
4026 @cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG}
4027 You may use the hook @code{targetm.calls.must_pass_in_stack}
4028 in the definition of this macro to determine if this argument is of a
4029 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4030 is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an
4031 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4032 defined, the argument will be computed in the stack and then loaded into
4036 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, const_tree @var{type})
4037 This target hook should return @code{true} if we should not pass @var{type}
4038 solely in registers. The file @file{expr.h} defines a
4039 definition that is usually appropriate, refer to @file{expr.h} for additional
4043 @deftypefn {Target Hook} rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t @var{ca}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4044 Define this hook if the target machine has ``register windows'', so
4045 that the register in which a function sees an arguments is not
4046 necessarily the same as the one in which the caller passed the
4049 For such machines, @code{TARGET_FUNCTION_ARG} computes the register in
4050 which the caller passes the value, and
4051 @code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar
4052 fashion to tell the function being called where the arguments will
4055 If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined,
4056 @code{TARGET_FUNCTION_ARG} serves both purposes.
4059 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4060 This target hook returns the number of bytes at the beginning of an
4061 argument that must be put in registers. The value must be zero for
4062 arguments that are passed entirely in registers or that are entirely
4063 pushed on the stack.
4065 On some machines, certain arguments must be passed partially in
4066 registers and partially in memory. On these machines, typically the
4067 first few words of arguments are passed in registers, and the rest
4068 on the stack. If a multi-word argument (a @code{double} or a
4069 structure) crosses that boundary, its first few words must be passed
4070 in registers and the rest must be pushed. This macro tells the
4071 compiler when this occurs, and how many bytes should go in registers.
4073 @code{TARGET_FUNCTION_ARG} for these arguments should return the first
4074 register to be used by the caller for this argument; likewise
4075 @code{TARGET_FUNCTION_INCOMING_ARG}, for the called function.
4078 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (cumulative_args_t @var{cum}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4079 This target hook should return @code{true} if an argument at the
4080 position indicated by @var{cum} should be passed by reference. This
4081 predicate is queried after target independent reasons for being
4082 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4084 If the hook returns true, a copy of that argument is made in memory and a
4085 pointer to the argument is passed instead of the argument itself.
4086 The pointer is passed in whatever way is appropriate for passing a pointer
4090 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4091 The function argument described by the parameters to this hook is
4092 known to be passed by reference. The hook should return true if the
4093 function argument should be copied by the callee instead of copied
4096 For any argument for which the hook returns true, if it can be
4097 determined that the argument is not modified, then a copy need
4100 The default version of this hook always returns false.
4103 @defmac CUMULATIVE_ARGS
4104 A C type for declaring a variable that is used as the first argument
4105 of @code{TARGET_FUNCTION_ARG} and other related values. For some
4106 target machines, the type @code{int} suffices and can hold the number
4107 of bytes of argument so far.
4109 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4110 arguments that have been passed on the stack. The compiler has other
4111 variables to keep track of that. For target machines on which all
4112 arguments are passed on the stack, there is no need to store anything in
4113 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4114 should not be empty, so use @code{int}.
4117 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4118 If defined, this macro is called before generating any code for a
4119 function, but after the @var{cfun} descriptor for the function has been
4120 created. The back end may use this macro to update @var{cfun} to
4121 reflect an ABI other than that which would normally be used by default.
4122 If the compiler is generating code for a compiler-generated function,
4123 @var{fndecl} may be @code{NULL}.
4126 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4127 A C statement (sans semicolon) for initializing the variable
4128 @var{cum} for the state at the beginning of the argument list. The
4129 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4130 is the tree node for the data type of the function which will receive
4131 the args, or 0 if the args are to a compiler support library function.
4132 For direct calls that are not libcalls, @var{fndecl} contain the
4133 declaration node of the function. @var{fndecl} is also set when
4134 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4135 being compiled. @var{n_named_args} is set to the number of named
4136 arguments, including a structure return address if it is passed as a
4137 parameter, when making a call. When processing incoming arguments,
4138 @var{n_named_args} is set to @minus{}1.
4140 When processing a call to a compiler support library function,
4141 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4142 contains the name of the function, as a string. @var{libname} is 0 when
4143 an ordinary C function call is being processed. Thus, each time this
4144 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4145 never both of them at once.
4148 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4149 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4150 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4151 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4152 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4153 0)} is used instead.
4156 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4157 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4158 finding the arguments for the function being compiled. If this macro is
4159 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4161 The value passed for @var{libname} is always 0, since library routines
4162 with special calling conventions are never compiled with GCC@. The
4163 argument @var{libname} exists for symmetry with
4164 @code{INIT_CUMULATIVE_ARGS}.
4165 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4166 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4169 @deftypefn {Target Hook} void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t @var{ca}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4170 This hook updates the summarizer variable pointed to by @var{ca} to
4171 advance past an argument in the argument list. The values @var{mode},
4172 @var{type} and @var{named} describe that argument. Once this is done,
4173 the variable @var{cum} is suitable for analyzing the @emph{following}
4174 argument with @code{TARGET_FUNCTION_ARG}, etc.
4176 This hook need not do anything if the argument in question was passed
4177 on the stack. The compiler knows how to track the amount of stack space
4178 used for arguments without any special help.
4181 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4182 If defined, a C expression that is the number of bytes to add to the
4183 offset of the argument passed in memory. This is needed for the SPU,
4184 which passes @code{char} and @code{short} arguments in the preferred
4185 slot that is in the middle of the quad word instead of starting at the
4189 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4190 If defined, a C expression which determines whether, and in which direction,
4191 to pad out an argument with extra space. The value should be of type
4192 @code{enum direction}: either @code{upward} to pad above the argument,
4193 @code{downward} to pad below, or @code{none} to inhibit padding.
4195 The @emph{amount} of padding is not controlled by this macro, but by the
4196 target hook @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is
4197 always just enough to reach the next multiple of that boundary.
4199 This macro has a default definition which is right for most systems.
4200 For little-endian machines, the default is to pad upward. For
4201 big-endian machines, the default is to pad downward for an argument of
4202 constant size shorter than an @code{int}, and upward otherwise.
4205 @defmac PAD_VARARGS_DOWN
4206 If defined, a C expression which determines whether the default
4207 implementation of va_arg will attempt to pad down before reading the
4208 next argument, if that argument is smaller than its aligned space as
4209 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4210 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4213 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4214 Specify padding for the last element of a block move between registers and
4215 memory. @var{first} is nonzero if this is the only element. Defining this
4216 macro allows better control of register function parameters on big-endian
4217 machines, without using @code{PARALLEL} rtl. In particular,
4218 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4219 registers, as there is no longer a "wrong" part of a register; For example,
4220 a three byte aggregate may be passed in the high part of a register if so
4224 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (enum machine_mode @var{mode}, const_tree @var{type})
4225 This hook returns the alignment boundary, in bits, of an argument
4226 with the specified mode and type. The default hook returns
4227 @code{PARM_BOUNDARY} for all arguments.
4230 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (enum machine_mode @var{mode}, const_tree @var{type})
4231 Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY},
4232 which is the default value for this hook. You can define this hook to
4233 return a different value if an argument size must be rounded to a larger
4237 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4238 A C expression that is nonzero if @var{regno} is the number of a hard
4239 register in which function arguments are sometimes passed. This does
4240 @emph{not} include implicit arguments such as the static chain and
4241 the structure-value address. On many machines, no registers can be
4242 used for this purpose since all function arguments are pushed on the
4246 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type})
4247 This hook should return true if parameter of type @var{type} are passed
4248 as two scalar parameters. By default, GCC will attempt to pack complex
4249 arguments into the target's word size. Some ABIs require complex arguments
4250 to be split and treated as their individual components. For example, on
4251 AIX64, complex floats should be passed in a pair of floating point
4252 registers, even though a complex float would fit in one 64-bit floating
4255 The default value of this hook is @code{NULL}, which is treated as always
4259 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4260 This hook returns a type node for @code{va_list} for the target.
4261 The default version of the hook returns @code{void*}.
4264 @deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree})
4265 This target hook is used in function @code{c_common_nodes_and_builtins}
4266 to iterate through the target specific builtin types for va_list. The
4267 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4268 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4270 The arguments @var{pname} and @var{ptree} are used to store the result of
4271 this macro and are set to the name of the va_list builtin type and its
4273 If the return value of this macro is zero, then there is no more element.
4274 Otherwise the @var{IDX} should be increased for the next call of this
4275 macro to iterate through all types.
4278 @deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl})
4279 This hook returns the va_list type of the calling convention specified by
4281 The default version of this hook returns @code{va_list_type_node}.
4284 @deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type})
4285 This hook returns the va_list type of the calling convention specified by the
4286 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4290 @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})
4291 This hook performs target-specific gimplification of
4292 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4293 arguments to @code{va_arg}; the latter two are as in
4294 @code{gimplify.c:gimplify_expr}.
4297 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
4298 Define this to return nonzero if the port can handle pointers
4299 with machine mode @var{mode}. The default version of this
4300 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4303 @deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (struct ao_ref_s *@var{ref})
4304 Define this to return nonzero if the memory reference @var{ref} may alias with the system C library errno location. The default version of this hook assumes the system C library errno location is either a declaration of type int or accessed by dereferencing a pointer to int.
4307 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4308 Define this to return nonzero if the port is prepared to handle
4309 insns involving scalar mode @var{mode}. For a scalar mode to be
4310 considered supported, all the basic arithmetic and comparisons
4313 The default version of this hook returns true for any mode
4314 required to handle the basic C types (as defined by the port).
4315 Included here are the double-word arithmetic supported by the
4316 code in @file{optabs.c}.
4319 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4320 Define this to return nonzero if the port is prepared to handle
4321 insns involving vector mode @var{mode}. At the very least, it
4322 must have move patterns for this mode.
4325 @deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (enum machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems})
4326 Return true if GCC should try to use a scalar mode to store an array
4327 of @var{nelems} elements, given that each element has mode @var{mode}.
4328 Returning true here overrides the usual @code{MAX_FIXED_MODE} limit
4329 and allows GCC to use any defined integer mode.
4331 One use of this hook is to support vector load and store operations
4332 that operate on several homogeneous vectors. For example, ARM NEON
4333 has operations like:
4336 int8x8x3_t vld3_s8 (const int8_t *)
4339 where the return type is defined as:
4342 typedef struct int8x8x3_t
4348 If this hook allows @code{val} to have a scalar mode, then
4349 @code{int8x8x3_t} can have the same mode. GCC can then store
4350 @code{int8x8x3_t}s in registers rather than forcing them onto the stack.
4353 @deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (enum machine_mode @var{mode})
4354 Define this to return nonzero for machine modes for which the port has
4355 small register classes. If this target hook returns nonzero for a given
4356 @var{mode}, the compiler will try to minimize the lifetime of registers
4357 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4358 In this case, the hook is expected to return nonzero if it returns nonzero
4361 On some machines, it is risky to let hard registers live across arbitrary
4362 insns. Typically, these machines have instructions that require values
4363 to be in specific registers (like an accumulator), and reload will fail
4364 if the required hard register is used for another purpose across such an
4367 Passes before reload do not know which hard registers will be used
4368 in an instruction, but the machine modes of the registers set or used in
4369 the instruction are already known. And for some machines, register
4370 classes are small for, say, integer registers but not for floating point
4371 registers. For example, the AMD x86-64 architecture requires specific
4372 registers for the legacy x86 integer instructions, but there are many
4373 SSE registers for floating point operations. On such targets, a good
4374 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4375 machine modes but zero for the SSE register classes.
4377 The default version of this hook returns false for any mode. It is always
4378 safe to redefine this hook to return with a nonzero value. But if you
4379 unnecessarily define it, you will reduce the amount of optimizations
4380 that can be performed in some cases. If you do not define this hook
4381 to return a nonzero value when it is required, the compiler will run out
4382 of spill registers and print a fatal error message.
4385 @deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM
4386 If the target has a dedicated flags register, and it needs to use the post-reload comparison elimination pass, then this value should be set appropriately.
4390 @subsection How Scalar Function Values Are Returned
4391 @cindex return values in registers
4392 @cindex values, returned by functions
4393 @cindex scalars, returned as values
4395 This section discusses the macros that control returning scalars as
4396 values---values that can fit in registers.
4398 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
4400 Define this to return an RTX representing the place where a function
4401 returns or receives a value of data type @var{ret_type}, a tree node
4402 representing a data type. @var{fn_decl_or_type} is a tree node
4403 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4404 function being called. If @var{outgoing} is false, the hook should
4405 compute the register in which the caller will see the return value.
4406 Otherwise, the hook should return an RTX representing the place where
4407 a function returns a value.
4409 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4410 (Actually, on most machines, scalar values are returned in the same
4411 place regardless of mode.) The value of the expression is usually a
4412 @code{reg} RTX for the hard register where the return value is stored.
4413 The value can also be a @code{parallel} RTX, if the return value is in
4414 multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the
4415 @code{parallel} form. Note that the callee will populate every
4416 location specified in the @code{parallel}, but if the first element of
4417 the @code{parallel} contains the whole return value, callers will use
4418 that element as the canonical location and ignore the others. The m68k
4419 port uses this type of @code{parallel} to return pointers in both
4420 @samp{%a0} (the canonical location) and @samp{%d0}.
4422 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4423 the same promotion rules specified in @code{PROMOTE_MODE} if
4424 @var{valtype} is a scalar type.
4426 If the precise function being called is known, @var{func} is a tree
4427 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4428 pointer. This makes it possible to use a different value-returning
4429 convention for specific functions when all their calls are
4432 Some target machines have ``register windows'' so that the register in
4433 which a function returns its value is not the same as the one in which
4434 the caller sees the value. For such machines, you should return
4435 different RTX depending on @var{outgoing}.
4437 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4438 aggregate data types, because these are returned in another way. See
4439 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4442 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4443 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4444 a new target instead.
4447 @defmac LIBCALL_VALUE (@var{mode})
4448 A C expression to create an RTX representing the place where a library
4449 function returns a value of mode @var{mode}.
4451 Note that ``library function'' in this context means a compiler
4452 support routine, used to perform arithmetic, whose name is known
4453 specially by the compiler and was not mentioned in the C code being
4457 @deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (enum machine_mode @var{mode}, const_rtx @var{fun})
4458 Define this hook if the back-end needs to know the name of the libcall
4459 function in order to determine where the result should be returned.
4461 The mode of the result is given by @var{mode} and the name of the called
4462 library function is given by @var{fun}. The hook should return an RTX
4463 representing the place where the library function result will be returned.
4465 If this hook is not defined, then LIBCALL_VALUE will be used.
4468 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4469 A C expression that is nonzero if @var{regno} is the number of a hard
4470 register in which the values of called function may come back.
4472 A register whose use for returning values is limited to serving as the
4473 second of a pair (for a value of type @code{double}, say) need not be
4474 recognized by this macro. So for most machines, this definition
4478 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4481 If the machine has register windows, so that the caller and the called
4482 function use different registers for the return value, this macro
4483 should recognize only the caller's register numbers.
4485 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4486 for a new target instead.
4489 @deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno})
4490 A target hook that return @code{true} if @var{regno} is the number of a hard
4491 register in which the values of called function may come back.
4493 A register whose use for returning values is limited to serving as the
4494 second of a pair (for a value of type @code{double}, say) need not be
4495 recognized by this target hook.
4497 If the machine has register windows, so that the caller and the called
4498 function use different registers for the return value, this target hook
4499 should recognize only the caller's register numbers.
4501 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4504 @defmac APPLY_RESULT_SIZE
4505 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4506 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4507 saving and restoring an arbitrary return value.
4510 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type})
4511 This hook should return true if values of type @var{type} are returned
4512 at the most significant end of a register (in other words, if they are
4513 padded at the least significant end). You can assume that @var{type}
4514 is returned in a register; the caller is required to check this.
4516 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4517 be able to hold the complete return value. For example, if a 1-, 2-
4518 or 3-byte structure is returned at the most significant end of a
4519 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4523 @node Aggregate Return
4524 @subsection How Large Values Are Returned
4525 @cindex aggregates as return values
4526 @cindex large return values
4527 @cindex returning aggregate values
4528 @cindex structure value address
4530 When a function value's mode is @code{BLKmode} (and in some other
4531 cases), the value is not returned according to
4532 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4533 caller passes the address of a block of memory in which the value
4534 should be stored. This address is called the @dfn{structure value
4537 This section describes how to control returning structure values in
4540 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype})
4541 This target hook should return a nonzero value to say to return the
4542 function value in memory, just as large structures are always returned.
4543 Here @var{type} will be the data type of the value, and @var{fntype}
4544 will be the type of the function doing the returning, or @code{NULL} for
4547 Note that values of mode @code{BLKmode} must be explicitly handled
4548 by this function. Also, the option @option{-fpcc-struct-return}
4549 takes effect regardless of this macro. On most systems, it is
4550 possible to leave the hook undefined; this causes a default
4551 definition to be used, whose value is the constant 1 for @code{BLKmode}
4552 values, and 0 otherwise.
4554 Do not use this hook to indicate that structures and unions should always
4555 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4559 @defmac DEFAULT_PCC_STRUCT_RETURN
4560 Define this macro to be 1 if all structure and union return values must be
4561 in memory. Since this results in slower code, this should be defined
4562 only if needed for compatibility with other compilers or with an ABI@.
4563 If you define this macro to be 0, then the conventions used for structure
4564 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4567 If not defined, this defaults to the value 1.
4570 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4571 This target hook should return the location of the structure value
4572 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4573 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4574 be @code{NULL}, for libcalls. You do not need to define this target
4575 hook if the address is always passed as an ``invisible'' first
4578 On some architectures the place where the structure value address
4579 is found by the called function is not the same place that the
4580 caller put it. This can be due to register windows, or it could
4581 be because the function prologue moves it to a different place.
4582 @var{incoming} is @code{1} or @code{2} when the location is needed in
4583 the context of the called function, and @code{0} in the context of
4586 If @var{incoming} is nonzero and the address is to be found on the
4587 stack, return a @code{mem} which refers to the frame pointer. If
4588 @var{incoming} is @code{2}, the result is being used to fetch the
4589 structure value address at the beginning of a function. If you need
4590 to emit adjusting code, you should do it at this point.
4593 @defmac PCC_STATIC_STRUCT_RETURN
4594 Define this macro if the usual system convention on the target machine
4595 for returning structures and unions is for the called function to return
4596 the address of a static variable containing the value.
4598 Do not define this if the usual system convention is for the caller to
4599 pass an address to the subroutine.
4601 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4602 nothing when you use @option{-freg-struct-return} mode.
4605 @deftypefn {Target Hook} {enum machine_mode} TARGET_GET_RAW_RESULT_MODE (int @var{regno})
4606 This target hook returns the mode to be used when accessing raw return registers in @code{__builtin_return}. Define this macro if the value in @var{reg_raw_mode} is not correct.
4609 @deftypefn {Target Hook} {enum machine_mode} TARGET_GET_RAW_ARG_MODE (int @var{regno})
4610 This target hook returns the mode to be used when accessing raw argument registers in @code{__builtin_apply_args}. Define this macro if the value in @var{reg_raw_mode} is not correct.
4614 @subsection Caller-Saves Register Allocation
4616 If you enable it, GCC can save registers around function calls. This
4617 makes it possible to use call-clobbered registers to hold variables that
4618 must live across calls.
4620 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4621 A C expression to determine whether it is worthwhile to consider placing
4622 a pseudo-register in a call-clobbered hard register and saving and
4623 restoring it around each function call. The expression should be 1 when
4624 this is worth doing, and 0 otherwise.
4626 If you don't define this macro, a default is used which is good on most
4627 machines: @code{4 * @var{calls} < @var{refs}}.
4630 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4631 A C expression specifying which mode is required for saving @var{nregs}
4632 of a pseudo-register in call-clobbered hard register @var{regno}. If
4633 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4634 returned. For most machines this macro need not be defined since GCC
4635 will select the smallest suitable mode.
4638 @node Function Entry
4639 @subsection Function Entry and Exit
4640 @cindex function entry and exit
4644 This section describes the macros that output function entry
4645 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4647 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4648 If defined, a function that outputs the assembler code for entry to a
4649 function. The prologue is responsible for setting up the stack frame,
4650 initializing the frame pointer register, saving registers that must be
4651 saved, and allocating @var{size} additional bytes of storage for the
4652 local variables. @var{size} is an integer. @var{file} is a stdio
4653 stream to which the assembler code should be output.
4655 The label for the beginning of the function need not be output by this
4656 macro. That has already been done when the macro is run.
4658 @findex regs_ever_live
4659 To determine which registers to save, the macro can refer to the array
4660 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4661 @var{r} is used anywhere within the function. This implies the function
4662 prologue should save register @var{r}, provided it is not one of the
4663 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4664 @code{regs_ever_live}.)
4666 On machines that have ``register windows'', the function entry code does
4667 not save on the stack the registers that are in the windows, even if
4668 they are supposed to be preserved by function calls; instead it takes
4669 appropriate steps to ``push'' the register stack, if any non-call-used
4670 registers are used in the function.
4672 @findex frame_pointer_needed
4673 On machines where functions may or may not have frame-pointers, the
4674 function entry code must vary accordingly; it must set up the frame
4675 pointer if one is wanted, and not otherwise. To determine whether a
4676 frame pointer is in wanted, the macro can refer to the variable
4677 @code{frame_pointer_needed}. The variable's value will be 1 at run
4678 time in a function that needs a frame pointer. @xref{Elimination}.
4680 The function entry code is responsible for allocating any stack space
4681 required for the function. This stack space consists of the regions
4682 listed below. In most cases, these regions are allocated in the
4683 order listed, with the last listed region closest to the top of the
4684 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4685 the highest address if it is not defined). You can use a different order
4686 for a machine if doing so is more convenient or required for
4687 compatibility reasons. Except in cases where required by standard
4688 or by a debugger, there is no reason why the stack layout used by GCC
4689 need agree with that used by other compilers for a machine.
4692 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4693 If defined, a function that outputs assembler code at the end of a
4694 prologue. This should be used when the function prologue is being
4695 emitted as RTL, and you have some extra assembler that needs to be
4696 emitted. @xref{prologue instruction pattern}.
4699 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4700 If defined, a function that outputs assembler code at the start of an
4701 epilogue. This should be used when the function epilogue is being
4702 emitted as RTL, and you have some extra assembler that needs to be
4703 emitted. @xref{epilogue instruction pattern}.
4706 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4707 If defined, a function that outputs the assembler code for exit from a
4708 function. The epilogue is responsible for restoring the saved
4709 registers and stack pointer to their values when the function was
4710 called, and returning control to the caller. This macro takes the
4711 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4712 registers to restore are determined from @code{regs_ever_live} and
4713 @code{CALL_USED_REGISTERS} in the same way.
4715 On some machines, there is a single instruction that does all the work
4716 of returning from the function. On these machines, give that
4717 instruction the name @samp{return} and do not define the macro
4718 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4720 Do not define a pattern named @samp{return} if you want the
4721 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4722 switches to control whether return instructions or epilogues are used,
4723 define a @samp{return} pattern with a validity condition that tests the
4724 target switches appropriately. If the @samp{return} pattern's validity
4725 condition is false, epilogues will be used.
4727 On machines where functions may or may not have frame-pointers, the
4728 function exit code must vary accordingly. Sometimes the code for these
4729 two cases is completely different. To determine whether a frame pointer
4730 is wanted, the macro can refer to the variable
4731 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4732 a function that needs a frame pointer.
4734 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4735 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4736 The C variable @code{current_function_is_leaf} is nonzero for such a
4737 function. @xref{Leaf Functions}.
4739 On some machines, some functions pop their arguments on exit while
4740 others leave that for the caller to do. For example, the 68020 when
4741 given @option{-mrtd} pops arguments in functions that take a fixed
4742 number of arguments.
4744 @findex current_function_pops_args
4745 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4746 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4747 needs to know what was decided. The number of bytes of the current
4748 function's arguments that this function should pop is available in
4749 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4754 @findex current_function_pretend_args_size
4755 A region of @code{current_function_pretend_args_size} bytes of
4756 uninitialized space just underneath the first argument arriving on the
4757 stack. (This may not be at the very start of the allocated stack region
4758 if the calling sequence has pushed anything else since pushing the stack
4759 arguments. But usually, on such machines, nothing else has been pushed
4760 yet, because the function prologue itself does all the pushing.) This
4761 region is used on machines where an argument may be passed partly in
4762 registers and partly in memory, and, in some cases to support the
4763 features in @code{<stdarg.h>}.
4766 An area of memory used to save certain registers used by the function.
4767 The size of this area, which may also include space for such things as
4768 the return address and pointers to previous stack frames, is
4769 machine-specific and usually depends on which registers have been used
4770 in the function. Machines with register windows often do not require
4774 A region of at least @var{size} bytes, possibly rounded up to an allocation
4775 boundary, to contain the local variables of the function. On some machines,
4776 this region and the save area may occur in the opposite order, with the
4777 save area closer to the top of the stack.
4780 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4781 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4782 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4783 argument lists of the function. @xref{Stack Arguments}.
4786 @defmac EXIT_IGNORE_STACK
4787 Define this macro as a C expression that is nonzero if the return
4788 instruction or the function epilogue ignores the value of the stack
4789 pointer; in other words, if it is safe to delete an instruction to
4790 adjust the stack pointer before a return from the function. The
4793 Note that this macro's value is relevant only for functions for which
4794 frame pointers are maintained. It is never safe to delete a final
4795 stack adjustment in a function that has no frame pointer, and the
4796 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4799 @defmac EPILOGUE_USES (@var{regno})
4800 Define this macro as a C expression that is nonzero for registers that are
4801 used by the epilogue or the @samp{return} pattern. The stack and frame
4802 pointer registers are already assumed to be used as needed.
4805 @defmac EH_USES (@var{regno})
4806 Define this macro as a C expression that is nonzero for registers that are
4807 used by the exception handling mechanism, and so should be considered live
4808 on entry to an exception edge.
4811 @defmac DELAY_SLOTS_FOR_EPILOGUE
4812 Define this macro if the function epilogue contains delay slots to which
4813 instructions from the rest of the function can be ``moved''. The
4814 definition should be a C expression whose value is an integer
4815 representing the number of delay slots there.
4818 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4819 A C expression that returns 1 if @var{insn} can be placed in delay
4820 slot number @var{n} of the epilogue.
4822 The argument @var{n} is an integer which identifies the delay slot now
4823 being considered (since different slots may have different rules of
4824 eligibility). It is never negative and is always less than the number
4825 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4826 If you reject a particular insn for a given delay slot, in principle, it
4827 may be reconsidered for a subsequent delay slot. Also, other insns may
4828 (at least in principle) be considered for the so far unfilled delay
4831 @findex current_function_epilogue_delay_list
4832 @findex final_scan_insn
4833 The insns accepted to fill the epilogue delay slots are put in an RTL
4834 list made with @code{insn_list} objects, stored in the variable
4835 @code{current_function_epilogue_delay_list}. The insn for the first
4836 delay slot comes first in the list. Your definition of the macro
4837 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4838 outputting the insns in this list, usually by calling
4839 @code{final_scan_insn}.
4841 You need not define this macro if you did not define
4842 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4845 @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})
4846 A function that outputs the assembler code for a thunk
4847 function, used to implement C++ virtual function calls with multiple
4848 inheritance. The thunk acts as a wrapper around a virtual function,
4849 adjusting the implicit object parameter before handing control off to
4852 First, emit code to add the integer @var{delta} to the location that
4853 contains the incoming first argument. Assume that this argument
4854 contains a pointer, and is the one used to pass the @code{this} pointer
4855 in C++. This is the incoming argument @emph{before} the function prologue,
4856 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4857 all other incoming arguments.
4859 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4860 made after adding @code{delta}. In particular, if @var{p} is the
4861 adjusted pointer, the following adjustment should be made:
4864 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4867 After the additions, emit code to jump to @var{function}, which is a
4868 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4869 not touch the return address. Hence returning from @var{FUNCTION} will
4870 return to whoever called the current @samp{thunk}.
4872 The effect must be as if @var{function} had been called directly with
4873 the adjusted first argument. This macro is responsible for emitting all
4874 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4875 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4877 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4878 have already been extracted from it.) It might possibly be useful on
4879 some targets, but probably not.
4881 If you do not define this macro, the target-independent code in the C++
4882 front end will generate a less efficient heavyweight thunk that calls
4883 @var{function} instead of jumping to it. The generic approach does
4884 not support varargs.
4887 @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})
4888 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4889 to output the assembler code for the thunk function specified by the
4890 arguments it is passed, and false otherwise. In the latter case, the
4891 generic approach will be used by the C++ front end, with the limitations
4896 @subsection Generating Code for Profiling
4897 @cindex profiling, code generation
4899 These macros will help you generate code for profiling.
4901 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4902 A C statement or compound statement to output to @var{file} some
4903 assembler code to call the profiling subroutine @code{mcount}.
4906 The details of how @code{mcount} expects to be called are determined by
4907 your operating system environment, not by GCC@. To figure them out,
4908 compile a small program for profiling using the system's installed C
4909 compiler and look at the assembler code that results.
4911 Older implementations of @code{mcount} expect the address of a counter
4912 variable to be loaded into some register. The name of this variable is
4913 @samp{LP} followed by the number @var{labelno}, so you would generate
4914 the name using @samp{LP%d} in a @code{fprintf}.
4917 @defmac PROFILE_HOOK
4918 A C statement or compound statement to output to @var{file} some assembly
4919 code to call the profiling subroutine @code{mcount} even the target does
4920 not support profiling.
4923 @defmac NO_PROFILE_COUNTERS
4924 Define this macro to be an expression with a nonzero value if the
4925 @code{mcount} subroutine on your system does not need a counter variable
4926 allocated for each function. This is true for almost all modern
4927 implementations. If you define this macro, you must not use the
4928 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4931 @defmac PROFILE_BEFORE_PROLOGUE
4932 Define this macro if the code for function profiling should come before
4933 the function prologue. Normally, the profiling code comes after.
4937 @subsection Permitting tail calls
4940 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4941 True if it is ok to do sibling call optimization for the specified
4942 call expression @var{exp}. @var{decl} will be the called function,
4943 or @code{NULL} if this is an indirect call.
4945 It is not uncommon for limitations of calling conventions to prevent
4946 tail calls to functions outside the current unit of translation, or
4947 during PIC compilation. The hook is used to enforce these restrictions,
4948 as the @code{sibcall} md pattern can not fail, or fall over to a
4949 ``normal'' call. The criteria for successful sibling call optimization
4950 may vary greatly between different architectures.
4953 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs})
4954 Add any hard registers to @var{regs} that are live on entry to the
4955 function. This hook only needs to be defined to provide registers that
4956 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4957 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4958 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4959 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4962 @node Stack Smashing Protection
4963 @subsection Stack smashing protection
4964 @cindex stack smashing protection
4966 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4967 This hook returns a @code{DECL} node for the external variable to use
4968 for the stack protection guard. This variable is initialized by the
4969 runtime to some random value and is used to initialize the guard value
4970 that is placed at the top of the local stack frame. The type of this
4971 variable must be @code{ptr_type_node}.
4973 The default version of this hook creates a variable called
4974 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4977 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4978 This hook returns a tree expression that alerts the runtime that the
4979 stack protect guard variable has been modified. This expression should
4980 involve a call to a @code{noreturn} function.
4982 The default version of this hook invokes a function called
4983 @samp{__stack_chk_fail}, taking no arguments. This function is
4984 normally defined in @file{libgcc2.c}.
4987 @deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts})
4988 Whether this target supports splitting the stack when the options described in @var{opts} have been passed. This is called after options have been parsed, so the target may reject splitting the stack in some configurations. The default version of this hook returns false. If @var{report} is true, this function may issue a warning or error; if @var{report} is false, it must simply return a value
4992 @section Implementing the Varargs Macros
4993 @cindex varargs implementation
4995 GCC comes with an implementation of @code{<varargs.h>} and
4996 @code{<stdarg.h>} that work without change on machines that pass arguments
4997 on the stack. Other machines require their own implementations of
4998 varargs, and the two machine independent header files must have
4999 conditionals to include it.
5001 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
5002 the calling convention for @code{va_start}. The traditional
5003 implementation takes just one argument, which is the variable in which
5004 to store the argument pointer. The ISO implementation of
5005 @code{va_start} takes an additional second argument. The user is
5006 supposed to write the last named argument of the function here.
5008 However, @code{va_start} should not use this argument. The way to find
5009 the end of the named arguments is with the built-in functions described
5012 @defmac __builtin_saveregs ()
5013 Use this built-in function to save the argument registers in memory so
5014 that the varargs mechanism can access them. Both ISO and traditional
5015 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5016 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5018 On some machines, @code{__builtin_saveregs} is open-coded under the
5019 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5020 other machines, it calls a routine written in assembler language,
5021 found in @file{libgcc2.c}.
5023 Code generated for the call to @code{__builtin_saveregs} appears at the
5024 beginning of the function, as opposed to where the call to
5025 @code{__builtin_saveregs} is written, regardless of what the code is.
5026 This is because the registers must be saved before the function starts
5027 to use them for its own purposes.
5028 @c i rewrote the first sentence above to fix an overfull hbox. --mew
5032 @defmac __builtin_next_arg (@var{lastarg})
5033 This builtin returns the address of the first anonymous stack
5034 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5035 returns the address of the location above the first anonymous stack
5036 argument. Use it in @code{va_start} to initialize the pointer for
5037 fetching arguments from the stack. Also use it in @code{va_start} to
5038 verify that the second parameter @var{lastarg} is the last named argument
5039 of the current function.
5042 @defmac __builtin_classify_type (@var{object})
5043 Since each machine has its own conventions for which data types are
5044 passed in which kind of register, your implementation of @code{va_arg}
5045 has to embody these conventions. The easiest way to categorize the
5046 specified data type is to use @code{__builtin_classify_type} together
5047 with @code{sizeof} and @code{__alignof__}.
5049 @code{__builtin_classify_type} ignores the value of @var{object},
5050 considering only its data type. It returns an integer describing what
5051 kind of type that is---integer, floating, pointer, structure, and so on.
5053 The file @file{typeclass.h} defines an enumeration that you can use to
5054 interpret the values of @code{__builtin_classify_type}.
5057 These machine description macros help implement varargs:
5059 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
5060 If defined, this hook produces the machine-specific code for a call to
5061 @code{__builtin_saveregs}. This code will be moved to the very
5062 beginning of the function, before any parameter access are made. The
5063 return value of this function should be an RTX that contains the value
5064 to use as the return of @code{__builtin_saveregs}.
5067 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t @var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time})
5068 This target hook offers an alternative to using
5069 @code{__builtin_saveregs} and defining the hook
5070 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5071 register arguments into the stack so that all the arguments appear to
5072 have been passed consecutively on the stack. Once this is done, you can
5073 use the standard implementation of varargs that works for machines that
5074 pass all their arguments on the stack.
5076 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5077 structure, containing the values that are obtained after processing the
5078 named arguments. The arguments @var{mode} and @var{type} describe the
5079 last named argument---its machine mode and its data type as a tree node.
5081 The target hook should do two things: first, push onto the stack all the
5082 argument registers @emph{not} used for the named arguments, and second,
5083 store the size of the data thus pushed into the @code{int}-valued
5084 variable pointed to by @var{pretend_args_size}. The value that you
5085 store here will serve as additional offset for setting up the stack
5088 Because you must generate code to push the anonymous arguments at
5089 compile time without knowing their data types,
5090 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5091 have just a single category of argument register and use it uniformly
5094 If the argument @var{second_time} is nonzero, it means that the
5095 arguments of the function are being analyzed for the second time. This
5096 happens for an inline function, which is not actually compiled until the
5097 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5098 not generate any instructions in this case.
5101 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca})
5102 Define this hook to return @code{true} if the location where a function
5103 argument is passed depends on whether or not it is a named argument.
5105 This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG}
5106 is set for varargs and stdarg functions. If this hook returns
5107 @code{true}, the @var{named} argument is always true for named
5108 arguments, and false for unnamed arguments. If it returns @code{false},
5109 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5110 then all arguments are treated as named. Otherwise, all named arguments
5111 except the last are treated as named.
5113 You need not define this hook if it always returns @code{false}.
5116 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca})
5117 If you need to conditionally change ABIs so that one works with
5118 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5119 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5120 defined, then define this hook to return @code{true} if
5121 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5122 Otherwise, you should not define this hook.
5126 @section Trampolines for Nested Functions
5127 @cindex trampolines for nested functions
5128 @cindex nested functions, trampolines for
5130 A @dfn{trampoline} is a small piece of code that is created at run time
5131 when the address of a nested function is taken. It normally resides on
5132 the stack, in the stack frame of the containing function. These macros
5133 tell GCC how to generate code to allocate and initialize a
5136 The instructions in the trampoline must do two things: load a constant
5137 address into the static chain register, and jump to the real address of
5138 the nested function. On CISC machines such as the m68k, this requires
5139 two instructions, a move immediate and a jump. Then the two addresses
5140 exist in the trampoline as word-long immediate operands. On RISC
5141 machines, it is often necessary to load each address into a register in
5142 two parts. Then pieces of each address form separate immediate
5145 The code generated to initialize the trampoline must store the variable
5146 parts---the static chain value and the function address---into the
5147 immediate operands of the instructions. On a CISC machine, this is
5148 simply a matter of copying each address to a memory reference at the
5149 proper offset from the start of the trampoline. On a RISC machine, it
5150 may be necessary to take out pieces of the address and store them
5153 @deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f})
5154 This hook is called by @code{assemble_trampoline_template} to output,
5155 on the stream @var{f}, assembler code for a block of data that contains
5156 the constant parts of a trampoline. This code should not include a
5157 label---the label is taken care of automatically.
5159 If you do not define this hook, it means no template is needed
5160 for the target. Do not define this hook on systems where the block move
5161 code to copy the trampoline into place would be larger than the code
5162 to generate it on the spot.
5165 @defmac TRAMPOLINE_SECTION
5166 Return the section into which the trampoline template is to be placed
5167 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5170 @defmac TRAMPOLINE_SIZE
5171 A C expression for the size in bytes of the trampoline, as an integer.
5174 @defmac TRAMPOLINE_ALIGNMENT
5175 Alignment required for trampolines, in bits.
5177 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5178 is used for aligning trampolines.
5181 @deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain})
5182 This hook is called to initialize a trampoline.
5183 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5184 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5185 RTX for the static chain value that should be passed to the function
5188 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5189 first thing this hook should do is emit a block move into @var{m_tramp}
5190 from the memory block returned by @code{assemble_trampoline_template}.
5191 Note that the block move need only cover the constant parts of the
5192 trampoline. If the target isolates the variable parts of the trampoline
5193 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5195 If the target requires any other actions, such as flushing caches or
5196 enabling stack execution, these actions should be performed after
5197 initializing the trampoline proper.
5200 @deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr})
5201 This hook should perform any machine-specific adjustment in
5202 the address of the trampoline. Its argument contains the address of the
5203 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5204 the address to be used for a function call should be different from the
5205 address at which the template was stored, the different address should
5206 be returned; otherwise @var{addr} should be returned unchanged.
5207 If this hook is not defined, @var{addr} will be used for function calls.
5210 Implementing trampolines is difficult on many machines because they have
5211 separate instruction and data caches. Writing into a stack location
5212 fails to clear the memory in the instruction cache, so when the program
5213 jumps to that location, it executes the old contents.
5215 Here are two possible solutions. One is to clear the relevant parts of
5216 the instruction cache whenever a trampoline is set up. The other is to
5217 make all trampolines identical, by having them jump to a standard
5218 subroutine. The former technique makes trampoline execution faster; the
5219 latter makes initialization faster.
5221 To clear the instruction cache when a trampoline is initialized, define
5222 the following macro.
5224 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5225 If defined, expands to a C expression clearing the @emph{instruction
5226 cache} in the specified interval. The definition of this macro would
5227 typically be a series of @code{asm} statements. Both @var{beg} and
5228 @var{end} are both pointer expressions.
5231 To use a standard subroutine, define the following macro. In addition,
5232 you must make sure that the instructions in a trampoline fill an entire
5233 cache line with identical instructions, or else ensure that the
5234 beginning of the trampoline code is always aligned at the same point in
5235 its cache line. Look in @file{m68k.h} as a guide.
5237 @defmac TRANSFER_FROM_TRAMPOLINE
5238 Define this macro if trampolines need a special subroutine to do their
5239 work. The macro should expand to a series of @code{asm} statements
5240 which will be compiled with GCC@. They go in a library function named
5241 @code{__transfer_from_trampoline}.
5243 If you need to avoid executing the ordinary prologue code of a compiled
5244 C function when you jump to the subroutine, you can do so by placing a
5245 special label of your own in the assembler code. Use one @code{asm}
5246 statement to generate an assembler label, and another to make the label
5247 global. Then trampolines can use that label to jump directly to your
5248 special assembler code.
5252 @section Implicit Calls to Library Routines
5253 @cindex library subroutine names
5254 @cindex @file{libgcc.a}
5256 @c prevent bad page break with this line
5257 Here is an explanation of implicit calls to library routines.
5259 @defmac DECLARE_LIBRARY_RENAMES
5260 This macro, if defined, should expand to a piece of C code that will get
5261 expanded when compiling functions for libgcc.a. It can be used to
5262 provide alternate names for GCC's internal library functions if there
5263 are ABI-mandated names that the compiler should provide.
5266 @findex set_optab_libfunc
5267 @findex init_one_libfunc
5268 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5269 This hook should declare additional library routines or rename
5270 existing ones, using the functions @code{set_optab_libfunc} and
5271 @code{init_one_libfunc} defined in @file{optabs.c}.
5272 @code{init_optabs} calls this macro after initializing all the normal
5275 The default is to do nothing. Most ports don't need to define this hook.
5278 @deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX
5279 If false (the default), internal library routines start with two
5280 underscores. If set to true, these routines start with @code{__gnu_}
5281 instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This
5282 currently only affects functions defined in @file{libgcc2.c}. If this
5283 is set to true, the @file{tm.h} file must also
5284 @code{#define LIBGCC2_GNU_PREFIX}.
5287 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5288 This macro should return @code{true} if the library routine that
5289 implements the floating point comparison operator @var{comparison} in
5290 mode @var{mode} will return a boolean, and @var{false} if it will
5293 GCC's own floating point libraries return tristates from the
5294 comparison operators, so the default returns false always. Most ports
5295 don't need to define this macro.
5298 @defmac TARGET_LIB_INT_CMP_BIASED
5299 This macro should evaluate to @code{true} if the integer comparison
5300 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5301 operand is smaller than the second, 1 to indicate that they are equal,
5302 and 2 to indicate that the first operand is greater than the second.
5303 If this macro evaluates to @code{false} the comparison functions return
5304 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5305 in @file{libgcc.a}, you do not need to define this macro.
5308 @cindex @code{EDOM}, implicit usage
5311 The value of @code{EDOM} on the target machine, as a C integer constant
5312 expression. If you don't define this macro, GCC does not attempt to
5313 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5314 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5317 If you do not define @code{TARGET_EDOM}, then compiled code reports
5318 domain errors by calling the library function and letting it report the
5319 error. If mathematical functions on your system use @code{matherr} when
5320 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5321 that @code{matherr} is used normally.
5324 @cindex @code{errno}, implicit usage
5325 @defmac GEN_ERRNO_RTX
5326 Define this macro as a C expression to create an rtl expression that
5327 refers to the global ``variable'' @code{errno}. (On certain systems,
5328 @code{errno} may not actually be a variable.) If you don't define this
5329 macro, a reasonable default is used.
5332 @cindex C99 math functions, implicit usage
5333 @defmac TARGET_C99_FUNCTIONS
5334 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5335 @code{sinf} and similarly for other functions defined by C99 standard. The
5336 default is zero because a number of existing systems lack support for these
5337 functions in their runtime so this macro needs to be redefined to one on
5338 systems that do support the C99 runtime.
5341 @cindex sincos math function, implicit usage
5342 @defmac TARGET_HAS_SINCOS
5343 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5344 and @code{cos} with the same argument to a call to @code{sincos}. The
5345 default is zero. The target has to provide the following functions:
5347 void sincos(double x, double *sin, double *cos);
5348 void sincosf(float x, float *sin, float *cos);
5349 void sincosl(long double x, long double *sin, long double *cos);
5353 @defmac NEXT_OBJC_RUNTIME
5354 Set this macro to 1 to use the "NeXT" Objective-C message sending conventions
5355 by default. This calling convention involves passing the object, the selector
5356 and the method arguments all at once to the method-lookup library function.
5357 This is the usual setting when targeting Darwin/Mac OS X systems, which have
5358 the NeXT runtime installed.
5360 If the macro is set to 0, the "GNU" Objective-C message sending convention
5361 will be used by default. This convention passes just the object and the
5362 selector to the method-lookup function, which returns a pointer to the method.
5364 In either case, it remains possible to select code-generation for the alternate
5365 scheme, by means of compiler command line switches.
5368 @node Addressing Modes
5369 @section Addressing Modes
5370 @cindex addressing modes
5372 @c prevent bad page break with this line
5373 This is about addressing modes.
5375 @defmac HAVE_PRE_INCREMENT
5376 @defmacx HAVE_PRE_DECREMENT
5377 @defmacx HAVE_POST_INCREMENT
5378 @defmacx HAVE_POST_DECREMENT
5379 A C expression that is nonzero if the machine supports pre-increment,
5380 pre-decrement, post-increment, or post-decrement addressing respectively.
5383 @defmac HAVE_PRE_MODIFY_DISP
5384 @defmacx HAVE_POST_MODIFY_DISP
5385 A C expression that is nonzero if the machine supports pre- or
5386 post-address side-effect generation involving constants other than
5387 the size of the memory operand.
5390 @defmac HAVE_PRE_MODIFY_REG
5391 @defmacx HAVE_POST_MODIFY_REG
5392 A C expression that is nonzero if the machine supports pre- or
5393 post-address side-effect generation involving a register displacement.
5396 @defmac CONSTANT_ADDRESS_P (@var{x})
5397 A C expression that is 1 if the RTX @var{x} is a constant which
5398 is a valid address. On most machines the default definition of
5399 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5400 is acceptable, but a few machines are more restrictive as to which
5401 constant addresses are supported.
5404 @defmac CONSTANT_P (@var{x})
5405 @code{CONSTANT_P}, which is defined by target-independent code,
5406 accepts integer-values expressions whose values are not explicitly
5407 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5408 expressions and @code{const} arithmetic expressions, in addition to
5409 @code{const_int} and @code{const_double} expressions.
5412 @defmac MAX_REGS_PER_ADDRESS
5413 A number, the maximum number of registers that can appear in a valid
5414 memory address. Note that it is up to you to specify a value equal to
5415 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5419 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (enum machine_mode @var{mode}, rtx @var{x}, bool @var{strict})
5420 A function that returns whether @var{x} (an RTX) is a legitimate memory
5421 address on the target machine for a memory operand of mode @var{mode}.
5423 Legitimate addresses are defined in two variants: a strict variant and a
5424 non-strict one. The @var{strict} parameter chooses which variant is
5425 desired by the caller.
5427 The strict variant is used in the reload pass. It must be defined so
5428 that any pseudo-register that has not been allocated a hard register is
5429 considered a memory reference. This is because in contexts where some
5430 kind of register is required, a pseudo-register with no hard register
5431 must be rejected. For non-hard registers, the strict variant should look
5432 up the @code{reg_renumber} array; it should then proceed using the hard
5433 register number in the array, or treat the pseudo as a memory reference
5434 if the array holds @code{-1}.
5436 The non-strict variant is used in other passes. It must be defined to
5437 accept all pseudo-registers in every context where some kind of
5438 register is required.
5440 Normally, constant addresses which are the sum of a @code{symbol_ref}
5441 and an integer are stored inside a @code{const} RTX to mark them as
5442 constant. Therefore, there is no need to recognize such sums
5443 specifically as legitimate addresses. Normally you would simply
5444 recognize any @code{const} as legitimate.
5446 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5447 sums that are not marked with @code{const}. It assumes that a naked
5448 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5449 naked constant sums as illegitimate addresses, so that none of them will
5450 be given to @code{PRINT_OPERAND_ADDRESS}.
5452 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5453 On some machines, whether a symbolic address is legitimate depends on
5454 the section that the address refers to. On these machines, define the
5455 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5456 into the @code{symbol_ref}, and then check for it here. When you see a
5457 @code{const}, you will have to look inside it to find the
5458 @code{symbol_ref} in order to determine the section. @xref{Assembler
5461 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5462 Some ports are still using a deprecated legacy substitute for
5463 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5467 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5471 and should @code{goto @var{label}} if the address @var{x} is a valid
5472 address on the target machine for a memory operand of mode @var{mode}.
5474 @findex REG_OK_STRICT
5475 Compiler source files that want to use the strict variant of this
5476 macro define the macro @code{REG_OK_STRICT}. You should use an
5477 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
5478 that case and the non-strict variant otherwise.
5480 Using the hook is usually simpler because it limits the number of
5481 files that are recompiled when changes are made.
5484 @defmac TARGET_MEM_CONSTRAINT
5485 A single character to be used instead of the default @code{'m'}
5486 character for general memory addresses. This defines the constraint
5487 letter which matches the memory addresses accepted by
5488 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5489 support new address formats in your back end without changing the
5490 semantics of the @code{'m'} constraint. This is necessary in order to
5491 preserve functionality of inline assembly constructs using the
5492 @code{'m'} constraint.
5495 @defmac FIND_BASE_TERM (@var{x})
5496 A C expression to determine the base term of address @var{x},
5497 or to provide a simplified version of @var{x} from which @file{alias.c}
5498 can easily find the base term. This macro is used in only two places:
5499 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5501 It is always safe for this macro to not be defined. It exists so
5502 that alias analysis can understand machine-dependent addresses.
5504 The typical use of this macro is to handle addresses containing
5505 a label_ref or symbol_ref within an UNSPEC@.
5508 @deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, enum machine_mode @var{mode})
5509 This hook is given an invalid memory address @var{x} for an
5510 operand of mode @var{mode} and should try to return a valid memory
5513 @findex break_out_memory_refs
5514 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5515 and @var{oldx} will be the operand that was given to that function to produce
5518 The code of the hook should not alter the substructure of
5519 @var{x}. If it transforms @var{x} into a more legitimate form, it
5520 should return the new @var{x}.
5522 It is not necessary for this hook to come up with a legitimate address.
5523 The compiler has standard ways of doing so in all cases. In fact, it
5524 is safe to omit this hook or make it return @var{x} if it cannot find
5525 a valid way to legitimize the address. But often a machine-dependent
5526 strategy can generate better code.
5529 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5530 A C compound statement that attempts to replace @var{x}, which is an address
5531 that needs reloading, with a valid memory address for an operand of mode
5532 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5533 It is not necessary to define this macro, but it might be useful for
5534 performance reasons.
5536 For example, on the i386, it is sometimes possible to use a single
5537 reload register instead of two by reloading a sum of two pseudo
5538 registers into a register. On the other hand, for number of RISC
5539 processors offsets are limited so that often an intermediate address
5540 needs to be generated in order to address a stack slot. By defining
5541 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5542 generated for adjacent some stack slots can be made identical, and thus
5545 @emph{Note}: This macro should be used with caution. It is necessary
5546 to know something of how reload works in order to effectively use this,
5547 and it is quite easy to produce macros that build in too much knowledge
5548 of reload internals.
5550 @emph{Note}: This macro must be able to reload an address created by a
5551 previous invocation of this macro. If it fails to handle such addresses
5552 then the compiler may generate incorrect code or abort.
5555 The macro definition should use @code{push_reload} to indicate parts that
5556 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5557 suitable to be passed unaltered to @code{push_reload}.
5559 The code generated by this macro must not alter the substructure of
5560 @var{x}. If it transforms @var{x} into a more legitimate form, it
5561 should assign @var{x} (which will always be a C variable) a new value.
5562 This also applies to parts that you change indirectly by calling
5565 @findex strict_memory_address_p
5566 The macro definition may use @code{strict_memory_address_p} to test if
5567 the address has become legitimate.
5570 If you want to change only a part of @var{x}, one standard way of doing
5571 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5572 single level of rtl. Thus, if the part to be changed is not at the
5573 top level, you'll need to replace first the top level.
5574 It is not necessary for this macro to come up with a legitimate
5575 address; but often a machine-dependent strategy can generate better code.
5578 @deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr})
5579 This hook returns @code{true} if memory address @var{addr} can have
5580 different meanings depending on the machine mode of the memory
5581 reference it is used for or if the address is valid for some modes
5584 Autoincrement and autodecrement addresses typically have mode-dependent
5585 effects because the amount of the increment or decrement is the size
5586 of the operand being addressed. Some machines have other mode-dependent
5587 addresses. Many RISC machines have no mode-dependent addresses.
5589 You may assume that @var{addr} is a valid address for the machine.
5591 The default version of this hook returns @code{false}.
5594 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5595 A C statement or compound statement with a conditional @code{goto
5596 @var{label};} executed if memory address @var{x} (an RTX) can have
5597 different meanings depending on the machine mode of the memory
5598 reference it is used for or if the address is valid for some modes
5601 Autoincrement and autodecrement addresses typically have mode-dependent
5602 effects because the amount of the increment or decrement is the size
5603 of the operand being addressed. Some machines have other mode-dependent
5604 addresses. Many RISC machines have no mode-dependent addresses.
5606 You may assume that @var{addr} is a valid address for the machine.
5608 These are obsolete macros, replaced by the
5609 @code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook.
5612 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x})
5613 This hook returns true if @var{x} is a legitimate constant for a
5614 @var{mode}-mode immediate operand on the target machine. You can assume that
5615 @var{x} satisfies @code{CONSTANT_P}, so you need not check this.
5617 The default definition returns true.
5620 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5621 This hook is used to undo the possibly obfuscating effects of the
5622 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5623 macros. Some backend implementations of these macros wrap symbol
5624 references inside an @code{UNSPEC} rtx to represent PIC or similar
5625 addressing modes. This target hook allows GCC's optimizers to understand
5626 the semantics of these opaque @code{UNSPEC}s by converting them back
5627 into their original form.
5630 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (enum machine_mode @var{mode}, rtx @var{x})
5631 This hook should return true if @var{x} is of a form that cannot (or
5632 should not) be spilled to the constant pool. @var{mode} is the mode
5635 The default version of this hook returns false.
5637 The primary reason to define this hook is to prevent reload from
5638 deciding that a non-legitimate constant would be better reloaded
5639 from the constant pool instead of spilling and reloading a register
5640 holding the constant. This restriction is often true of addresses
5641 of TLS symbols for various targets.
5644 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, const_rtx @var{x})
5645 This hook should return true if pool entries for constant @var{x} can
5646 be placed in an @code{object_block} structure. @var{mode} is the mode
5649 The default version returns false for all constants.
5652 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (unsigned @var{fn}, bool @var{md_fn}, bool @var{sqrt})
5653 This hook should return the DECL of a function that implements reciprocal of
5654 the builtin function with builtin function code @var{fn}, or
5655 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5656 when @var{fn} is a code of a machine-dependent builtin function. When
5657 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5658 of a square root function are performed, and only reciprocals of @code{sqrt}
5662 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5663 This hook should return the DECL of a function @var{f} that given an
5664 address @var{addr} as an argument returns a mask @var{m} that can be
5665 used to extract from two vectors the relevant data that resides in
5666 @var{addr} in case @var{addr} is not properly aligned.
5668 The autovectorizer, when vectorizing a load operation from an address
5669 @var{addr} that may be unaligned, will generate two vector loads from
5670 the two aligned addresses around @var{addr}. It then generates a
5671 @code{REALIGN_LOAD} operation to extract the relevant data from the
5672 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5673 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5674 the third argument, @var{OFF}, defines how the data will be extracted
5675 from these two vectors: if @var{OFF} is 0, then the returned vector is
5676 @var{v2}; otherwise, the returned vector is composed from the last
5677 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5678 @var{OFF} elements of @var{v2}.
5680 If this hook is defined, the autovectorizer will generate a call
5681 to @var{f} (using the DECL tree that this hook returns) and will
5682 use the return value of @var{f} as the argument @var{OFF} to
5683 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5684 should comply with the semantics expected by @code{REALIGN_LOAD}
5686 If this hook is not defined, then @var{addr} will be used as
5687 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5688 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5691 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN (tree @var{x})
5692 This hook should return the DECL of a function @var{f} that implements
5693 widening multiplication of the even elements of two input vectors of type @var{x}.
5695 If this hook is defined, the autovectorizer will use it along with the
5696 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD} target hook when vectorizing
5697 widening multiplication in cases that the order of the results does not have to be
5698 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5699 @code{widen_mult_hi/lo} idioms will be used.
5702 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_ODD (tree @var{x})
5703 This hook should return the DECL of a function @var{f} that implements
5704 widening multiplication of the odd elements of two input vectors of type @var{x}.
5706 If this hook is defined, the autovectorizer will use it along with the
5707 @code{TARGET_VECTORIZE_BUILTIN_MUL_WIDEN_EVEN} target hook when vectorizing
5708 widening multiplication in cases that the order of the results does not have to be
5709 preserved (e.g.@: used only by a reduction computation). Otherwise, the
5710 @code{widen_mult_hi/lo} idioms will be used.
5713 @deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign})
5714 Returns cost of different scalar or vector statements for vectorization cost model.
5715 For vector memory operations the cost may depend on type (@var{vectype}) and
5716 misalignment value (@var{misalign}).
5719 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed})
5720 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5723 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST_OK (enum @var{machine_mode}, const unsigned char *@var{sel})
5724 Return true if a vector created for @code{vec_perm_const} is valid.
5727 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (unsigned @var{code}, tree @var{dest_type}, tree @var{src_type})
5728 This hook should return the DECL of a function that implements conversion of the
5729 input vector of type @var{src_type} to type @var{dest_type}.
5730 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5731 specifies how the conversion is to be applied
5732 (truncation, rounding, etc.).
5734 If this hook is defined, the autovectorizer will use the
5735 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5736 conversion. Otherwise, it will return @code{NULL_TREE}.
5739 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in})
5740 This hook should return the decl of a function that implements the
5741 vectorized variant of the builtin function with builtin function code
5742 @var{code} or @code{NULL_TREE} if such a function is not available.
5743 The value of @var{fndecl} is the builtin function declaration. The
5744 return type of the vectorized function shall be of vector type
5745 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5748 @deftypefn {Target Hook} bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT (enum machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed})
5749 This hook should return true if the target supports misaligned vector
5750 store/load of a specific factor denoted in the @var{misalignment}
5751 parameter. The vector store/load should be of machine mode @var{mode} and
5752 the elements in the vectors should be of type @var{type}. @var{is_packed}
5753 parameter is true if the memory access is defined in a packed struct.
5756 @deftypefn {Target Hook} {enum machine_mode} TARGET_VECTORIZE_PREFERRED_SIMD_MODE (enum machine_mode @var{mode})
5757 This hook should return the preferred mode for vectorizing scalar
5758 mode @var{mode}. The default is
5759 equal to @code{word_mode}, because the vectorizer can do some
5760 transformations even in absence of specialized @acronym{SIMD} hardware.
5763 @deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES (void)
5764 This hook should return a mask of sizes that should be iterated over
5765 after trying to autovectorize using the vector size derived from the
5766 mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5767 The default is zero which means to not iterate over other vector sizes.
5770 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_LOAD (tree)
5771 This hook should return the built-in decl needed to load a vector of the given type within a transaction.
5774 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_STORE (tree)
5775 This hook should return the built-in decl needed to store a vector of the given type within a transaction.
5778 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale})
5779 Target builtin that implements vector gather operation. @var{mem_vectype}
5780 is the vector type of the load and @var{index_type} is scalar type of
5781 the index, scaled by @var{scale}.
5782 The default is @code{NULL_TREE} which means to not vectorize gather
5786 @node Anchored Addresses
5787 @section Anchored Addresses
5788 @cindex anchored addresses
5789 @cindex @option{-fsection-anchors}
5791 GCC usually addresses every static object as a separate entity.
5792 For example, if we have:
5796 int foo (void) @{ return a + b + c; @}
5799 the code for @code{foo} will usually calculate three separate symbolic
5800 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5801 it would be better to calculate just one symbolic address and access
5802 the three variables relative to it. The equivalent pseudocode would
5808 register int *xr = &x;
5809 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5813 (which isn't valid C). We refer to shared addresses like @code{x} as
5814 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5816 The hooks below describe the target properties that GCC needs to know
5817 in order to make effective use of section anchors. It won't use
5818 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5819 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5821 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5822 The minimum offset that should be applied to a section anchor.
5823 On most targets, it should be the smallest offset that can be
5824 applied to a base register while still giving a legitimate address
5825 for every mode. The default value is 0.
5828 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5829 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5830 offset that should be applied to section anchors. The default
5834 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5835 Write the assembly code to define section anchor @var{x}, which is a
5836 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5837 The hook is called with the assembly output position set to the beginning
5838 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5840 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5841 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5842 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5843 is @code{NULL}, which disables the use of section anchors altogether.
5846 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x})
5847 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5848 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5849 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5851 The default version is correct for most targets, but you might need to
5852 intercept this hook to handle things like target-specific attributes
5853 or target-specific sections.
5856 @node Condition Code
5857 @section Condition Code Status
5858 @cindex condition code status
5860 The macros in this section can be split in two families, according to the
5861 two ways of representing condition codes in GCC.
5863 The first representation is the so called @code{(cc0)} representation
5864 (@pxref{Jump Patterns}), where all instructions can have an implicit
5865 clobber of the condition codes. The second is the condition code
5866 register representation, which provides better schedulability for
5867 architectures that do have a condition code register, but on which
5868 most instructions do not affect it. The latter category includes
5871 The implicit clobbering poses a strong restriction on the placement of
5872 the definition and use of the condition code, which need to be in adjacent
5873 insns for machines using @code{(cc0)}. This can prevent important
5874 optimizations on some machines. For example, on the IBM RS/6000, there
5875 is a delay for taken branches unless the condition code register is set
5876 three instructions earlier than the conditional branch. The instruction
5877 scheduler cannot perform this optimization if it is not permitted to
5878 separate the definition and use of the condition code register.
5880 For this reason, it is possible and suggested to use a register to
5881 represent the condition code for new ports. If there is a specific
5882 condition code register in the machine, use a hard register. If the
5883 condition code or comparison result can be placed in any general register,
5884 or if there are multiple condition registers, use a pseudo register.
5885 Registers used to store the condition code value will usually have a mode
5886 that is in class @code{MODE_CC}.
5888 Alternatively, you can use @code{BImode} if the comparison operator is
5889 specified already in the compare instruction. In this case, you are not
5890 interested in most macros in this section.
5893 * CC0 Condition Codes:: Old style representation of condition codes.
5894 * MODE_CC Condition Codes:: Modern representation of condition codes.
5895 * Cond Exec Macros:: Macros to control conditional execution.
5898 @node CC0 Condition Codes
5899 @subsection Representation of condition codes using @code{(cc0)}
5903 The file @file{conditions.h} defines a variable @code{cc_status} to
5904 describe how the condition code was computed (in case the interpretation of
5905 the condition code depends on the instruction that it was set by). This
5906 variable contains the RTL expressions on which the condition code is
5907 currently based, and several standard flags.
5909 Sometimes additional machine-specific flags must be defined in the machine
5910 description header file. It can also add additional machine-specific
5911 information by defining @code{CC_STATUS_MDEP}.
5913 @defmac CC_STATUS_MDEP
5914 C code for a data type which is used for declaring the @code{mdep}
5915 component of @code{cc_status}. It defaults to @code{int}.
5917 This macro is not used on machines that do not use @code{cc0}.
5920 @defmac CC_STATUS_MDEP_INIT
5921 A C expression to initialize the @code{mdep} field to ``empty''.
5922 The default definition does nothing, since most machines don't use
5923 the field anyway. If you want to use the field, you should probably
5924 define this macro to initialize it.
5926 This macro is not used on machines that do not use @code{cc0}.
5929 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5930 A C compound statement to set the components of @code{cc_status}
5931 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5932 this macro's responsibility to recognize insns that set the condition
5933 code as a byproduct of other activity as well as those that explicitly
5936 This macro is not used on machines that do not use @code{cc0}.
5938 If there are insns that do not set the condition code but do alter
5939 other machine registers, this macro must check to see whether they
5940 invalidate the expressions that the condition code is recorded as
5941 reflecting. For example, on the 68000, insns that store in address
5942 registers do not set the condition code, which means that usually
5943 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5944 insns. But suppose that the previous insn set the condition code
5945 based on location @samp{a4@@(102)} and the current insn stores a new
5946 value in @samp{a4}. Although the condition code is not changed by
5947 this, it will no longer be true that it reflects the contents of
5948 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5949 @code{cc_status} in this case to say that nothing is known about the
5950 condition code value.
5952 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5953 with the results of peephole optimization: insns whose patterns are
5954 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5955 constants which are just the operands. The RTL structure of these
5956 insns is not sufficient to indicate what the insns actually do. What
5957 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5958 @code{CC_STATUS_INIT}.
5960 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5961 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5962 @samp{cc}. This avoids having detailed information about patterns in
5963 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5966 @node MODE_CC Condition Codes
5967 @subsection Representation of condition codes using registers
5971 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5972 On many machines, the condition code may be produced by other instructions
5973 than compares, for example the branch can use directly the condition
5974 code set by a subtract instruction. However, on some machines
5975 when the condition code is set this way some bits (such as the overflow
5976 bit) are not set in the same way as a test instruction, so that a different
5977 branch instruction must be used for some conditional branches. When
5978 this happens, use the machine mode of the condition code register to
5979 record different formats of the condition code register. Modes can
5980 also be used to record which compare instruction (e.g. a signed or an
5981 unsigned comparison) produced the condition codes.
5983 If other modes than @code{CCmode} are required, add them to
5984 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
5985 a mode given an operand of a compare. This is needed because the modes
5986 have to be chosen not only during RTL generation but also, for example,
5987 by instruction combination. The result of @code{SELECT_CC_MODE} should
5988 be consistent with the mode used in the patterns; for example to support
5989 the case of the add on the SPARC discussed above, we have the pattern
5993 [(set (reg:CC_NOOV 0)
5995 (plus:SI (match_operand:SI 0 "register_operand" "%r")
5996 (match_operand:SI 1 "arith_operand" "rI"))
6003 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
6004 for comparisons whose argument is a @code{plus}:
6007 #define SELECT_CC_MODE(OP,X,Y) \
6008 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
6009 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
6010 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
6011 || GET_CODE (X) == NEG) \
6012 ? CC_NOOVmode : CCmode))
6015 Another reason to use modes is to retain information on which operands
6016 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6019 You should define this macro if and only if you define extra CC modes
6020 in @file{@var{machine}-modes.def}.
6023 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
6024 On some machines not all possible comparisons are defined, but you can
6025 convert an invalid comparison into a valid one. For example, the Alpha
6026 does not have a @code{GT} comparison, but you can use an @code{LT}
6027 comparison instead and swap the order of the operands.
6029 On such machines, define this macro to be a C statement to do any
6030 required conversions. @var{code} is the initial comparison code
6031 and @var{op0} and @var{op1} are the left and right operands of the
6032 comparison, respectively. You should modify @var{code}, @var{op0}, and
6033 @var{op1} as required.
6035 GCC will not assume that the comparison resulting from this macro is
6036 valid but will see if the resulting insn matches a pattern in the
6039 You need not define this macro if it would never change the comparison
6043 @defmac REVERSIBLE_CC_MODE (@var{mode})
6044 A C expression whose value is one if it is always safe to reverse a
6045 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6046 can ever return @var{mode} for a floating-point inequality comparison,
6047 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6049 You need not define this macro if it would always returns zero or if the
6050 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6051 For example, here is the definition used on the SPARC, where floating-point
6052 inequality comparisons are always given @code{CCFPEmode}:
6055 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
6059 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6060 A C expression whose value is reversed condition code of the @var{code} for
6061 comparison done in CC_MODE @var{mode}. The macro is used only in case
6062 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6063 machine has some non-standard way how to reverse certain conditionals. For
6064 instance in case all floating point conditions are non-trapping, compiler may
6065 freely convert unordered compares to ordered one. Then definition may look
6069 #define REVERSE_CONDITION(CODE, MODE) \
6070 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6071 : reverse_condition_maybe_unordered (CODE))
6075 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2})
6076 On targets which do not use @code{(cc0)}, and which use a hard
6077 register rather than a pseudo-register to hold condition codes, the
6078 regular CSE passes are often not able to identify cases in which the
6079 hard register is set to a common value. Use this hook to enable a
6080 small pass which optimizes such cases. This hook should return true
6081 to enable this pass, and it should set the integers to which its
6082 arguments point to the hard register numbers used for condition codes.
6083 When there is only one such register, as is true on most systems, the
6084 integer pointed to by @var{p2} should be set to
6085 @code{INVALID_REGNUM}.
6087 The default version of this hook returns false.
6090 @deftypefn {Target Hook} {enum machine_mode} TARGET_CC_MODES_COMPATIBLE (enum machine_mode @var{m1}, enum machine_mode @var{m2})
6091 On targets which use multiple condition code modes in class
6092 @code{MODE_CC}, it is sometimes the case that a comparison can be
6093 validly done in more than one mode. On such a system, define this
6094 target hook to take two mode arguments and to return a mode in which
6095 both comparisons may be validly done. If there is no such mode,
6096 return @code{VOIDmode}.
6098 The default version of this hook checks whether the modes are the
6099 same. If they are, it returns that mode. If they are different, it
6100 returns @code{VOIDmode}.
6103 @node Cond Exec Macros
6104 @subsection Macros to control conditional execution
6105 @findex conditional execution
6108 There is one macro that may need to be defined for targets
6109 supporting conditional execution, independent of how they
6110 represent conditional branches.
6112 @defmac REVERSE_CONDEXEC_PREDICATES_P (@var{op1}, @var{op2})
6113 A C expression that returns true if the conditional execution predicate
6114 @var{op1}, a comparison operation, is the inverse of @var{op2} and vice
6115 versa. Define this to return 0 if the target has conditional execution
6116 predicates that cannot be reversed safely. There is no need to validate
6117 that the arguments of op1 and op2 are the same, this is done separately.
6118 If no expansion is specified, this macro is defined as follows:
6121 #define REVERSE_CONDEXEC_PREDICATES_P (x, y) \
6122 (GET_CODE ((x)) == reversed_comparison_code ((y), NULL))
6127 @section Describing Relative Costs of Operations
6128 @cindex costs of instructions
6129 @cindex relative costs
6130 @cindex speed of instructions
6132 These macros let you describe the relative speed of various operations
6133 on the target machine.
6135 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6136 A C expression for the cost of moving data of mode @var{mode} from a
6137 register in class @var{from} to one in class @var{to}. The classes are
6138 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6139 value of 2 is the default; other values are interpreted relative to
6142 It is not required that the cost always equal 2 when @var{from} is the
6143 same as @var{to}; on some machines it is expensive to move between
6144 registers if they are not general registers.
6146 If reload sees an insn consisting of a single @code{set} between two
6147 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6148 classes returns a value of 2, reload does not check to ensure that the
6149 constraints of the insn are met. Setting a cost of other than 2 will
6150 allow reload to verify that the constraints are met. You should do this
6151 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6153 These macros are obsolete, new ports should use the target hook
6154 @code{TARGET_REGISTER_MOVE_COST} instead.
6157 @deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (enum machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to})
6158 This target hook should return the cost of moving data of mode @var{mode}
6159 from a register in class @var{from} to one in class @var{to}. The classes
6160 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6161 A value of 2 is the default; other values are interpreted relative to
6164 It is not required that the cost always equal 2 when @var{from} is the
6165 same as @var{to}; on some machines it is expensive to move between
6166 registers if they are not general registers.
6168 If reload sees an insn consisting of a single @code{set} between two
6169 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6170 classes returns a value of 2, reload does not check to ensure that the
6171 constraints of the insn are met. Setting a cost of other than 2 will
6172 allow reload to verify that the constraints are met. You should do this
6173 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6175 The default version of this function returns 2.
6178 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6179 A C expression for the cost of moving data of mode @var{mode} between a
6180 register of class @var{class} and memory; @var{in} is zero if the value
6181 is to be written to memory, nonzero if it is to be read in. This cost
6182 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6183 registers and memory is more expensive than between two registers, you
6184 should define this macro to express the relative cost.
6186 If you do not define this macro, GCC uses a default cost of 4 plus
6187 the cost of copying via a secondary reload register, if one is
6188 needed. If your machine requires a secondary reload register to copy
6189 between memory and a register of @var{class} but the reload mechanism is
6190 more complex than copying via an intermediate, define this macro to
6191 reflect the actual cost of the move.
6193 GCC defines the function @code{memory_move_secondary_cost} if
6194 secondary reloads are needed. It computes the costs due to copying via
6195 a secondary register. If your machine copies from memory using a
6196 secondary register in the conventional way but the default base value of
6197 4 is not correct for your machine, define this macro to add some other
6198 value to the result of that function. The arguments to that function
6199 are the same as to this macro.
6201 These macros are obsolete, new ports should use the target hook
6202 @code{TARGET_MEMORY_MOVE_COST} instead.
6205 @deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (enum machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in})
6206 This target hook should return the cost of moving data of mode @var{mode}
6207 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6208 if the value is to be written to memory, @code{true} if it is to be read in.
6209 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6210 If moving between registers and memory is more expensive than between two
6211 registers, you should add this target hook to express the relative cost.
6213 If you do not add this target hook, GCC uses a default cost of 4 plus
6214 the cost of copying via a secondary reload register, if one is
6215 needed. If your machine requires a secondary reload register to copy
6216 between memory and a register of @var{rclass} but the reload mechanism is
6217 more complex than copying via an intermediate, use this target hook to
6218 reflect the actual cost of the move.
6220 GCC defines the function @code{memory_move_secondary_cost} if
6221 secondary reloads are needed. It computes the costs due to copying via
6222 a secondary register. If your machine copies from memory using a
6223 secondary register in the conventional way but the default base value of
6224 4 is not correct for your machine, use this target hook to add some other
6225 value to the result of that function. The arguments to that function
6226 are the same as to this target hook.
6229 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6230 A C expression for the cost of a branch instruction. A value of 1 is
6231 the default; other values are interpreted relative to that. Parameter
6232 @var{speed_p} is true when the branch in question should be optimized
6233 for speed. When it is false, @code{BRANCH_COST} should return a value
6234 optimal for code size rather than performance. @var{predictable_p} is
6235 true for well-predicted branches. On many architectures the
6236 @code{BRANCH_COST} can be reduced then.
6239 Here are additional macros which do not specify precise relative costs,
6240 but only that certain actions are more expensive than GCC would
6243 @defmac SLOW_BYTE_ACCESS
6244 Define this macro as a C expression which is nonzero if accessing less
6245 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6246 faster than accessing a word of memory, i.e., if such access
6247 require more than one instruction or if there is no difference in cost
6248 between byte and (aligned) word loads.
6250 When this macro is not defined, the compiler will access a field by
6251 finding the smallest containing object; when it is defined, a fullword
6252 load will be used if alignment permits. Unless bytes accesses are
6253 faster than word accesses, using word accesses is preferable since it
6254 may eliminate subsequent memory access if subsequent accesses occur to
6255 other fields in the same word of the structure, but to different bytes.
6258 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6259 Define this macro to be the value 1 if memory accesses described by the
6260 @var{mode} and @var{alignment} parameters have a cost many times greater
6261 than aligned accesses, for example if they are emulated in a trap
6264 When this macro is nonzero, the compiler will act as if
6265 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6266 moves. This can cause significantly more instructions to be produced.
6267 Therefore, do not set this macro nonzero if unaligned accesses only add a
6268 cycle or two to the time for a memory access.
6270 If the value of this macro is always zero, it need not be defined. If
6271 this macro is defined, it should produce a nonzero value when
6272 @code{STRICT_ALIGNMENT} is nonzero.
6275 @defmac MOVE_RATIO (@var{speed})
6276 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6277 which a sequence of insns should be generated instead of a
6278 string move insn or a library call. Increasing the value will always
6279 make code faster, but eventually incurs high cost in increased code size.
6281 Note that on machines where the corresponding move insn is a
6282 @code{define_expand} that emits a sequence of insns, this macro counts
6283 the number of such sequences.
6285 The parameter @var{speed} is true if the code is currently being
6286 optimized for speed rather than size.
6288 If you don't define this, a reasonable default is used.
6291 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6292 A C expression used to determine whether @code{move_by_pieces} will be used to
6293 copy a chunk of memory, or whether some other block move mechanism
6294 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6295 than @code{MOVE_RATIO}.
6298 @defmac MOVE_MAX_PIECES
6299 A C expression used by @code{move_by_pieces} to determine the largest unit
6300 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6303 @defmac CLEAR_RATIO (@var{speed})
6304 The threshold of number of scalar move insns, @emph{below} which a sequence
6305 of insns should be generated to clear memory instead of a string clear insn
6306 or a library call. Increasing the value will always make code faster, but
6307 eventually incurs high cost in increased code size.
6309 The parameter @var{speed} is true if the code is currently being
6310 optimized for speed rather than size.
6312 If you don't define this, a reasonable default is used.
6315 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6316 A C expression used to determine whether @code{clear_by_pieces} will be used
6317 to clear a chunk of memory, or whether some other block clear mechanism
6318 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6319 than @code{CLEAR_RATIO}.
6322 @defmac SET_RATIO (@var{speed})
6323 The threshold of number of scalar move insns, @emph{below} which a sequence
6324 of insns should be generated to set memory to a constant value, instead of
6325 a block set insn or a library call.
6326 Increasing the value will always make code faster, but
6327 eventually incurs high cost in increased code size.
6329 The parameter @var{speed} is true if the code is currently being
6330 optimized for speed rather than size.
6332 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6335 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6336 A C expression used to determine whether @code{store_by_pieces} will be
6337 used to set a chunk of memory to a constant value, or whether some
6338 other mechanism will be used. Used by @code{__builtin_memset} when
6339 storing values other than constant zero.
6340 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6341 than @code{SET_RATIO}.
6344 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6345 A C expression used to determine whether @code{store_by_pieces} will be
6346 used to set a chunk of memory to a constant string value, or whether some
6347 other mechanism will be used. Used by @code{__builtin_strcpy} when
6348 called with a constant source string.
6349 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6350 than @code{MOVE_RATIO}.
6353 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6354 A C expression used to determine whether a load postincrement is a good
6355 thing to use for a given mode. Defaults to the value of
6356 @code{HAVE_POST_INCREMENT}.
6359 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6360 A C expression used to determine whether a load postdecrement is a good
6361 thing to use for a given mode. Defaults to the value of
6362 @code{HAVE_POST_DECREMENT}.
6365 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6366 A C expression used to determine whether a load preincrement is a good
6367 thing to use for a given mode. Defaults to the value of
6368 @code{HAVE_PRE_INCREMENT}.
6371 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6372 A C expression used to determine whether a load predecrement is a good
6373 thing to use for a given mode. Defaults to the value of
6374 @code{HAVE_PRE_DECREMENT}.
6377 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6378 A C expression used to determine whether a store postincrement is a good
6379 thing to use for a given mode. Defaults to the value of
6380 @code{HAVE_POST_INCREMENT}.
6383 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6384 A C expression used to determine whether a store postdecrement is a good
6385 thing to use for a given mode. Defaults to the value of
6386 @code{HAVE_POST_DECREMENT}.
6389 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6390 This macro is used to determine whether a store preincrement is a good
6391 thing to use for a given mode. Defaults to the value of
6392 @code{HAVE_PRE_INCREMENT}.
6395 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6396 This macro is used to determine whether a store predecrement is a good
6397 thing to use for a given mode. Defaults to the value of
6398 @code{HAVE_PRE_DECREMENT}.
6401 @defmac NO_FUNCTION_CSE
6402 Define this macro if it is as good or better to call a constant
6403 function address than to call an address kept in a register.
6406 @defmac RANGE_TEST_NON_SHORT_CIRCUIT
6407 Define this macro if a non-short-circuit operation produced by
6408 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6409 @code{BRANCH_COST} is greater than or equal to the value 2.
6412 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int @var{opno}, int *@var{total}, bool @var{speed})
6413 This target hook describes the relative costs of RTL expressions.
6415 The cost may depend on the precise form of the expression, which is
6416 available for examination in @var{x}, and the fact that @var{x} appears
6417 as operand @var{opno} of an expression with rtx code @var{outer_code}.
6418 That is, the hook can assume that there is some rtx @var{y} such
6419 that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that
6420 either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or
6421 (b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}.
6423 @var{code} is @var{x}'s expression code---redundant, since it can be
6424 obtained with @code{GET_CODE (@var{x})}.
6426 In implementing this hook, you can use the construct
6427 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6430 On entry to the hook, @code{*@var{total}} contains a default estimate
6431 for the cost of the expression. The hook should modify this value as
6432 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6433 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6434 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6436 When optimizing for code size, i.e.@: when @code{speed} is
6437 false, this target hook should be used to estimate the relative
6438 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6440 The hook returns true when all subexpressions of @var{x} have been
6441 processed, and false when @code{rtx_cost} should recurse.
6444 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, bool @var{speed})
6445 This hook computes the cost of an addressing mode that contains
6446 @var{address}. If not defined, the cost is computed from
6447 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6449 For most CISC machines, the default cost is a good approximation of the
6450 true cost of the addressing mode. However, on RISC machines, all
6451 instructions normally have the same length and execution time. Hence
6452 all addresses will have equal costs.
6454 In cases where more than one form of an address is known, the form with
6455 the lowest cost will be used. If multiple forms have the same, lowest,
6456 cost, the one that is the most complex will be used.
6458 For example, suppose an address that is equal to the sum of a register
6459 and a constant is used twice in the same basic block. When this macro
6460 is not defined, the address will be computed in a register and memory
6461 references will be indirect through that register. On machines where
6462 the cost of the addressing mode containing the sum is no higher than
6463 that of a simple indirect reference, this will produce an additional
6464 instruction and possibly require an additional register. Proper
6465 specification of this macro eliminates this overhead for such machines.
6467 This hook is never called with an invalid address.
6469 On machines where an address involving more than one register is as
6470 cheap as an address computation involving only one register, defining
6471 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6472 be live over a region of code where only one would have been if
6473 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6474 should be considered in the definition of this macro. Equivalent costs
6475 should probably only be given to addresses with different numbers of
6476 registers on machines with lots of registers.
6480 @section Adjusting the Instruction Scheduler
6482 The instruction scheduler may need a fair amount of machine-specific
6483 adjustment in order to produce good code. GCC provides several target
6484 hooks for this purpose. It is usually enough to define just a few of
6485 them: try the first ones in this list first.
6487 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
6488 This hook returns the maximum number of instructions that can ever
6489 issue at the same time on the target machine. The default is one.
6490 Although the insn scheduler can define itself the possibility of issue
6491 an insn on the same cycle, the value can serve as an additional
6492 constraint to issue insns on the same simulated processor cycle (see
6493 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6494 This value must be constant over the entire compilation. If you need
6495 it to vary depending on what the instructions are, you must use
6496 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6499 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
6500 This hook is executed by the scheduler after it has scheduled an insn
6501 from the ready list. It should return the number of insns which can
6502 still be issued in the current cycle. The default is
6503 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6504 @code{USE}, which normally are not counted against the issue rate.
6505 You should define this hook if some insns take more machine resources
6506 than others, so that fewer insns can follow them in the same cycle.
6507 @var{file} is either a null pointer, or a stdio stream to write any
6508 debug output to. @var{verbose} is the verbose level provided by
6509 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6513 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
6514 This function corrects the value of @var{cost} based on the
6515 relationship between @var{insn} and @var{dep_insn} through the
6516 dependence @var{link}. It should return the new value. The default
6517 is to make no adjustment to @var{cost}. This can be used for example
6518 to specify to the scheduler using the traditional pipeline description
6519 that an output- or anti-dependence does not incur the same cost as a
6520 data-dependence. If the scheduler using the automaton based pipeline
6521 description, the cost of anti-dependence is zero and the cost of
6522 output-dependence is maximum of one and the difference of latency
6523 times of the first and the second insns. If these values are not
6524 acceptable, you could use the hook to modify them too. See also
6525 @pxref{Processor pipeline description}.
6528 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
6529 This hook adjusts the integer scheduling priority @var{priority} of
6530 @var{insn}. It should return the new priority. Increase the priority to
6531 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6532 later. Do not define this hook if you do not need to adjust the
6533 scheduling priorities of insns.
6536 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6537 This hook is executed by the scheduler after it has scheduled the ready
6538 list, to allow the machine description to reorder it (for example to
6539 combine two small instructions together on @samp{VLIW} machines).
6540 @var{file} is either a null pointer, or a stdio stream to write any
6541 debug output to. @var{verbose} is the verbose level provided by
6542 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6543 list of instructions that are ready to be scheduled. @var{n_readyp} is
6544 a pointer to the number of elements in the ready list. The scheduler
6545 reads the ready list in reverse order, starting with
6546 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6547 is the timer tick of the scheduler. You may modify the ready list and
6548 the number of ready insns. The return value is the number of insns that
6549 can issue this cycle; normally this is just @code{issue_rate}. See also
6550 @samp{TARGET_SCHED_REORDER2}.
6553 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6554 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6555 function is called whenever the scheduler starts a new cycle. This one
6556 is called once per iteration over a cycle, immediately after
6557 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6558 return the number of insns to be scheduled in the same cycle. Defining
6559 this hook can be useful if there are frequent situations where
6560 scheduling one insn causes other insns to become ready in the same
6561 cycle. These other insns can then be taken into account properly.
6564 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
6565 This hook is called after evaluation forward dependencies of insns in
6566 chain given by two parameter values (@var{head} and @var{tail}
6567 correspondingly) but before insns scheduling of the insn chain. For
6568 example, it can be used for better insn classification if it requires
6569 analysis of dependencies. This hook can use backward and forward
6570 dependencies of the insn scheduler because they are already
6574 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
6575 This hook is executed by the scheduler at the beginning of each block of
6576 instructions that are to be scheduled. @var{file} is either a null
6577 pointer, or a stdio stream to write any debug output to. @var{verbose}
6578 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6579 @var{max_ready} is the maximum number of insns in the current scheduling
6580 region that can be live at the same time. This can be used to allocate
6581 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6584 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
6585 This hook is executed by the scheduler at the end of each block of
6586 instructions that are to be scheduled. It can be used to perform
6587 cleanup of any actions done by the other scheduling hooks. @var{file}
6588 is either a null pointer, or a stdio stream to write any debug output
6589 to. @var{verbose} is the verbose level provided by
6590 @option{-fsched-verbose-@var{n}}.
6593 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
6594 This hook is executed by the scheduler after function level initializations.
6595 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6596 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6597 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6600 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
6601 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6602 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6603 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6606 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
6607 The hook returns an RTL insn. The automaton state used in the
6608 pipeline hazard recognizer is changed as if the insn were scheduled
6609 when the new simulated processor cycle starts. Usage of the hook may
6610 simplify the automaton pipeline description for some @acronym{VLIW}
6611 processors. If the hook is defined, it is used only for the automaton
6612 based pipeline description. The default is not to change the state
6613 when the new simulated processor cycle starts.
6616 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
6617 The hook can be used to initialize data used by the previous hook.
6620 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
6621 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6622 to changed the state as if the insn were scheduled when the new
6623 simulated processor cycle finishes.
6626 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
6627 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6628 used to initialize data used by the previous hook.
6631 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)
6632 The hook to notify target that the current simulated cycle is about to finish.
6633 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6634 to change the state in more complicated situations - e.g., when advancing
6635 state on a single insn is not enough.
6638 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)
6639 The hook to notify target that new simulated cycle has just started.
6640 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6641 to change the state in more complicated situations - e.g., when advancing
6642 state on a single insn is not enough.
6645 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
6646 This hook controls better choosing an insn from the ready insn queue
6647 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6648 chooses the first insn from the queue. If the hook returns a positive
6649 value, an additional scheduler code tries all permutations of
6650 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6651 subsequent ready insns to choose an insn whose issue will result in
6652 maximal number of issued insns on the same cycle. For the
6653 @acronym{VLIW} processor, the code could actually solve the problem of
6654 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6655 rules of @acronym{VLIW} packing are described in the automaton.
6657 This code also could be used for superscalar @acronym{RISC}
6658 processors. Let us consider a superscalar @acronym{RISC} processor
6659 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6660 @var{B}, some insns can be executed only in pipelines @var{B} or
6661 @var{C}, and one insn can be executed in pipeline @var{B}. The
6662 processor may issue the 1st insn into @var{A} and the 2nd one into
6663 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6664 until the next cycle. If the scheduler issues the 3rd insn the first,
6665 the processor could issue all 3 insns per cycle.
6667 Actually this code demonstrates advantages of the automaton based
6668 pipeline hazard recognizer. We try quickly and easy many insn
6669 schedules to choose the best one.
6671 The default is no multipass scheduling.
6674 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx @var{insn})
6676 This hook controls what insns from the ready insn queue will be
6677 considered for the multipass insn scheduling. If the hook returns
6678 zero for @var{insn}, the insn will be not chosen to
6681 The default is that any ready insns can be chosen to be issued.
6684 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void *@var{data}, char *@var{ready_try}, int @var{n_ready}, bool @var{first_cycle_insn_p})
6685 This hook prepares the target backend for a new round of multipass
6689 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void *@var{data}, char *@var{ready_try}, int @var{n_ready}, rtx @var{insn}, const void *@var{prev_data})
6690 This hook is called when multipass scheduling evaluates instruction INSN.
6693 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, char *@var{ready_try}, int @var{n_ready})
6694 This is called when multipass scheduling backtracks from evaluation of
6698 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data})
6699 This hook notifies the target about the result of the concluded current
6700 round of multipass scheduling.
6703 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data})
6704 This hook initializes target-specific data used in multipass scheduling.
6707 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data})
6708 This hook finalizes target-specific data used in multipass scheduling.
6711 @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})
6712 This hook is called by the insn scheduler before issuing @var{insn}
6713 on cycle @var{clock}. If the hook returns nonzero,
6714 @var{insn} is not issued on this processor cycle. Instead,
6715 the processor cycle is advanced. If *@var{sort_p}
6716 is zero, the insn ready queue is not sorted on the new cycle
6717 start as usually. @var{dump} and @var{verbose} specify the file and
6718 verbosity level to use for debugging output.
6719 @var{last_clock} and @var{clock} are, respectively, the
6720 processor cycle on which the previous insn has been issued,
6721 and the current processor cycle.
6724 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance})
6725 This hook is used to define which dependences are considered costly by
6726 the target, so costly that it is not advisable to schedule the insns that
6727 are involved in the dependence too close to one another. The parameters
6728 to this hook are as follows: The first parameter @var{_dep} is the dependence
6729 being evaluated. The second parameter @var{cost} is the cost of the
6730 dependence as estimated by the scheduler, and the third
6731 parameter @var{distance} is the distance in cycles between the two insns.
6732 The hook returns @code{true} if considering the distance between the two
6733 insns the dependence between them is considered costly by the target,
6734 and @code{false} otherwise.
6736 Defining this hook can be useful in multiple-issue out-of-order machines,
6737 where (a) it's practically hopeless to predict the actual data/resource
6738 delays, however: (b) there's a better chance to predict the actual grouping
6739 that will be formed, and (c) correctly emulating the grouping can be very
6740 important. In such targets one may want to allow issuing dependent insns
6741 closer to one another---i.e., closer than the dependence distance; however,
6742 not in cases of ``costly dependences'', which this hooks allows to define.
6745 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6746 This hook is called by the insn scheduler after emitting a new instruction to
6747 the instruction stream. The hook notifies a target backend to extend its
6748 per instruction data structures.
6751 @deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
6752 Return a pointer to a store large enough to hold target scheduling context.
6755 @deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p})
6756 Initialize store pointed to by @var{tc} to hold target scheduling context.
6757 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6758 beginning of the block. Otherwise, copy the current context into @var{tc}.
6761 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc})
6762 Copy target scheduling context pointed to by @var{tc} to the current context.
6765 @deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc})
6766 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6769 @deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc})
6770 Deallocate a store for target scheduling context pointed to by @var{tc}.
6773 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat})
6774 This hook is called by the insn scheduler when @var{insn} has only
6775 speculative dependencies and therefore can be scheduled speculatively.
6776 The hook is used to check if the pattern of @var{insn} has a speculative
6777 version and, in case of successful check, to generate that speculative
6778 pattern. The hook should return 1, if the instruction has a speculative form,
6779 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6780 speculation. If the return value equals 1 then @var{new_pat} is assigned
6781 the generated speculative pattern.
6784 @deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (int @var{dep_status})
6785 This hook is called by the insn scheduler during generation of recovery code
6786 for @var{insn}. It should return @code{true}, if the corresponding check
6787 instruction should branch to recovery code, or @code{false} otherwise.
6790 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p})
6791 This hook is called by the insn scheduler to generate a pattern for recovery
6792 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6793 speculative instruction for which the check should be generated.
6794 @var{label} is either a label of a basic block, where recovery code should
6795 be emitted, or a null pointer, when requested check doesn't branch to
6796 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6797 a pattern for a branchy check corresponding to a simple check denoted by
6798 @var{insn} should be generated. In this case @var{label} can't be null.
6801 @deftypefn {Target Hook} bool TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (const_rtx @var{insn})
6802 This hook is used as a workaround for
6803 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6804 called on the first instruction of the ready list. The hook is used to
6805 discard speculative instructions that stand first in the ready list from
6806 being scheduled on the current cycle. If the hook returns @code{false},
6807 @var{insn} will not be chosen to be issued.
6808 For non-speculative instructions,
6809 the hook should always return @code{true}. For example, in the ia64 backend
6810 the hook is used to cancel data speculative insns when the ALAT table
6814 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info})
6815 This hook is used by the insn scheduler to find out what features should be
6817 The structure *@var{spec_info} should be filled in by the target.
6818 The structure describes speculation types that can be used in the scheduler.
6821 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
6822 This hook is called by the swing modulo scheduler to calculate a
6823 resource-based lower bound which is based on the resources available in
6824 the machine and the resources required by each instruction. The target
6825 backend can use @var{g} to calculate such bound. A very simple lower
6826 bound will be used in case this hook is not implemented: the total number
6827 of instructions divided by the issue rate.
6830 @deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx @var{insn}, int @var{x})
6831 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6832 is supported in hardware and the condition specified in the parameter is true.
6835 @deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx @var{insn}, int @var{x})
6836 This hook is called by Haifa Scheduler. It performs the operation specified
6837 in its second parameter.
6840 @deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE
6841 True if the processor has an exposed pipeline, which means that not just
6842 the order of instructions is important for correctness when scheduling, but
6843 also the latencies of operations.
6846 @deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, enum machine_mode @var{mode})
6847 This hook is called by tree reassociator to determine a level of
6848 parallelism required in output calculations chain.
6852 @section Dividing the Output into Sections (Texts, Data, @dots{})
6853 @c the above section title is WAY too long. maybe cut the part between
6854 @c the (...)? --mew 10feb93
6856 An object file is divided into sections containing different types of
6857 data. In the most common case, there are three sections: the @dfn{text
6858 section}, which holds instructions and read-only data; the @dfn{data
6859 section}, which holds initialized writable data; and the @dfn{bss
6860 section}, which holds uninitialized data. Some systems have other kinds
6863 @file{varasm.c} provides several well-known sections, such as
6864 @code{text_section}, @code{data_section} and @code{bss_section}.
6865 The normal way of controlling a @code{@var{foo}_section} variable
6866 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6867 as described below. The macros are only read once, when @file{varasm.c}
6868 initializes itself, so their values must be run-time constants.
6869 They may however depend on command-line flags.
6871 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6872 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6873 to be string literals.
6875 Some assemblers require a different string to be written every time a
6876 section is selected. If your assembler falls into this category, you
6877 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6878 @code{get_unnamed_section} to set up the sections.
6880 You must always create a @code{text_section}, either by defining
6881 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6882 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6883 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6884 create a distinct @code{readonly_data_section}, the default is to
6885 reuse @code{text_section}.
6887 All the other @file{varasm.c} sections are optional, and are null
6888 if the target does not provide them.
6890 @defmac TEXT_SECTION_ASM_OP
6891 A C expression whose value is a string, including spacing, containing the
6892 assembler operation that should precede instructions and read-only data.
6893 Normally @code{"\t.text"} is right.
6896 @defmac HOT_TEXT_SECTION_NAME
6897 If defined, a C string constant for the name of the section containing most
6898 frequently executed functions of the program. If not defined, GCC will provide
6899 a default definition if the target supports named sections.
6902 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6903 If defined, a C string constant for the name of the section containing unlikely
6904 executed functions in the program.
6907 @defmac DATA_SECTION_ASM_OP
6908 A C expression whose value is a string, including spacing, containing the
6909 assembler operation to identify the following data as writable initialized
6910 data. Normally @code{"\t.data"} is right.
6913 @defmac SDATA_SECTION_ASM_OP
6914 If defined, a C expression whose value is a string, including spacing,
6915 containing the assembler operation to identify the following data as
6916 initialized, writable small data.
6919 @defmac READONLY_DATA_SECTION_ASM_OP
6920 A C expression whose value is a string, including spacing, containing the
6921 assembler operation to identify the following data as read-only initialized
6925 @defmac BSS_SECTION_ASM_OP
6926 If defined, a C expression whose value is a string, including spacing,
6927 containing the assembler operation to identify the following data as
6928 uninitialized global data. If not defined, and
6929 @code{ASM_OUTPUT_ALIGNED_BSS} not defined,
6930 uninitialized global data will be output in the data section if
6931 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6935 @defmac SBSS_SECTION_ASM_OP
6936 If defined, a C expression whose value is a string, including spacing,
6937 containing the assembler operation to identify the following data as
6938 uninitialized, writable small data.
6941 @defmac TLS_COMMON_ASM_OP
6942 If defined, a C expression whose value is a string containing the
6943 assembler operation to identify the following data as thread-local
6944 common data. The default is @code{".tls_common"}.
6947 @defmac TLS_SECTION_ASM_FLAG
6948 If defined, a C expression whose value is a character constant
6949 containing the flag used to mark a section as a TLS section. The
6950 default is @code{'T'}.
6953 @defmac INIT_SECTION_ASM_OP
6954 If defined, a C expression whose value is a string, including spacing,
6955 containing the assembler operation to identify the following data as
6956 initialization code. If not defined, GCC will assume such a section does
6957 not exist. This section has no corresponding @code{init_section}
6958 variable; it is used entirely in runtime code.
6961 @defmac FINI_SECTION_ASM_OP
6962 If defined, a C expression whose value is a string, including spacing,
6963 containing the assembler operation to identify the following data as
6964 finalization code. If not defined, GCC will assume such a section does
6965 not exist. This section has no corresponding @code{fini_section}
6966 variable; it is used entirely in runtime code.
6969 @defmac INIT_ARRAY_SECTION_ASM_OP
6970 If defined, a C expression whose value is a string, including spacing,
6971 containing the assembler operation to identify the following data as
6972 part of the @code{.init_array} (or equivalent) section. If not
6973 defined, GCC will assume such a section does not exist. Do not define
6974 both this macro and @code{INIT_SECTION_ASM_OP}.
6977 @defmac FINI_ARRAY_SECTION_ASM_OP
6978 If defined, a C expression whose value is a string, including spacing,
6979 containing the assembler operation to identify the following data as
6980 part of the @code{.fini_array} (or equivalent) section. If not
6981 defined, GCC will assume such a section does not exist. Do not define
6982 both this macro and @code{FINI_SECTION_ASM_OP}.
6985 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6986 If defined, an ASM statement that switches to a different section
6987 via @var{section_op}, calls @var{function}, and switches back to
6988 the text section. This is used in @file{crtstuff.c} if
6989 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6990 to initialization and finalization functions from the init and fini
6991 sections. By default, this macro uses a simple function call. Some
6992 ports need hand-crafted assembly code to avoid dependencies on
6993 registers initialized in the function prologue or to ensure that
6994 constant pools don't end up too far way in the text section.
6997 @defmac TARGET_LIBGCC_SDATA_SECTION
6998 If defined, a string which names the section into which small
6999 variables defined in crtstuff and libgcc should go. This is useful
7000 when the target has options for optimizing access to small data, and
7001 you want the crtstuff and libgcc routines to be conservative in what
7002 they expect of your application yet liberal in what your application
7003 expects. For example, for targets with a @code{.sdata} section (like
7004 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
7005 require small data support from your application, but use this macro
7006 to put small data into @code{.sdata} so that your application can
7007 access these variables whether it uses small data or not.
7010 @defmac FORCE_CODE_SECTION_ALIGN
7011 If defined, an ASM statement that aligns a code section to some
7012 arbitrary boundary. This is used to force all fragments of the
7013 @code{.init} and @code{.fini} sections to have to same alignment
7014 and thus prevent the linker from having to add any padding.
7017 @defmac JUMP_TABLES_IN_TEXT_SECTION
7018 Define this macro to be an expression with a nonzero value if jump
7019 tables (for @code{tablejump} insns) should be output in the text
7020 section, along with the assembler instructions. Otherwise, the
7021 readonly data section is used.
7023 This macro is irrelevant if there is no separate readonly data section.
7026 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
7027 Define this hook if you need to do something special to set up the
7028 @file{varasm.c} sections, or if your target has some special sections
7029 of its own that you need to create.
7031 GCC calls this hook after processing the command line, but before writing
7032 any assembly code, and before calling any of the section-returning hooks
7036 @deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void)
7037 Return a mask describing how relocations should be treated when
7038 selecting sections. Bit 1 should be set if global relocations
7039 should be placed in a read-write section; bit 0 should be set if
7040 local relocations should be placed in a read-write section.
7042 The default version of this function returns 3 when @option{-fpic}
7043 is in effect, and 0 otherwise. The hook is typically redefined
7044 when the target cannot support (some kinds of) dynamic relocations
7045 in read-only sections even in executables.
7048 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
7049 Return the section into which @var{exp} should be placed. You can
7050 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
7051 some sort. @var{reloc} indicates whether the initial value of @var{exp}
7052 requires link-time relocations. Bit 0 is set when variable contains
7053 local relocations only, while bit 1 is set for global relocations.
7054 @var{align} is the constant alignment in bits.
7056 The default version of this function takes care of putting read-only
7057 variables in @code{readonly_data_section}.
7059 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7062 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7063 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7064 for @code{FUNCTION_DECL}s as well as for variables and constants.
7066 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7067 function has been determined to be likely to be called, and nonzero if
7068 it is unlikely to be called.
7071 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
7072 Build up a unique section name, expressed as a @code{STRING_CST} node,
7073 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7074 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7075 the initial value of @var{exp} requires link-time relocations.
7077 The default version of this function appends the symbol name to the
7078 ELF section name that would normally be used for the symbol. For
7079 example, the function @code{foo} would be placed in @code{.text.foo}.
7080 Whatever the actual target object format, this is often good enough.
7083 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
7084 Return the readonly data section associated with
7085 @samp{DECL_SECTION_NAME (@var{decl})}.
7086 The default version of this function selects @code{.gnu.linkonce.r.name} if
7087 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7088 if function is in @code{.text.name}, and the normal readonly-data section
7092 @deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX
7093 Usually, the compiler uses the prefix @code{".rodata"} to construct
7094 section names for mergeable constant data. Define this macro to override
7095 the string if a different section name should be used.
7098 @deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void)
7099 Return the section that should be used for transactional memory clone tables.
7102 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
7103 Return the section into which a constant @var{x}, of mode @var{mode},
7104 should be placed. You can assume that @var{x} is some kind of
7105 constant in RTL@. The argument @var{mode} is redundant except in the
7106 case of a @code{const_int} rtx. @var{align} is the constant alignment
7109 The default version of this function takes care of putting symbolic
7110 constants in @code{flag_pic} mode in @code{data_section} and everything
7111 else in @code{readonly_data_section}.
7114 @deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
7115 Define this hook if you need to postprocess the assembler name generated
7116 by target-independent code. The @var{id} provided to this hook will be
7117 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7118 or the mangled name of the @var{decl} in C++). The return value of the
7119 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7120 your target system. The default implementation of this hook just
7121 returns the @var{id} provided.
7124 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
7125 Define this hook if references to a symbol or a constant must be
7126 treated differently depending on something about the variable or
7127 function named by the symbol (such as what section it is in).
7129 The hook is executed immediately after rtl has been created for
7130 @var{decl}, which may be a variable or function declaration or
7131 an entry in the constant pool. In either case, @var{rtl} is the
7132 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7133 in this hook; that field may not have been initialized yet.
7135 In the case of a constant, it is safe to assume that the rtl is
7136 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7137 will also have this form, but that is not guaranteed. Global
7138 register variables, for instance, will have a @code{reg} for their
7139 rtl. (Normally the right thing to do with such unusual rtl is
7142 The @var{new_decl_p} argument will be true if this is the first time
7143 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7144 be false for subsequent invocations, which will happen for duplicate
7145 declarations. Whether or not anything must be done for the duplicate
7146 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7147 @var{new_decl_p} is always true when the hook is called for a constant.
7149 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7150 The usual thing for this hook to do is to record flags in the
7151 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7152 Historically, the name string was modified if it was necessary to
7153 encode more than one bit of information, but this practice is now
7154 discouraged; use @code{SYMBOL_REF_FLAGS}.
7156 The default definition of this hook, @code{default_encode_section_info}
7157 in @file{varasm.c}, sets a number of commonly-useful bits in
7158 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7159 before overriding it.
7162 @deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name})
7163 Decode @var{name} and return the real name part, sans
7164 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7168 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp})
7169 Returns true if @var{exp} should be placed into a ``small data'' section.
7170 The default version of this hook always returns false.
7173 @deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
7174 Contains the value true if the target places read-only
7175 ``small data'' into a separate section. The default value is false.
7178 @deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void)
7179 It returns true if target wants profile code emitted before prologue.
7181 The default version of this hook use the target macro
7182 @code{PROFILE_BEFORE_PROLOGUE}.
7185 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp})
7186 Returns true if @var{exp} names an object for which name resolution
7187 rules must resolve to the current ``module'' (dynamic shared library
7188 or executable image).
7190 The default version of this hook implements the name resolution rules
7191 for ELF, which has a looser model of global name binding than other
7192 currently supported object file formats.
7195 @deftypevr {Target Hook} bool TARGET_HAVE_TLS
7196 Contains the value true if the target supports thread-local storage.
7197 The default value is false.
7202 @section Position Independent Code
7203 @cindex position independent code
7206 This section describes macros that help implement generation of position
7207 independent code. Simply defining these macros is not enough to
7208 generate valid PIC; you must also add support to the hook
7209 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7210 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7211 must modify the definition of @samp{movsi} to do something appropriate
7212 when the source operand contains a symbolic address. You may also
7213 need to alter the handling of switch statements so that they use
7215 @c i rearranged the order of the macros above to try to force one of
7216 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7218 @defmac PIC_OFFSET_TABLE_REGNUM
7219 The register number of the register used to address a table of static
7220 data addresses in memory. In some cases this register is defined by a
7221 processor's ``application binary interface'' (ABI)@. When this macro
7222 is defined, RTL is generated for this register once, as with the stack
7223 pointer and frame pointer registers. If this macro is not defined, it
7224 is up to the machine-dependent files to allocate such a register (if
7225 necessary). Note that this register must be fixed when in use (e.g.@:
7226 when @code{flag_pic} is true).
7229 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7230 A C expression that is nonzero if the register defined by
7231 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7232 the default is zero. Do not define
7233 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7236 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7237 A C expression that is nonzero if @var{x} is a legitimate immediate
7238 operand on the target machine when generating position independent code.
7239 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7240 check this. You can also assume @var{flag_pic} is true, so you need not
7241 check it either. You need not define this macro if all constants
7242 (including @code{SYMBOL_REF}) can be immediate operands when generating
7243 position independent code.
7246 @node Assembler Format
7247 @section Defining the Output Assembler Language
7249 This section describes macros whose principal purpose is to describe how
7250 to write instructions in assembler language---rather than what the
7254 * File Framework:: Structural information for the assembler file.
7255 * Data Output:: Output of constants (numbers, strings, addresses).
7256 * Uninitialized Data:: Output of uninitialized variables.
7257 * Label Output:: Output and generation of labels.
7258 * Initialization:: General principles of initialization
7259 and termination routines.
7260 * Macros for Initialization::
7261 Specific macros that control the handling of
7262 initialization and termination routines.
7263 * Instruction Output:: Output of actual instructions.
7264 * Dispatch Tables:: Output of jump tables.
7265 * Exception Region Output:: Output of exception region code.
7266 * Alignment Output:: Pseudo ops for alignment and skipping data.
7269 @node File Framework
7270 @subsection The Overall Framework of an Assembler File
7271 @cindex assembler format
7272 @cindex output of assembler code
7274 @c prevent bad page break with this line
7275 This describes the overall framework of an assembly file.
7277 @findex default_file_start
7278 @deftypefn {Target Hook} void TARGET_ASM_FILE_START (void)
7279 Output to @code{asm_out_file} any text which the assembler expects to
7280 find at the beginning of a file. The default behavior is controlled
7281 by two flags, documented below. Unless your target's assembler is
7282 quite unusual, if you override the default, you should call
7283 @code{default_file_start} at some point in your target hook. This
7284 lets other target files rely on these variables.
7287 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
7288 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7289 printed as the very first line in the assembly file, unless
7290 @option{-fverbose-asm} is in effect. (If that macro has been defined
7291 to the empty string, this variable has no effect.) With the normal
7292 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7293 assembler that it need not bother stripping comments or extra
7294 whitespace from its input. This allows it to work a bit faster.
7296 The default is false. You should not set it to true unless you have
7297 verified that your port does not generate any extra whitespace or
7298 comments that will cause GAS to issue errors in NO_APP mode.
7301 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
7302 If this flag is true, @code{output_file_directive} will be called
7303 for the primary source file, immediately after printing
7304 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7305 this to be done. The default is false.
7308 @deftypefn {Target Hook} void TARGET_ASM_FILE_END (void)
7309 Output to @code{asm_out_file} any text which the assembler expects
7310 to find at the end of a file. The default is to output nothing.
7313 @deftypefun void file_end_indicate_exec_stack ()
7314 Some systems use a common convention, the @samp{.note.GNU-stack}
7315 special section, to indicate whether or not an object file relies on
7316 the stack being executable. If your system uses this convention, you
7317 should define @code{TARGET_ASM_FILE_END} to this function. If you
7318 need to do other things in that hook, have your hook function call
7322 @deftypefn {Target Hook} void TARGET_ASM_LTO_START (void)
7323 Output to @code{asm_out_file} any text which the assembler expects
7324 to find at the start of an LTO section. The default is to output
7328 @deftypefn {Target Hook} void TARGET_ASM_LTO_END (void)
7329 Output to @code{asm_out_file} any text which the assembler expects
7330 to find at the end of an LTO section. The default is to output
7334 @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void)
7335 Output to @code{asm_out_file} any text which is needed before emitting
7336 unwind info and debug info at the end of a file. Some targets emit
7337 here PIC setup thunks that cannot be emitted at the end of file,
7338 because they couldn't have unwind info then. The default is to output
7342 @defmac ASM_COMMENT_START
7343 A C string constant describing how to begin a comment in the target
7344 assembler language. The compiler assumes that the comment will end at
7345 the end of the line.
7349 A C string constant for text to be output before each @code{asm}
7350 statement or group of consecutive ones. Normally this is
7351 @code{"#APP"}, which is a comment that has no effect on most
7352 assemblers but tells the GNU assembler that it must check the lines
7353 that follow for all valid assembler constructs.
7357 A C string constant for text to be output after each @code{asm}
7358 statement or group of consecutive ones. Normally this is
7359 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7360 time-saving assumptions that are valid for ordinary compiler output.
7363 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7364 A C statement to output COFF information or DWARF debugging information
7365 which indicates that filename @var{name} is the current source file to
7366 the stdio stream @var{stream}.
7368 This macro need not be defined if the standard form of output
7369 for the file format in use is appropriate.
7372 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name})
7373 Output COFF information or DWARF debugging information which indicates that filename @var{name} is the current source file to the stdio stream @var{file}.
7375 This target hook need not be defined if the standard form of output for the file format in use is appropriate.
7378 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7379 A C statement to output the string @var{string} to the stdio stream
7380 @var{stream}. If you do not call the function @code{output_quoted_string}
7381 in your config files, GCC will only call it to output filenames to
7382 the assembler source. So you can use it to canonicalize the format
7383 of the filename using this macro.
7386 @defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string})
7387 A C statement to output something to the assembler file to handle a
7388 @samp{#ident} directive containing the text @var{string}. If this
7389 macro is not defined, nothing is output for a @samp{#ident} directive.
7392 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl})
7393 Output assembly directives to switch to section @var{name}. The section
7394 should have attributes as specified by @var{flags}, which is a bit mask
7395 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7396 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7397 this section is associated.
7400 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit})
7401 Return preferred text (sub)section for function @var{decl}.
7402 Main purpose of this function is to separate cold, normal and hot
7403 functions. @var{startup} is true when function is known to be used only
7404 at startup (from static constructors or it is @code{main()}).
7405 @var{exit} is true when function is known to be used only at exit
7406 (from static destructors).
7407 Return NULL if function should go to default text section.
7410 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold})
7411 Used by the target to emit any assembler directives or additional labels needed when a function is partitioned between different sections. Output should be written to @var{file}. The function decl is available as @var{decl} and the new section is `cold' if @var{new_is_cold} is @code{true}.
7414 @deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
7415 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7416 It must not be modified by command-line option processing.
7419 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7420 @deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7421 This flag is true if we can create zeroed data by switching to a BSS
7422 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7423 This is true on most ELF targets.
7426 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
7427 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7428 based on a variable or function decl, a section name, and whether or not the
7429 declaration's initializer may contain runtime relocations. @var{decl} may be
7430 null, in which case read-write data should be assumed.
7432 The default version of this function handles choosing code vs data,
7433 read-only vs read-write data, and @code{flag_pic}. You should only
7434 need to override this if your target has special flags that might be
7435 set via @code{__attribute__}.
7438 @deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text})
7439 Provides the target with the ability to record the gcc command line
7440 switches that have been passed to the compiler, and options that are
7441 enabled. The @var{type} argument specifies what is being recorded.
7442 It can take the following values:
7445 @item SWITCH_TYPE_PASSED
7446 @var{text} is a command line switch that has been set by the user.
7448 @item SWITCH_TYPE_ENABLED
7449 @var{text} is an option which has been enabled. This might be as a
7450 direct result of a command line switch, or because it is enabled by
7451 default or because it has been enabled as a side effect of a different
7452 command line switch. For example, the @option{-O2} switch enables
7453 various different individual optimization passes.
7455 @item SWITCH_TYPE_DESCRIPTIVE
7456 @var{text} is either NULL or some descriptive text which should be
7457 ignored. If @var{text} is NULL then it is being used to warn the
7458 target hook that either recording is starting or ending. The first
7459 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7460 warning is for start up and the second time the warning is for
7461 wind down. This feature is to allow the target hook to make any
7462 necessary preparations before it starts to record switches and to
7463 perform any necessary tidying up after it has finished recording
7466 @item SWITCH_TYPE_LINE_START
7467 This option can be ignored by this target hook.
7469 @item SWITCH_TYPE_LINE_END
7470 This option can be ignored by this target hook.
7473 The hook's return value must be zero. Other return values may be
7474 supported in the future.
7476 By default this hook is set to NULL, but an example implementation is
7477 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7478 it records the switches as ASCII text inside a new, string mergeable
7479 section in the assembler output file. The name of the new section is
7480 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7484 @deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7485 This is the name of the section that will be created by the example
7486 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7492 @subsection Output of Data
7495 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
7496 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7497 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7498 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7499 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7500 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7501 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7502 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7503 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7504 These hooks specify assembly directives for creating certain kinds
7505 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7506 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7507 aligned two-byte object, and so on. Any of the hooks may be
7508 @code{NULL}, indicating that no suitable directive is available.
7510 The compiler will print these strings at the start of a new line,
7511 followed immediately by the object's initial value. In most cases,
7512 the string should contain a tab, a pseudo-op, and then another tab.
7515 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
7516 The @code{assemble_integer} function uses this hook to output an
7517 integer object. @var{x} is the object's value, @var{size} is its size
7518 in bytes and @var{aligned_p} indicates whether it is aligned. The
7519 function should return @code{true} if it was able to output the
7520 object. If it returns false, @code{assemble_integer} will try to
7521 split the object into smaller parts.
7523 The default implementation of this hook will use the
7524 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7525 when the relevant string is @code{NULL}.
7528 @deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x})
7529 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7530 can't deal with, and output assembly code to @var{file} corresponding to
7531 the pattern @var{x}. This may be used to allow machine-dependent
7532 @code{UNSPEC}s to appear within constants.
7534 If target hook fails to recognize a pattern, it must return @code{false},
7535 so that a standard error message is printed. If it prints an error message
7536 itself, by calling, for example, @code{output_operand_lossage}, it may just
7540 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7541 A C statement to output to the stdio stream @var{stream} an assembler
7542 instruction to assemble a string constant containing the @var{len}
7543 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7544 @code{char *} and @var{len} a C expression of type @code{int}.
7546 If the assembler has a @code{.ascii} pseudo-op as found in the
7547 Berkeley Unix assembler, do not define the macro
7548 @code{ASM_OUTPUT_ASCII}.
7551 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7552 A C statement to output word @var{n} of a function descriptor for
7553 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7554 is defined, and is otherwise unused.
7557 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7558 You may define this macro as a C expression. You should define the
7559 expression to have a nonzero value if GCC should output the constant
7560 pool for a function before the code for the function, or a zero value if
7561 GCC should output the constant pool after the function. If you do
7562 not define this macro, the usual case, GCC will output the constant
7563 pool before the function.
7566 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7567 A C statement to output assembler commands to define the start of the
7568 constant pool for a function. @var{funname} is a string giving
7569 the name of the function. Should the return type of the function
7570 be required, it can be obtained via @var{fundecl}. @var{size}
7571 is the size, in bytes, of the constant pool that will be written
7572 immediately after this call.
7574 If no constant-pool prefix is required, the usual case, this macro need
7578 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7579 A C statement (with or without semicolon) to output a constant in the
7580 constant pool, if it needs special treatment. (This macro need not do
7581 anything for RTL expressions that can be output normally.)
7583 The argument @var{file} is the standard I/O stream to output the
7584 assembler code on. @var{x} is the RTL expression for the constant to
7585 output, and @var{mode} is the machine mode (in case @var{x} is a
7586 @samp{const_int}). @var{align} is the required alignment for the value
7587 @var{x}; you should output an assembler directive to force this much
7590 The argument @var{labelno} is a number to use in an internal label for
7591 the address of this pool entry. The definition of this macro is
7592 responsible for outputting the label definition at the proper place.
7593 Here is how to do this:
7596 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7599 When you output a pool entry specially, you should end with a
7600 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7601 entry from being output a second time in the usual manner.
7603 You need not define this macro if it would do nothing.
7606 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7607 A C statement to output assembler commands to at the end of the constant
7608 pool for a function. @var{funname} is a string giving the name of the
7609 function. Should the return type of the function be required, you can
7610 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7611 constant pool that GCC wrote immediately before this call.
7613 If no constant-pool epilogue is required, the usual case, you need not
7617 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7618 Define this macro as a C expression which is nonzero if @var{C} is
7619 used as a logical line separator by the assembler. @var{STR} points
7620 to the position in the string where @var{C} was found; this can be used if
7621 a line separator uses multiple characters.
7623 If you do not define this macro, the default is that only
7624 the character @samp{;} is treated as a logical line separator.
7627 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
7628 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
7629 These target hooks are C string constants, describing the syntax in the
7630 assembler for grouping arithmetic expressions. If not overridden, they
7631 default to normal parentheses, which is correct for most assemblers.
7634 These macros are provided by @file{real.h} for writing the definitions
7635 of @code{ASM_OUTPUT_DOUBLE} and the like:
7637 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7638 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7639 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7640 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7641 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7642 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7643 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7644 target's floating point representation, and store its bit pattern in
7645 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7646 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7647 simple @code{long int}. For the others, it should be an array of
7648 @code{long int}. The number of elements in this array is determined
7649 by the size of the desired target floating point data type: 32 bits of
7650 it go in each @code{long int} array element. Each array element holds
7651 32 bits of the result, even if @code{long int} is wider than 32 bits
7652 on the host machine.
7654 The array element values are designed so that you can print them out
7655 using @code{fprintf} in the order they should appear in the target
7659 @node Uninitialized Data
7660 @subsection Output of Uninitialized Variables
7662 Each of the macros in this section is used to do the whole job of
7663 outputting a single uninitialized variable.
7665 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7666 A C statement (sans semicolon) to output to the stdio stream
7667 @var{stream} the assembler definition of a common-label named
7668 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7669 is the size rounded up to whatever alignment the caller wants. It is
7670 possible that @var{size} may be zero, for instance if a struct with no
7671 other member than a zero-length array is defined. In this case, the
7672 backend must output a symbol definition that allocates at least one
7673 byte, both so that the address of the resulting object does not compare
7674 equal to any other, and because some object formats cannot even express
7675 the concept of a zero-sized common symbol, as that is how they represent
7676 an ordinary undefined external.
7678 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7679 output the name itself; before and after that, output the additional
7680 assembler syntax for defining the name, and a newline.
7682 This macro controls how the assembler definitions of uninitialized
7683 common global variables are output.
7686 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7687 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7688 separate, explicit argument. If you define this macro, it is used in
7689 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7690 handling the required alignment of the variable. The alignment is specified
7691 as the number of bits.
7694 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7695 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7696 variable to be output, if there is one, or @code{NULL_TREE} if there
7697 is no corresponding variable. If you define this macro, GCC will use it
7698 in place of both @code{ASM_OUTPUT_COMMON} and
7699 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7700 the variable's decl in order to chose what to output.
7703 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7704 A C statement (sans semicolon) to output to the stdio stream
7705 @var{stream} the assembler definition of uninitialized global @var{decl} named
7706 @var{name} whose size is @var{size} bytes. The variable @var{alignment}
7707 is the alignment specified as the number of bits.
7709 Try to use function @code{asm_output_aligned_bss} defined in file
7710 @file{varasm.c} when defining this macro. If unable, use the expression
7711 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7712 before and after that, output the additional assembler syntax for defining
7713 the name, and a newline.
7715 There are two ways of handling global BSS@. One is to define this macro.
7716 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7717 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7718 You do not need to do both.
7720 Some languages do not have @code{common} data, and require a
7721 non-common form of global BSS in order to handle uninitialized globals
7722 efficiently. C++ is one example of this. However, if the target does
7723 not support global BSS, the front end may choose to make globals
7724 common in order to save space in the object file.
7727 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7728 A C statement (sans semicolon) to output to the stdio stream
7729 @var{stream} the assembler definition of a local-common-label named
7730 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7731 is the size rounded up to whatever alignment the caller wants.
7733 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7734 output the name itself; before and after that, output the additional
7735 assembler syntax for defining the name, and a newline.
7737 This macro controls how the assembler definitions of uninitialized
7738 static variables are output.
7741 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7742 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7743 separate, explicit argument. If you define this macro, it is used in
7744 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7745 handling the required alignment of the variable. The alignment is specified
7746 as the number of bits.
7749 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7750 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7751 variable to be output, if there is one, or @code{NULL_TREE} if there
7752 is no corresponding variable. If you define this macro, GCC will use it
7753 in place of both @code{ASM_OUTPUT_DECL} and
7754 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7755 the variable's decl in order to chose what to output.
7759 @subsection Output and Generation of Labels
7761 @c prevent bad page break with this line
7762 This is about outputting labels.
7764 @findex assemble_name
7765 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7766 A C statement (sans semicolon) to output to the stdio stream
7767 @var{stream} the assembler definition of a label named @var{name}.
7768 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7769 output the name itself; before and after that, output the additional
7770 assembler syntax for defining the name, and a newline. A default
7771 definition of this macro is provided which is correct for most systems.
7774 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7775 A C statement (sans semicolon) to output to the stdio stream
7776 @var{stream} the assembler definition of a label named @var{name} of
7778 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7779 output the name itself; before and after that, output the additional
7780 assembler syntax for defining the name, and a newline. A default
7781 definition of this macro is provided which is correct for most systems.
7783 If this macro is not defined, then the function name is defined in the
7784 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7787 @findex assemble_name_raw
7788 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7789 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7790 to refer to a compiler-generated label. The default definition uses
7791 @code{assemble_name_raw}, which is like @code{assemble_name} except
7792 that it is more efficient.
7796 A C string containing the appropriate assembler directive to specify the
7797 size of a symbol, without any arguments. On systems that use ELF, the
7798 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7799 systems, the default is not to define this macro.
7801 Define this macro only if it is correct to use the default definitions
7802 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7803 for your system. If you need your own custom definitions of those
7804 macros, or if you do not need explicit symbol sizes at all, do not
7808 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7809 A C statement (sans semicolon) to output to the stdio stream
7810 @var{stream} a directive telling the assembler that the size of the
7811 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7812 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7816 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7817 A C statement (sans semicolon) to output to the stdio stream
7818 @var{stream} a directive telling the assembler to calculate the size of
7819 the symbol @var{name} by subtracting its address from the current
7822 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7823 provided. The default assumes that the assembler recognizes a special
7824 @samp{.} symbol as referring to the current address, and can calculate
7825 the difference between this and another symbol. If your assembler does
7826 not recognize @samp{.} or cannot do calculations with it, you will need
7827 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7831 A C string containing the appropriate assembler directive to specify the
7832 type of a symbol, without any arguments. On systems that use ELF, the
7833 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7834 systems, the default is not to define this macro.
7836 Define this macro only if it is correct to use the default definition of
7837 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7838 custom definition of this macro, or if you do not need explicit symbol
7839 types at all, do not define this macro.
7842 @defmac TYPE_OPERAND_FMT
7843 A C string which specifies (using @code{printf} syntax) the format of
7844 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7845 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7846 the default is not to define this macro.
7848 Define this macro only if it is correct to use the default definition of
7849 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7850 custom definition of this macro, or if you do not need explicit symbol
7851 types at all, do not define this macro.
7854 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7855 A C statement (sans semicolon) to output to the stdio stream
7856 @var{stream} a directive telling the assembler that the type of the
7857 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7858 that string is always either @samp{"function"} or @samp{"object"}, but
7859 you should not count on this.
7861 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7862 definition of this macro is provided.
7865 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7866 A C statement (sans semicolon) to output to the stdio stream
7867 @var{stream} any text necessary for declaring the name @var{name} of a
7868 function which is being defined. This macro is responsible for
7869 outputting the label definition (perhaps using
7870 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
7871 @code{FUNCTION_DECL} tree node representing the function.
7873 If this macro is not defined, then the function name is defined in the
7874 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
7876 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7880 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7881 A C statement (sans semicolon) to output to the stdio stream
7882 @var{stream} any text necessary for declaring the size of a function
7883 which is being defined. The argument @var{name} is the name of the
7884 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7885 representing the function.
7887 If this macro is not defined, then the function size is not defined.
7889 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7893 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7894 A C statement (sans semicolon) to output to the stdio stream
7895 @var{stream} any text necessary for declaring the name @var{name} of an
7896 initialized variable which is being defined. This macro must output the
7897 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7898 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7900 If this macro is not defined, then the variable name is defined in the
7901 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7903 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7904 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7907 @deftypefn {Target Hook} void TARGET_ASM_DECLARE_CONSTANT_NAME (FILE *@var{file}, const char *@var{name}, const_tree @var{expr}, HOST_WIDE_INT @var{size})
7908 A target hook to output to the stdio stream @var{file} any text necessary
7909 for declaring the name @var{name} of a constant which is being defined. This
7910 target hook is responsible for outputting the label definition (perhaps using
7911 @code{assemble_label}). The argument @var{exp} is the value of the constant,
7912 and @var{size} is the size of the constant in bytes. The @var{name}
7913 will be an internal label.
7915 The default version of this target hook, define the @var{name} in the
7916 usual manner as a label (by means of @code{assemble_label}).
7918 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7921 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7922 A C statement (sans semicolon) to output to the stdio stream
7923 @var{stream} any text necessary for claiming a register @var{regno}
7924 for a global variable @var{decl} with name @var{name}.
7926 If you don't define this macro, that is equivalent to defining it to do
7930 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7931 A C statement (sans semicolon) to finish up declaring a variable name
7932 once the compiler has processed its initializer fully and thus has had a
7933 chance to determine the size of an array when controlled by an
7934 initializer. This is used on systems where it's necessary to declare
7935 something about the size of the object.
7937 If you don't define this macro, that is equivalent to defining it to do
7940 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7941 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7944 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
7945 This target hook is a function to output to the stdio stream
7946 @var{stream} some commands that will make the label @var{name} global;
7947 that is, available for reference from other files.
7949 The default implementation relies on a proper definition of
7950 @code{GLOBAL_ASM_OP}.
7953 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
7954 This target hook is a function to output to the stdio stream
7955 @var{stream} some commands that will make the name associated with @var{decl}
7956 global; that is, available for reference from other files.
7958 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7961 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7962 A C statement (sans semicolon) to output to the stdio stream
7963 @var{stream} some commands that will make the label @var{name} weak;
7964 that is, available for reference from other files but only used if
7965 no other definition is available. Use the expression
7966 @code{assemble_name (@var{stream}, @var{name})} to output the name
7967 itself; before and after that, output the additional assembler syntax
7968 for making that name weak, and a newline.
7970 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7971 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7975 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7976 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7977 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7978 or variable decl. If @var{value} is not @code{NULL}, this C statement
7979 should output to the stdio stream @var{stream} assembler code which
7980 defines (equates) the weak symbol @var{name} to have the value
7981 @var{value}. If @var{value} is @code{NULL}, it should output commands
7982 to make @var{name} weak.
7985 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7986 Outputs a directive that enables @var{name} to be used to refer to
7987 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7988 declaration of @code{name}.
7991 @defmac SUPPORTS_WEAK
7992 A preprocessor constant expression which evaluates to true if the target
7993 supports weak symbols.
7995 If you don't define this macro, @file{defaults.h} provides a default
7996 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
7997 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
8000 @defmac TARGET_SUPPORTS_WEAK
8001 A C expression which evaluates to true if the target supports weak symbols.
8003 If you don't define this macro, @file{defaults.h} provides a default
8004 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
8005 this macro if you want to control weak symbol support with a compiler
8006 flag such as @option{-melf}.
8009 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
8010 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
8011 public symbol such that extra copies in multiple translation units will
8012 be discarded by the linker. Define this macro if your object file
8013 format provides support for this concept, such as the @samp{COMDAT}
8014 section flags in the Microsoft Windows PE/COFF format, and this support
8015 requires changes to @var{decl}, such as putting it in a separate section.
8018 @defmac SUPPORTS_ONE_ONLY
8019 A C expression which evaluates to true if the target supports one-only
8022 If you don't define this macro, @file{varasm.c} provides a default
8023 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
8024 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
8025 you want to control one-only symbol support with a compiler flag, or if
8026 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
8027 be emitted as one-only.
8030 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility})
8031 This target hook is a function to output to @var{asm_out_file} some
8032 commands that will make the symbol(s) associated with @var{decl} have
8033 hidden, protected or internal visibility as specified by @var{visibility}.
8036 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
8037 A C expression that evaluates to true if the target's linker expects
8038 that weak symbols do not appear in a static archive's table of contents.
8039 The default is @code{0}.
8041 Leaving weak symbols out of an archive's table of contents means that,
8042 if a symbol will only have a definition in one translation unit and
8043 will have undefined references from other translation units, that
8044 symbol should not be weak. Defining this macro to be nonzero will
8045 thus have the effect that certain symbols that would normally be weak
8046 (explicit template instantiations, and vtables for polymorphic classes
8047 with noninline key methods) will instead be nonweak.
8049 The C++ ABI requires this macro to be zero. Define this macro for
8050 targets where full C++ ABI compliance is impossible and where linker
8051 restrictions require weak symbols to be left out of a static archive's
8055 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
8056 A C statement (sans semicolon) to output to the stdio stream
8057 @var{stream} any text necessary for declaring the name of an external
8058 symbol named @var{name} which is referenced in this compilation but
8059 not defined. The value of @var{decl} is the tree node for the
8062 This macro need not be defined if it does not need to output anything.
8063 The GNU assembler and most Unix assemblers don't require anything.
8066 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
8067 This target hook is a function to output to @var{asm_out_file} an assembler
8068 pseudo-op to declare a library function name external. The name of the
8069 library function is given by @var{symref}, which is a @code{symbol_ref}.
8072 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol})
8073 This target hook is a function to output to @var{asm_out_file} an assembler
8074 directive to annotate @var{symbol} as used. The Darwin target uses the
8075 .no_dead_code_strip directive.
8078 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8079 A C statement (sans semicolon) to output to the stdio stream
8080 @var{stream} a reference in assembler syntax to a label named
8081 @var{name}. This should add @samp{_} to the front of the name, if that
8082 is customary on your operating system, as it is in most Berkeley Unix
8083 systems. This macro is used in @code{assemble_name}.
8086 @deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name})
8087 Given a symbol @var{name}, perform same mangling as @code{varasm.c}'s @code{assemble_name}, but in memory rather than to a file stream, returning result as an @code{IDENTIFIER_NODE}. Required for correct LTO symtabs. The default implementation calls the @code{TARGET_STRIP_NAME_ENCODING} hook and then prepends the @code{USER_LABEL_PREFIX}, if any.
8090 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8091 A C statement (sans semicolon) to output a reference to
8092 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8093 will be used to output the name of the symbol. This macro may be used
8094 to modify the way a symbol is referenced depending on information
8095 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8098 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8099 A C statement (sans semicolon) to output a reference to @var{buf}, the
8100 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8101 @code{assemble_name} will be used to output the name of the symbol.
8102 This macro is not used by @code{output_asm_label}, or the @code{%l}
8103 specifier that calls it; the intention is that this macro should be set
8104 when it is necessary to output a label differently when its address is
8108 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
8109 A function to output to the stdio stream @var{stream} a label whose
8110 name is made from the string @var{prefix} and the number @var{labelno}.
8112 It is absolutely essential that these labels be distinct from the labels
8113 used for user-level functions and variables. Otherwise, certain programs
8114 will have name conflicts with internal labels.
8116 It is desirable to exclude internal labels from the symbol table of the
8117 object file. Most assemblers have a naming convention for labels that
8118 should be excluded; on many systems, the letter @samp{L} at the
8119 beginning of a label has this effect. You should find out what
8120 convention your system uses, and follow it.
8122 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8125 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8126 A C statement to output to the stdio stream @var{stream} a debug info
8127 label whose name is made from the string @var{prefix} and the number
8128 @var{num}. This is useful for VLIW targets, where debug info labels
8129 may need to be treated differently than branch target labels. On some
8130 systems, branch target labels must be at the beginning of instruction
8131 bundles, but debug info labels can occur in the middle of instruction
8134 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8138 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8139 A C statement to store into the string @var{string} a label whose name
8140 is made from the string @var{prefix} and the number @var{num}.
8142 This string, when output subsequently by @code{assemble_name}, should
8143 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8144 with the same @var{prefix} and @var{num}.
8146 If the string begins with @samp{*}, then @code{assemble_name} will
8147 output the rest of the string unchanged. It is often convenient for
8148 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8149 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8150 to output the string, and may change it. (Of course,
8151 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8152 you should know what it does on your machine.)
8155 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8156 A C expression to assign to @var{outvar} (which is a variable of type
8157 @code{char *}) a newly allocated string made from the string
8158 @var{name} and the number @var{number}, with some suitable punctuation
8159 added. Use @code{alloca} to get space for the string.
8161 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8162 produce an assembler label for an internal static variable whose name is
8163 @var{name}. Therefore, the string must be such as to result in valid
8164 assembler code. The argument @var{number} is different each time this
8165 macro is executed; it prevents conflicts between similarly-named
8166 internal static variables in different scopes.
8168 Ideally this string should not be a valid C identifier, to prevent any
8169 conflict with the user's own symbols. Most assemblers allow periods
8170 or percent signs in assembler symbols; putting at least one of these
8171 between the name and the number will suffice.
8173 If this macro is not defined, a default definition will be provided
8174 which is correct for most systems.
8177 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8178 A C statement to output to the stdio stream @var{stream} assembler code
8179 which defines (equates) the symbol @var{name} to have the value @var{value}.
8182 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8183 correct for most systems.
8186 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8187 A C statement to output to the stdio stream @var{stream} assembler code
8188 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8189 to have the value of the tree node @var{decl_of_value}. This macro will
8190 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8191 the tree nodes are available.
8194 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8195 correct for most systems.
8198 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8199 A C statement that evaluates to true if the assembler code which defines
8200 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8201 of the tree node @var{decl_of_value} should be emitted near the end of the
8202 current compilation unit. The default is to not defer output of defines.
8203 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8204 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8207 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8208 A C statement to output to the stdio stream @var{stream} assembler code
8209 which defines (equates) the weak symbol @var{name} to have the value
8210 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8211 an undefined weak symbol.
8213 Define this macro if the target only supports weak aliases; define
8214 @code{ASM_OUTPUT_DEF} instead if possible.
8217 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8218 Define this macro to override the default assembler names used for
8219 Objective-C methods.
8221 The default name is a unique method number followed by the name of the
8222 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8223 the category is also included in the assembler name (e.g.@:
8226 These names are safe on most systems, but make debugging difficult since
8227 the method's selector is not present in the name. Therefore, particular
8228 systems define other ways of computing names.
8230 @var{buf} is an expression of type @code{char *} which gives you a
8231 buffer in which to store the name; its length is as long as
8232 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8233 50 characters extra.
8235 The argument @var{is_inst} specifies whether the method is an instance
8236 method or a class method; @var{class_name} is the name of the class;
8237 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8238 in a category); and @var{sel_name} is the name of the selector.
8240 On systems where the assembler can handle quoted names, you can use this
8241 macro to provide more human-readable names.
8244 @defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name})
8245 A C statement (sans semicolon) to output to the stdio stream
8246 @var{stream} commands to declare that the label @var{name} is an
8247 Objective-C class reference. This is only needed for targets whose
8248 linkers have special support for NeXT-style runtimes.
8251 @defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name})
8252 A C statement (sans semicolon) to output to the stdio stream
8253 @var{stream} commands to declare that the label @var{name} is an
8254 unresolved Objective-C class reference. This is only needed for targets
8255 whose linkers have special support for NeXT-style runtimes.
8258 @node Initialization
8259 @subsection How Initialization Functions Are Handled
8260 @cindex initialization routines
8261 @cindex termination routines
8262 @cindex constructors, output of
8263 @cindex destructors, output of
8265 The compiled code for certain languages includes @dfn{constructors}
8266 (also called @dfn{initialization routines})---functions to initialize
8267 data in the program when the program is started. These functions need
8268 to be called before the program is ``started''---that is to say, before
8269 @code{main} is called.
8271 Compiling some languages generates @dfn{destructors} (also called
8272 @dfn{termination routines}) that should be called when the program
8275 To make the initialization and termination functions work, the compiler
8276 must output something in the assembler code to cause those functions to
8277 be called at the appropriate time. When you port the compiler to a new
8278 system, you need to specify how to do this.
8280 There are two major ways that GCC currently supports the execution of
8281 initialization and termination functions. Each way has two variants.
8282 Much of the structure is common to all four variations.
8284 @findex __CTOR_LIST__
8285 @findex __DTOR_LIST__
8286 The linker must build two lists of these functions---a list of
8287 initialization functions, called @code{__CTOR_LIST__}, and a list of
8288 termination functions, called @code{__DTOR_LIST__}.
8290 Each list always begins with an ignored function pointer (which may hold
8291 0, @minus{}1, or a count of the function pointers after it, depending on
8292 the environment). This is followed by a series of zero or more function
8293 pointers to constructors (or destructors), followed by a function
8294 pointer containing zero.
8296 Depending on the operating system and its executable file format, either
8297 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8298 time and exit time. Constructors are called in reverse order of the
8299 list; destructors in forward order.
8301 The best way to handle static constructors works only for object file
8302 formats which provide arbitrarily-named sections. A section is set
8303 aside for a list of constructors, and another for a list of destructors.
8304 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8305 object file that defines an initialization function also puts a word in
8306 the constructor section to point to that function. The linker
8307 accumulates all these words into one contiguous @samp{.ctors} section.
8308 Termination functions are handled similarly.
8310 This method will be chosen as the default by @file{target-def.h} if
8311 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8312 support arbitrary sections, but does support special designated
8313 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8314 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8316 When arbitrary sections are available, there are two variants, depending
8317 upon how the code in @file{crtstuff.c} is called. On systems that
8318 support a @dfn{.init} section which is executed at program startup,
8319 parts of @file{crtstuff.c} are compiled into that section. The
8320 program is linked by the @command{gcc} driver like this:
8323 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8326 The prologue of a function (@code{__init}) appears in the @code{.init}
8327 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8328 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8329 files are provided by the operating system or by the GNU C library, but
8330 are provided by GCC for a few targets.
8332 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8333 compiled from @file{crtstuff.c}. They contain, among other things, code
8334 fragments within the @code{.init} and @code{.fini} sections that branch
8335 to routines in the @code{.text} section. The linker will pull all parts
8336 of a section together, which results in a complete @code{__init} function
8337 that invokes the routines we need at startup.
8339 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8342 If no init section is available, when GCC compiles any function called
8343 @code{main} (or more accurately, any function designated as a program
8344 entry point by the language front end calling @code{expand_main_function}),
8345 it inserts a procedure call to @code{__main} as the first executable code
8346 after the function prologue. The @code{__main} function is defined
8347 in @file{libgcc2.c} and runs the global constructors.
8349 In file formats that don't support arbitrary sections, there are again
8350 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8351 and an `a.out' format must be used. In this case,
8352 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8353 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8354 and with the address of the void function containing the initialization
8355 code as its value. The GNU linker recognizes this as a request to add
8356 the value to a @dfn{set}; the values are accumulated, and are eventually
8357 placed in the executable as a vector in the format described above, with
8358 a leading (ignored) count and a trailing zero element.
8359 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8360 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8361 the compilation of @code{main} to call @code{__main} as above, starting
8362 the initialization process.
8364 The last variant uses neither arbitrary sections nor the GNU linker.
8365 This is preferable when you want to do dynamic linking and when using
8366 file formats which the GNU linker does not support, such as `ECOFF'@. In
8367 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8368 termination functions are recognized simply by their names. This requires
8369 an extra program in the linkage step, called @command{collect2}. This program
8370 pretends to be the linker, for use with GCC; it does its job by running
8371 the ordinary linker, but also arranges to include the vectors of
8372 initialization and termination functions. These functions are called
8373 via @code{__main} as described above. In order to use this method,
8374 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8377 The following section describes the specific macros that control and
8378 customize the handling of initialization and termination functions.
8381 @node Macros for Initialization
8382 @subsection Macros Controlling Initialization Routines
8384 Here are the macros that control how the compiler handles initialization
8385 and termination functions:
8387 @defmac INIT_SECTION_ASM_OP
8388 If defined, a C string constant, including spacing, for the assembler
8389 operation to identify the following data as initialization code. If not
8390 defined, GCC will assume such a section does not exist. When you are
8391 using special sections for initialization and termination functions, this
8392 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8393 run the initialization functions.
8396 @defmac HAS_INIT_SECTION
8397 If defined, @code{main} will not call @code{__main} as described above.
8398 This macro should be defined for systems that control start-up code
8399 on a symbol-by-symbol basis, such as OSF/1, and should not
8400 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8403 @defmac LD_INIT_SWITCH
8404 If defined, a C string constant for a switch that tells the linker that
8405 the following symbol is an initialization routine.
8408 @defmac LD_FINI_SWITCH
8409 If defined, a C string constant for a switch that tells the linker that
8410 the following symbol is a finalization routine.
8413 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8414 If defined, a C statement that will write a function that can be
8415 automatically called when a shared library is loaded. The function
8416 should call @var{func}, which takes no arguments. If not defined, and
8417 the object format requires an explicit initialization function, then a
8418 function called @code{_GLOBAL__DI} will be generated.
8420 This function and the following one are used by collect2 when linking a
8421 shared library that needs constructors or destructors, or has DWARF2
8422 exception tables embedded in the code.
8425 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8426 If defined, a C statement that will write a function that can be
8427 automatically called when a shared library is unloaded. The function
8428 should call @var{func}, which takes no arguments. If not defined, and
8429 the object format requires an explicit finalization function, then a
8430 function called @code{_GLOBAL__DD} will be generated.
8433 @defmac INVOKE__main
8434 If defined, @code{main} will call @code{__main} despite the presence of
8435 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8436 where the init section is not actually run automatically, but is still
8437 useful for collecting the lists of constructors and destructors.
8440 @defmac SUPPORTS_INIT_PRIORITY
8441 If nonzero, the C++ @code{init_priority} attribute is supported and the
8442 compiler should emit instructions to control the order of initialization
8443 of objects. If zero, the compiler will issue an error message upon
8444 encountering an @code{init_priority} attribute.
8447 @deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS
8448 This value is true if the target supports some ``native'' method of
8449 collecting constructors and destructors to be run at startup and exit.
8450 It is false if we must use @command{collect2}.
8453 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
8454 If defined, a function that outputs assembler code to arrange to call
8455 the function referenced by @var{symbol} at initialization time.
8457 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8458 no arguments and with no return value. If the target supports initialization
8459 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8460 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8462 If this macro is not defined by the target, a suitable default will
8463 be chosen if (1) the target supports arbitrary section names, (2) the
8464 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8468 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
8469 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8470 functions rather than initialization functions.
8473 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8474 generated for the generated object file will have static linkage.
8476 If your system uses @command{collect2} as the means of processing
8477 constructors, then that program normally uses @command{nm} to scan
8478 an object file for constructor functions to be called.
8480 On certain kinds of systems, you can define this macro to make
8481 @command{collect2} work faster (and, in some cases, make it work at all):
8483 @defmac OBJECT_FORMAT_COFF
8484 Define this macro if the system uses COFF (Common Object File Format)
8485 object files, so that @command{collect2} can assume this format and scan
8486 object files directly for dynamic constructor/destructor functions.
8488 This macro is effective only in a native compiler; @command{collect2} as
8489 part of a cross compiler always uses @command{nm} for the target machine.
8492 @defmac REAL_NM_FILE_NAME
8493 Define this macro as a C string constant containing the file name to use
8494 to execute @command{nm}. The default is to search the path normally for
8499 @command{collect2} calls @command{nm} to scan object files for static
8500 constructors and destructors and LTO info. By default, @option{-n} is
8501 passed. Define @code{NM_FLAGS} to a C string constant if other options
8502 are needed to get the same output format as GNU @command{nm -n}
8506 If your system supports shared libraries and has a program to list the
8507 dynamic dependencies of a given library or executable, you can define
8508 these macros to enable support for running initialization and
8509 termination functions in shared libraries:
8512 Define this macro to a C string constant containing the name of the program
8513 which lists dynamic dependencies, like @command{ldd} under SunOS 4.
8516 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8517 Define this macro to be C code that extracts filenames from the output
8518 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8519 of type @code{char *} that points to the beginning of a line of output
8520 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8521 code must advance @var{ptr} to the beginning of the filename on that
8522 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8525 @defmac SHLIB_SUFFIX
8526 Define this macro to a C string constant containing the default shared
8527 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8528 strips version information after this suffix when generating global
8529 constructor and destructor names. This define is only needed on targets
8530 that use @command{collect2} to process constructors and destructors.
8533 @node Instruction Output
8534 @subsection Output of Assembler Instructions
8536 @c prevent bad page break with this line
8537 This describes assembler instruction output.
8539 @defmac REGISTER_NAMES
8540 A C initializer containing the assembler's names for the machine
8541 registers, each one as a C string constant. This is what translates
8542 register numbers in the compiler into assembler language.
8545 @defmac ADDITIONAL_REGISTER_NAMES
8546 If defined, a C initializer for an array of structures containing a name
8547 and a register number. This macro defines additional names for hard
8548 registers, thus allowing the @code{asm} option in declarations to refer
8549 to registers using alternate names.
8552 @defmac OVERLAPPING_REGISTER_NAMES
8553 If defined, a C initializer for an array of structures containing a
8554 name, a register number and a count of the number of consecutive
8555 machine registers the name overlaps. This macro defines additional
8556 names for hard registers, thus allowing the @code{asm} option in
8557 declarations to refer to registers using alternate names. Unlike
8558 @code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the
8559 register name implies multiple underlying registers.
8561 This macro should be used when it is important that a clobber in an
8562 @code{asm} statement clobbers all the underlying values implied by the
8563 register name. For example, on ARM, clobbering the double-precision
8564 VFP register ``d0'' implies clobbering both single-precision registers
8568 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8569 Define this macro if you are using an unusual assembler that
8570 requires different names for the machine instructions.
8572 The definition is a C statement or statements which output an
8573 assembler instruction opcode to the stdio stream @var{stream}. The
8574 macro-operand @var{ptr} is a variable of type @code{char *} which
8575 points to the opcode name in its ``internal'' form---the form that is
8576 written in the machine description. The definition should output the
8577 opcode name to @var{stream}, performing any translation you desire, and
8578 increment the variable @var{ptr} to point at the end of the opcode
8579 so that it will not be output twice.
8581 In fact, your macro definition may process less than the entire opcode
8582 name, or more than the opcode name; but if you want to process text
8583 that includes @samp{%}-sequences to substitute operands, you must take
8584 care of the substitution yourself. Just be sure to increment
8585 @var{ptr} over whatever text should not be output normally.
8587 @findex recog_data.operand
8588 If you need to look at the operand values, they can be found as the
8589 elements of @code{recog_data.operand}.
8591 If the macro definition does nothing, the instruction is output
8595 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8596 If defined, a C statement to be executed just prior to the output of
8597 assembler code for @var{insn}, to modify the extracted operands so
8598 they will be output differently.
8600 Here the argument @var{opvec} is the vector containing the operands
8601 extracted from @var{insn}, and @var{noperands} is the number of
8602 elements of the vector which contain meaningful data for this insn.
8603 The contents of this vector are what will be used to convert the insn
8604 template into assembler code, so you can change the assembler output
8605 by changing the contents of the vector.
8607 This macro is useful when various assembler syntaxes share a single
8608 file of instruction patterns; by defining this macro differently, you
8609 can cause a large class of instructions to be output differently (such
8610 as with rearranged operands). Naturally, variations in assembler
8611 syntax affecting individual insn patterns ought to be handled by
8612 writing conditional output routines in those patterns.
8614 If this macro is not defined, it is equivalent to a null statement.
8617 @deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx @var{insn}, rtx *@var{opvec}, int @var{noperands})
8618 If defined, this target hook is a function which is executed just after the
8619 output of assembler code for @var{insn}, to change the mode of the assembler
8622 Here the argument @var{opvec} is the vector containing the operands
8623 extracted from @var{insn}, and @var{noperands} is the number of
8624 elements of the vector which contain meaningful data for this insn.
8625 The contents of this vector are what was used to convert the insn
8626 template into assembler code, so you can change the assembler mode
8627 by checking the contents of the vector.
8630 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8631 A C compound statement to output to stdio stream @var{stream} the
8632 assembler syntax for an instruction operand @var{x}. @var{x} is an
8635 @var{code} is a value that can be used to specify one of several ways
8636 of printing the operand. It is used when identical operands must be
8637 printed differently depending on the context. @var{code} comes from
8638 the @samp{%} specification that was used to request printing of the
8639 operand. If the specification was just @samp{%@var{digit}} then
8640 @var{code} is 0; if the specification was @samp{%@var{ltr}
8641 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8644 If @var{x} is a register, this macro should print the register's name.
8645 The names can be found in an array @code{reg_names} whose type is
8646 @code{char *[]}. @code{reg_names} is initialized from
8647 @code{REGISTER_NAMES}.
8649 When the machine description has a specification @samp{%@var{punct}}
8650 (a @samp{%} followed by a punctuation character), this macro is called
8651 with a null pointer for @var{x} and the punctuation character for
8655 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8656 A C expression which evaluates to true if @var{code} is a valid
8657 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8658 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8659 punctuation characters (except for the standard one, @samp{%}) are used
8663 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8664 A C compound statement to output to stdio stream @var{stream} the
8665 assembler syntax for an instruction operand that is a memory reference
8666 whose address is @var{x}. @var{x} is an RTL expression.
8668 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8669 On some machines, the syntax for a symbolic address depends on the
8670 section that the address refers to. On these machines, define the hook
8671 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8672 @code{symbol_ref}, and then check for it here. @xref{Assembler
8676 @findex dbr_sequence_length
8677 @defmac DBR_OUTPUT_SEQEND (@var{file})
8678 A C statement, to be executed after all slot-filler instructions have
8679 been output. If necessary, call @code{dbr_sequence_length} to
8680 determine the number of slots filled in a sequence (zero if not
8681 currently outputting a sequence), to decide how many no-ops to output,
8684 Don't define this macro if it has nothing to do, but it is helpful in
8685 reading assembly output if the extent of the delay sequence is made
8686 explicit (e.g.@: with white space).
8689 @findex final_sequence
8690 Note that output routines for instructions with delay slots must be
8691 prepared to deal with not being output as part of a sequence
8692 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8693 found.) The variable @code{final_sequence} is null when not
8694 processing a sequence, otherwise it contains the @code{sequence} rtx
8698 @defmac REGISTER_PREFIX
8699 @defmacx LOCAL_LABEL_PREFIX
8700 @defmacx USER_LABEL_PREFIX
8701 @defmacx IMMEDIATE_PREFIX
8702 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8703 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8704 @file{final.c}). These are useful when a single @file{md} file must
8705 support multiple assembler formats. In that case, the various @file{tm.h}
8706 files can define these macros differently.
8709 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8710 If defined this macro should expand to a series of @code{case}
8711 statements which will be parsed inside the @code{switch} statement of
8712 the @code{asm_fprintf} function. This allows targets to define extra
8713 printf formats which may useful when generating their assembler
8714 statements. Note that uppercase letters are reserved for future
8715 generic extensions to asm_fprintf, and so are not available to target
8716 specific code. The output file is given by the parameter @var{file}.
8717 The varargs input pointer is @var{argptr} and the rest of the format
8718 string, starting the character after the one that is being switched
8719 upon, is pointed to by @var{format}.
8722 @defmac ASSEMBLER_DIALECT
8723 If your target supports multiple dialects of assembler language (such as
8724 different opcodes), define this macro as a C expression that gives the
8725 numeric index of the assembler language dialect to use, with zero as the
8728 If this macro is defined, you may use constructs of the form
8730 @samp{@{option0|option1|option2@dots{}@}}
8733 in the output templates of patterns (@pxref{Output Template}) or in the
8734 first argument of @code{asm_fprintf}. This construct outputs
8735 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8736 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8737 within these strings retain their usual meaning. If there are fewer
8738 alternatives within the braces than the value of
8739 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8741 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8742 @samp{@}} do not have any special meaning when used in templates or
8743 operands to @code{asm_fprintf}.
8745 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8746 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8747 the variations in assembler language syntax with that mechanism. Define
8748 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8749 if the syntax variant are larger and involve such things as different
8750 opcodes or operand order.
8753 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8754 A C expression to output to @var{stream} some assembler code
8755 which will push hard register number @var{regno} onto the stack.
8756 The code need not be optimal, since this macro is used only when
8760 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8761 A C expression to output to @var{stream} some assembler code
8762 which will pop hard register number @var{regno} off of the stack.
8763 The code need not be optimal, since this macro is used only when
8767 @node Dispatch Tables
8768 @subsection Output of Dispatch Tables
8770 @c prevent bad page break with this line
8771 This concerns dispatch tables.
8773 @cindex dispatch table
8774 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8775 A C statement to output to the stdio stream @var{stream} an assembler
8776 pseudo-instruction to generate a difference between two labels.
8777 @var{value} and @var{rel} are the numbers of two internal labels. The
8778 definitions of these labels are output using
8779 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8780 way here. For example,
8783 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8784 @var{value}, @var{rel})
8787 You must provide this macro on machines where the addresses in a
8788 dispatch table are relative to the table's own address. If defined, GCC
8789 will also use this macro on all machines when producing PIC@.
8790 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8791 mode and flags can be read.
8794 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8795 This macro should be provided on machines where the addresses
8796 in a dispatch table are absolute.
8798 The definition should be a C statement to output to the stdio stream
8799 @var{stream} an assembler pseudo-instruction to generate a reference to
8800 a label. @var{value} is the number of an internal label whose
8801 definition is output using @code{(*targetm.asm_out.internal_label)}.
8805 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8809 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8810 Define this if the label before a jump-table needs to be output
8811 specially. The first three arguments are the same as for
8812 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8813 jump-table which follows (a @code{jump_insn} containing an
8814 @code{addr_vec} or @code{addr_diff_vec}).
8816 This feature is used on system V to output a @code{swbeg} statement
8819 If this macro is not defined, these labels are output with
8820 @code{(*targetm.asm_out.internal_label)}.
8823 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8824 Define this if something special must be output at the end of a
8825 jump-table. The definition should be a C statement to be executed
8826 after the assembler code for the table is written. It should write
8827 the appropriate code to stdio stream @var{stream}. The argument
8828 @var{table} is the jump-table insn, and @var{num} is the label-number
8829 of the preceding label.
8831 If this macro is not defined, nothing special is output at the end of
8835 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty})
8836 This target hook emits a label at the beginning of each FDE@. It
8837 should be defined on targets where FDEs need special labels, and it
8838 should write the appropriate label, for the FDE associated with the
8839 function declaration @var{decl}, to the stdio stream @var{stream}.
8840 The third argument, @var{for_eh}, is a boolean: true if this is for an
8841 exception table. The fourth argument, @var{empty}, is a boolean:
8842 true if this is a placeholder label for an omitted FDE@.
8844 The default is that FDEs are not given nonlocal labels.
8847 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream})
8848 This target hook emits a label at the beginning of the exception table.
8849 It should be defined on targets where it is desirable for the table
8850 to be broken up according to function.
8852 The default is that no label is emitted.
8855 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality})
8856 If the target implements @code{TARGET_ASM_UNWIND_EMIT}, this hook may be used to emit a directive to install a personality hook into the unwind info. This hook should not be used if dwarf2 unwind info is used.
8859 @deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx @var{insn})
8860 This target hook emits assembly directives required to unwind the
8861 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8862 returns @code{UI_TARGET}.
8865 @deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8866 True if the @code{TARGET_ASM_UNWIND_EMIT} hook should be called before the assembly for @var{insn} has been emitted, false if the hook should be called afterward.
8869 @node Exception Region Output
8870 @subsection Assembler Commands for Exception Regions
8872 @c prevent bad page break with this line
8874 This describes commands marking the start and the end of an exception
8877 @defmac EH_FRAME_SECTION_NAME
8878 If defined, a C string constant for the name of the section containing
8879 exception handling frame unwind information. If not defined, GCC will
8880 provide a default definition if the target supports named sections.
8881 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8883 You should define this symbol if your target supports DWARF 2 frame
8884 unwind information and the default definition does not work.
8887 @defmac EH_FRAME_IN_DATA_SECTION
8888 If defined, DWARF 2 frame unwind information will be placed in the
8889 data section even though the target supports named sections. This
8890 might be necessary, for instance, if the system linker does garbage
8891 collection and sections cannot be marked as not to be collected.
8893 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8897 @defmac EH_TABLES_CAN_BE_READ_ONLY
8898 Define this macro to 1 if your target is such that no frame unwind
8899 information encoding used with non-PIC code will ever require a
8900 runtime relocation, but the linker may not support merging read-only
8901 and read-write sections into a single read-write section.
8904 @defmac MASK_RETURN_ADDR
8905 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8906 that it does not contain any extraneous set bits in it.
8909 @defmac DWARF2_UNWIND_INFO
8910 Define this macro to 0 if your target supports DWARF 2 frame unwind
8911 information, but it does not yet work with exception handling.
8912 Otherwise, if your target supports this information (if it defines
8913 @code{INCOMING_RETURN_ADDR_RTX} and either @code{UNALIGNED_INT_ASM_OP}
8914 or @code{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 1.
8917 @deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts})
8918 This hook defines the mechanism that will be used for exception handling
8919 by the target. If the target has ABI specified unwind tables, the hook
8920 should return @code{UI_TARGET}. If the target is to use the
8921 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8922 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8923 information, the hook should return @code{UI_DWARF2}.
8925 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8926 This may end up simplifying other parts of target-specific code. The
8927 default implementation of this hook never returns @code{UI_NONE}.
8929 Note that the value returned by this hook should be constant. It should
8930 not depend on anything except the command-line switches described by
8931 @var{opts}. In particular, the
8932 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8933 macros and builtin functions related to exception handling are set up
8934 depending on this setting.
8936 The default implementation of the hook first honors the
8937 @option{--enable-sjlj-exceptions} configure option, then
8938 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If
8939 @code{DWARF2_UNWIND_INFO} depends on command-line options, the target
8940 must define this hook so that @var{opts} is used correctly.
8943 @deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
8944 This variable should be set to @code{true} if the target ABI requires unwinding
8945 tables even when exceptions are not used. It must not be modified by
8946 command-line option processing.
8949 @defmac DONT_USE_BUILTIN_SETJMP
8950 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8951 should use the @code{setjmp}/@code{longjmp} functions from the C library
8952 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8955 @defmac DWARF_CIE_DATA_ALIGNMENT
8956 This macro need only be defined if the target might save registers in the
8957 function prologue at an offset to the stack pointer that is not aligned to
8958 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8959 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8960 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8961 the target supports DWARF 2 frame unwind information.
8964 @deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
8965 Contains the value true if the target should add a zero word onto the
8966 end of a Dwarf-2 frame info section when used for exception handling.
8967 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8971 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
8972 Given a register, this hook should return a parallel of registers to
8973 represent where to find the register pieces. Define this hook if the
8974 register and its mode are represented in Dwarf in non-contiguous
8975 locations, or if the register should be represented in more than one
8976 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8977 If not defined, the default is to return @code{NULL_RTX}.
8980 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
8981 If some registers are represented in Dwarf-2 unwind information in
8982 multiple pieces, define this hook to fill in information about the
8983 sizes of those pieces in the table used by the unwinder at runtime.
8984 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8985 filling in a single size corresponding to each hard register;
8986 @var{address} is the address of the table.
8989 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
8990 This hook is used to output a reference from a frame unwinding table to
8991 the type_info object identified by @var{sym}. It should return @code{true}
8992 if the reference was output. Returning @code{false} will cause the
8993 reference to be output using the normal Dwarf2 routines.
8996 @deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER
8997 This flag should be set to @code{true} on targets that use an ARM EABI
8998 based unwinding library, and @code{false} on other targets. This effects
8999 the format of unwinding tables, and how the unwinder in entered after
9000 running a cleanup. The default is @code{false}.
9003 @node Alignment Output
9004 @subsection Assembler Commands for Alignment
9006 @c prevent bad page break with this line
9007 This describes commands for alignment.
9009 @defmac JUMP_ALIGN (@var{label})
9010 The alignment (log base 2) to put in front of @var{label}, which is
9011 a common destination of jumps and has no fallthru incoming edge.
9013 This macro need not be defined if you don't want any special alignment
9014 to be done at such a time. Most machine descriptions do not currently
9017 Unless it's necessary to inspect the @var{label} parameter, it is better
9018 to set the variable @var{align_jumps} in the target's
9019 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9020 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
9023 @deftypefn {Target Hook} int TARGET_ASM_JUMP_ALIGN_MAX_SKIP (rtx @var{label})
9024 The maximum number of bytes to skip before @var{label} when applying
9025 @code{JUMP_ALIGN}. This works only if
9026 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9029 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
9030 The alignment (log base 2) to put in front of @var{label}, which follows
9033 This macro need not be defined if you don't want any special alignment
9034 to be done at such a time. Most machine descriptions do not currently
9038 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP (rtx @var{label})
9039 The maximum number of bytes to skip before @var{label} when applying
9040 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
9041 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9044 @defmac LOOP_ALIGN (@var{label})
9045 The alignment (log base 2) to put in front of @var{label}, which follows
9046 a @code{NOTE_INSN_LOOP_BEG} note.
9048 This macro need not be defined if you don't want any special alignment
9049 to be done at such a time. Most machine descriptions do not currently
9052 Unless it's necessary to inspect the @var{label} parameter, it is better
9053 to set the variable @code{align_loops} in the target's
9054 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9055 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
9058 @deftypefn {Target Hook} int TARGET_ASM_LOOP_ALIGN_MAX_SKIP (rtx @var{label})
9059 The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to
9060 @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is
9064 @defmac LABEL_ALIGN (@var{label})
9065 The alignment (log base 2) to put in front of @var{label}.
9066 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
9067 the maximum of the specified values is used.
9069 Unless it's necessary to inspect the @var{label} parameter, it is better
9070 to set the variable @code{align_labels} in the target's
9071 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9072 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
9075 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_MAX_SKIP (rtx @var{label})
9076 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}
9077 to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN}
9081 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
9082 A C statement to output to the stdio stream @var{stream} an assembler
9083 instruction to advance the location counter by @var{nbytes} bytes.
9084 Those bytes should be zero when loaded. @var{nbytes} will be a C
9085 expression of type @code{unsigned HOST_WIDE_INT}.
9088 @defmac ASM_NO_SKIP_IN_TEXT
9089 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
9090 text section because it fails to put zeros in the bytes that are skipped.
9091 This is true on many Unix systems, where the pseudo--op to skip bytes
9092 produces no-op instructions rather than zeros when used in the text
9096 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9097 A C statement to output to the stdio stream @var{stream} an assembler
9098 command to advance the location counter to a multiple of 2 to the
9099 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9102 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9103 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9104 for padding, if necessary.
9107 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9108 A C statement to output to the stdio stream @var{stream} an assembler
9109 command to advance the location counter to a multiple of 2 to the
9110 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9111 satisfy the alignment request. @var{power} and @var{max_skip} will be
9112 a C expression of type @code{int}.
9116 @node Debugging Info
9117 @section Controlling Debugging Information Format
9119 @c prevent bad page break with this line
9120 This describes how to specify debugging information.
9123 * All Debuggers:: Macros that affect all debugging formats uniformly.
9124 * DBX Options:: Macros enabling specific options in DBX format.
9125 * DBX Hooks:: Hook macros for varying DBX format.
9126 * File Names and DBX:: Macros controlling output of file names in DBX format.
9127 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
9128 * VMS Debug:: Macros for VMS debug format.
9132 @subsection Macros Affecting All Debugging Formats
9134 @c prevent bad page break with this line
9135 These macros affect all debugging formats.
9137 @defmac DBX_REGISTER_NUMBER (@var{regno})
9138 A C expression that returns the DBX register number for the compiler
9139 register number @var{regno}. In the default macro provided, the value
9140 of this expression will be @var{regno} itself. But sometimes there are
9141 some registers that the compiler knows about and DBX does not, or vice
9142 versa. In such cases, some register may need to have one number in the
9143 compiler and another for DBX@.
9145 If two registers have consecutive numbers inside GCC, and they can be
9146 used as a pair to hold a multiword value, then they @emph{must} have
9147 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9148 Otherwise, debuggers will be unable to access such a pair, because they
9149 expect register pairs to be consecutive in their own numbering scheme.
9151 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9152 does not preserve register pairs, then what you must do instead is
9153 redefine the actual register numbering scheme.
9156 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9157 A C expression that returns the integer offset value for an automatic
9158 variable having address @var{x} (an RTL expression). The default
9159 computation assumes that @var{x} is based on the frame-pointer and
9160 gives the offset from the frame-pointer. This is required for targets
9161 that produce debugging output for DBX or COFF-style debugging output
9162 for SDB and allow the frame-pointer to be eliminated when the
9163 @option{-g} options is used.
9166 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9167 A C expression that returns the integer offset value for an argument
9168 having address @var{x} (an RTL expression). The nominal offset is
9172 @defmac PREFERRED_DEBUGGING_TYPE
9173 A C expression that returns the type of debugging output GCC should
9174 produce when the user specifies just @option{-g}. Define
9175 this if you have arranged for GCC to support more than one format of
9176 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9177 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9178 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9180 When the user specifies @option{-ggdb}, GCC normally also uses the
9181 value of this macro to select the debugging output format, but with two
9182 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9183 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9184 defined, GCC uses @code{DBX_DEBUG}.
9186 The value of this macro only affects the default debugging output; the
9187 user can always get a specific type of output by using @option{-gstabs},
9188 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9192 @subsection Specific Options for DBX Output
9194 @c prevent bad page break with this line
9195 These are specific options for DBX output.
9197 @defmac DBX_DEBUGGING_INFO
9198 Define this macro if GCC should produce debugging output for DBX
9199 in response to the @option{-g} option.
9202 @defmac XCOFF_DEBUGGING_INFO
9203 Define this macro if GCC should produce XCOFF format debugging output
9204 in response to the @option{-g} option. This is a variant of DBX format.
9207 @defmac DEFAULT_GDB_EXTENSIONS
9208 Define this macro to control whether GCC should by default generate
9209 GDB's extended version of DBX debugging information (assuming DBX-format
9210 debugging information is enabled at all). If you don't define the
9211 macro, the default is 1: always generate the extended information
9212 if there is any occasion to.
9215 @defmac DEBUG_SYMS_TEXT
9216 Define this macro if all @code{.stabs} commands should be output while
9217 in the text section.
9220 @defmac ASM_STABS_OP
9221 A C string constant, including spacing, naming the assembler pseudo op to
9222 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9223 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9224 applies only to DBX debugging information format.
9227 @defmac ASM_STABD_OP
9228 A C string constant, including spacing, naming the assembler pseudo op to
9229 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9230 value is the current location. If you don't define this macro,
9231 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9235 @defmac ASM_STABN_OP
9236 A C string constant, including spacing, naming the assembler pseudo op to
9237 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9238 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9239 macro applies only to DBX debugging information format.
9242 @defmac DBX_NO_XREFS
9243 Define this macro if DBX on your system does not support the construct
9244 @samp{xs@var{tagname}}. On some systems, this construct is used to
9245 describe a forward reference to a structure named @var{tagname}.
9246 On other systems, this construct is not supported at all.
9249 @defmac DBX_CONTIN_LENGTH
9250 A symbol name in DBX-format debugging information is normally
9251 continued (split into two separate @code{.stabs} directives) when it
9252 exceeds a certain length (by default, 80 characters). On some
9253 operating systems, DBX requires this splitting; on others, splitting
9254 must not be done. You can inhibit splitting by defining this macro
9255 with the value zero. You can override the default splitting-length by
9256 defining this macro as an expression for the length you desire.
9259 @defmac DBX_CONTIN_CHAR
9260 Normally continuation is indicated by adding a @samp{\} character to
9261 the end of a @code{.stabs} string when a continuation follows. To use
9262 a different character instead, define this macro as a character
9263 constant for the character you want to use. Do not define this macro
9264 if backslash is correct for your system.
9267 @defmac DBX_STATIC_STAB_DATA_SECTION
9268 Define this macro if it is necessary to go to the data section before
9269 outputting the @samp{.stabs} pseudo-op for a non-global static
9273 @defmac DBX_TYPE_DECL_STABS_CODE
9274 The value to use in the ``code'' field of the @code{.stabs} directive
9275 for a typedef. The default is @code{N_LSYM}.
9278 @defmac DBX_STATIC_CONST_VAR_CODE
9279 The value to use in the ``code'' field of the @code{.stabs} directive
9280 for a static variable located in the text section. DBX format does not
9281 provide any ``right'' way to do this. The default is @code{N_FUN}.
9284 @defmac DBX_REGPARM_STABS_CODE
9285 The value to use in the ``code'' field of the @code{.stabs} directive
9286 for a parameter passed in registers. DBX format does not provide any
9287 ``right'' way to do this. The default is @code{N_RSYM}.
9290 @defmac DBX_REGPARM_STABS_LETTER
9291 The letter to use in DBX symbol data to identify a symbol as a parameter
9292 passed in registers. DBX format does not customarily provide any way to
9293 do this. The default is @code{'P'}.
9296 @defmac DBX_FUNCTION_FIRST
9297 Define this macro if the DBX information for a function and its
9298 arguments should precede the assembler code for the function. Normally,
9299 in DBX format, the debugging information entirely follows the assembler
9303 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9304 Define this macro, with value 1, if the value of a symbol describing
9305 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9306 relative to the start of the enclosing function. Normally, GCC uses
9307 an absolute address.
9310 @defmac DBX_LINES_FUNCTION_RELATIVE
9311 Define this macro, with value 1, if the value of a symbol indicating
9312 the current line number (@code{N_SLINE}) should be relative to the
9313 start of the enclosing function. Normally, GCC uses an absolute address.
9316 @defmac DBX_USE_BINCL
9317 Define this macro if GCC should generate @code{N_BINCL} and
9318 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9319 macro also directs GCC to output a type number as a pair of a file
9320 number and a type number within the file. Normally, GCC does not
9321 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9322 number for a type number.
9326 @subsection Open-Ended Hooks for DBX Format
9328 @c prevent bad page break with this line
9329 These are hooks for DBX format.
9331 @defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name})
9332 Define this macro to say how to output to @var{stream} the debugging
9333 information for the start of a scope level for variable names. The
9334 argument @var{name} is the name of an assembler symbol (for use with
9335 @code{assemble_name}) whose value is the address where the scope begins.
9338 @defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name})
9339 Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level.
9342 @defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl})
9343 Define this macro if the target machine requires special handling to
9344 output an @code{N_FUN} entry for the function @var{decl}.
9347 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9348 A C statement to output DBX debugging information before code for line
9349 number @var{line} of the current source file to the stdio stream
9350 @var{stream}. @var{counter} is the number of time the macro was
9351 invoked, including the current invocation; it is intended to generate
9352 unique labels in the assembly output.
9354 This macro should not be defined if the default output is correct, or
9355 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9358 @defmac NO_DBX_FUNCTION_END
9359 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9360 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9361 On those machines, define this macro to turn this feature off without
9362 disturbing the rest of the gdb extensions.
9365 @defmac NO_DBX_BNSYM_ENSYM
9366 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9367 extension construct. On those machines, define this macro to turn this
9368 feature off without disturbing the rest of the gdb extensions.
9371 @node File Names and DBX
9372 @subsection File Names in DBX Format
9374 @c prevent bad page break with this line
9375 This describes file names in DBX format.
9377 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9378 A C statement to output DBX debugging information to the stdio stream
9379 @var{stream}, which indicates that file @var{name} is the main source
9380 file---the file specified as the input file for compilation.
9381 This macro is called only once, at the beginning of compilation.
9383 This macro need not be defined if the standard form of output
9384 for DBX debugging information is appropriate.
9386 It may be necessary to refer to a label equal to the beginning of the
9387 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9388 to do so. If you do this, you must also set the variable
9389 @var{used_ltext_label_name} to @code{true}.
9392 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9393 Define this macro, with value 1, if GCC should not emit an indication
9394 of the current directory for compilation and current source language at
9395 the beginning of the file.
9398 @defmac NO_DBX_GCC_MARKER
9399 Define this macro, with value 1, if GCC should not emit an indication
9400 that this object file was compiled by GCC@. The default is to emit
9401 an @code{N_OPT} stab at the beginning of every source file, with
9402 @samp{gcc2_compiled.} for the string and value 0.
9405 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9406 A C statement to output DBX debugging information at the end of
9407 compilation of the main source file @var{name}. Output should be
9408 written to the stdio stream @var{stream}.
9410 If you don't define this macro, nothing special is output at the end
9411 of compilation, which is correct for most machines.
9414 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9415 Define this macro @emph{instead of} defining
9416 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9417 the end of compilation is an @code{N_SO} stab with an empty string,
9418 whose value is the highest absolute text address in the file.
9423 @subsection Macros for SDB and DWARF Output
9425 @c prevent bad page break with this line
9426 Here are macros for SDB and DWARF output.
9428 @defmac SDB_DEBUGGING_INFO
9429 Define this macro if GCC should produce COFF-style debugging output
9430 for SDB in response to the @option{-g} option.
9433 @defmac DWARF2_DEBUGGING_INFO
9434 Define this macro if GCC should produce dwarf version 2 format
9435 debugging output in response to the @option{-g} option.
9437 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function})
9438 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9439 be emitted for each function. Instead of an integer return the enum
9440 value for the @code{DW_CC_} tag.
9443 To support optional call frame debugging information, you must also
9444 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9445 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9446 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9447 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9450 @defmac DWARF2_FRAME_INFO
9451 Define this macro to a nonzero value if GCC should always output
9452 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9453 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9454 exceptions are enabled, GCC will output this information not matter
9455 how you define @code{DWARF2_FRAME_INFO}.
9458 @deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void)
9459 This hook defines the mechanism that will be used for describing frame
9460 unwind information to the debugger. Normally the hook will return
9461 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9462 return @code{UI_NONE} otherwise.
9464 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9465 is disabled in order to always output DWARF 2 frame information.
9467 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9468 This will suppress generation of the normal debug frame unwind information.
9471 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9472 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9473 line debug info sections. This will result in much more compact line number
9474 tables, and hence is desirable if it works.
9477 @deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS
9478 True if the @code{.debug_pubtypes} and @code{.debug_pubnames} sections should be emitted. These sections are not used on most platforms, and in particular GDB does not use them.
9481 @deftypevr {Target Hook} bool TARGET_DELAY_SCHED2
9482 True if sched2 is not to be run at its normal place. This usually means it will be run as part of machine-specific reorg.
9485 @deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK
9486 True if vartrack is not to be run at its normal place. This usually means it will be run as part of machine-specific reorg.
9489 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9490 A C statement to issue assembly directives that create a difference
9491 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9494 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9495 A C statement to issue assembly directives that create a difference
9496 between the two given labels in system defined units, e.g. instruction
9497 slots on IA64 VMS, using an integer of the given size.
9500 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9501 A C statement to issue assembly directives that create a
9502 section-relative reference to the given @var{label}, using an integer of the
9503 given @var{size}. The label is known to be defined in the given @var{section}.
9506 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9507 A C statement to issue assembly directives that create a self-relative
9508 reference to the given @var{label}, using an integer of the given @var{size}.
9511 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9512 A C statement to issue assembly directives that create a reference to
9513 the DWARF table identifier @var{label} from the current section. This
9514 is used on some systems to avoid garbage collecting a DWARF table which
9515 is referenced by a function.
9518 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x})
9519 If defined, this target hook is a function which outputs a DTP-relative
9520 reference to the given TLS symbol of the specified size.
9523 @defmac PUT_SDB_@dots{}
9524 Define these macros to override the assembler syntax for the special
9525 SDB assembler directives. See @file{sdbout.c} for a list of these
9526 macros and their arguments. If the standard syntax is used, you need
9527 not define them yourself.
9531 Some assemblers do not support a semicolon as a delimiter, even between
9532 SDB assembler directives. In that case, define this macro to be the
9533 delimiter to use (usually @samp{\n}). It is not necessary to define
9534 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9538 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9539 Define this macro to allow references to unknown structure,
9540 union, or enumeration tags to be emitted. Standard COFF does not
9541 allow handling of unknown references, MIPS ECOFF has support for
9545 @defmac SDB_ALLOW_FORWARD_REFERENCES
9546 Define this macro to allow references to structure, union, or
9547 enumeration tags that have not yet been seen to be handled. Some
9548 assemblers choke if forward tags are used, while some require it.
9551 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9552 A C statement to output SDB debugging information before code for line
9553 number @var{line} of the current source file to the stdio stream
9554 @var{stream}. The default is to emit an @code{.ln} directive.
9559 @subsection Macros for VMS Debug Format
9561 @c prevent bad page break with this line
9562 Here are macros for VMS debug format.
9564 @defmac VMS_DEBUGGING_INFO
9565 Define this macro if GCC should produce debugging output for VMS
9566 in response to the @option{-g} option. The default behavior for VMS
9567 is to generate minimal debug info for a traceback in the absence of
9568 @option{-g} unless explicitly overridden with @option{-g0}. This
9569 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
9570 @code{TARGET_OPTION_OVERRIDE}.
9573 @node Floating Point
9574 @section Cross Compilation and Floating Point
9575 @cindex cross compilation and floating point
9576 @cindex floating point and cross compilation
9578 While all modern machines use twos-complement representation for integers,
9579 there are a variety of representations for floating point numbers. This
9580 means that in a cross-compiler the representation of floating point numbers
9581 in the compiled program may be different from that used in the machine
9582 doing the compilation.
9584 Because different representation systems may offer different amounts of
9585 range and precision, all floating point constants must be represented in
9586 the target machine's format. Therefore, the cross compiler cannot
9587 safely use the host machine's floating point arithmetic; it must emulate
9588 the target's arithmetic. To ensure consistency, GCC always uses
9589 emulation to work with floating point values, even when the host and
9590 target floating point formats are identical.
9592 The following macros are provided by @file{real.h} for the compiler to
9593 use. All parts of the compiler which generate or optimize
9594 floating-point calculations must use these macros. They may evaluate
9595 their operands more than once, so operands must not have side effects.
9597 @defmac REAL_VALUE_TYPE
9598 The C data type to be used to hold a floating point value in the target
9599 machine's format. Typically this is a @code{struct} containing an
9600 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9604 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9605 Compares for equality the two values, @var{x} and @var{y}. If the target
9606 floating point format supports negative zeroes and/or NaNs,
9607 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9608 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9611 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9612 Tests whether @var{x} is less than @var{y}.
9615 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9616 Truncates @var{x} to a signed integer, rounding toward zero.
9619 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9620 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9621 @var{x} is negative, returns zero.
9624 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9625 Converts @var{string} into a floating point number in the target machine's
9626 representation for mode @var{mode}. This routine can handle both
9627 decimal and hexadecimal floating point constants, using the syntax
9628 defined by the C language for both.
9631 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9632 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9635 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9636 Determines whether @var{x} represents infinity (positive or negative).
9639 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9640 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9643 @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})
9644 Calculates an arithmetic operation on the two floating point values
9645 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9648 The operation to be performed is specified by @var{code}. Only the
9649 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9650 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9652 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9653 target's floating point format cannot represent infinity, it will call
9654 @code{abort}. Callers should check for this situation first, using
9655 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9658 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9659 Returns the negative of the floating point value @var{x}.
9662 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9663 Returns the absolute value of @var{x}.
9666 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x})
9667 Truncates the floating point value @var{x} to fit in @var{mode}. The
9668 return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an
9669 appropriate bit pattern to be output as a floating constant whose
9670 precision accords with mode @var{mode}.
9673 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9674 Converts a floating point value @var{x} into a double-precision integer
9675 which is then stored into @var{low} and @var{high}. If the value is not
9676 integral, it is truncated.
9679 @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})
9680 Converts a double-precision integer found in @var{low} and @var{high},
9681 into a floating point value which is then stored into @var{x}. The
9682 value is truncated to fit in mode @var{mode}.
9685 @node Mode Switching
9686 @section Mode Switching Instructions
9687 @cindex mode switching
9688 The following macros control mode switching optimizations:
9690 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9691 Define this macro if the port needs extra instructions inserted for mode
9692 switching in an optimizing compilation.
9694 For an example, the SH4 can perform both single and double precision
9695 floating point operations, but to perform a single precision operation,
9696 the FPSCR PR bit has to be cleared, while for a double precision
9697 operation, this bit has to be set. Changing the PR bit requires a general
9698 purpose register as a scratch register, hence these FPSCR sets have to
9699 be inserted before reload, i.e.@: you can't put this into instruction emitting
9700 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9702 You can have multiple entities that are mode-switched, and select at run time
9703 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9704 return nonzero for any @var{entity} that needs mode-switching.
9705 If you define this macro, you also have to define
9706 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9707 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9708 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9712 @defmac NUM_MODES_FOR_MODE_SWITCHING
9713 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9714 initializer for an array of integers. Each initializer element
9715 N refers to an entity that needs mode switching, and specifies the number
9716 of different modes that might need to be set for this entity.
9717 The position of the initializer in the initializer---starting counting at
9718 zero---determines the integer that is used to refer to the mode-switched
9720 In macros that take mode arguments / yield a mode result, modes are
9721 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9722 switch is needed / supplied.
9725 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9726 @var{entity} is an integer specifying a mode-switched entity. If
9727 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9728 return an integer value not larger than the corresponding element in
9729 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9730 be switched into prior to the execution of @var{insn}.
9733 @defmac MODE_AFTER (@var{mode}, @var{insn})
9734 If this macro is defined, it is evaluated for every @var{insn} during
9735 mode switching. It determines the mode that an insn results in (if
9736 different from the incoming mode).
9739 @defmac MODE_ENTRY (@var{entity})
9740 If this macro is defined, it is evaluated for every @var{entity} that needs
9741 mode switching. It should evaluate to an integer, which is a mode that
9742 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9743 is defined then @code{MODE_EXIT} must be defined.
9746 @defmac MODE_EXIT (@var{entity})
9747 If this macro is defined, it is evaluated for every @var{entity} that needs
9748 mode switching. It should evaluate to an integer, which is a mode that
9749 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9750 is defined then @code{MODE_ENTRY} must be defined.
9753 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9754 This macro specifies the order in which modes for @var{entity} are processed.
9755 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9756 lowest. The value of the macro should be an integer designating a mode
9757 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9758 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9759 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9762 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9763 Generate one or more insns to set @var{entity} to @var{mode}.
9764 @var{hard_reg_live} is the set of hard registers live at the point where
9765 the insn(s) are to be inserted.
9768 @node Target Attributes
9769 @section Defining target-specific uses of @code{__attribute__}
9770 @cindex target attributes
9771 @cindex machine attributes
9772 @cindex attributes, target-specific
9774 Target-specific attributes may be defined for functions, data and types.
9775 These are described using the following target hooks; they also need to
9776 be documented in @file{extend.texi}.
9778 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
9779 If defined, this target hook points to an array of @samp{struct
9780 attribute_spec} (defined in @file{tree.h}) specifying the machine
9781 specific attributes for this target and some of the restrictions on the
9782 entities to which these attributes are applied and the arguments they
9786 @deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name})
9787 If defined, this target hook is a function which returns true if the
9788 machine-specific attribute named @var{name} expects an identifier
9789 given as its first argument to be passed on as a plain identifier, not
9790 subjected to name lookup. If this is not defined, the default is
9791 false for all machine-specific attributes.
9794 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2})
9795 If defined, this target hook is a function which returns zero if the attributes on
9796 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9797 and two if they are nearly compatible (which causes a warning to be
9798 generated). If this is not defined, machine-specific attributes are
9799 supposed always to be compatible.
9802 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
9803 If defined, this target hook is a function which assigns default attributes to
9804 the newly defined @var{type}.
9807 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9808 Define this target hook if the merging of type attributes needs special
9809 handling. If defined, the result is a list of the combined
9810 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9811 that @code{comptypes} has already been called and returned 1. This
9812 function may call @code{merge_attributes} to handle machine-independent
9816 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
9817 Define this target hook if the merging of decl attributes needs special
9818 handling. If defined, the result is a list of the combined
9819 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9820 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9821 when this is needed are when one attribute overrides another, or when an
9822 attribute is nullified by a subsequent definition. This function may
9823 call @code{merge_attributes} to handle machine-independent merging.
9825 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9826 If the only target-specific handling you require is @samp{dllimport}
9827 for Microsoft Windows targets, you should define the macro
9828 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9829 will then define a function called
9830 @code{merge_dllimport_decl_attributes} which can then be defined as
9831 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9832 add @code{handle_dll_attribute} in the attribute table for your port
9833 to perform initial processing of the @samp{dllimport} and
9834 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9835 @file{i386/i386.c}, for example.
9838 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl})
9839 @var{decl} is a variable or function with @code{__attribute__((dllimport))} specified. Use this hook if the target needs to add extra validation checks to @code{handle_dll_attribute}.
9842 @defmac TARGET_DECLSPEC
9843 Define this macro to a nonzero value if you want to treat
9844 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9845 default, this behavior is enabled only for targets that define
9846 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9847 of @code{__declspec} is via a built-in macro, but you should not rely
9848 on this implementation detail.
9851 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
9852 Define this target hook if you want to be able to add attributes to a decl
9853 when it is being created. This is normally useful for back ends which
9854 wish to implement a pragma by using the attributes which correspond to
9855 the pragma's effect. The @var{node} argument is the decl which is being
9856 created. The @var{attr_ptr} argument is a pointer to the attribute list
9857 for this decl. The list itself should not be modified, since it may be
9858 shared with other decls, but attributes may be chained on the head of
9859 the list and @code{*@var{attr_ptr}} modified to point to the new
9860 attributes, or a copy of the list may be made if further changes are
9864 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl})
9866 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9867 into the current function, despite its having target-specific
9868 attributes, @code{false} otherwise. By default, if a function has a
9869 target specific attribute attached to it, it will not be inlined.
9872 @deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags})
9873 This hook is called to parse the @code{attribute(option("..."))}, and
9874 it allows the function to set different target machine compile time
9875 options for the current function that might be different than the
9876 options specified on the command line. The hook should return
9877 @code{true} if the options are valid.
9879 The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9880 the function declaration to hold a pointer to a target specific
9881 @var{struct cl_target_option} structure.
9884 @deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr})
9885 This hook is called to save any additional target specific information
9886 in the @var{struct cl_target_option} structure for function specific
9888 @xref{Option file format}.
9891 @deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct cl_target_option *@var{ptr})
9892 This hook is called to restore any additional target specific
9893 information in the @var{struct cl_target_option} structure for
9894 function specific options.
9897 @deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr})
9898 This hook is called to print any additional target specific
9899 information in the @var{struct cl_target_option} structure for
9900 function specific options.
9903 @deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target})
9904 This target hook parses the options for @code{#pragma GCC option} to
9905 set the machine specific options for functions that occur later in the
9906 input stream. The options should be the same as handled by the
9907 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
9910 @deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void)
9911 Sometimes certain combinations of command options do not make sense on
9912 a particular target machine. You can override the hook
9913 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9914 once just after all the command options have been parsed.
9916 Don't use this hook to turn on various extra optimizations for
9917 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
9919 If you need to do something whenever the optimization level is
9920 changed via the optimize attribute or pragma, see
9921 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9924 @deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee})
9925 This target hook returns @code{false} if the @var{caller} function
9926 cannot inline @var{callee}, based on target specific information. By
9927 default, inlining is not allowed if the callee function has function
9928 specific target options and the caller does not use the same options.
9932 @section Emulating TLS
9933 @cindex Emulated TLS
9935 For targets whose psABI does not provide Thread Local Storage via
9936 specific relocations and instruction sequences, an emulation layer is
9937 used. A set of target hooks allows this emulation layer to be
9938 configured for the requirements of a particular target. For instance
9939 the psABI may in fact specify TLS support in terms of an emulation
9942 The emulation layer works by creating a control object for every TLS
9943 object. To access the TLS object, a lookup function is provided
9944 which, when given the address of the control object, will return the
9945 address of the current thread's instance of the TLS object.
9947 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
9948 Contains the name of the helper function that uses a TLS control
9949 object to locate a TLS instance. The default causes libgcc's
9950 emulated TLS helper function to be used.
9953 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
9954 Contains the name of the helper function that should be used at
9955 program startup to register TLS objects that are implicitly
9956 initialized to zero. If this is @code{NULL}, all TLS objects will
9957 have explicit initializers. The default causes libgcc's emulated TLS
9958 registration function to be used.
9961 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
9962 Contains the name of the section in which TLS control variables should
9963 be placed. The default of @code{NULL} allows these to be placed in
9967 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
9968 Contains the name of the section in which TLS initializers should be
9969 placed. The default of @code{NULL} allows these to be placed in any
9973 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
9974 Contains the prefix to be prepended to TLS control variable names.
9975 The default of @code{NULL} uses a target-specific prefix.
9978 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
9979 Contains the prefix to be prepended to TLS initializer objects. The
9980 default of @code{NULL} uses a target-specific prefix.
9983 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
9984 Specifies a function that generates the FIELD_DECLs for a TLS control
9985 object type. @var{type} is the RECORD_TYPE the fields are for and
9986 @var{name} should be filled with the structure tag, if the default of
9987 @code{__emutls_object} is unsuitable. The default creates a type suitable
9988 for libgcc's emulated TLS function.
9991 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
9992 Specifies a function that generates the CONSTRUCTOR to initialize a
9993 TLS control object. @var{var} is the TLS control object, @var{decl}
9994 is the TLS object and @var{tmpl_addr} is the address of the
9995 initializer. The default initializes libgcc's emulated TLS control object.
9998 @deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED
9999 Specifies whether the alignment of TLS control variable objects is
10000 fixed and should not be increased as some backends may do to optimize
10001 single objects. The default is false.
10004 @deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
10005 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
10006 may be used to describe emulated TLS control objects.
10009 @node MIPS Coprocessors
10010 @section Defining coprocessor specifics for MIPS targets.
10011 @cindex MIPS coprocessor-definition macros
10013 The MIPS specification allows MIPS implementations to have as many as 4
10014 coprocessors, each with as many as 32 private registers. GCC supports
10015 accessing these registers and transferring values between the registers
10016 and memory using asm-ized variables. For example:
10019 register unsigned int cp0count asm ("c0r1");
10025 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
10026 names may be added as described below, or the default names may be
10027 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
10029 Coprocessor registers are assumed to be epilogue-used; sets to them will
10030 be preserved even if it does not appear that the register is used again
10031 later in the function.
10033 Another note: according to the MIPS spec, coprocessor 1 (if present) is
10034 the FPU@. One accesses COP1 registers through standard mips
10035 floating-point support; they are not included in this mechanism.
10037 There is one macro used in defining the MIPS coprocessor interface which
10038 you may want to override in subtargets; it is described below.
10040 @defmac ALL_COP_ADDITIONAL_REGISTER_NAMES
10041 A comma-separated list (with leading comma) of pairs describing the
10042 alternate names of coprocessor registers. The format of each entry should be
10044 @{ @var{alternatename}, @var{register_number}@}
10050 @section Parameters for Precompiled Header Validity Checking
10051 @cindex parameters, precompiled headers
10053 @deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
10054 This hook returns a pointer to the data needed by
10055 @code{TARGET_PCH_VALID_P} and sets
10056 @samp{*@var{sz}} to the size of the data in bytes.
10059 @deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
10060 This hook checks whether the options used to create a PCH file are
10061 compatible with the current settings. It returns @code{NULL}
10062 if so and a suitable error message if not. Error messages will
10063 be presented to the user and must be localized using @samp{_(@var{msg})}.
10065 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
10066 when the PCH file was created and @var{sz} is the size of that data in bytes.
10067 It's safe to assume that the data was created by the same version of the
10068 compiler, so no format checking is needed.
10070 The default definition of @code{default_pch_valid_p} should be
10071 suitable for most targets.
10074 @deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
10075 If this hook is nonnull, the default implementation of
10076 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
10077 of @code{target_flags}. @var{pch_flags} specifies the value that
10078 @code{target_flags} had when the PCH file was created. The return
10079 value is the same as for @code{TARGET_PCH_VALID_P}.
10082 @deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void)
10083 Called before writing out a PCH file. If the target has some
10084 garbage-collected data that needs to be in a particular state on PCH loads,
10085 it can use this hook to enforce that state. Very few targets need
10086 to do anything here.
10090 @section C++ ABI parameters
10091 @cindex parameters, c++ abi
10093 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
10094 Define this hook to override the integer type used for guard variables.
10095 These are used to implement one-time construction of static objects. The
10096 default is long_long_integer_type_node.
10099 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
10100 This hook determines how guard variables are used. It should return
10101 @code{false} (the default) if the first byte should be used. A return value of
10102 @code{true} indicates that only the least significant bit should be used.
10105 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
10106 This hook returns the size of the cookie to use when allocating an array
10107 whose elements have the indicated @var{type}. Assumes that it is already
10108 known that a cookie is needed. The default is
10109 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
10110 IA64/Generic C++ ABI@.
10113 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
10114 This hook should return @code{true} if the element size should be stored in
10115 array cookies. The default is to return @code{false}.
10118 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
10119 If defined by a backend this hook allows the decision made to export
10120 class @var{type} to be overruled. Upon entry @var{import_export}
10121 will contain 1 if the class is going to be exported, @minus{}1 if it is going
10122 to be imported and 0 otherwise. This function should return the
10123 modified value and perform any other actions necessary to support the
10124 backend's targeted operating system.
10127 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
10128 This hook should return @code{true} if constructors and destructors return
10129 the address of the object created/destroyed. The default is to return
10133 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
10134 This hook returns true if the key method for a class (i.e., the method
10135 which, if defined in the current translation unit, causes the virtual
10136 table to be emitted) may be an inline function. Under the standard
10137 Itanium C++ ABI the key method may be an inline function so long as
10138 the function is not declared inline in the class definition. Under
10139 some variants of the ABI, an inline function can never be the key
10140 method. The default is to return @code{true}.
10143 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
10144 @var{decl} is a virtual table, virtual table table, typeinfo object, or other similar implicit class data object that will be emitted with external linkage in this translation unit. No ELF visibility has been explicitly specified. If the target needs to specify a visibility other than that of the containing class, use this hook to set @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
10147 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
10148 This hook returns true (the default) if virtual tables and other
10149 similar implicit class data objects are always COMDAT if they have
10150 external linkage. If this hook returns false, then class data for
10151 classes whose virtual table will be emitted in only one translation
10152 unit will not be COMDAT.
10155 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
10156 This hook returns true (the default) if the RTTI information for
10157 the basic types which is defined in the C++ runtime should always
10158 be COMDAT, false if it should not be COMDAT.
10161 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
10162 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10163 should be used to register static destructors when @option{-fuse-cxa-atexit}
10164 is in effect. The default is to return false to use @code{__cxa_atexit}.
10167 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
10168 This hook returns true if the target @code{atexit} function can be used
10169 in the same manner as @code{__cxa_atexit} to register C++ static
10170 destructors. This requires that @code{atexit}-registered functions in
10171 shared libraries are run in the correct order when the libraries are
10172 unloaded. The default is to return false.
10175 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
10176 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been defined. Use this hook to make adjustments to the class (eg, tweak visibility or perform any other required target modifications).
10179 @deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl})
10180 Return target-specific mangling context of @var{decl} or @code{NULL_TREE}.
10183 @node Named Address Spaces
10184 @section Adding support for named address spaces
10185 @cindex named address spaces
10187 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10188 standards committee, @cite{Programming Languages - C - Extensions to
10189 support embedded processors}, specifies a syntax for embedded
10190 processors to specify alternate address spaces. You can configure a
10191 GCC port to support section 5.1 of the draft report to add support for
10192 address spaces other than the default address space. These address
10193 spaces are new keywords that are similar to the @code{volatile} and
10194 @code{const} type attributes.
10196 Pointers to named address spaces can have a different size than
10197 pointers to the generic address space.
10199 For example, the SPU port uses the @code{__ea} address space to refer
10200 to memory in the host processor, rather than memory local to the SPU
10201 processor. Access to memory in the @code{__ea} address space involves
10202 issuing DMA operations to move data between the host processor and the
10203 local processor memory address space. Pointers in the @code{__ea}
10204 address space are either 32 bits or 64 bits based on the
10205 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10208 Internally, address spaces are represented as a small integer in the
10209 range 0 to 15 with address space 0 being reserved for the generic
10212 To register a named address space qualifier keyword with the C front end,
10213 the target may call the @code{c_register_addr_space} routine. For example,
10214 the SPU port uses the following to declare @code{__ea} as the keyword for
10215 named address space #1:
10217 #define ADDR_SPACE_EA 1
10218 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10221 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space})
10222 Define this to return the machine mode to use for pointers to
10223 @var{address_space} if the target supports named address spaces.
10224 The default version of this hook returns @code{ptr_mode} for the
10225 generic address space only.
10228 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space})
10229 Define this to return the machine mode to use for addresses in
10230 @var{address_space} if the target supports named address spaces.
10231 The default version of this hook returns @code{Pmode} for the
10232 generic address space only.
10235 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (enum machine_mode @var{mode}, addr_space_t @var{as})
10236 Define this to return nonzero if the port can handle pointers
10237 with machine mode @var{mode} to address space @var{as}. This target
10238 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10239 except that it includes explicit named address space support. The default
10240 version of this hook returns true for the modes returned by either the
10241 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10242 target hooks for the given address space.
10245 @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})
10246 Define this to return true if @var{exp} is a valid address for mode
10247 @var{mode} in the named address space @var{as}. The @var{strict}
10248 parameter says whether strict addressing is in effect after reload has
10249 finished. This target hook is the same as the
10250 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10251 explicit named address space support.
10254 @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})
10255 Define this to modify an invalid address @var{x} to be a valid address
10256 with mode @var{mode} in the named address space @var{as}. This target
10257 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10258 except that it includes explicit named address space support.
10261 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset})
10262 Define this to return whether the @var{subset} named address space is
10263 contained within the @var{superset} named address space. Pointers to
10264 a named address space that is a subset of another named address space
10265 will be converted automatically without a cast if used together in
10266 arithmetic operations. Pointers to a superset address space can be
10267 converted to pointers to a subset address space via explicit casts.
10270 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type})
10271 Define this to convert the pointer expression represented by the RTL
10272 @var{op} with type @var{from_type} that points to a named address
10273 space to a new pointer expression with type @var{to_type} that points
10274 to a different named address space. When this hook it called, it is
10275 guaranteed that one of the two address spaces is a subset of the other,
10276 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10280 @section Miscellaneous Parameters
10281 @cindex parameters, miscellaneous
10283 @c prevent bad page break with this line
10284 Here are several miscellaneous parameters.
10286 @defmac HAS_LONG_COND_BRANCH
10287 Define this boolean macro to indicate whether or not your architecture
10288 has conditional branches that can span all of memory. It is used in
10289 conjunction with an optimization that partitions hot and cold basic
10290 blocks into separate sections of the executable. If this macro is
10291 set to false, gcc will convert any conditional branches that attempt
10292 to cross between sections into unconditional branches or indirect jumps.
10295 @defmac HAS_LONG_UNCOND_BRANCH
10296 Define this boolean macro to indicate whether or not your architecture
10297 has unconditional branches that can span all of memory. It is used in
10298 conjunction with an optimization that partitions hot and cold basic
10299 blocks into separate sections of the executable. If this macro is
10300 set to false, gcc will convert any unconditional branches that attempt
10301 to cross between sections into indirect jumps.
10304 @defmac CASE_VECTOR_MODE
10305 An alias for a machine mode name. This is the machine mode that
10306 elements of a jump-table should have.
10309 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10310 Optional: return the preferred mode for an @code{addr_diff_vec}
10311 when the minimum and maximum offset are known. If you define this,
10312 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10313 To make this work, you also have to define @code{INSN_ALIGN} and
10314 make the alignment for @code{addr_diff_vec} explicit.
10315 The @var{body} argument is provided so that the offset_unsigned and scale
10316 flags can be updated.
10319 @defmac CASE_VECTOR_PC_RELATIVE
10320 Define this macro to be a C expression to indicate when jump-tables
10321 should contain relative addresses. You need not define this macro if
10322 jump-tables never contain relative addresses, or jump-tables should
10323 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10327 @deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void)
10328 This function return the smallest number of different values for which it
10329 is best to use a jump-table instead of a tree of conditional branches.
10330 The default is four for machines with a @code{casesi} instruction and
10331 five otherwise. This is best for most machines.
10334 @defmac CASE_USE_BIT_TESTS
10335 Define this macro to be a C expression to indicate whether C switch
10336 statements may be implemented by a sequence of bit tests. This is
10337 advantageous on processors that can efficiently implement left shift
10338 of 1 by the number of bits held in a register, but inappropriate on
10339 targets that would require a loop. By default, this macro returns
10340 @code{true} if the target defines an @code{ashlsi3} pattern, and
10341 @code{false} otherwise.
10344 @defmac WORD_REGISTER_OPERATIONS
10345 Define this macro if operations between registers with integral mode
10346 smaller than a word are always performed on the entire register.
10347 Most RISC machines have this property and most CISC machines do not.
10350 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10351 Define this macro to be a C expression indicating when insns that read
10352 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10353 bits outside of @var{mem_mode} to be either the sign-extension or the
10354 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10355 of @var{mem_mode} for which the
10356 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10357 @code{UNKNOWN} for other modes.
10359 This macro is not called with @var{mem_mode} non-integral or with a width
10360 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10361 value in this case. Do not define this macro if it would always return
10362 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10363 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10365 You may return a non-@code{UNKNOWN} value even if for some hard registers
10366 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10367 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10368 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10369 integral mode larger than this but not larger than @code{word_mode}.
10371 You must return @code{UNKNOWN} if for some hard registers that allow this
10372 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10373 @code{word_mode}, but that they can change to another integral mode that
10374 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10377 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10378 Define this macro if loading short immediate values into registers sign
10382 @defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC
10383 Define this macro if the same instructions that convert a floating
10384 point number to a signed fixed point number also convert validly to an
10388 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
10389 When @option{-ffast-math} is in effect, GCC tries to optimize
10390 divisions by the same divisor, by turning them into multiplications by
10391 the reciprocal. This target hook specifies the minimum number of divisions
10392 that should be there for GCC to perform the optimization for a variable
10393 of mode @var{mode}. The default implementation returns 3 if the machine
10394 has an instruction for the division, and 2 if it does not.
10398 The maximum number of bytes that a single instruction can move quickly
10399 between memory and registers or between two memory locations.
10402 @defmac MAX_MOVE_MAX
10403 The maximum number of bytes that a single instruction can move quickly
10404 between memory and registers or between two memory locations. If this
10405 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10406 constant value that is the largest value that @code{MOVE_MAX} can have
10410 @defmac SHIFT_COUNT_TRUNCATED
10411 A C expression that is nonzero if on this machine the number of bits
10412 actually used for the count of a shift operation is equal to the number
10413 of bits needed to represent the size of the object being shifted. When
10414 this macro is nonzero, the compiler will assume that it is safe to omit
10415 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10416 truncates the count of a shift operation. On machines that have
10417 instructions that act on bit-fields at variable positions, which may
10418 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10419 also enables deletion of truncations of the values that serve as
10420 arguments to bit-field instructions.
10422 If both types of instructions truncate the count (for shifts) and
10423 position (for bit-field operations), or if no variable-position bit-field
10424 instructions exist, you should define this macro.
10426 However, on some machines, such as the 80386 and the 680x0, truncation
10427 only applies to shift operations and not the (real or pretended)
10428 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10429 such machines. Instead, add patterns to the @file{md} file that include
10430 the implied truncation of the shift instructions.
10432 You need not define this macro if it would always have the value of zero.
10435 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10436 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
10437 This function describes how the standard shift patterns for @var{mode}
10438 deal with shifts by negative amounts or by more than the width of the mode.
10439 @xref{shift patterns}.
10441 On many machines, the shift patterns will apply a mask @var{m} to the
10442 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10443 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10444 this is true for mode @var{mode}, the function should return @var{m},
10445 otherwise it should return 0. A return value of 0 indicates that no
10446 particular behavior is guaranteed.
10448 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10449 @emph{not} apply to general shift rtxes; it applies only to instructions
10450 that are generated by the named shift patterns.
10452 The default implementation of this function returns
10453 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10454 and 0 otherwise. This definition is always safe, but if
10455 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10456 nevertheless truncate the shift count, you may get better code
10460 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10461 A C expression which is nonzero if on this machine it is safe to
10462 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10463 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10464 operating on it as if it had only @var{outprec} bits.
10466 On many machines, this expression can be 1.
10468 @c rearranged this, removed the phrase "it is reported that". this was
10469 @c to fix an overfull hbox. --mew 10feb93
10470 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10471 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10472 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10473 such cases may improve things.
10476 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode})
10477 The representation of an integral mode can be such that the values
10478 are always extended to a wider integral mode. Return
10479 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10480 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10481 otherwise. (Currently, none of the targets use zero-extended
10482 representation this way so unlike @code{LOAD_EXTEND_OP},
10483 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10484 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10485 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10486 widest integral mode and currently we take advantage of this fact.)
10488 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10489 value even if the extension is not performed on certain hard registers
10490 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10491 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10493 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10494 describe two related properties. If you define
10495 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10496 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10499 In order to enforce the representation of @code{mode},
10500 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10504 @defmac STORE_FLAG_VALUE
10505 A C expression describing the value returned by a comparison operator
10506 with an integral mode and stored by a store-flag instruction
10507 (@samp{cstore@var{mode}4}) when the condition is true. This description must
10508 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10509 comparison operators whose results have a @code{MODE_INT} mode.
10511 A value of 1 or @minus{}1 means that the instruction implementing the
10512 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10513 and 0 when the comparison is false. Otherwise, the value indicates
10514 which bits of the result are guaranteed to be 1 when the comparison is
10515 true. This value is interpreted in the mode of the comparison
10516 operation, which is given by the mode of the first operand in the
10517 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10518 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10521 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10522 generate code that depends only on the specified bits. It can also
10523 replace comparison operators with equivalent operations if they cause
10524 the required bits to be set, even if the remaining bits are undefined.
10525 For example, on a machine whose comparison operators return an
10526 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10527 @samp{0x80000000}, saying that just the sign bit is relevant, the
10531 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10535 can be converted to
10538 (ashift:SI @var{x} (const_int @var{n}))
10542 where @var{n} is the appropriate shift count to move the bit being
10543 tested into the sign bit.
10545 There is no way to describe a machine that always sets the low-order bit
10546 for a true value, but does not guarantee the value of any other bits,
10547 but we do not know of any machine that has such an instruction. If you
10548 are trying to port GCC to such a machine, include an instruction to
10549 perform a logical-and of the result with 1 in the pattern for the
10550 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10552 Often, a machine will have multiple instructions that obtain a value
10553 from a comparison (or the condition codes). Here are rules to guide the
10554 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10559 Use the shortest sequence that yields a valid definition for
10560 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10561 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10562 comparison operators to do so because there may be opportunities to
10563 combine the normalization with other operations.
10566 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10567 slightly preferred on machines with expensive jumps and 1 preferred on
10571 As a second choice, choose a value of @samp{0x80000001} if instructions
10572 exist that set both the sign and low-order bits but do not define the
10576 Otherwise, use a value of @samp{0x80000000}.
10579 Many machines can produce both the value chosen for
10580 @code{STORE_FLAG_VALUE} and its negation in the same number of
10581 instructions. On those machines, you should also define a pattern for
10582 those cases, e.g., one matching
10585 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10588 Some machines can also perform @code{and} or @code{plus} operations on
10589 condition code values with less instructions than the corresponding
10590 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10591 machines, define the appropriate patterns. Use the names @code{incscc}
10592 and @code{decscc}, respectively, for the patterns which perform
10593 @code{plus} or @code{minus} operations on condition code values. See
10594 @file{rs6000.md} for some examples. The GNU Superoptimizer can be used to
10595 find such instruction sequences on other machines.
10597 If this macro is not defined, the default value, 1, is used. You need
10598 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10599 instructions, or if the value generated by these instructions is 1.
10602 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10603 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10604 returned when comparison operators with floating-point results are true.
10605 Define this macro on machines that have comparison operations that return
10606 floating-point values. If there are no such operations, do not define
10610 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10611 A C expression that gives a rtx representing the nonzero true element
10612 for vector comparisons. The returned rtx should be valid for the inner
10613 mode of @var{mode} which is guaranteed to be a vector mode. Define
10614 this macro on machines that have vector comparison operations that
10615 return a vector result. If there are no such operations, do not define
10616 this macro. Typically, this macro is defined as @code{const1_rtx} or
10617 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10618 the compiler optimizing such vector comparison operations for the
10622 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10623 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10624 A C expression that indicates whether the architecture defines a value
10625 for @code{clz} or @code{ctz} with a zero operand.
10626 A result of @code{0} indicates the value is undefined.
10627 If the value is defined for only the RTL expression, the macro should
10628 evaluate to @code{1}; if the value applies also to the corresponding optab
10629 entry (which is normally the case if it expands directly into
10630 the corresponding RTL), then the macro should evaluate to @code{2}.
10631 In the cases where the value is defined, @var{value} should be set to
10634 If this macro is not defined, the value of @code{clz} or
10635 @code{ctz} at zero is assumed to be undefined.
10637 This macro must be defined if the target's expansion for @code{ffs}
10638 relies on a particular value to get correct results. Otherwise it
10639 is not necessary, though it may be used to optimize some corner cases, and
10640 to provide a default expansion for the @code{ffs} optab.
10642 Note that regardless of this macro the ``definedness'' of @code{clz}
10643 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10644 visible to the user. Thus one may be free to adjust the value at will
10645 to match the target expansion of these operations without fear of
10650 An alias for the machine mode for pointers. On most machines, define
10651 this to be the integer mode corresponding to the width of a hardware
10652 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10653 On some machines you must define this to be one of the partial integer
10654 modes, such as @code{PSImode}.
10656 The width of @code{Pmode} must be at least as large as the value of
10657 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10658 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10662 @defmac FUNCTION_MODE
10663 An alias for the machine mode used for memory references to functions
10664 being called, in @code{call} RTL expressions. On most CISC machines,
10665 where an instruction can begin at any byte address, this should be
10666 @code{QImode}. On most RISC machines, where all instructions have fixed
10667 size and alignment, this should be a mode with the same size and alignment
10668 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10671 @defmac STDC_0_IN_SYSTEM_HEADERS
10672 In normal operation, the preprocessor expands @code{__STDC__} to the
10673 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10674 hosts, like Solaris, the system compiler uses a different convention,
10675 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10676 strict conformance to the C Standard.
10678 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10679 convention when processing system header files, but when processing user
10680 files @code{__STDC__} will always expand to 1.
10683 @defmac NO_IMPLICIT_EXTERN_C
10684 Define this macro if the system header files support C++ as well as C@.
10685 This macro inhibits the usual method of using system header files in
10686 C++, which is to pretend that the file's contents are enclosed in
10687 @samp{extern "C" @{@dots{}@}}.
10692 @defmac REGISTER_TARGET_PRAGMAS ()
10693 Define this macro if you want to implement any target-specific pragmas.
10694 If defined, it is a C expression which makes a series of calls to
10695 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10696 for each pragma. The macro may also do any
10697 setup required for the pragmas.
10699 The primary reason to define this macro is to provide compatibility with
10700 other compilers for the same target. In general, we discourage
10701 definition of target-specific pragmas for GCC@.
10703 If the pragma can be implemented by attributes then you should consider
10704 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10706 Preprocessor macros that appear on pragma lines are not expanded. All
10707 @samp{#pragma} directives that do not match any registered pragma are
10708 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10711 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10712 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10714 Each call to @code{c_register_pragma} or
10715 @code{c_register_pragma_with_expansion} establishes one pragma. The
10716 @var{callback} routine will be called when the preprocessor encounters a
10720 #pragma [@var{space}] @var{name} @dots{}
10723 @var{space} is the case-sensitive namespace of the pragma, or
10724 @code{NULL} to put the pragma in the global namespace. The callback
10725 routine receives @var{pfile} as its first argument, which can be passed
10726 on to cpplib's functions if necessary. You can lex tokens after the
10727 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10728 callback will be silently ignored. The end of the line is indicated by
10729 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10730 arguments of pragmas registered with
10731 @code{c_register_pragma_with_expansion} but not on the arguments of
10732 pragmas registered with @code{c_register_pragma}.
10734 Note that the use of @code{pragma_lex} is specific to the C and C++
10735 compilers. It will not work in the Java or Fortran compilers, or any
10736 other language compilers for that matter. Thus if @code{pragma_lex} is going
10737 to be called from target-specific code, it must only be done so when
10738 building the C and C++ compilers. This can be done by defining the
10739 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10740 target entry in the @file{config.gcc} file. These variables should name
10741 the target-specific, language-specific object file which contains the
10742 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10743 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10744 how to build this object file.
10747 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10748 Define this macro if macros should be expanded in the
10749 arguments of @samp{#pragma pack}.
10752 @deftypevr {Target Hook} bool TARGET_HANDLE_PRAGMA_EXTERN_PREFIX
10753 True if @code{#pragma extern_prefix} is to be supported.
10756 @defmac TARGET_DEFAULT_PACK_STRUCT
10757 If your target requires a structure packing default other than 0 (meaning
10758 the machine default), define this macro to the necessary value (in bytes).
10759 This must be a value that would also be valid to use with
10760 @samp{#pragma pack()} (that is, a small power of two).
10763 @defmac DOLLARS_IN_IDENTIFIERS
10764 Define this macro to control use of the character @samp{$} in
10765 identifier names for the C family of languages. 0 means @samp{$} is
10766 not allowed by default; 1 means it is allowed. 1 is the default;
10767 there is no need to define this macro in that case.
10770 @defmac NO_DOLLAR_IN_LABEL
10771 Define this macro if the assembler does not accept the character
10772 @samp{$} in label names. By default constructors and destructors in
10773 G++ have @samp{$} in the identifiers. If this macro is defined,
10774 @samp{.} is used instead.
10777 @defmac NO_DOT_IN_LABEL
10778 Define this macro if the assembler does not accept the character
10779 @samp{.} in label names. By default constructors and destructors in G++
10780 have names that use @samp{.}. If this macro is defined, these names
10781 are rewritten to avoid @samp{.}.
10784 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10785 Define this macro as a C expression that is nonzero if it is safe for the
10786 delay slot scheduler to place instructions in the delay slot of @var{insn},
10787 even if they appear to use a resource set or clobbered in @var{insn}.
10788 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10789 every @code{call_insn} has this behavior. On machines where some @code{insn}
10790 or @code{jump_insn} is really a function call and hence has this behavior,
10791 you should define this macro.
10793 You need not define this macro if it would always return zero.
10796 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10797 Define this macro as a C expression that is nonzero if it is safe for the
10798 delay slot scheduler to place instructions in the delay slot of @var{insn},
10799 even if they appear to set or clobber a resource referenced in @var{insn}.
10800 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10801 some @code{insn} or @code{jump_insn} is really a function call and its operands
10802 are registers whose use is actually in the subroutine it calls, you should
10803 define this macro. Doing so allows the delay slot scheduler to move
10804 instructions which copy arguments into the argument registers into the delay
10805 slot of @var{insn}.
10807 You need not define this macro if it would always return zero.
10810 @defmac MULTIPLE_SYMBOL_SPACES
10811 Define this macro as a C expression that is nonzero if, in some cases,
10812 global symbols from one translation unit may not be bound to undefined
10813 symbols in another translation unit without user intervention. For
10814 instance, under Microsoft Windows symbols must be explicitly imported
10815 from shared libraries (DLLs).
10817 You need not define this macro if it would always evaluate to zero.
10820 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
10821 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10822 any hard regs the port wishes to automatically clobber for an asm.
10823 It should return the result of the last @code{tree_cons} used to add a
10824 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10825 corresponding parameters to the asm and may be inspected to avoid
10826 clobbering a register that is an input or output of the asm. You can use
10827 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10828 for overlap with regards to asm-declared registers.
10831 @defmac MATH_LIBRARY
10832 Define this macro as a C string constant for the linker argument to link
10833 in the system math library, minus the initial @samp{"-l"}, or
10834 @samp{""} if the target does not have a
10835 separate math library.
10837 You need only define this macro if the default of @samp{"m"} is wrong.
10840 @defmac LIBRARY_PATH_ENV
10841 Define this macro as a C string constant for the environment variable that
10842 specifies where the linker should look for libraries.
10844 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10848 @defmac TARGET_POSIX_IO
10849 Define this macro if the target supports the following POSIX@ file
10850 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10851 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10852 to use file locking when exiting a program, which avoids race conditions
10853 if the program has forked. It will also create directories at run-time
10854 for cross-profiling.
10857 @defmac MAX_CONDITIONAL_EXECUTE
10859 A C expression for the maximum number of instructions to execute via
10860 conditional execution instructions instead of a branch. A value of
10861 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10862 1 if it does use cc0.
10865 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10866 Used if the target needs to perform machine-dependent modifications on the
10867 conditionals used for turning basic blocks into conditionally executed code.
10868 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10869 contains information about the currently processed blocks. @var{true_expr}
10870 and @var{false_expr} are the tests that are used for converting the
10871 then-block and the else-block, respectively. Set either @var{true_expr} or
10872 @var{false_expr} to a null pointer if the tests cannot be converted.
10875 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10876 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10877 if-statements into conditions combined by @code{and} and @code{or} operations.
10878 @var{bb} contains the basic block that contains the test that is currently
10879 being processed and about to be turned into a condition.
10882 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10883 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10884 be converted to conditional execution format. @var{ce_info} points to
10885 a data structure, @code{struct ce_if_block}, which contains information
10886 about the currently processed blocks.
10889 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10890 A C expression to perform any final machine dependent modifications in
10891 converting code to conditional execution. The involved basic blocks
10892 can be found in the @code{struct ce_if_block} structure that is pointed
10893 to by @var{ce_info}.
10896 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10897 A C expression to cancel any machine dependent modifications in
10898 converting code to conditional execution. The involved basic blocks
10899 can be found in the @code{struct ce_if_block} structure that is pointed
10900 to by @var{ce_info}.
10903 @defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info})
10904 A C expression to initialize any extra fields in a @code{struct ce_if_block}
10905 structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro.
10908 @defmac IFCVT_EXTRA_FIELDS
10909 If defined, it should expand to a set of field declarations that will be
10910 added to the @code{struct ce_if_block} structure. These should be initialized
10911 by the @code{IFCVT_INIT_EXTRA_FIELDS} macro.
10914 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void)
10915 If non-null, this hook performs a target-specific pass over the
10916 instruction stream. The compiler will run it at all optimization levels,
10917 just before the point at which it normally does delayed-branch scheduling.
10919 The exact purpose of the hook varies from target to target. Some use
10920 it to do transformations that are necessary for correctness, such as
10921 laying out in-function constant pools or avoiding hardware hazards.
10922 Others use it as an opportunity to do some machine-dependent optimizations.
10924 You need not implement the hook if it has nothing to do. The default
10925 definition is null.
10928 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void)
10929 Define this hook if you have any machine-specific built-in functions
10930 that need to be defined. It should be a function that performs the
10933 Machine specific built-in functions can be useful to expand special machine
10934 instructions that would otherwise not normally be generated because
10935 they have no equivalent in the source language (for example, SIMD vector
10936 instructions or prefetch instructions).
10938 To create a built-in function, call the function
10939 @code{lang_hooks.builtin_function}
10940 which is defined by the language front end. You can use any type nodes set
10941 up by @code{build_common_tree_nodes};
10942 only language front ends that use those two functions will call
10943 @samp{TARGET_INIT_BUILTINS}.
10946 @deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p})
10947 Define this hook if you have any machine-specific built-in functions
10948 that need to be defined. It should be a function that returns the
10949 builtin function declaration for the builtin function code @var{code}.
10950 If there is no such builtin and it cannot be initialized at this time
10951 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10952 If @var{code} is out of range the function should return
10953 @code{error_mark_node}.
10956 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
10958 Expand a call to a machine specific built-in function that was set up by
10959 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10960 function call; the result should go to @var{target} if that is
10961 convenient, and have mode @var{mode} if that is convenient.
10962 @var{subtarget} may be used as the target for computing one of
10963 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10964 ignored. This function should return the result of the call to the
10968 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist})
10969 Select a replacement for a machine specific built-in function that
10970 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10971 @emph{before} regular type checking, and so allows the target to
10972 implement a crude form of function overloading. @var{fndecl} is the
10973 declaration of the built-in function. @var{arglist} is the list of
10974 arguments passed to the built-in function. The result is a
10975 complete expression that implements the operation, usually
10976 another @code{CALL_EXPR}.
10977 @var{arglist} really has type @samp{VEC(tree,gc)*}
10980 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore})
10981 Fold a call to a machine specific built-in function that was set up by
10982 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10983 built-in function. @var{n_args} is the number of arguments passed to
10984 the function; the arguments themselves are pointed to by @var{argp}.
10985 The result is another tree containing a simplified expression for the
10986 call's result. If @var{ignore} is true the value will be ignored.
10989 @deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const_rtx @var{insn})
10991 Take an instruction in @var{insn} and return NULL if it is valid within a
10992 low-overhead loop, otherwise return a string explaining why doloop
10993 could not be applied.
10995 Many targets use special registers for low-overhead looping. For any
10996 instruction that clobbers these this function should return a string indicating
10997 the reason why the doloop could not be applied.
10998 By default, the RTL loop optimizer does not use a present doloop pattern for
10999 loops containing function calls or branch on table instructions.
11002 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
11004 Take a branch insn in @var{branch1} and another in @var{branch2}.
11005 Return true if redirecting @var{branch1} to the destination of
11006 @var{branch2} is possible.
11008 On some targets, branches may have a limited range. Optimizing the
11009 filling of delay slots can result in branches being redirected, and this
11010 may in turn cause a branch offset to overflow.
11013 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code})
11014 This target hook returns @code{true} if @var{x} is considered to be commutative.
11015 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
11016 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
11017 of the enclosing rtl, if known, otherwise it is UNKNOWN.
11020 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
11022 When the initial value of a hard register has been copied in a pseudo
11023 register, it is often not necessary to actually allocate another register
11024 to this pseudo register, because the original hard register or a stack slot
11025 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
11026 is called at the start of register allocation once for each hard register
11027 that had its initial value copied by using
11028 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
11029 Possible values are @code{NULL_RTX}, if you don't want
11030 to do any special allocation, a @code{REG} rtx---that would typically be
11031 the hard register itself, if it is known not to be clobbered---or a
11033 If you are returning a @code{MEM}, this is only a hint for the allocator;
11034 it might decide to use another register anyways.
11035 You may use @code{current_function_leaf_function} in the hook, functions
11036 that use @code{REG_N_SETS}, to determine if the hard
11037 register in question will not be clobbered.
11038 The default value of this hook is @code{NULL}, which disables any special
11042 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
11043 This target hook returns nonzero if @var{x}, an @code{unspec} or
11044 @code{unspec_volatile} operation, might cause a trap. Targets can use
11045 this hook to enhance precision of analysis for @code{unspec} and
11046 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
11047 to analyze inner elements of @var{x} in which case @var{flags} should be
11051 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
11052 The compiler invokes this hook whenever it changes its current function
11053 context (@code{cfun}). You can define this function if
11054 the back end needs to perform any initialization or reset actions on a
11055 per-function basis. For example, it may be used to implement function
11056 attributes that affect register usage or code generation patterns.
11057 The argument @var{decl} is the declaration for the new function context,
11058 and may be null to indicate that the compiler has left a function context
11059 and is returning to processing at the top level.
11060 The default hook function does nothing.
11062 GCC sets @code{cfun} to a dummy function context during initialization of
11063 some parts of the back end. The hook function is not invoked in this
11064 situation; you need not worry about the hook being invoked recursively,
11065 or when the back end is in a partially-initialized state.
11066 @code{cfun} might be @code{NULL} to indicate processing at top level,
11067 outside of any function scope.
11070 @defmac TARGET_OBJECT_SUFFIX
11071 Define this macro to be a C string representing the suffix for object
11072 files on your target machine. If you do not define this macro, GCC will
11073 use @samp{.o} as the suffix for object files.
11076 @defmac TARGET_EXECUTABLE_SUFFIX
11077 Define this macro to be a C string representing the suffix to be
11078 automatically added to executable files on your target machine. If you
11079 do not define this macro, GCC will use the null string as the suffix for
11083 @defmac COLLECT_EXPORT_LIST
11084 If defined, @code{collect2} will scan the individual object files
11085 specified on its command line and create an export list for the linker.
11086 Define this macro for systems like AIX, where the linker discards
11087 object files that are not referenced from @code{main} and uses export
11091 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
11092 Define this macro to a C expression representing a variant of the
11093 method call @var{mdecl}, if Java Native Interface (JNI) methods
11094 must be invoked differently from other methods on your target.
11095 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
11096 the @code{stdcall} calling convention and this macro is then
11097 defined as this expression:
11100 build_type_attribute_variant (@var{mdecl},
11102 (get_identifier ("stdcall"),
11107 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
11108 This target hook returns @code{true} past the point in which new jump
11109 instructions could be created. On machines that require a register for
11110 every jump such as the SHmedia ISA of SH5, this point would typically be
11111 reload, so this target hook should be defined to a function such as:
11115 cannot_modify_jumps_past_reload_p ()
11117 return (reload_completed || reload_in_progress);
11122 @deftypefn {Target Hook} reg_class_t TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
11123 This target hook returns a register class for which branch target register
11124 optimizations should be applied. All registers in this class should be
11125 usable interchangeably. After reload, registers in this class will be
11126 re-allocated and loads will be hoisted out of loops and be subjected
11127 to inter-block scheduling.
11130 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
11131 Branch target register optimization will by default exclude callee-saved
11133 that are not already live during the current function; if this target hook
11134 returns true, they will be included. The target code must than make sure
11135 that all target registers in the class returned by
11136 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
11137 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
11138 epilogues have already been generated. Note, even if you only return
11139 true when @var{after_prologue_epilogue_gen} is false, you still are likely
11140 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
11141 to reserve space for caller-saved target registers.
11144 @deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
11145 This target hook returns true if the target supports conditional execution.
11146 This target hook is required only when the target has several different
11147 modes and they have different conditional execution capability, such as ARM.
11150 @deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, struct loop *@var{loop})
11151 This target hook returns a new value for the number of times @var{loop}
11152 should be unrolled. The parameter @var{nunroll} is the number of times
11153 the loop is to be unrolled. The parameter @var{loop} is a pointer to
11154 the loop, which is going to be checked for unrolling. This target hook
11155 is required only when the target has special constraints like maximum
11156 number of memory accesses.
11159 @defmac POWI_MAX_MULTS
11160 If defined, this macro is interpreted as a signed integer C expression
11161 that specifies the maximum number of floating point multiplications
11162 that should be emitted when expanding exponentiation by an integer
11163 constant inline. When this value is defined, exponentiation requiring
11164 more than this number of multiplications is implemented by calling the
11165 system library's @code{pow}, @code{powf} or @code{powl} routines.
11166 The default value places no upper bound on the multiplication count.
11169 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11170 This target hook should register any extra include files for the
11171 target. The parameter @var{stdinc} indicates if normal include files
11172 are present. The parameter @var{sysroot} is the system root directory.
11173 The parameter @var{iprefix} is the prefix for the gcc directory.
11176 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11177 This target hook should register any extra include files for the
11178 target before any standard headers. The parameter @var{stdinc}
11179 indicates if normal include files are present. The parameter
11180 @var{sysroot} is the system root directory. The parameter
11181 @var{iprefix} is the prefix for the gcc directory.
11184 @deftypefn Macro void TARGET_OPTF (char *@var{path})
11185 This target hook should register special include paths for the target.
11186 The parameter @var{path} is the include to register. On Darwin
11187 systems, this is used for Framework includes, which have semantics
11188 that are different from @option{-I}.
11191 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11192 This target macro returns @code{true} if it is safe to use a local alias
11193 for a virtual function @var{fndecl} when constructing thunks,
11194 @code{false} otherwise. By default, the macro returns @code{true} for all
11195 functions, if a target supports aliases (i.e.@: defines
11196 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11199 @defmac TARGET_FORMAT_TYPES
11200 If defined, this macro is the name of a global variable containing
11201 target-specific format checking information for the @option{-Wformat}
11202 option. The default is to have no target-specific format checks.
11205 @defmac TARGET_N_FORMAT_TYPES
11206 If defined, this macro is the number of entries in
11207 @code{TARGET_FORMAT_TYPES}.
11210 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11211 If defined, this macro is the name of a global variable containing
11212 target-specific format overrides for the @option{-Wformat} option. The
11213 default is to have no target-specific format overrides. If defined,
11214 @code{TARGET_FORMAT_TYPES} must be defined, too.
11217 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11218 If defined, this macro specifies the number of entries in
11219 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11222 @defmac TARGET_OVERRIDES_FORMAT_INIT
11223 If defined, this macro specifies the optional initialization
11224 routine for target specific customizations of the system printf
11225 and scanf formatter settings.
11228 @deftypevr {Target Hook} bool TARGET_RELAXED_ORDERING
11229 If set to @code{true}, means that the target's memory model does not
11230 guarantee that loads which do not depend on one another will access
11231 main memory in the order of the instruction stream; if ordering is
11232 important, an explicit memory barrier must be used. This is true of
11233 many recent processors which implement a policy of ``relaxed,''
11234 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11235 and ia64. The default is @code{false}.
11238 @deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val})
11239 If defined, this macro returns the diagnostic message when it is
11240 illegal to pass argument @var{val} to function @var{funcdecl}
11241 with prototype @var{typelist}.
11244 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype})
11245 If defined, this macro returns the diagnostic message when it is
11246 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11247 if validity should be determined by the front end.
11250 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type})
11251 If defined, this macro returns the diagnostic message when it is
11252 invalid to apply operation @var{op} (where unary plus is denoted by
11253 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11254 if validity should be determined by the front end.
11257 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2})
11258 If defined, this macro returns the diagnostic message when it is
11259 invalid to apply operation @var{op} to operands of types @var{type1}
11260 and @var{type2}, or @code{NULL} if validity should be determined by
11264 @deftypefn {Target Hook} {const char *} TARGET_INVALID_PARAMETER_TYPE (const_tree @var{type})
11265 If defined, this macro returns the diagnostic message when it is
11266 invalid for functions to include parameters of type @var{type},
11267 or @code{NULL} if validity should be determined by
11268 the front end. This is currently used only by the C and C++ front ends.
11271 @deftypefn {Target Hook} {const char *} TARGET_INVALID_RETURN_TYPE (const_tree @var{type})
11272 If defined, this macro returns the diagnostic message when it is
11273 invalid for functions to have return type @var{type},
11274 or @code{NULL} if validity should be determined by
11275 the front end. This is currently used only by the C and C++ front ends.
11278 @deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type})
11279 If defined, this target hook returns the type to which values of
11280 @var{type} should be promoted when they appear in expressions,
11281 analogous to the integer promotions, or @code{NULL_TREE} to use the
11282 front end's normal promotion rules. This hook is useful when there are
11283 target-specific types with special promotion rules.
11284 This is currently used only by the C and C++ front ends.
11287 @deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr})
11288 If defined, this hook returns the result of converting @var{expr} to
11289 @var{type}. It should return the converted expression,
11290 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11291 This hook is useful when there are target-specific types with special
11293 This is currently used only by the C and C++ front ends.
11296 @defmac TARGET_USE_JCR_SECTION
11297 This macro determines whether to use the JCR section to register Java
11298 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11299 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11303 This macro determines the size of the objective C jump buffer for the
11304 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11307 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11308 Define this macro if any target-specific attributes need to be attached
11309 to the functions in @file{libgcc} that provide low-level support for
11310 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11311 and the associated definitions of those functions.
11314 @deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void)
11315 Define this macro to update the current function stack boundary if
11319 @deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void)
11320 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11321 different argument pointer register is needed to access the function's
11322 argument list due to stack realignment. Return @code{NULL} if no DRAP
11326 @deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
11327 When optimization is disabled, this hook indicates whether or not
11328 arguments should be allocated to stack slots. Normally, GCC allocates
11329 stacks slots for arguments when not optimizing in order to make
11330 debugging easier. However, when a function is declared with
11331 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11332 cannot safely move arguments from the registers in which they are passed
11333 to the stack. Therefore, this hook should return true in general, but
11334 false for naked functions. The default implementation always returns true.
11337 @deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR
11338 On some architectures it can take multiple instructions to synthesize
11339 a constant. If there is another constant already in a register that
11340 is close enough in value then it is preferable that the new constant
11341 is computed from this register using immediate addition or
11342 subtraction. We accomplish this through CSE. Besides the value of
11343 the constant we also add a lower and an upper constant anchor to the
11344 available expressions. These are then queried when encountering new
11345 constants. The anchors are computed by rounding the constant up and
11346 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11347 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11348 accepted by immediate-add plus one. We currently assume that the
11349 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11350 MIPS, where add-immediate takes a 16-bit signed value,
11351 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11352 is zero, which disables this optimization. @end deftypevr