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, 2012
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 @deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT
391 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.
394 @defmac MULTILIB_DEFAULTS
395 Define this macro as a C expression for the initializer of an array of
396 string to tell the driver program which options are defaults for this
397 target and thus do not need to be handled specially when using
398 @code{MULTILIB_OPTIONS}.
400 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
401 the target makefile fragment or if none of the options listed in
402 @code{MULTILIB_OPTIONS} are set by default.
403 @xref{Target Fragment}.
406 @defmac RELATIVE_PREFIX_NOT_LINKDIR
407 Define this macro to tell @command{gcc} that it should only translate
408 a @option{-B} prefix into a @option{-L} linker option if the prefix
409 indicates an absolute file name.
412 @defmac MD_EXEC_PREFIX
413 If defined, this macro is an additional prefix to try after
414 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
415 when the compiler is built as a cross
416 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
417 to the list of directories used to find the assembler in @file{configure.in}.
420 @defmac STANDARD_STARTFILE_PREFIX
421 Define this macro as a C string constant if you wish to override the
422 standard choice of @code{libdir} as the default prefix to
423 try when searching for startup files such as @file{crt0.o}.
424 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
425 is built as a cross compiler.
428 @defmac STANDARD_STARTFILE_PREFIX_1
429 Define this macro as a C string constant if you wish to override the
430 standard choice of @code{/lib} as a prefix to try after the default prefix
431 when searching for startup files such as @file{crt0.o}.
432 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
433 is built as a cross compiler.
436 @defmac STANDARD_STARTFILE_PREFIX_2
437 Define this macro as a C string constant if you wish to override the
438 standard choice of @code{/lib} as yet another prefix to try after the
439 default prefix when searching for startup files such as @file{crt0.o}.
440 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
441 is built as a cross compiler.
444 @defmac MD_STARTFILE_PREFIX
445 If defined, this macro supplies an additional prefix to try after the
446 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
447 compiler is built as a cross compiler.
450 @defmac MD_STARTFILE_PREFIX_1
451 If defined, this macro supplies yet another prefix to try after the
452 standard prefixes. It is not searched when the compiler is built as a
456 @defmac INIT_ENVIRONMENT
457 Define this macro as a C string constant if you wish to set environment
458 variables for programs called by the driver, such as the assembler and
459 loader. The driver passes the value of this macro to @code{putenv} to
460 initialize the necessary environment variables.
463 @defmac LOCAL_INCLUDE_DIR
464 Define this macro as a C string constant if you wish to override the
465 standard choice of @file{/usr/local/include} as the default prefix to
466 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
467 comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in
468 @file{config.gcc}, normally @file{/usr/include}) in the search order.
470 Cross compilers do not search either @file{/usr/local/include} or its
474 @defmac NATIVE_SYSTEM_HEADER_COMPONENT
475 The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}.
476 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
477 If you do not define this macro, no component is used.
480 @defmac INCLUDE_DEFAULTS
481 Define this macro if you wish to override the entire default search path
482 for include files. For a native compiler, the default search path
483 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
484 @code{GPLUSPLUS_INCLUDE_DIR}, and
485 @code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
486 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
487 and specify private search areas for GCC@. The directory
488 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
490 The definition should be an initializer for an array of structures.
491 Each array element should have four elements: the directory name (a
492 string constant), the component name (also a string constant), a flag
493 for C++-only directories,
494 and a flag showing that the includes in the directory don't need to be
495 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
496 the array with a null element.
498 The component name denotes what GNU package the include file is part of,
499 if any, in all uppercase letters. For example, it might be @samp{GCC}
500 or @samp{BINUTILS}. If the package is part of a vendor-supplied
501 operating system, code the component name as @samp{0}.
503 For example, here is the definition used for VAX/VMS:
506 #define INCLUDE_DEFAULTS \
508 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
509 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
510 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
517 Here is the order of prefixes tried for exec files:
521 Any prefixes specified by the user with @option{-B}.
524 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
525 is not set and the compiler has not been installed in the configure-time
526 @var{prefix}, the location in which the compiler has actually been installed.
529 The directories specified by the environment variable @code{COMPILER_PATH}.
532 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
533 in the configured-time @var{prefix}.
536 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
539 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
542 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
546 Here is the order of prefixes tried for startfiles:
550 Any prefixes specified by the user with @option{-B}.
553 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
554 value based on the installed toolchain location.
557 The directories specified by the environment variable @code{LIBRARY_PATH}
558 (or port-specific name; native only, cross compilers do not use this).
561 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
562 in the configured @var{prefix} or this is a native compiler.
565 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
568 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
572 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
573 native compiler, or we have a target system root.
576 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
577 native compiler, or we have a target system root.
580 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
581 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
582 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
585 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
586 compiler, or we have a target system root. The default for this macro is
590 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
591 compiler, or we have a target system root. The default for this macro is
595 @node Run-time Target
596 @section Run-time Target Specification
597 @cindex run-time target specification
598 @cindex predefined macros
599 @cindex target specifications
601 @c prevent bad page break with this line
602 Here are run-time target specifications.
604 @defmac TARGET_CPU_CPP_BUILTINS ()
605 This function-like macro expands to a block of code that defines
606 built-in preprocessor macros and assertions for the target CPU, using
607 the functions @code{builtin_define}, @code{builtin_define_std} and
608 @code{builtin_assert}. When the front end
609 calls this macro it provides a trailing semicolon, and since it has
610 finished command line option processing your code can use those
613 @code{builtin_assert} takes a string in the form you pass to the
614 command-line option @option{-A}, such as @code{cpu=mips}, and creates
615 the assertion. @code{builtin_define} takes a string in the form
616 accepted by option @option{-D} and unconditionally defines the macro.
618 @code{builtin_define_std} takes a string representing the name of an
619 object-like macro. If it doesn't lie in the user's namespace,
620 @code{builtin_define_std} defines it unconditionally. Otherwise, it
621 defines a version with two leading underscores, and another version
622 with two leading and trailing underscores, and defines the original
623 only if an ISO standard was not requested on the command line. For
624 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
625 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
626 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
627 defines only @code{_ABI64}.
629 You can also test for the C dialect being compiled. The variable
630 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
631 or @code{clk_objective_c}. Note that if we are preprocessing
632 assembler, this variable will be @code{clk_c} but the function-like
633 macro @code{preprocessing_asm_p()} will return true, so you might want
634 to check for that first. If you need to check for strict ANSI, the
635 variable @code{flag_iso} can be used. The function-like macro
636 @code{preprocessing_trad_p()} can be used to check for traditional
640 @defmac TARGET_OS_CPP_BUILTINS ()
641 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
642 and is used for the target operating system instead.
645 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
646 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
647 and is used for the target object format. @file{elfos.h} uses this
648 macro to define @code{__ELF__}, so you probably do not need to define
652 @deftypevar {extern int} target_flags
653 This variable is declared in @file{options.h}, which is included before
654 any target-specific headers.
657 @deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
658 This variable specifies the initial value of @code{target_flags}.
659 Its default setting is 0.
662 @cindex optional hardware or system features
663 @cindex features, optional, in system conventions
665 @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})
666 This hook is called whenever the user specifies one of the
667 target-specific options described by the @file{.opt} definition files
668 (@pxref{Options}). It has the opportunity to do some option-specific
669 processing and should return true if the option is valid. The default
670 definition does nothing but return true.
672 @var{decoded} specifies the option and its arguments. @var{opts} and
673 @var{opts_set} are the @code{gcc_options} structures to be used for
674 storing option state, and @var{loc} is the location at which the
675 option was passed (@code{UNKNOWN_LOCATION} except for options passed
679 @deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
680 This target hook is called whenever the user specifies one of the
681 target-specific C language family options described by the @file{.opt}
682 definition files(@pxref{Options}). It has the opportunity to do some
683 option-specific processing and should return true if the option is
684 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
685 default definition does nothing but return false.
687 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
688 options. However, if processing an option requires routines that are
689 only available in the C (and related language) front ends, then you
690 should use @code{TARGET_HANDLE_C_OPTION} instead.
693 @deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string})
694 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.
697 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE (const char *@var{classname})
698 Declare that Objective C class @var{classname} is referenced by the current TU.
701 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char *@var{classname})
702 Declare that Objective C class @var{classname} is defined by the current TU.
705 @deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref})
706 If a target implements string objects then this hook should return @code{true} if @var{stringref} is a valid reference to such an object.
709 @deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list})
710 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.
713 @deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)
714 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
715 but is called when the optimize level is changed via an attribute or
716 pragma or when it is reset at the end of the code affected by the
717 attribute or pragma. It is not called at the beginning of compilation
718 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
719 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
720 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
723 @defmac C_COMMON_OVERRIDE_OPTIONS
724 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
725 but is only used in the C
726 language frontends (C, Objective-C, C++, Objective-C++) and so can be
727 used to alter option flag variables which only exist in those
731 @deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE
732 Some machines may desire to change what optimizations are performed for
733 various optimization levels. This variable, if defined, describes
734 options to enable at particular sets of optimization levels. These
735 options are processed once
736 just after the optimization level is determined and before the remainder
737 of the command options have been parsed, so may be overridden by other
738 options passed explicitly.
740 This processing is run once at program startup and when the optimization
741 options are changed via @code{#pragma GCC optimize} or by using the
742 @code{optimize} attribute.
745 @deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts})
746 Set target-dependent initial values of fields in @var{opts}.
749 @deftypefn {Common Target Hook} void TARGET_OPTION_DEFAULT_PARAMS (void)
750 Set target-dependent default values for @option{--param} settings, using calls to @code{set_default_param_value}.
753 @defmac SWITCHABLE_TARGET
754 Some targets need to switch between substantially different subtargets
755 during compilation. For example, the MIPS target has one subtarget for
756 the traditional MIPS architecture and another for MIPS16. Source code
757 can switch between these two subarchitectures using the @code{mips16}
758 and @code{nomips16} attributes.
760 Such subtargets can differ in things like the set of available
761 registers, the set of available instructions, the costs of various
762 operations, and so on. GCC caches a lot of this type of information
763 in global variables, and recomputing them for each subtarget takes a
764 significant amount of time. The compiler therefore provides a facility
765 for maintaining several versions of the global variables and quickly
766 switching between them; see @file{target-globals.h} for details.
768 Define this macro to 1 if your target needs this facility. The default
772 @node Per-Function Data
773 @section Defining data structures for per-function information.
774 @cindex per-function data
775 @cindex data structures
777 If the target needs to store information on a per-function basis, GCC
778 provides a macro and a couple of variables to allow this. Note, just
779 using statics to store the information is a bad idea, since GCC supports
780 nested functions, so you can be halfway through encoding one function
781 when another one comes along.
783 GCC defines a data structure called @code{struct function} which
784 contains all of the data specific to an individual function. This
785 structure contains a field called @code{machine} whose type is
786 @code{struct machine_function *}, which can be used by targets to point
787 to their own specific data.
789 If a target needs per-function specific data it should define the type
790 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
791 This macro should be used to initialize the function pointer
792 @code{init_machine_status}. This pointer is explained below.
794 One typical use of per-function, target specific data is to create an
795 RTX to hold the register containing the function's return address. This
796 RTX can then be used to implement the @code{__builtin_return_address}
797 function, for level 0.
799 Note---earlier implementations of GCC used a single data area to hold
800 all of the per-function information. Thus when processing of a nested
801 function began the old per-function data had to be pushed onto a
802 stack, and when the processing was finished, it had to be popped off the
803 stack. GCC used to provide function pointers called
804 @code{save_machine_status} and @code{restore_machine_status} to handle
805 the saving and restoring of the target specific information. Since the
806 single data area approach is no longer used, these pointers are no
809 @defmac INIT_EXPANDERS
810 Macro called to initialize any target specific information. This macro
811 is called once per function, before generation of any RTL has begun.
812 The intention of this macro is to allow the initialization of the
813 function pointer @code{init_machine_status}.
816 @deftypevar {void (*)(struct function *)} init_machine_status
817 If this function pointer is non-@code{NULL} it will be called once per
818 function, before function compilation starts, in order to allow the
819 target to perform any target specific initialization of the
820 @code{struct function} structure. It is intended that this would be
821 used to initialize the @code{machine} of that structure.
823 @code{struct machine_function} structures are expected to be freed by GC@.
824 Generally, any memory that they reference must be allocated by using
825 GC allocation, including the structure itself.
829 @section Storage Layout
830 @cindex storage layout
832 Note that the definitions of the macros in this table which are sizes or
833 alignments measured in bits do not need to be constant. They can be C
834 expressions that refer to static variables, such as the @code{target_flags}.
835 @xref{Run-time Target}.
837 @defmac BITS_BIG_ENDIAN
838 Define this macro to have the value 1 if the most significant bit in a
839 byte has the lowest number; otherwise define it to have the value zero.
840 This means that bit-field instructions count from the most significant
841 bit. If the machine has no bit-field instructions, then this must still
842 be defined, but it doesn't matter which value it is defined to. This
843 macro need not be a constant.
845 This macro does not affect the way structure fields are packed into
846 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
849 @defmac BYTES_BIG_ENDIAN
850 Define this macro to have the value 1 if the most significant byte in a
851 word has the lowest number. This macro need not be a constant.
854 @defmac WORDS_BIG_ENDIAN
855 Define this macro to have the value 1 if, in a multiword object, the
856 most significant word has the lowest number. This applies to both
857 memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the
858 order of words in memory is not the same as the order in registers. This
859 macro need not be a constant.
862 @defmac REG_WORDS_BIG_ENDIAN
863 On some machines, the order of words in a multiword object differs between
864 registers in memory. In such a situation, define this macro to describe
865 the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls
866 the order of words in memory.
869 @defmac FLOAT_WORDS_BIG_ENDIAN
870 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
871 @code{TFmode} floating point numbers are stored in memory with the word
872 containing the sign bit at the lowest address; otherwise define it to
873 have the value 0. This macro need not be a constant.
875 You need not define this macro if the ordering is the same as for
879 @defmac BITS_PER_UNIT
880 Define this macro to be the number of bits in an addressable storage
881 unit (byte). If you do not define this macro the default is 8.
884 @defmac BITS_PER_WORD
885 Number of bits in a word. If you do not define this macro, the default
886 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
889 @defmac MAX_BITS_PER_WORD
890 Maximum number of bits in a word. If this is undefined, the default is
891 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
892 largest value that @code{BITS_PER_WORD} can have at run-time.
895 @defmac UNITS_PER_WORD
896 Number of storage units in a word; normally the size of a general-purpose
897 register, a power of two from 1 or 8.
900 @defmac MIN_UNITS_PER_WORD
901 Minimum number of units in a word. If this is undefined, the default is
902 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
903 smallest value that @code{UNITS_PER_WORD} can have at run-time.
907 Width of a pointer, in bits. You must specify a value no wider than the
908 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
909 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
910 a value the default is @code{BITS_PER_WORD}.
913 @defmac POINTERS_EXTEND_UNSIGNED
914 A C expression that determines how pointers should be extended from
915 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
916 greater than zero if pointers should be zero-extended, zero if they
917 should be sign-extended, and negative if some other sort of conversion
918 is needed. In the last case, the extension is done by the target's
919 @code{ptr_extend} instruction.
921 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
922 and @code{word_mode} are all the same width.
925 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
926 A macro to update @var{m} and @var{unsignedp} when an object whose type
927 is @var{type} and which has the specified mode and signedness is to be
928 stored in a register. This macro is only called when @var{type} is a
931 On most RISC machines, which only have operations that operate on a full
932 register, define this macro to set @var{m} to @code{word_mode} if
933 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
934 cases, only integer modes should be widened because wider-precision
935 floating-point operations are usually more expensive than their narrower
938 For most machines, the macro definition does not change @var{unsignedp}.
939 However, some machines, have instructions that preferentially handle
940 either signed or unsigned quantities of certain modes. For example, on
941 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
942 sign-extend the result to 64 bits. On such machines, set
943 @var{unsignedp} according to which kind of extension is more efficient.
945 Do not define this macro if it would never modify @var{m}.
948 @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})
949 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
950 function return values. The target hook should return the new mode
951 and possibly change @code{*@var{punsignedp}} if the promotion should
952 change signedness. This function is called only for scalar @emph{or
955 @var{for_return} allows to distinguish the promotion of arguments and
956 return values. If it is @code{1}, a return value is being promoted and
957 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
958 If it is @code{2}, the returned mode should be that of the register in
959 which an incoming parameter is copied, or the outgoing result is computed;
960 then the hook should return the same mode as @code{promote_mode}, though
961 the signedness may be different.
963 @var{type} can be NULL when promoting function arguments of libcalls.
965 The default is to not promote arguments and return values. You can
966 also define the hook to @code{default_promote_function_mode_always_promote}
967 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
970 @defmac PARM_BOUNDARY
971 Normal alignment required for function parameters on the stack, in
972 bits. All stack parameters receive at least this much alignment
973 regardless of data type. On most machines, this is the same as the
977 @defmac STACK_BOUNDARY
978 Define this macro to the minimum alignment enforced by hardware for the
979 stack pointer on this machine. The definition is a C expression for the
980 desired alignment (measured in bits). This value is used as a default
981 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
982 this should be the same as @code{PARM_BOUNDARY}.
985 @defmac PREFERRED_STACK_BOUNDARY
986 Define this macro if you wish to preserve a certain alignment for the
987 stack pointer, greater than what the hardware enforces. The definition
988 is a C expression for the desired alignment (measured in bits). This
989 macro must evaluate to a value equal to or larger than
990 @code{STACK_BOUNDARY}.
993 @defmac INCOMING_STACK_BOUNDARY
994 Define this macro if the incoming stack boundary may be different
995 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
996 to a value equal to or larger than @code{STACK_BOUNDARY}.
999 @defmac FUNCTION_BOUNDARY
1000 Alignment required for a function entry point, in bits.
1003 @defmac BIGGEST_ALIGNMENT
1004 Biggest alignment that any data type can require on this machine, in
1005 bits. Note that this is not the biggest alignment that is supported,
1006 just the biggest alignment that, when violated, may cause a fault.
1009 @defmac MALLOC_ABI_ALIGNMENT
1010 Alignment, in bits, a C conformant malloc implementation has to
1011 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1014 @defmac ATTRIBUTE_ALIGNED_VALUE
1015 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1016 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1019 @defmac MINIMUM_ATOMIC_ALIGNMENT
1020 If defined, the smallest alignment, in bits, that can be given to an
1021 object that can be referenced in one operation, without disturbing any
1022 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1023 on machines that don't have byte or half-word store operations.
1026 @defmac BIGGEST_FIELD_ALIGNMENT
1027 Biggest alignment that any structure or union field can require on this
1028 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1029 structure and union fields only, unless the field alignment has been set
1030 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1033 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed})
1034 An expression for the alignment of a structure field @var{field} if the
1035 alignment computed in the usual way (including applying of
1036 @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1037 alignment) is @var{computed}. It overrides alignment only if the
1038 field alignment has not been set by the
1039 @code{__attribute__ ((aligned (@var{n})))} construct.
1042 @defmac MAX_STACK_ALIGNMENT
1043 Biggest stack alignment guaranteed by the backend. Use this macro
1044 to specify the maximum alignment of a variable on stack.
1046 If not defined, the default value is @code{STACK_BOUNDARY}.
1048 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1049 @c But the fix for PR 32893 indicates that we can only guarantee
1050 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1051 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1054 @defmac MAX_OFILE_ALIGNMENT
1055 Biggest alignment supported by the object file format of this machine.
1056 Use this macro to limit the alignment which can be specified using the
1057 @code{__attribute__ ((aligned (@var{n})))} construct. If not defined,
1058 the default value is @code{BIGGEST_ALIGNMENT}.
1060 On systems that use ELF, the default (in @file{config/elfos.h}) is
1061 the largest supported 32-bit ELF section alignment representable on
1062 a 32-bit host e.g. @samp{(((unsigned HOST_WIDEST_INT) 1 << 28) * 8)}.
1063 On 32-bit ELF the largest supported section alignment in bits is
1064 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1067 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1068 If defined, a C expression to compute the alignment for a variable in
1069 the static store. @var{type} is the data type, and @var{basic-align} is
1070 the alignment that the object would ordinarily have. The value of this
1071 macro is used instead of that alignment to align the object.
1073 If this macro is not defined, then @var{basic-align} is used.
1076 One use of this macro is to increase alignment of medium-size data to
1077 make it all fit in fewer cache lines. Another is to cause character
1078 arrays to be word-aligned so that @code{strcpy} calls that copy
1079 constants to character arrays can be done inline.
1082 @defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align})
1083 If defined, a C expression to compute the alignment given to a constant
1084 that is being placed in memory. @var{constant} is the constant and
1085 @var{basic-align} is the alignment that the object would ordinarily
1086 have. The value of this macro is used instead of that alignment to
1089 If this macro is not defined, then @var{basic-align} is used.
1091 The typical use of this macro is to increase alignment for string
1092 constants to be word aligned so that @code{strcpy} calls that copy
1093 constants can be done inline.
1096 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1097 If defined, a C expression to compute the alignment for a variable in
1098 the local store. @var{type} is the data type, and @var{basic-align} is
1099 the alignment that the object would ordinarily have. The value of this
1100 macro is used instead of that alignment to align the object.
1102 If this macro is not defined, then @var{basic-align} is used.
1104 One use of this macro is to increase alignment of medium-size data to
1105 make it all fit in fewer cache lines.
1107 If the value of this macro has a type, it should be an unsigned type.
1110 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1111 If defined, a C expression to compute the alignment for stack slot.
1112 @var{type} is the data type, @var{mode} is the widest mode available,
1113 and @var{basic-align} is the alignment that the slot would ordinarily
1114 have. The value of this macro is used instead of that alignment to
1117 If this macro is not defined, then @var{basic-align} is used when
1118 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1121 This macro is to set alignment of stack slot to the maximum alignment
1122 of all possible modes which the slot may have.
1124 If the value of this macro has a type, it should be an unsigned type.
1127 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1128 If defined, a C expression to compute the alignment for a local
1129 variable @var{decl}.
1131 If this macro is not defined, then
1132 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1135 One use of this macro is to increase alignment of medium-size data to
1136 make it all fit in fewer cache lines.
1138 If the value of this macro has a type, it should be an unsigned type.
1141 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1142 If defined, a C expression to compute the minimum required alignment
1143 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1144 @var{mode}, assuming normal alignment @var{align}.
1146 If this macro is not defined, then @var{align} will be used.
1149 @defmac EMPTY_FIELD_BOUNDARY
1150 Alignment in bits to be given to a structure bit-field that follows an
1151 empty field such as @code{int : 0;}.
1153 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1156 @defmac STRUCTURE_SIZE_BOUNDARY
1157 Number of bits which any structure or union's size must be a multiple of.
1158 Each structure or union's size is rounded up to a multiple of this.
1160 If you do not define this macro, the default is the same as
1161 @code{BITS_PER_UNIT}.
1164 @defmac STRICT_ALIGNMENT
1165 Define this macro to be the value 1 if instructions will fail to work
1166 if given data not on the nominal alignment. If instructions will merely
1167 go slower in that case, define this macro as 0.
1170 @defmac PCC_BITFIELD_TYPE_MATTERS
1171 Define this if you wish to imitate the way many other C compilers handle
1172 alignment of bit-fields and the structures that contain them.
1174 The behavior is that the type written for a named bit-field (@code{int},
1175 @code{short}, or other integer type) imposes an alignment for the entire
1176 structure, as if the structure really did contain an ordinary field of
1177 that type. In addition, the bit-field is placed within the structure so
1178 that it would fit within such a field, not crossing a boundary for it.
1180 Thus, on most machines, a named bit-field whose type is written as
1181 @code{int} would not cross a four-byte boundary, and would force
1182 four-byte alignment for the whole structure. (The alignment used may
1183 not be four bytes; it is controlled by the other alignment parameters.)
1185 An unnamed bit-field will not affect the alignment of the containing
1188 If the macro is defined, its definition should be a C expression;
1189 a nonzero value for the expression enables this behavior.
1191 Note that if this macro is not defined, or its value is zero, some
1192 bit-fields may cross more than one alignment boundary. The compiler can
1193 support such references if there are @samp{insv}, @samp{extv}, and
1194 @samp{extzv} insns that can directly reference memory.
1196 The other known way of making bit-fields work is to define
1197 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1198 Then every structure can be accessed with fullwords.
1200 Unless the machine has bit-field instructions or you define
1201 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1202 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1204 If your aim is to make GCC use the same conventions for laying out
1205 bit-fields as are used by another compiler, here is how to investigate
1206 what the other compiler does. Compile and run this program:
1225 printf ("Size of foo1 is %d\n",
1226 sizeof (struct foo1));
1227 printf ("Size of foo2 is %d\n",
1228 sizeof (struct foo2));
1233 If this prints 2 and 5, then the compiler's behavior is what you would
1234 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1237 @defmac BITFIELD_NBYTES_LIMITED
1238 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1239 to aligning a bit-field within the structure.
1242 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void)
1243 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1244 whether unnamed bitfields affect the alignment of the containing
1245 structure. The hook should return true if the structure should inherit
1246 the alignment requirements of an unnamed bitfield's type.
1249 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void)
1250 This target hook should return @code{true} if accesses to volatile bitfields
1251 should use the narrowest mode possible. It should return @code{false} if
1252 these accesses should use the bitfield container type.
1254 The default is @code{!TARGET_STRICT_ALIGN}.
1257 @defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode})
1258 Return 1 if a structure or array containing @var{field} should be accessed using
1261 If @var{field} is the only field in the structure, @var{mode} is its
1262 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1263 case where structures of one field would require the structure's mode to
1264 retain the field's mode.
1266 Normally, this is not needed.
1269 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1270 Define this macro as an expression for the alignment of a type (given
1271 by @var{type} as a tree node) if the alignment computed in the usual
1272 way is @var{computed} and the alignment explicitly specified was
1275 The default is to use @var{specified} if it is larger; otherwise, use
1276 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1279 @defmac MAX_FIXED_MODE_SIZE
1280 An integer expression for the size in bits of the largest integer
1281 machine mode that should actually be used. All integer machine modes of
1282 this size or smaller can be used for structures and unions with the
1283 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1284 (DImode)} is assumed.
1287 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1288 If defined, an expression of type @code{enum machine_mode} that
1289 specifies the mode of the save area operand of a
1290 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1291 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1292 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1293 having its mode specified.
1295 You need not define this macro if it always returns @code{Pmode}. You
1296 would most commonly define this macro if the
1297 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1301 @defmac STACK_SIZE_MODE
1302 If defined, an expression of type @code{enum machine_mode} that
1303 specifies the mode of the size increment operand of an
1304 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1306 You need not define this macro if it always returns @code{word_mode}.
1307 You would most commonly define this macro if the @code{allocate_stack}
1308 pattern needs to support both a 32- and a 64-bit mode.
1311 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_CMP_RETURN_MODE (void)
1312 This target hook should return the mode to be used for the return value
1313 of compare instructions expanded to libgcc calls. If not defined
1314 @code{word_mode} is returned which is the right choice for a majority of
1318 @deftypefn {Target Hook} {enum machine_mode} TARGET_LIBGCC_SHIFT_COUNT_MODE (void)
1319 This target hook should return the mode to be used for the shift count operand
1320 of shift instructions expanded to libgcc calls. If not defined
1321 @code{word_mode} is returned which is the right choice for a majority of
1325 @deftypefn {Target Hook} {enum machine_mode} TARGET_UNWIND_WORD_MODE (void)
1326 Return machine mode to be used for @code{_Unwind_Word} type.
1327 The default is to use @code{word_mode}.
1330 @defmac ROUND_TOWARDS_ZERO
1331 If defined, this macro should be true if the prevailing rounding
1332 mode is towards zero.
1334 Defining this macro only affects the way @file{libgcc.a} emulates
1335 floating-point arithmetic.
1337 Not defining this macro is equivalent to returning zero.
1340 @defmac LARGEST_EXPONENT_IS_NORMAL (@var{size})
1341 This macro should return true if floats with @var{size}
1342 bits do not have a NaN or infinity representation, but use the largest
1343 exponent for normal numbers instead.
1345 Defining this macro only affects the way @file{libgcc.a} emulates
1346 floating-point arithmetic.
1348 The default definition of this macro returns false for all sizes.
1351 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type})
1352 This target hook returns @code{true} if bit-fields in the given
1353 @var{record_type} are to be laid out following the rules of Microsoft
1354 Visual C/C++, namely: (i) a bit-field won't share the same storage
1355 unit with the previous bit-field if their underlying types have
1356 different sizes, and the bit-field will be aligned to the highest
1357 alignment of the underlying types of itself and of the previous
1358 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1359 the whole enclosing structure, even if it is unnamed; except that
1360 (iii) a zero-sized bit-field will be disregarded unless it follows
1361 another bit-field of nonzero size. If this hook returns @code{true},
1362 other macros that control bit-field layout are ignored.
1364 When a bit-field is inserted into a packed record, the whole size
1365 of the underlying type is used by one or more same-size adjacent
1366 bit-fields (that is, if its long:3, 32 bits is used in the record,
1367 and any additional adjacent long bit-fields are packed into the same
1368 chunk of 32 bits. However, if the size changes, a new field of that
1369 size is allocated). In an unpacked record, this is the same as using
1370 alignment, but not equivalent when packing.
1372 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1373 the latter will take precedence. If @samp{__attribute__((packed))} is
1374 used on a single field when MS bit-fields are in use, it will take
1375 precedence for that field, but the alignment of the rest of the structure
1376 may affect its placement.
1379 @deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1380 Returns true if the target supports decimal floating point.
1383 @deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void)
1384 Returns true if the target supports fixed-point arithmetic.
1387 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1388 This hook is called just before expansion into rtl, allowing the target
1389 to perform additional initializations or analysis before the expansion.
1390 For example, the rs6000 port uses it to allocate a scratch stack slot
1391 for use in copying SDmode values between memory and floating point
1392 registers whenever the function being expanded has any SDmode
1396 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1397 This hook allows the backend to perform additional instantiations on rtl
1398 that are not actually in any insns yet, but will be later.
1401 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type})
1402 If your target defines any fundamental types, or any types your target
1403 uses should be mangled differently from the default, define this hook
1404 to return the appropriate encoding for these types as part of a C++
1405 mangled name. The @var{type} argument is the tree structure representing
1406 the type to be mangled. The hook may be applied to trees which are
1407 not target-specific fundamental types; it should return @code{NULL}
1408 for all such types, as well as arguments it does not recognize. If the
1409 return value is not @code{NULL}, it must point to a statically-allocated
1412 Target-specific fundamental types might be new fundamental types or
1413 qualified versions of ordinary fundamental types. Encode new
1414 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1415 is the name used for the type in source code, and @var{n} is the
1416 length of @var{name} in decimal. Encode qualified versions of
1417 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1418 @var{name} is the name used for the type qualifier in source code,
1419 @var{n} is the length of @var{name} as above, and @var{code} is the
1420 code used to represent the unqualified version of this type. (See
1421 @code{write_builtin_type} in @file{cp/mangle.c} for the list of
1422 codes.) In both cases the spaces are for clarity; do not include any
1423 spaces in your string.
1425 This hook is applied to types prior to typedef resolution. If the mangled
1426 name for a particular type depends only on that type's main variant, you
1427 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1430 The default version of this hook always returns @code{NULL}, which is
1431 appropriate for a target that does not define any new fundamental
1436 @section Layout of Source Language Data Types
1438 These macros define the sizes and other characteristics of the standard
1439 basic data types used in programs being compiled. Unlike the macros in
1440 the previous section, these apply to specific features of C and related
1441 languages, rather than to fundamental aspects of storage layout.
1443 @defmac INT_TYPE_SIZE
1444 A C expression for the size in bits of the type @code{int} on the
1445 target machine. If you don't define this, the default is one word.
1448 @defmac SHORT_TYPE_SIZE
1449 A C expression for the size in bits of the type @code{short} on the
1450 target machine. If you don't define this, the default is half a word.
1451 (If this would be less than one storage unit, it is rounded up to one
1455 @defmac LONG_TYPE_SIZE
1456 A C expression for the size in bits of the type @code{long} on the
1457 target machine. If you don't define this, the default is one word.
1460 @defmac ADA_LONG_TYPE_SIZE
1461 On some machines, the size used for the Ada equivalent of the type
1462 @code{long} by a native Ada compiler differs from that used by C@. In
1463 that situation, define this macro to be a C expression to be used for
1464 the size of that type. If you don't define this, the default is the
1465 value of @code{LONG_TYPE_SIZE}.
1468 @defmac LONG_LONG_TYPE_SIZE
1469 A C expression for the size in bits of the type @code{long long} on the
1470 target machine. If you don't define this, the default is two
1471 words. If you want to support GNU Ada on your machine, the value of this
1472 macro must be at least 64.
1475 @defmac CHAR_TYPE_SIZE
1476 A C expression for the size in bits of the type @code{char} on the
1477 target machine. If you don't define this, the default is
1478 @code{BITS_PER_UNIT}.
1481 @defmac BOOL_TYPE_SIZE
1482 A C expression for the size in bits of the C++ type @code{bool} and
1483 C99 type @code{_Bool} on the target machine. If you don't define
1484 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1487 @defmac FLOAT_TYPE_SIZE
1488 A C expression for the size in bits of the type @code{float} on the
1489 target machine. If you don't define this, the default is one word.
1492 @defmac DOUBLE_TYPE_SIZE
1493 A C expression for the size in bits of the type @code{double} on the
1494 target machine. If you don't define this, the default is two
1498 @defmac LONG_DOUBLE_TYPE_SIZE
1499 A C expression for the size in bits of the type @code{long double} on
1500 the target machine. If you don't define this, the default is two
1504 @defmac SHORT_FRACT_TYPE_SIZE
1505 A C expression for the size in bits of the type @code{short _Fract} on
1506 the target machine. If you don't define this, the default is
1507 @code{BITS_PER_UNIT}.
1510 @defmac FRACT_TYPE_SIZE
1511 A C expression for the size in bits of the type @code{_Fract} on
1512 the target machine. If you don't define this, the default is
1513 @code{BITS_PER_UNIT * 2}.
1516 @defmac LONG_FRACT_TYPE_SIZE
1517 A C expression for the size in bits of the type @code{long _Fract} on
1518 the target machine. If you don't define this, the default is
1519 @code{BITS_PER_UNIT * 4}.
1522 @defmac LONG_LONG_FRACT_TYPE_SIZE
1523 A C expression for the size in bits of the type @code{long long _Fract} on
1524 the target machine. If you don't define this, the default is
1525 @code{BITS_PER_UNIT * 8}.
1528 @defmac SHORT_ACCUM_TYPE_SIZE
1529 A C expression for the size in bits of the type @code{short _Accum} on
1530 the target machine. If you don't define this, the default is
1531 @code{BITS_PER_UNIT * 2}.
1534 @defmac ACCUM_TYPE_SIZE
1535 A C expression for the size in bits of the type @code{_Accum} on
1536 the target machine. If you don't define this, the default is
1537 @code{BITS_PER_UNIT * 4}.
1540 @defmac LONG_ACCUM_TYPE_SIZE
1541 A C expression for the size in bits of the type @code{long _Accum} on
1542 the target machine. If you don't define this, the default is
1543 @code{BITS_PER_UNIT * 8}.
1546 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1547 A C expression for the size in bits of the type @code{long long _Accum} on
1548 the target machine. If you don't define this, the default is
1549 @code{BITS_PER_UNIT * 16}.
1552 @defmac LIBGCC2_LONG_DOUBLE_TYPE_SIZE
1553 Define this macro if @code{LONG_DOUBLE_TYPE_SIZE} is not constant or
1554 if you want routines in @file{libgcc2.a} for a size other than
1555 @code{LONG_DOUBLE_TYPE_SIZE}. If you don't define this, the
1556 default is @code{LONG_DOUBLE_TYPE_SIZE}.
1559 @defmac LIBGCC2_HAS_DF_MODE
1560 Define this macro if neither @code{DOUBLE_TYPE_SIZE} nor
1561 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is
1562 @code{DFmode} but you want @code{DFmode} routines in @file{libgcc2.a}
1563 anyway. If you don't define this and either @code{DOUBLE_TYPE_SIZE}
1564 or @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64 then the default is 1,
1568 @defmac LIBGCC2_HAS_XF_MODE
1569 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1570 @code{XFmode} but you want @code{XFmode} routines in @file{libgcc2.a}
1571 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1572 is 80 then the default is 1, otherwise it is 0.
1575 @defmac LIBGCC2_HAS_TF_MODE
1576 Define this macro if @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is not
1577 @code{TFmode} but you want @code{TFmode} routines in @file{libgcc2.a}
1578 anyway. If you don't define this and @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE}
1579 is 128 then the default is 1, otherwise it is 0.
1582 @defmac LIBGCC2_GNU_PREFIX
1583 This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target
1584 hook and should be defined if that hook is overriden to be true. It
1585 causes function names in libgcc to be changed to use a @code{__gnu_}
1586 prefix for their name rather than the default @code{__}. A port which
1587 uses this macro should also arrange to use @file{t-gnu-prefix} in
1588 the libgcc @file{config.host}.
1595 Define these macros to be the size in bits of the mantissa of
1596 @code{SFmode}, @code{DFmode}, @code{XFmode} and @code{TFmode} values,
1597 if the defaults in @file{libgcc2.h} are inappropriate. By default,
1598 @code{FLT_MANT_DIG} is used for @code{SF_SIZE}, @code{LDBL_MANT_DIG}
1599 for @code{XF_SIZE} and @code{TF_SIZE}, and @code{DBL_MANT_DIG} or
1600 @code{LDBL_MANT_DIG} for @code{DF_SIZE} according to whether
1601 @code{DOUBLE_TYPE_SIZE} or
1602 @code{LIBGCC2_LONG_DOUBLE_TYPE_SIZE} is 64.
1605 @defmac TARGET_FLT_EVAL_METHOD
1606 A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h},
1607 assuming, if applicable, that the floating-point control word is in its
1608 default state. If you do not define this macro the value of
1609 @code{FLT_EVAL_METHOD} will be zero.
1612 @defmac WIDEST_HARDWARE_FP_SIZE
1613 A C expression for the size in bits of the widest floating-point format
1614 supported by the hardware. If you define this macro, you must specify a
1615 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1616 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1620 @defmac DEFAULT_SIGNED_CHAR
1621 An expression whose value is 1 or 0, according to whether the type
1622 @code{char} should be signed or unsigned by default. The user can
1623 always override this default with the options @option{-fsigned-char}
1624 and @option{-funsigned-char}.
1627 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1628 This target hook should return true if the compiler should give an
1629 @code{enum} type only as many bytes as it takes to represent the range
1630 of possible values of that type. It should return false if all
1631 @code{enum} types should be allocated like @code{int}.
1633 The default is to return false.
1637 A C expression for a string describing the name of the data type to use
1638 for size values. The typedef name @code{size_t} is defined using the
1639 contents of the string.
1641 The string can contain more than one keyword. If so, separate them with
1642 spaces, and write first any length keyword, then @code{unsigned} if
1643 appropriate, and finally @code{int}. The string must exactly match one
1644 of the data type names defined in the function
1645 @code{c_common_nodes_and_builtins} in the file @file{c-family/c-common.c}.
1646 You may not omit @code{int} or change the order---that would cause the
1647 compiler to crash on startup.
1649 If you don't define this macro, the default is @code{"long unsigned
1654 GCC defines internal types (@code{sizetype}, @code{ssizetype},
1655 @code{bitsizetype} and @code{sbitsizetype}) for expressions
1656 dealing with size. This macro is a C expression for a string describing
1657 the name of the data type from which the precision of @code{sizetype}
1660 The string has the same restrictions as @code{SIZE_TYPE} string.
1662 If you don't define this macro, the default is @code{SIZE_TYPE}.
1665 @defmac PTRDIFF_TYPE
1666 A C expression for a string describing the name of the data type to use
1667 for the result of subtracting two pointers. The typedef name
1668 @code{ptrdiff_t} is defined using the contents of the string. See
1669 @code{SIZE_TYPE} above for more information.
1671 If you don't define this macro, the default is @code{"long int"}.
1675 A C expression for a string describing the name of the data type to use
1676 for wide characters. The typedef name @code{wchar_t} is defined using
1677 the contents of the string. See @code{SIZE_TYPE} above for more
1680 If you don't define this macro, the default is @code{"int"}.
1683 @defmac WCHAR_TYPE_SIZE
1684 A C expression for the size in bits of the data type for wide
1685 characters. This is used in @code{cpp}, which cannot make use of
1690 A C expression for a string describing the name of the data type to
1691 use for wide characters passed to @code{printf} and returned from
1692 @code{getwc}. The typedef name @code{wint_t} is defined using the
1693 contents of the string. See @code{SIZE_TYPE} above for more
1696 If you don't define this macro, the default is @code{"unsigned int"}.
1700 A C expression for a string describing the name of the data type that
1701 can represent any value of any standard or extended signed integer type.
1702 The typedef name @code{intmax_t} is defined using the contents of the
1703 string. See @code{SIZE_TYPE} above for more information.
1705 If you don't define this macro, the default is the first of
1706 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1707 much precision as @code{long long int}.
1710 @defmac UINTMAX_TYPE
1711 A C expression for a string describing the name of the data type that
1712 can represent any value of any standard or extended unsigned integer
1713 type. The typedef name @code{uintmax_t} is defined using the contents
1714 of the string. See @code{SIZE_TYPE} above for more information.
1716 If you don't define this macro, the default is the first of
1717 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1718 unsigned int"} that has as much precision as @code{long long unsigned
1722 @defmac SIG_ATOMIC_TYPE
1728 @defmacx UINT16_TYPE
1729 @defmacx UINT32_TYPE
1730 @defmacx UINT64_TYPE
1731 @defmacx INT_LEAST8_TYPE
1732 @defmacx INT_LEAST16_TYPE
1733 @defmacx INT_LEAST32_TYPE
1734 @defmacx INT_LEAST64_TYPE
1735 @defmacx UINT_LEAST8_TYPE
1736 @defmacx UINT_LEAST16_TYPE
1737 @defmacx UINT_LEAST32_TYPE
1738 @defmacx UINT_LEAST64_TYPE
1739 @defmacx INT_FAST8_TYPE
1740 @defmacx INT_FAST16_TYPE
1741 @defmacx INT_FAST32_TYPE
1742 @defmacx INT_FAST64_TYPE
1743 @defmacx UINT_FAST8_TYPE
1744 @defmacx UINT_FAST16_TYPE
1745 @defmacx UINT_FAST32_TYPE
1746 @defmacx UINT_FAST64_TYPE
1747 @defmacx INTPTR_TYPE
1748 @defmacx UINTPTR_TYPE
1749 C expressions for the standard types @code{sig_atomic_t},
1750 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1751 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1752 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1753 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1754 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1755 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1756 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1757 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1758 @code{SIZE_TYPE} above for more information.
1760 If any of these macros evaluates to a null pointer, the corresponding
1761 type is not supported; if GCC is configured to provide
1762 @code{<stdint.h>} in such a case, the header provided may not conform
1763 to C99, depending on the type in question. The defaults for all of
1764 these macros are null pointers.
1767 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1768 The C++ compiler represents a pointer-to-member-function with a struct
1775 ptrdiff_t vtable_index;
1782 The C++ compiler must use one bit to indicate whether the function that
1783 will be called through a pointer-to-member-function is virtual.
1784 Normally, we assume that the low-order bit of a function pointer must
1785 always be zero. Then, by ensuring that the vtable_index is odd, we can
1786 distinguish which variant of the union is in use. But, on some
1787 platforms function pointers can be odd, and so this doesn't work. In
1788 that case, we use the low-order bit of the @code{delta} field, and shift
1789 the remainder of the @code{delta} field to the left.
1791 GCC will automatically make the right selection about where to store
1792 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1793 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1794 set such that functions always start at even addresses, but the lowest
1795 bit of pointers to functions indicate whether the function at that
1796 address is in ARM or Thumb mode. If this is the case of your
1797 architecture, you should define this macro to
1798 @code{ptrmemfunc_vbit_in_delta}.
1800 In general, you should not have to define this macro. On architectures
1801 in which function addresses are always even, according to
1802 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1803 @code{ptrmemfunc_vbit_in_pfn}.
1806 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1807 Normally, the C++ compiler uses function pointers in vtables. This
1808 macro allows the target to change to use ``function descriptors''
1809 instead. Function descriptors are found on targets for whom a
1810 function pointer is actually a small data structure. Normally the
1811 data structure consists of the actual code address plus a data
1812 pointer to which the function's data is relative.
1814 If vtables are used, the value of this macro should be the number
1815 of words that the function descriptor occupies.
1818 @defmac TARGET_VTABLE_ENTRY_ALIGN
1819 By default, the vtable entries are void pointers, the so the alignment
1820 is the same as pointer alignment. The value of this macro specifies
1821 the alignment of the vtable entry in bits. It should be defined only
1822 when special alignment is necessary. */
1825 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1826 There are a few non-descriptor entries in the vtable at offsets below
1827 zero. If these entries must be padded (say, to preserve the alignment
1828 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1829 of words in each data entry.
1833 @section Register Usage
1834 @cindex register usage
1836 This section explains how to describe what registers the target machine
1837 has, and how (in general) they can be used.
1839 The description of which registers a specific instruction can use is
1840 done with register classes; see @ref{Register Classes}. For information
1841 on using registers to access a stack frame, see @ref{Frame Registers}.
1842 For passing values in registers, see @ref{Register Arguments}.
1843 For returning values in registers, see @ref{Scalar Return}.
1846 * Register Basics:: Number and kinds of registers.
1847 * Allocation Order:: Order in which registers are allocated.
1848 * Values in Registers:: What kinds of values each reg can hold.
1849 * Leaf Functions:: Renumbering registers for leaf functions.
1850 * Stack Registers:: Handling a register stack such as 80387.
1853 @node Register Basics
1854 @subsection Basic Characteristics of Registers
1856 @c prevent bad page break with this line
1857 Registers have various characteristics.
1859 @defmac FIRST_PSEUDO_REGISTER
1860 Number of hardware registers known to the compiler. They receive
1861 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1862 pseudo register's number really is assigned the number
1863 @code{FIRST_PSEUDO_REGISTER}.
1866 @defmac FIXED_REGISTERS
1867 @cindex fixed register
1868 An initializer that says which registers are used for fixed purposes
1869 all throughout the compiled code and are therefore not available for
1870 general allocation. These would include the stack pointer, the frame
1871 pointer (except on machines where that can be used as a general
1872 register when no frame pointer is needed), the program counter on
1873 machines where that is considered one of the addressable registers,
1874 and any other numbered register with a standard use.
1876 This information is expressed as a sequence of numbers, separated by
1877 commas and surrounded by braces. The @var{n}th number is 1 if
1878 register @var{n} is fixed, 0 otherwise.
1880 The table initialized from this macro, and the table initialized by
1881 the following one, may be overridden at run time either automatically,
1882 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1883 the user with the command options @option{-ffixed-@var{reg}},
1884 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1887 @defmac CALL_USED_REGISTERS
1888 @cindex call-used register
1889 @cindex call-clobbered register
1890 @cindex call-saved register
1891 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1892 clobbered (in general) by function calls as well as for fixed
1893 registers. This macro therefore identifies the registers that are not
1894 available for general allocation of values that must live across
1897 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1898 automatically saves it on function entry and restores it on function
1899 exit, if the register is used within the function.
1902 @defmac CALL_REALLY_USED_REGISTERS
1903 @cindex call-used register
1904 @cindex call-clobbered register
1905 @cindex call-saved register
1906 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1907 that the entire set of @code{FIXED_REGISTERS} be included.
1908 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1909 This macro is optional. If not specified, it defaults to the value
1910 of @code{CALL_USED_REGISTERS}.
1913 @defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode})
1914 @cindex call-used register
1915 @cindex call-clobbered register
1916 @cindex call-saved register
1917 A C expression that is nonzero if it is not permissible to store a
1918 value of mode @var{mode} in hard register number @var{regno} across a
1919 call without some part of it being clobbered. For most machines this
1920 macro need not be defined. It is only required for machines that do not
1921 preserve the entire contents of a register across a call.
1925 @findex call_used_regs
1928 @findex reg_class_contents
1929 @deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void)
1930 This hook may conditionally modify five variables
1931 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
1932 @code{reg_names}, and @code{reg_class_contents}, to take into account
1933 any dependence of these register sets on target flags. The first three
1934 of these are of type @code{char []} (interpreted as Boolean vectors).
1935 @code{global_regs} is a @code{const char *[]}, and
1936 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
1937 called, @code{fixed_regs}, @code{call_used_regs},
1938 @code{reg_class_contents}, and @code{reg_names} have been initialized
1939 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
1940 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
1941 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
1942 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
1943 command options have been applied.
1945 @cindex disabling certain registers
1946 @cindex controlling register usage
1947 If the usage of an entire class of registers depends on the target
1948 flags, you may indicate this to GCC by using this macro to modify
1949 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
1950 registers in the classes which should not be used by GCC@. Also define
1951 the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT}
1952 to return @code{NO_REGS} if it
1953 is called with a letter for a class that shouldn't be used.
1955 (However, if this class is not included in @code{GENERAL_REGS} and all
1956 of the insn patterns whose constraints permit this class are
1957 controlled by target switches, then GCC will automatically avoid using
1958 these registers when the target switches are opposed to them.)
1961 @defmac INCOMING_REGNO (@var{out})
1962 Define this macro if the target machine has register windows. This C
1963 expression returns the register number as seen by the called function
1964 corresponding to the register number @var{out} as seen by the calling
1965 function. Return @var{out} if register number @var{out} is not an
1969 @defmac OUTGOING_REGNO (@var{in})
1970 Define this macro if the target machine has register windows. This C
1971 expression returns the register number as seen by the calling function
1972 corresponding to the register number @var{in} as seen by the called
1973 function. Return @var{in} if register number @var{in} is not an inbound
1977 @defmac LOCAL_REGNO (@var{regno})
1978 Define this macro if the target machine has register windows. This C
1979 expression returns true if the register is call-saved but is in the
1980 register window. Unlike most call-saved registers, such registers
1981 need not be explicitly restored on function exit or during non-local
1986 If the program counter has a register number, define this as that
1987 register number. Otherwise, do not define it.
1990 @node Allocation Order
1991 @subsection Order of Allocation of Registers
1992 @cindex order of register allocation
1993 @cindex register allocation order
1995 @c prevent bad page break with this line
1996 Registers are allocated in order.
1998 @defmac REG_ALLOC_ORDER
1999 If defined, an initializer for a vector of integers, containing the
2000 numbers of hard registers in the order in which GCC should prefer
2001 to use them (from most preferred to least).
2003 If this macro is not defined, registers are used lowest numbered first
2004 (all else being equal).
2006 One use of this macro is on machines where the highest numbered
2007 registers must always be saved and the save-multiple-registers
2008 instruction supports only sequences of consecutive registers. On such
2009 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2010 the highest numbered allocable register first.
2013 @defmac ADJUST_REG_ALLOC_ORDER
2014 A C statement (sans semicolon) to choose the order in which to allocate
2015 hard registers for pseudo-registers local to a basic block.
2017 Store the desired register order in the array @code{reg_alloc_order}.
2018 Element 0 should be the register to allocate first; element 1, the next
2019 register; and so on.
2021 The macro body should not assume anything about the contents of
2022 @code{reg_alloc_order} before execution of the macro.
2024 On most machines, it is not necessary to define this macro.
2027 @defmac HONOR_REG_ALLOC_ORDER
2028 Normally, IRA tries to estimate the costs for saving a register in the
2029 prologue and restoring it in the epilogue. This discourages it from
2030 using call-saved registers. If a machine wants to ensure that IRA
2031 allocates registers in the order given by REG_ALLOC_ORDER even if some
2032 call-saved registers appear earlier than call-used ones, this macro
2036 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2037 In some case register allocation order is not enough for the
2038 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2039 If this macro is defined, it should return a floating point value
2040 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2041 be increased by approximately the pseudo's usage frequency times the
2042 value returned by this macro. Not defining this macro is equivalent
2043 to having it always return @code{0.0}.
2045 On most machines, it is not necessary to define this macro.
2048 @node Values in Registers
2049 @subsection How Values Fit in Registers
2051 This section discusses the macros that describe which kinds of values
2052 (specifically, which machine modes) each register can hold, and how many
2053 consecutive registers are needed for a given mode.
2055 @defmac HARD_REGNO_NREGS (@var{regno}, @var{mode})
2056 A C expression for the number of consecutive hard registers, starting
2057 at register number @var{regno}, required to hold a value of mode
2058 @var{mode}. This macro must never return zero, even if a register
2059 cannot hold the requested mode - indicate that with HARD_REGNO_MODE_OK
2060 and/or CANNOT_CHANGE_MODE_CLASS instead.
2062 On a machine where all registers are exactly one word, a suitable
2063 definition of this macro is
2066 #define HARD_REGNO_NREGS(REGNO, MODE) \
2067 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \
2072 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2073 A C expression that is nonzero if a value of mode @var{mode}, stored
2074 in memory, ends with padding that causes it to take up more space than
2075 in registers starting at register number @var{regno} (as determined by
2076 multiplying GCC's notion of the size of the register when containing
2077 this mode by the number of registers returned by
2078 @code{HARD_REGNO_NREGS}). By default this is zero.
2080 For example, if a floating-point value is stored in three 32-bit
2081 registers but takes up 128 bits in memory, then this would be
2084 This macros only needs to be defined if there are cases where
2085 @code{subreg_get_info}
2086 would otherwise wrongly determine that a @code{subreg} can be
2087 represented by an offset to the register number, when in fact such a
2088 @code{subreg} would contain some of the padding not stored in
2089 registers and so not be representable.
2092 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2093 For values of @var{regno} and @var{mode} for which
2094 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2095 returning the greater number of registers required to hold the value
2096 including any padding. In the example above, the value would be four.
2099 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2100 Define this macro if the natural size of registers that hold values
2101 of mode @var{mode} is not the word size. It is a C expression that
2102 should give the natural size in bytes for the specified mode. It is
2103 used by the register allocator to try to optimize its results. This
2104 happens for example on SPARC 64-bit where the natural size of
2105 floating-point registers is still 32-bit.
2108 @defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode})
2109 A C expression that is nonzero if it is permissible to store a value
2110 of mode @var{mode} in hard register number @var{regno} (or in several
2111 registers starting with that one). For a machine where all registers
2112 are equivalent, a suitable definition is
2115 #define HARD_REGNO_MODE_OK(REGNO, MODE) 1
2118 You need not include code to check for the numbers of fixed registers,
2119 because the allocation mechanism considers them to be always occupied.
2121 @cindex register pairs
2122 On some machines, double-precision values must be kept in even/odd
2123 register pairs. You can implement that by defining this macro to reject
2124 odd register numbers for such modes.
2126 The minimum requirement for a mode to be OK in a register is that the
2127 @samp{mov@var{mode}} instruction pattern support moves between the
2128 register and other hard register in the same class and that moving a
2129 value into the register and back out not alter it.
2131 Since the same instruction used to move @code{word_mode} will work for
2132 all narrower integer modes, it is not necessary on any machine for
2133 @code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided
2134 you define patterns @samp{movhi}, etc., to take advantage of this. This
2135 is useful because of the interaction between @code{HARD_REGNO_MODE_OK}
2136 and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes
2139 Many machines have special registers for floating point arithmetic.
2140 Often people assume that floating point machine modes are allowed only
2141 in floating point registers. This is not true. Any registers that
2142 can hold integers can safely @emph{hold} a floating point machine
2143 mode, whether or not floating arithmetic can be done on it in those
2144 registers. Integer move instructions can be used to move the values.
2146 On some machines, though, the converse is true: fixed-point machine
2147 modes may not go in floating registers. This is true if the floating
2148 registers normalize any value stored in them, because storing a
2149 non-floating value there would garble it. In this case,
2150 @code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2151 floating registers. But if the floating registers do not automatically
2152 normalize, if you can store any bit pattern in one and retrieve it
2153 unchanged without a trap, then any machine mode may go in a floating
2154 register, so you can define this macro to say so.
2156 The primary significance of special floating registers is rather that
2157 they are the registers acceptable in floating point arithmetic
2158 instructions. However, this is of no concern to
2159 @code{HARD_REGNO_MODE_OK}. You handle it by writing the proper
2160 constraints for those instructions.
2162 On some machines, the floating registers are especially slow to access,
2163 so that it is better to store a value in a stack frame than in such a
2164 register if floating point arithmetic is not being done. As long as the
2165 floating registers are not in class @code{GENERAL_REGS}, they will not
2166 be used unless some pattern's constraint asks for one.
2169 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2170 A C expression that is nonzero if it is OK to rename a hard register
2171 @var{from} to another hard register @var{to}.
2173 One common use of this macro is to prevent renaming of a register to
2174 another register that is not saved by a prologue in an interrupt
2177 The default is always nonzero.
2180 @defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2})
2181 A C expression that is nonzero if a value of mode
2182 @var{mode1} is accessible in mode @var{mode2} without copying.
2184 If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2185 @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for
2186 any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2187 should be nonzero. If they differ for any @var{r}, you should define
2188 this macro to return zero unless some other mechanism ensures the
2189 accessibility of the value in a narrower mode.
2191 You should define this macro to return nonzero in as many cases as
2192 possible since doing so will allow GCC to perform better register
2196 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno})
2197 This target hook should return @code{true} if it is OK to use a hard register
2198 @var{regno} as scratch reg in peephole2.
2200 One common use of this macro is to prevent using of a register that
2201 is not saved by a prologue in an interrupt handler.
2203 The default version of this hook always returns @code{true}.
2206 @defmac AVOID_CCMODE_COPIES
2207 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2208 registers. You should only define this macro if support for copying to/from
2209 @code{CCmode} is incomplete.
2212 @node Leaf Functions
2213 @subsection Handling Leaf Functions
2215 @cindex leaf functions
2216 @cindex functions, leaf
2217 On some machines, a leaf function (i.e., one which makes no calls) can run
2218 more efficiently if it does not make its own register window. Often this
2219 means it is required to receive its arguments in the registers where they
2220 are passed by the caller, instead of the registers where they would
2223 The special treatment for leaf functions generally applies only when
2224 other conditions are met; for example, often they may use only those
2225 registers for its own variables and temporaries. We use the term ``leaf
2226 function'' to mean a function that is suitable for this special
2227 handling, so that functions with no calls are not necessarily ``leaf
2230 GCC assigns register numbers before it knows whether the function is
2231 suitable for leaf function treatment. So it needs to renumber the
2232 registers in order to output a leaf function. The following macros
2235 @defmac LEAF_REGISTERS
2236 Name of a char vector, indexed by hard register number, which
2237 contains 1 for a register that is allowable in a candidate for leaf
2240 If leaf function treatment involves renumbering the registers, then the
2241 registers marked here should be the ones before renumbering---those that
2242 GCC would ordinarily allocate. The registers which will actually be
2243 used in the assembler code, after renumbering, should not be marked with 1
2246 Define this macro only if the target machine offers a way to optimize
2247 the treatment of leaf functions.
2250 @defmac LEAF_REG_REMAP (@var{regno})
2251 A C expression whose value is the register number to which @var{regno}
2252 should be renumbered, when a function is treated as a leaf function.
2254 If @var{regno} is a register number which should not appear in a leaf
2255 function before renumbering, then the expression should yield @minus{}1, which
2256 will cause the compiler to abort.
2258 Define this macro only if the target machine offers a way to optimize the
2259 treatment of leaf functions, and registers need to be renumbered to do
2263 @findex current_function_is_leaf
2264 @findex current_function_uses_only_leaf_regs
2265 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2266 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2267 specially. They can test the C variable @code{current_function_is_leaf}
2268 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2269 set prior to local register allocation and is valid for the remaining
2270 compiler passes. They can also test the C variable
2271 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2272 functions which only use leaf registers.
2273 @code{current_function_uses_only_leaf_regs} is valid after all passes
2274 that modify the instructions have been run and is only useful if
2275 @code{LEAF_REGISTERS} is defined.
2276 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2277 @c of the next paragraph?! --mew 2feb93
2279 @node Stack Registers
2280 @subsection Registers That Form a Stack
2282 There are special features to handle computers where some of the
2283 ``registers'' form a stack. Stack registers are normally written by
2284 pushing onto the stack, and are numbered relative to the top of the
2287 Currently, GCC can only handle one group of stack-like registers, and
2288 they must be consecutively numbered. Furthermore, the existing
2289 support for stack-like registers is specific to the 80387 floating
2290 point coprocessor. If you have a new architecture that uses
2291 stack-like registers, you will need to do substantial work on
2292 @file{reg-stack.c} and write your machine description to cooperate
2293 with it, as well as defining these macros.
2296 Define this if the machine has any stack-like registers.
2299 @defmac STACK_REG_COVER_CLASS
2300 This is a cover class containing the stack registers. Define this if
2301 the machine has any stack-like registers.
2304 @defmac FIRST_STACK_REG
2305 The number of the first stack-like register. This one is the top
2309 @defmac LAST_STACK_REG
2310 The number of the last stack-like register. This one is the bottom of
2314 @node Register Classes
2315 @section Register Classes
2316 @cindex register class definitions
2317 @cindex class definitions, register
2319 On many machines, the numbered registers are not all equivalent.
2320 For example, certain registers may not be allowed for indexed addressing;
2321 certain registers may not be allowed in some instructions. These machine
2322 restrictions are described to the compiler using @dfn{register classes}.
2324 You define a number of register classes, giving each one a name and saying
2325 which of the registers belong to it. Then you can specify register classes
2326 that are allowed as operands to particular instruction patterns.
2330 In general, each register will belong to several classes. In fact, one
2331 class must be named @code{ALL_REGS} and contain all the registers. Another
2332 class must be named @code{NO_REGS} and contain no registers. Often the
2333 union of two classes will be another class; however, this is not required.
2335 @findex GENERAL_REGS
2336 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2337 terribly special about the name, but the operand constraint letters
2338 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2339 the same as @code{ALL_REGS}, just define it as a macro which expands
2342 Order the classes so that if class @var{x} is contained in class @var{y}
2343 then @var{x} has a lower class number than @var{y}.
2345 The way classes other than @code{GENERAL_REGS} are specified in operand
2346 constraints is through machine-dependent operand constraint letters.
2347 You can define such letters to correspond to various classes, then use
2348 them in operand constraints.
2350 You must define the narrowest register classes for allocatable
2351 registers, so that each class either has no subclasses, or that for
2352 some mode, the move cost between registers within the class is
2353 cheaper than moving a register in the class to or from memory
2356 You should define a class for the union of two classes whenever some
2357 instruction allows both classes. For example, if an instruction allows
2358 either a floating point (coprocessor) register or a general register for a
2359 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2360 which includes both of them. Otherwise you will get suboptimal code,
2361 or even internal compiler errors when reload cannot find a register in the
2362 class computed via @code{reg_class_subunion}.
2364 You must also specify certain redundant information about the register
2365 classes: for each class, which classes contain it and which ones are
2366 contained in it; for each pair of classes, the largest class contained
2369 When a value occupying several consecutive registers is expected in a
2370 certain class, all the registers used must belong to that class.
2371 Therefore, register classes cannot be used to enforce a requirement for
2372 a register pair to start with an even-numbered register. The way to
2373 specify this requirement is with @code{HARD_REGNO_MODE_OK}.
2375 Register classes used for input-operands of bitwise-and or shift
2376 instructions have a special requirement: each such class must have, for
2377 each fixed-point machine mode, a subclass whose registers can transfer that
2378 mode to or from memory. For example, on some machines, the operations for
2379 single-byte values (@code{QImode}) are limited to certain registers. When
2380 this is so, each register class that is used in a bitwise-and or shift
2381 instruction must have a subclass consisting of registers from which
2382 single-byte values can be loaded or stored. This is so that
2383 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2385 @deftp {Data type} {enum reg_class}
2386 An enumerated type that must be defined with all the register class names
2387 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2388 must be the last register class, followed by one more enumerated value,
2389 @code{LIM_REG_CLASSES}, which is not a register class but rather
2390 tells how many classes there are.
2392 Each register class has a number, which is the value of casting
2393 the class name to type @code{int}. The number serves as an index
2394 in many of the tables described below.
2397 @defmac N_REG_CLASSES
2398 The number of distinct register classes, defined as follows:
2401 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2405 @defmac REG_CLASS_NAMES
2406 An initializer containing the names of the register classes as C string
2407 constants. These names are used in writing some of the debugging dumps.
2410 @defmac REG_CLASS_CONTENTS
2411 An initializer containing the contents of the register classes, as integers
2412 which are bit masks. The @var{n}th integer specifies the contents of class
2413 @var{n}. The way the integer @var{mask} is interpreted is that
2414 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2416 When the machine has more than 32 registers, an integer does not suffice.
2417 Then the integers are replaced by sub-initializers, braced groupings containing
2418 several integers. Each sub-initializer must be suitable as an initializer
2419 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2420 In this situation, the first integer in each sub-initializer corresponds to
2421 registers 0 through 31, the second integer to registers 32 through 63, and
2425 @defmac REGNO_REG_CLASS (@var{regno})
2426 A C expression whose value is a register class containing hard register
2427 @var{regno}. In general there is more than one such class; choose a class
2428 which is @dfn{minimal}, meaning that no smaller class also contains the
2432 @defmac BASE_REG_CLASS
2433 A macro whose definition is the name of the class to which a valid
2434 base register must belong. A base register is one used in an address
2435 which is the register value plus a displacement.
2438 @defmac MODE_BASE_REG_CLASS (@var{mode})
2439 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2440 the selection of a base register in a mode dependent manner. If
2441 @var{mode} is VOIDmode then it should return the same value as
2442 @code{BASE_REG_CLASS}.
2445 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2446 A C expression whose value is the register class to which a valid
2447 base register must belong in order to be used in a base plus index
2448 register address. You should define this macro if base plus index
2449 addresses have different requirements than other base register uses.
2452 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2453 A C expression whose value is the register class to which a valid
2454 base register for a memory reference in mode @var{mode} to address
2455 space @var{address_space} must belong. @var{outer_code} and @var{index_code}
2456 define the context in which the base register occurs. @var{outer_code} is
2457 the code of the immediately enclosing expression (@code{MEM} for the top level
2458 of an address, @code{ADDRESS} for something that occurs in an
2459 @code{address_operand}). @var{index_code} is the code of the corresponding
2460 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2463 @defmac INDEX_REG_CLASS
2464 A macro whose definition is the name of the class to which a valid
2465 index register must belong. An index register is one used in an
2466 address where its value is either multiplied by a scale factor or
2467 added to another register (as well as added to a displacement).
2470 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2471 A C expression which is nonzero if register number @var{num} is
2472 suitable for use as a base register in operand addresses.
2475 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2476 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2477 that expression may examine the mode of the memory reference in
2478 @var{mode}. You should define this macro if the mode of the memory
2479 reference affects whether a register may be used as a base register. If
2480 you define this macro, the compiler will use it instead of
2481 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2482 addresses that appear outside a @code{MEM}, i.e., as an
2483 @code{address_operand}.
2486 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2487 A C expression which is nonzero if register number @var{num} is suitable for
2488 use as a base register in base plus index operand addresses, accessing
2489 memory in mode @var{mode}. It may be either a suitable hard register or a
2490 pseudo register that has been allocated such a hard register. You should
2491 define this macro if base plus index addresses have different requirements
2492 than other base register uses.
2494 Use of this macro is deprecated; please use the more general
2495 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2498 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2499 A C expression which is nonzero if register number @var{num} is
2500 suitable for use as a base register in operand addresses, accessing
2501 memory in mode @var{mode} in address space @var{address_space}.
2502 This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except
2503 that that expression may examine the context in which the register
2504 appears in the memory reference. @var{outer_code} is the code of the
2505 immediately enclosing expression (@code{MEM} if at the top level of the
2506 address, @code{ADDRESS} for something that occurs in an
2507 @code{address_operand}). @var{index_code} is the code of the
2508 corresponding index expression if @var{outer_code} is @code{PLUS};
2509 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2510 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2513 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2514 A C expression which is nonzero if register number @var{num} is
2515 suitable for use as an index register in operand addresses. It may be
2516 either a suitable hard register or a pseudo register that has been
2517 allocated such a hard register.
2519 The difference between an index register and a base register is that
2520 the index register may be scaled. If an address involves the sum of
2521 two registers, neither one of them scaled, then either one may be
2522 labeled the ``base'' and the other the ``index''; but whichever
2523 labeling is used must fit the machine's constraints of which registers
2524 may serve in each capacity. The compiler will try both labelings,
2525 looking for one that is valid, and will reload one or both registers
2526 only if neither labeling works.
2529 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass})
2530 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.
2533 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2534 A target hook that places additional restrictions on the register class
2535 to use when it is necessary to copy value @var{x} into a register in class
2536 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2537 another, smaller class.
2539 The default version of this hook always returns value of @code{rclass} argument.
2541 Sometimes returning a more restrictive class makes better code. For
2542 example, on the 68000, when @var{x} is an integer constant that is in range
2543 for a @samp{moveq} instruction, the value of this macro is always
2544 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2545 Requiring a data register guarantees that a @samp{moveq} will be used.
2547 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2548 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2549 loaded into some register class. By returning @code{NO_REGS} you can
2550 force @var{x} into a memory location. For example, rs6000 can load
2551 immediate values into general-purpose registers, but does not have an
2552 instruction for loading an immediate value into a floating-point
2553 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2554 @var{x} is a floating-point constant. If the constant can't be loaded
2555 into any kind of register, code generation will be better if
2556 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2557 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2559 If an insn has pseudos in it after register allocation, reload will go
2560 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2561 to find the best one. Returning @code{NO_REGS}, in this case, makes
2562 reload add a @code{!} in front of the constraint: the x86 back-end uses
2563 this feature to discourage usage of 387 registers when math is done in
2564 the SSE registers (and vice versa).
2567 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2568 A C expression that places additional restrictions on the register class
2569 to use when it is necessary to copy value @var{x} into a register in class
2570 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2571 another, smaller class. On many machines, the following definition is
2575 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2578 Sometimes returning a more restrictive class makes better code. For
2579 example, on the 68000, when @var{x} is an integer constant that is in range
2580 for a @samp{moveq} instruction, the value of this macro is always
2581 @code{DATA_REGS} as long as @var{class} includes the data registers.
2582 Requiring a data register guarantees that a @samp{moveq} will be used.
2584 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2585 @var{class} is if @var{x} is a legitimate constant which cannot be
2586 loaded into some register class. By returning @code{NO_REGS} you can
2587 force @var{x} into a memory location. For example, rs6000 can load
2588 immediate values into general-purpose registers, but does not have an
2589 instruction for loading an immediate value into a floating-point
2590 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2591 @var{x} is a floating-point constant. If the constant can't be loaded
2592 into any kind of register, code generation will be better if
2593 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2594 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2596 If an insn has pseudos in it after register allocation, reload will go
2597 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2598 to find the best one. Returning @code{NO_REGS}, in this case, makes
2599 reload add a @code{!} in front of the constraint: the x86 back-end uses
2600 this feature to discourage usage of 387 registers when math is done in
2601 the SSE registers (and vice versa).
2604 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2605 Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2608 The default version of this hook always returns value of @code{rclass}
2611 You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2612 reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2615 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2616 A C expression that places additional restrictions on the register class
2617 to use when it is necessary to be able to hold a value of mode
2618 @var{mode} in a reload register for which class @var{class} would
2621 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2622 there are certain modes that simply can't go in certain reload classes.
2624 The value is a register class; perhaps @var{class}, or perhaps another,
2627 Don't define this macro unless the target machine has limitations which
2628 require the macro to do something nontrivial.
2631 @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})
2632 Many machines have some registers that cannot be copied directly to or
2633 from memory or even from other types of registers. An example is the
2634 @samp{MQ} register, which on most machines, can only be copied to or
2635 from general registers, but not memory. Below, we shall be using the
2636 term 'intermediate register' when a move operation cannot be performed
2637 directly, but has to be done by copying the source into the intermediate
2638 register first, and then copying the intermediate register to the
2639 destination. An intermediate register always has the same mode as
2640 source and destination. Since it holds the actual value being copied,
2641 reload might apply optimizations to re-use an intermediate register
2642 and eliding the copy from the source when it can determine that the
2643 intermediate register still holds the required value.
2645 Another kind of secondary reload is required on some machines which
2646 allow copying all registers to and from memory, but require a scratch
2647 register for stores to some memory locations (e.g., those with symbolic
2648 address on the RT, and those with certain symbolic address on the SPARC
2649 when compiling PIC)@. Scratch registers need not have the same mode
2650 as the value being copied, and usually hold a different value than
2651 that being copied. Special patterns in the md file are needed to
2652 describe how the copy is performed with the help of the scratch register;
2653 these patterns also describe the number, register class(es) and mode(s)
2654 of the scratch register(s).
2656 In some cases, both an intermediate and a scratch register are required.
2658 For input reloads, this target hook is called with nonzero @var{in_p},
2659 and @var{x} is an rtx that needs to be copied to a register of class
2660 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2661 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2662 needs to be copied to rtx @var{x} in @var{reload_mode}.
2664 If copying a register of @var{reload_class} from/to @var{x} requires
2665 an intermediate register, the hook @code{secondary_reload} should
2666 return the register class required for this intermediate register.
2667 If no intermediate register is required, it should return NO_REGS.
2668 If more than one intermediate register is required, describe the one
2669 that is closest in the copy chain to the reload register.
2671 If scratch registers are needed, you also have to describe how to
2672 perform the copy from/to the reload register to/from this
2673 closest intermediate register. Or if no intermediate register is
2674 required, but still a scratch register is needed, describe the
2675 copy from/to the reload register to/from the reload operand @var{x}.
2677 You do this by setting @code{sri->icode} to the instruction code of a pattern
2678 in the md file which performs the move. Operands 0 and 1 are the output
2679 and input of this copy, respectively. Operands from operand 2 onward are
2680 for scratch operands. These scratch operands must have a mode, and a
2681 single-register-class
2682 @c [later: or memory]
2685 When an intermediate register is used, the @code{secondary_reload}
2686 hook will be called again to determine how to copy the intermediate
2687 register to/from the reload operand @var{x}, so your hook must also
2688 have code to handle the register class of the intermediate operand.
2690 @c [For later: maybe we'll allow multi-alternative reload patterns -
2691 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2692 @c and match the constraints of input and output to determine the required
2693 @c alternative. A restriction would be that constraints used to match
2694 @c against reloads registers would have to be written as register class
2695 @c constraints, or we need a new target macro / hook that tells us if an
2696 @c arbitrary constraint can match an unknown register of a given class.
2697 @c Such a macro / hook would also be useful in other places.]
2700 @var{x} might be a pseudo-register or a @code{subreg} of a
2701 pseudo-register, which could either be in a hard register or in memory.
2702 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2703 in memory and the hard register number if it is in a register.
2705 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2706 currently not supported. For the time being, you will have to continue
2707 to use @code{SECONDARY_MEMORY_NEEDED} for that purpose.
2709 @code{copy_cost} also uses this target hook to find out how values are
2710 copied. If you want it to include some extra cost for the need to allocate
2711 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2712 Or if two dependent moves are supposed to have a lower cost than the sum
2713 of the individual moves due to expected fortuitous scheduling and/or special
2714 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2717 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2718 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2719 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2720 These macros are obsolete, new ports should use the target hook
2721 @code{TARGET_SECONDARY_RELOAD} instead.
2723 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2724 target hook. Older ports still define these macros to indicate to the
2725 reload phase that it may
2726 need to allocate at least one register for a reload in addition to the
2727 register to contain the data. Specifically, if copying @var{x} to a
2728 register @var{class} in @var{mode} requires an intermediate register,
2729 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2730 largest register class all of whose registers can be used as
2731 intermediate registers or scratch registers.
2733 If copying a register @var{class} in @var{mode} to @var{x} requires an
2734 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2735 was supposed to be defined be defined to return the largest register
2736 class required. If the
2737 requirements for input and output reloads were the same, the macro
2738 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2741 The values returned by these macros are often @code{GENERAL_REGS}.
2742 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2743 can be directly copied to or from a register of @var{class} in
2744 @var{mode} without requiring a scratch register. Do not define this
2745 macro if it would always return @code{NO_REGS}.
2747 If a scratch register is required (either with or without an
2748 intermediate register), you were supposed to define patterns for
2749 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2750 (@pxref{Standard Names}. These patterns, which were normally
2751 implemented with a @code{define_expand}, should be similar to the
2752 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2755 These patterns need constraints for the reload register and scratch
2757 contain a single register class. If the original reload register (whose
2758 class is @var{class}) can meet the constraint given in the pattern, the
2759 value returned by these macros is used for the class of the scratch
2760 register. Otherwise, two additional reload registers are required.
2761 Their classes are obtained from the constraints in the insn pattern.
2763 @var{x} might be a pseudo-register or a @code{subreg} of a
2764 pseudo-register, which could either be in a hard register or in memory.
2765 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2766 in memory and the hard register number if it is in a register.
2768 These macros should not be used in the case where a particular class of
2769 registers can only be copied to memory and not to another class of
2770 registers. In that case, secondary reload registers are not needed and
2771 would not be helpful. Instead, a stack location must be used to perform
2772 the copy and the @code{mov@var{m}} pattern should use memory as an
2773 intermediate storage. This case often occurs between floating-point and
2777 @defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m})
2778 Certain machines have the property that some registers cannot be copied
2779 to some other registers without using memory. Define this macro on
2780 those machines to be a C expression that is nonzero if objects of mode
2781 @var{m} in registers of @var{class1} can only be copied to registers of
2782 class @var{class2} by storing a register of @var{class1} into memory
2783 and loading that memory location into a register of @var{class2}.
2785 Do not define this macro if its value would always be zero.
2788 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2789 Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler
2790 allocates a stack slot for a memory location needed for register copies.
2791 If this macro is defined, the compiler instead uses the memory location
2792 defined by this macro.
2794 Do not define this macro if you do not define
2795 @code{SECONDARY_MEMORY_NEEDED}.
2798 @defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode})
2799 When the compiler needs a secondary memory location to copy between two
2800 registers of mode @var{mode}, it normally allocates sufficient memory to
2801 hold a quantity of @code{BITS_PER_WORD} bits and performs the store and
2802 load operations in a mode that many bits wide and whose class is the
2803 same as that of @var{mode}.
2805 This is right thing to do on most machines because it ensures that all
2806 bits of the register are copied and prevents accesses to the registers
2807 in a narrower mode, which some machines prohibit for floating-point
2810 However, this default behavior is not correct on some machines, such as
2811 the DEC Alpha, that store short integers in floating-point registers
2812 differently than in integer registers. On those machines, the default
2813 widening will not work correctly and you must define this macro to
2814 suppress that widening in some cases. See the file @file{alpha.h} for
2817 Do not define this macro if you do not define
2818 @code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that
2819 is @code{BITS_PER_WORD} bits wide is correct for your machine.
2822 @deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass})
2823 A target hook which returns @code{true} if pseudos that have been assigned
2824 to registers of class @var{rclass} would likely be spilled because
2825 registers of @var{rclass} are needed for spill registers.
2827 The default version of this target hook returns @code{true} if @var{rclass}
2828 has exactly one register and @code{false} otherwise. On most machines, this
2829 default should be used. Only use this target hook to some other expression
2830 if pseudos allocated by @file{local-alloc.c} end up in memory because their
2831 hard registers were needed for spill registers. If this target hook returns
2832 @code{false} for those classes, those pseudos will only be allocated by
2833 @file{global.c}, which knows how to reallocate the pseudo to another
2834 register. If there would not be another register available for reallocation,
2835 you should not change the implementation of this target hook since
2836 the only effect of such implementation would be to slow down register
2840 @deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, enum machine_mode @var{mode})
2841 A target hook returns the maximum number of consecutive registers
2842 of class @var{rclass} needed to hold a value of mode @var{mode}.
2844 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2845 the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass},
2846 @var{mode})} target hook should be the maximum value of
2847 @code{HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno}
2848 values in the class @var{rclass}.
2850 This target hook helps control the handling of multiple-word values
2853 The default version of this target hook returns the size of @var{mode}
2857 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2858 A C expression for the maximum number of consecutive registers
2859 of class @var{class} needed to hold a value of mode @var{mode}.
2861 This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact,
2862 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2863 should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno},
2864 @var{mode})} for all @var{regno} values in the class @var{class}.
2866 This macro helps control the handling of multiple-word values
2870 @defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class})
2871 If defined, a C expression that returns nonzero for a @var{class} for which
2872 a change from mode @var{from} to mode @var{to} is invalid.
2874 For the example, loading 32-bit integer or floating-point objects into
2875 floating-point registers on the Alpha extends them to 64 bits.
2876 Therefore loading a 64-bit object and then storing it as a 32-bit object
2877 does not store the low-order 32 bits, as would be the case for a normal
2878 register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS}
2882 #define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \
2883 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \
2884 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0)
2888 @node Old Constraints
2889 @section Obsolete Macros for Defining Constraints
2890 @cindex defining constraints, obsolete method
2891 @cindex constraints, defining, obsolete method
2893 Machine-specific constraints can be defined with these macros instead
2894 of the machine description constructs described in @ref{Define
2895 Constraints}. This mechanism is obsolete. New ports should not use
2896 it; old ports should convert to the new mechanism.
2898 @defmac CONSTRAINT_LEN (@var{char}, @var{str})
2899 For the constraint at the start of @var{str}, which starts with the letter
2900 @var{c}, return the length. This allows you to have register class /
2901 constant / extra constraints that are longer than a single letter;
2902 you don't need to define this macro if you can do with single-letter
2903 constraints only. The definition of this macro should use
2904 DEFAULT_CONSTRAINT_LEN for all the characters that you don't want
2905 to handle specially.
2906 There are some sanity checks in genoutput.c that check the constraint lengths
2907 for the md file, so you can also use this macro to help you while you are
2908 transitioning from a byzantine single-letter-constraint scheme: when you
2909 return a negative length for a constraint you want to re-use, genoutput
2910 will complain about every instance where it is used in the md file.
2913 @defmac REG_CLASS_FROM_LETTER (@var{char})
2914 A C expression which defines the machine-dependent operand constraint
2915 letters for register classes. If @var{char} is such a letter, the
2916 value should be the register class corresponding to it. Otherwise,
2917 the value should be @code{NO_REGS}. The register letter @samp{r},
2918 corresponding to class @code{GENERAL_REGS}, will not be passed
2919 to this macro; you do not need to handle it.
2922 @defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str})
2923 Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string
2924 passed in @var{str}, so that you can use suffixes to distinguish between
2928 @defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c})
2929 A C expression that defines the machine-dependent operand constraint
2930 letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify
2931 particular ranges of integer values. If @var{c} is one of those
2932 letters, the expression should check that @var{value}, an integer, is in
2933 the appropriate range and return 1 if so, 0 otherwise. If @var{c} is
2934 not one of those letters, the value should be 0 regardless of
2938 @defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2939 Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint
2940 string passed in @var{str}, so that you can use suffixes to distinguish
2941 between different variants.
2944 @defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c})
2945 A C expression that defines the machine-dependent operand constraint
2946 letters that specify particular ranges of @code{const_double} values
2947 (@samp{G} or @samp{H}).
2949 If @var{c} is one of those letters, the expression should check that
2950 @var{value}, an RTX of code @code{const_double}, is in the appropriate
2951 range and return 1 if so, 0 otherwise. If @var{c} is not one of those
2952 letters, the value should be 0 regardless of @var{value}.
2954 @code{const_double} is used for all floating-point constants and for
2955 @code{DImode} fixed-point constants. A given letter can accept either
2956 or both kinds of values. It can use @code{GET_MODE} to distinguish
2957 between these kinds.
2960 @defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str})
2961 Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint
2962 string passed in @var{str}, so that you can use suffixes to distinguish
2963 between different variants.
2966 @defmac EXTRA_CONSTRAINT (@var{value}, @var{c})
2967 A C expression that defines the optional machine-dependent constraint
2968 letters that can be used to segregate specific types of operands, usually
2969 memory references, for the target machine. Any letter that is not
2970 elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} /
2971 @code{REG_CLASS_FROM_CONSTRAINT}
2972 may be used. Normally this macro will not be defined.
2974 If it is required for a particular target machine, it should return 1
2975 if @var{value} corresponds to the operand type represented by the
2976 constraint letter @var{c}. If @var{c} is not defined as an extra
2977 constraint, the value returned should be 0 regardless of @var{value}.
2979 For example, on the ROMP, load instructions cannot have their output
2980 in r0 if the memory reference contains a symbolic address. Constraint
2981 letter @samp{Q} is defined as representing a memory address that does
2982 @emph{not} contain a symbolic address. An alternative is specified with
2983 a @samp{Q} constraint on the input and @samp{r} on the output. The next
2984 alternative specifies @samp{m} on the input and a register class that
2985 does not include r0 on the output.
2988 @defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str})
2989 Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed
2990 in @var{str}, so that you can use suffixes to distinguish between different
2994 @defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str})
2995 A C expression that defines the optional machine-dependent constraint
2996 letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should
2997 be treated like memory constraints by the reload pass.
2999 It should return 1 if the operand type represented by the constraint
3000 at the start of @var{str}, the first letter of which is the letter @var{c},
3001 comprises a subset of all memory references including
3002 all those whose address is simply a base register. This allows the reload
3003 pass to reload an operand, if it does not directly correspond to the operand
3004 type of @var{c}, by copying its address into a base register.
3006 For example, on the S/390, some instructions do not accept arbitrary
3007 memory references, but only those that do not make use of an index
3008 register. The constraint letter @samp{Q} is defined via
3009 @code{EXTRA_CONSTRAINT} as representing a memory address of this type.
3010 If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT},
3011 a @samp{Q} constraint can handle any memory operand, because the
3012 reload pass knows it can be reloaded by copying the memory address
3013 into a base register if required. This is analogous to the way
3014 an @samp{o} constraint can handle any memory operand.
3017 @defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str})
3018 A C expression that defines the optional machine-dependent constraint
3019 letters, amongst those accepted by @code{EXTRA_CONSTRAINT} /
3020 @code{EXTRA_CONSTRAINT_STR}, that should
3021 be treated like address constraints by the reload pass.
3023 It should return 1 if the operand type represented by the constraint
3024 at the start of @var{str}, which starts with the letter @var{c}, comprises
3025 a subset of all memory addresses including
3026 all those that consist of just a base register. This allows the reload
3027 pass to reload an operand, if it does not directly correspond to the operand
3028 type of @var{str}, by copying it into a base register.
3030 Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only
3031 be used with the @code{address_operand} predicate. It is treated
3032 analogously to the @samp{p} constraint.
3035 @node Stack and Calling
3036 @section Stack Layout and Calling Conventions
3037 @cindex calling conventions
3039 @c prevent bad page break with this line
3040 This describes the stack layout and calling conventions.
3044 * Exception Handling::
3049 * Register Arguments::
3051 * Aggregate Return::
3056 * Stack Smashing Protection::
3060 @subsection Basic Stack Layout
3061 @cindex stack frame layout
3062 @cindex frame layout
3064 @c prevent bad page break with this line
3065 Here is the basic stack layout.
3067 @defmac STACK_GROWS_DOWNWARD
3068 Define this macro if pushing a word onto the stack moves the stack
3069 pointer to a smaller address.
3071 When we say, ``define this macro if @dots{}'', it means that the
3072 compiler checks this macro only with @code{#ifdef} so the precise
3073 definition used does not matter.
3076 @defmac STACK_PUSH_CODE
3077 This macro defines the operation used when something is pushed
3078 on the stack. In RTL, a push operation will be
3079 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3081 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3082 and @code{POST_INC}. Which of these is correct depends on
3083 the stack direction and on whether the stack pointer points
3084 to the last item on the stack or whether it points to the
3085 space for the next item on the stack.
3087 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3088 defined, which is almost always right, and @code{PRE_INC} otherwise,
3089 which is often wrong.
3092 @defmac FRAME_GROWS_DOWNWARD
3093 Define this macro to nonzero value if the addresses of local variable slots
3094 are at negative offsets from the frame pointer.
3097 @defmac ARGS_GROW_DOWNWARD
3098 Define this macro if successive arguments to a function occupy decreasing
3099 addresses on the stack.
3102 @defmac STARTING_FRAME_OFFSET
3103 Offset from the frame pointer to the first local variable slot to be allocated.
3105 If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by
3106 subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}.
3107 Otherwise, it is found by adding the length of the first slot to the
3108 value @code{STARTING_FRAME_OFFSET}.
3109 @c i'm not sure if the above is still correct.. had to change it to get
3110 @c rid of an overfull. --mew 2feb93
3113 @defmac STACK_ALIGNMENT_NEEDED
3114 Define to zero to disable final alignment of the stack during reload.
3115 The nonzero default for this macro is suitable for most ports.
3117 On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there
3118 is a register save block following the local block that doesn't require
3119 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3120 stack alignment and do it in the backend.
3123 @defmac STACK_POINTER_OFFSET
3124 Offset from the stack pointer register to the first location at which
3125 outgoing arguments are placed. If not specified, the default value of
3126 zero is used. This is the proper value for most machines.
3128 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3129 the first location at which outgoing arguments are placed.
3132 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3133 Offset from the argument pointer register to the first argument's
3134 address. On some machines it may depend on the data type of the
3137 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3138 the first argument's address.
3141 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3142 Offset from the stack pointer register to an item dynamically allocated
3143 on the stack, e.g., by @code{alloca}.
3145 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3146 length of the outgoing arguments. The default is correct for most
3147 machines. See @file{function.c} for details.
3150 @defmac INITIAL_FRAME_ADDRESS_RTX
3151 A C expression whose value is RTL representing the address of the initial
3152 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3153 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3154 default value will be used. Define this macro in order to make frame pointer
3155 elimination work in the presence of @code{__builtin_frame_address (count)} and
3156 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3159 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3160 A C expression whose value is RTL representing the address in a stack
3161 frame where the pointer to the caller's frame is stored. Assume that
3162 @var{frameaddr} is an RTL expression for the address of the stack frame
3165 If you don't define this macro, the default is to return the value
3166 of @var{frameaddr}---that is, the stack frame address is also the
3167 address of the stack word that points to the previous frame.
3170 @defmac SETUP_FRAME_ADDRESSES
3171 If defined, a C expression that produces the machine-specific code to
3172 setup the stack so that arbitrary frames can be accessed. For example,
3173 on the SPARC, we must flush all of the register windows to the stack
3174 before we can access arbitrary stack frames. You will seldom need to
3178 @deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)
3179 This target hook should return an rtx that is used to store
3180 the address of the current frame into the built in @code{setjmp} buffer.
3181 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3182 machines. One reason you may need to define this target hook is if
3183 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3186 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3187 A C expression whose value is RTL representing the value of the frame
3188 address for the current frame. @var{frameaddr} is the frame pointer
3189 of the current frame. This is used for __builtin_frame_address.
3190 You need only define this macro if the frame address is not the same
3191 as the frame pointer. Most machines do not need to define it.
3194 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3195 A C expression whose value is RTL representing the value of the return
3196 address for the frame @var{count} steps up from the current frame, after
3197 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3198 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3199 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined.
3201 The value of the expression must always be the correct address when
3202 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3203 determine the return address of other frames.
3206 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3207 Define this if the return address of a particular stack frame is accessed
3208 from the frame pointer of the previous stack frame.
3211 @defmac INCOMING_RETURN_ADDR_RTX
3212 A C expression whose value is RTL representing the location of the
3213 incoming return address at the beginning of any function, before the
3214 prologue. This RTL is either a @code{REG}, indicating that the return
3215 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3218 You only need to define this macro if you want to support call frame
3219 debugging information like that provided by DWARF 2.
3221 If this RTL is a @code{REG}, you should also define
3222 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3225 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3226 A C expression whose value is an integer giving a DWARF 2 column
3227 number that may be used as an alternative return column. The column
3228 must not correspond to any gcc hard register (that is, it must not
3229 be in the range of @code{DWARF_FRAME_REGNUM}).
3231 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3232 general register, but an alternative column needs to be used for signal
3233 frames. Some targets have also used different frame return columns
3237 @defmac DWARF_ZERO_REG
3238 A C expression whose value is an integer giving a DWARF 2 register
3239 number that is considered to always have the value zero. This should
3240 only be defined if the target has an architected zero register, and
3241 someone decided it was a good idea to use that register number to
3242 terminate the stack backtrace. New ports should avoid this.
3245 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3246 This target hook allows the backend to emit frame-related insns that
3247 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3248 info engine will invoke it on insns of the form
3250 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3254 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3256 to let the backend emit the call frame instructions. @var{label} is
3257 the CFI label attached to the insn, @var{pattern} is the pattern of
3258 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3261 @defmac INCOMING_FRAME_SP_OFFSET
3262 A C expression whose value is an integer giving the offset, in bytes,
3263 from the value of the stack pointer register to the top of the stack
3264 frame at the beginning of any function, before the prologue. The top of
3265 the frame is defined to be the value of the stack pointer in the
3266 previous frame, just before the call instruction.
3268 You only need to define this macro if you want to support call frame
3269 debugging information like that provided by DWARF 2.
3272 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3273 A C expression whose value is an integer giving the offset, in bytes,
3274 from the argument pointer to the canonical frame address (cfa). The
3275 final value should coincide with that calculated by
3276 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3277 during virtual register instantiation.
3279 The default value for this macro is
3280 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3281 which is correct for most machines; in general, the arguments are found
3282 immediately before the stack frame. Note that this is not the case on
3283 some targets that save registers into the caller's frame, such as SPARC
3284 and rs6000, and so such targets need to define this macro.
3286 You only need to define this macro if the default is incorrect, and you
3287 want to support call frame debugging information like that provided by
3291 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3292 If defined, a C expression whose value is an integer giving the offset
3293 in bytes from the frame pointer to the canonical frame address (cfa).
3294 The final value should coincide with that calculated by
3295 @code{INCOMING_FRAME_SP_OFFSET}.
3297 Normally the CFA is calculated as an offset from the argument pointer,
3298 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3299 variable due to the ABI, this may not be possible. If this macro is
3300 defined, it implies that the virtual register instantiation should be
3301 based on the frame pointer instead of the argument pointer. Only one
3302 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3306 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3307 If defined, a C expression whose value is an integer giving the offset
3308 in bytes from the canonical frame address (cfa) to the frame base used
3309 in DWARF 2 debug information. The default is zero. A different value
3310 may reduce the size of debug information on some ports.
3313 @node Exception Handling
3314 @subsection Exception Handling Support
3315 @cindex exception handling
3317 @defmac EH_RETURN_DATA_REGNO (@var{N})
3318 A C expression whose value is the @var{N}th register number used for
3319 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3320 @var{N} registers are usable.
3322 The exception handling library routines communicate with the exception
3323 handlers via a set of agreed upon registers. Ideally these registers
3324 should be call-clobbered; it is possible to use call-saved registers,
3325 but may negatively impact code size. The target must support at least
3326 2 data registers, but should define 4 if there are enough free registers.
3328 You must define this macro if you want to support call frame exception
3329 handling like that provided by DWARF 2.
3332 @defmac EH_RETURN_STACKADJ_RTX
3333 A C expression whose value is RTL representing a location in which
3334 to store a stack adjustment to be applied before function return.
3335 This is used to unwind the stack to an exception handler's call frame.
3336 It will be assigned zero on code paths that return normally.
3338 Typically this is a call-clobbered hard register that is otherwise
3339 untouched by the epilogue, but could also be a stack slot.
3341 Do not define this macro if the stack pointer is saved and restored
3342 by the regular prolog and epilog code in the call frame itself; in
3343 this case, the exception handling library routines will update the
3344 stack location to be restored in place. Otherwise, you must define
3345 this macro if you want to support call frame exception handling like
3346 that provided by DWARF 2.
3349 @defmac EH_RETURN_HANDLER_RTX
3350 A C expression whose value is RTL representing a location in which
3351 to store the address of an exception handler to which we should
3352 return. It will not be assigned on code paths that return normally.
3354 Typically this is the location in the call frame at which the normal
3355 return address is stored. For targets that return by popping an
3356 address off the stack, this might be a memory address just below
3357 the @emph{target} call frame rather than inside the current call
3358 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3359 been assigned, so it may be used to calculate the location of the
3362 Some targets have more complex requirements than storing to an
3363 address calculable during initial code generation. In that case
3364 the @code{eh_return} instruction pattern should be used instead.
3366 If you want to support call frame exception handling, you must
3367 define either this macro or the @code{eh_return} instruction pattern.
3370 @defmac RETURN_ADDR_OFFSET
3371 If defined, an integer-valued C expression for which rtl will be generated
3372 to add it to the exception handler address before it is searched in the
3373 exception handling tables, and to subtract it again from the address before
3374 using it to return to the exception handler.
3377 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3378 This macro chooses the encoding of pointers embedded in the exception
3379 handling sections. If at all possible, this should be defined such
3380 that the exception handling section will not require dynamic relocations,
3381 and so may be read-only.
3383 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3384 @var{global} is true if the symbol may be affected by dynamic relocations.
3385 The macro should return a combination of the @code{DW_EH_PE_*} defines
3386 as found in @file{dwarf2.h}.
3388 If this macro is not defined, pointers will not be encoded but
3389 represented directly.
3392 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3393 This macro allows the target to emit whatever special magic is required
3394 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3395 Generic code takes care of pc-relative and indirect encodings; this must
3396 be defined if the target uses text-relative or data-relative encodings.
3398 This is a C statement that branches to @var{done} if the format was
3399 handled. @var{encoding} is the format chosen, @var{size} is the number
3400 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3404 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3405 This macro allows the target to add CPU and operating system specific
3406 code to the call-frame unwinder for use when there is no unwind data
3407 available. The most common reason to implement this macro is to unwind
3408 through signal frames.
3410 This macro is called from @code{uw_frame_state_for} in
3411 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3412 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3413 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3414 for the address of the code being executed and @code{context->cfa} for
3415 the stack pointer value. If the frame can be decoded, the register
3416 save addresses should be updated in @var{fs} and the macro should
3417 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3418 the macro should evaluate to @code{_URC_END_OF_STACK}.
3420 For proper signal handling in Java this macro is accompanied by
3421 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3424 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3425 This macro allows the target to add operating system specific code to the
3426 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3427 usually used for signal or interrupt frames.
3429 This macro is called from @code{uw_update_context} in libgcc's
3430 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3431 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3432 for the abi and context in the @code{.unwabi} directive. If the
3433 @code{.unwabi} directive can be handled, the register save addresses should
3434 be updated in @var{fs}.
3437 @defmac TARGET_USES_WEAK_UNWIND_INFO
3438 A C expression that evaluates to true if the target requires unwind
3439 info to be given comdat linkage. Define it to be @code{1} if comdat
3440 linkage is necessary. The default is @code{0}.
3443 @node Stack Checking
3444 @subsection Specifying How Stack Checking is Done
3446 GCC will check that stack references are within the boundaries of the
3447 stack, if the option @option{-fstack-check} is specified, in one of
3452 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3453 will assume that you have arranged for full stack checking to be done
3454 at appropriate places in the configuration files. GCC will not do
3455 other special processing.
3458 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3459 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3460 that you have arranged for static stack checking (checking of the
3461 static stack frame of functions) to be done at appropriate places
3462 in the configuration files. GCC will only emit code to do dynamic
3463 stack checking (checking on dynamic stack allocations) using the third
3467 If neither of the above are true, GCC will generate code to periodically
3468 ``probe'' the stack pointer using the values of the macros defined below.
3471 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3472 GCC will change its allocation strategy for large objects if the option
3473 @option{-fstack-check} is specified: they will always be allocated
3474 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3476 @defmac STACK_CHECK_BUILTIN
3477 A nonzero value if stack checking is done by the configuration files in a
3478 machine-dependent manner. You should define this macro if stack checking
3479 is required by the ABI of your machine or if you would like to do stack
3480 checking in some more efficient way than the generic approach. The default
3481 value of this macro is zero.
3484 @defmac STACK_CHECK_STATIC_BUILTIN
3485 A nonzero value if static stack checking is done by the configuration files
3486 in a machine-dependent manner. You should define this macro if you would
3487 like to do static stack checking in some more efficient way than the generic
3488 approach. The default value of this macro is zero.
3491 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3492 An integer specifying the interval at which GCC must generate stack probe
3493 instructions, defined as 2 raised to this integer. You will normally
3494 define this macro so that the interval be no larger than the size of
3495 the ``guard pages'' at the end of a stack area. The default value
3496 of 12 (4096-byte interval) is suitable for most systems.
3499 @defmac STACK_CHECK_MOVING_SP
3500 An integer which is nonzero if GCC should move the stack pointer page by page
3501 when doing probes. This can be necessary on systems where the stack pointer
3502 contains the bottom address of the memory area accessible to the executing
3503 thread at any point in time. In this situation an alternate signal stack
3504 is required in order to be able to recover from a stack overflow. The
3505 default value of this macro is zero.
3508 @defmac STACK_CHECK_PROTECT
3509 The number of bytes of stack needed to recover from a stack overflow, for
3510 languages where such a recovery is supported. The default value of 75 words
3511 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3512 8192 bytes with other exception handling mechanisms should be adequate for
3516 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3517 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3518 in the opposite case.
3520 @defmac STACK_CHECK_MAX_FRAME_SIZE
3521 The maximum size of a stack frame, in bytes. GCC will generate probe
3522 instructions in non-leaf functions to ensure at least this many bytes of
3523 stack are available. If a stack frame is larger than this size, stack
3524 checking will not be reliable and GCC will issue a warning. The
3525 default is chosen so that GCC only generates one instruction on most
3526 systems. You should normally not change the default value of this macro.
3529 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3530 GCC uses this value to generate the above warning message. It
3531 represents the amount of fixed frame used by a function, not including
3532 space for any callee-saved registers, temporaries and user variables.
3533 You need only specify an upper bound for this amount and will normally
3534 use the default of four words.
3537 @defmac STACK_CHECK_MAX_VAR_SIZE
3538 The maximum size, in bytes, of an object that GCC will place in the
3539 fixed area of the stack frame when the user specifies
3540 @option{-fstack-check}.
3541 GCC computed the default from the values of the above macros and you will
3542 normally not need to override that default.
3546 @node Frame Registers
3547 @subsection Registers That Address the Stack Frame
3549 @c prevent bad page break with this line
3550 This discusses registers that address the stack frame.
3552 @defmac STACK_POINTER_REGNUM
3553 The register number of the stack pointer register, which must also be a
3554 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3555 the hardware determines which register this is.
3558 @defmac FRAME_POINTER_REGNUM
3559 The register number of the frame pointer register, which is used to
3560 access automatic variables in the stack frame. On some machines, the
3561 hardware determines which register this is. On other machines, you can
3562 choose any register you wish for this purpose.
3565 @defmac HARD_FRAME_POINTER_REGNUM
3566 On some machines the offset between the frame pointer and starting
3567 offset of the automatic variables is not known until after register
3568 allocation has been done (for example, because the saved registers are
3569 between these two locations). On those machines, define
3570 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3571 be used internally until the offset is known, and define
3572 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3573 used for the frame pointer.
3575 You should define this macro only in the very rare circumstances when it
3576 is not possible to calculate the offset between the frame pointer and
3577 the automatic variables until after register allocation has been
3578 completed. When this macro is defined, you must also indicate in your
3579 definition of @code{ELIMINABLE_REGS} how to eliminate
3580 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3581 or @code{STACK_POINTER_REGNUM}.
3583 Do not define this macro if it would be the same as
3584 @code{FRAME_POINTER_REGNUM}.
3587 @defmac ARG_POINTER_REGNUM
3588 The register number of the arg pointer register, which is used to access
3589 the function's argument list. On some machines, this is the same as the
3590 frame pointer register. On some machines, the hardware determines which
3591 register this is. On other machines, you can choose any register you
3592 wish for this purpose. If this is not the same register as the frame
3593 pointer register, then you must mark it as a fixed register according to
3594 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3595 (@pxref{Elimination}).
3598 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3599 Define this to a preprocessor constant that is nonzero if
3600 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3601 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3602 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3603 definition is not suitable for use in preprocessor conditionals.
3606 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3607 Define this to a preprocessor constant that is nonzero if
3608 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3609 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3610 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3611 definition is not suitable for use in preprocessor conditionals.
3614 @defmac RETURN_ADDRESS_POINTER_REGNUM
3615 The register number of the return address pointer register, which is used to
3616 access the current function's return address from the stack. On some
3617 machines, the return address is not at a fixed offset from the frame
3618 pointer or stack pointer or argument pointer. This register can be defined
3619 to point to the return address on the stack, and then be converted by
3620 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3622 Do not define this macro unless there is no other way to get the return
3623 address from the stack.
3626 @defmac STATIC_CHAIN_REGNUM
3627 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3628 Register numbers used for passing a function's static chain pointer. If
3629 register windows are used, the register number as seen by the called
3630 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3631 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3632 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3635 The static chain register need not be a fixed register.
3637 If the static chain is passed in memory, these macros should not be
3638 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3641 @deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl}, bool @var{incoming_p})
3642 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3643 targets that may use different static chain locations for different
3644 nested functions. This may be required if the target has function
3645 attributes that affect the calling conventions of the function and
3646 those calling conventions use different static chain locations.
3648 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3650 If the static chain is passed in memory, this hook should be used to
3651 provide rtx giving @code{mem} expressions that denote where they are stored.
3652 Often the @code{mem} expression as seen by the caller will be at an offset
3653 from the stack pointer and the @code{mem} expression as seen by the callee
3654 will be at an offset from the frame pointer.
3655 @findex stack_pointer_rtx
3656 @findex frame_pointer_rtx
3657 @findex arg_pointer_rtx
3658 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3659 @code{arg_pointer_rtx} will have been initialized and should be used
3660 to refer to those items.
3663 @defmac DWARF_FRAME_REGISTERS
3664 This macro specifies the maximum number of hard registers that can be
3665 saved in a call frame. This is used to size data structures used in
3666 DWARF2 exception handling.
3668 Prior to GCC 3.0, this macro was needed in order to establish a stable
3669 exception handling ABI in the face of adding new hard registers for ISA
3670 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3671 in the number of hard registers. Nevertheless, this macro can still be
3672 used to reduce the runtime memory requirements of the exception handling
3673 routines, which can be substantial if the ISA contains a lot of
3674 registers that are not call-saved.
3676 If this macro is not defined, it defaults to
3677 @code{FIRST_PSEUDO_REGISTER}.
3680 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3682 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3683 for backward compatibility in pre GCC 3.0 compiled code.
3685 If this macro is not defined, it defaults to
3686 @code{DWARF_FRAME_REGISTERS}.
3689 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3691 Define this macro if the target's representation for dwarf registers
3692 is different than the internal representation for unwind column.
3693 Given a dwarf register, this macro should return the internal unwind
3694 column number to use instead.
3696 See the PowerPC's SPE target for an example.
3699 @defmac DWARF_FRAME_REGNUM (@var{regno})
3701 Define this macro if the target's representation for dwarf registers
3702 used in .eh_frame or .debug_frame is different from that used in other
3703 debug info sections. Given a GCC hard register number, this macro
3704 should return the .eh_frame register number. The default is
3705 @code{DBX_REGISTER_NUMBER (@var{regno})}.
3709 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3711 Define this macro to map register numbers held in the call frame info
3712 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3713 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3714 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3715 return @code{@var{regno}}.
3719 @defmac REG_VALUE_IN_UNWIND_CONTEXT
3721 Define this macro if the target stores register values as
3722 @code{_Unwind_Word} type in unwind context. It should be defined if
3723 target register size is larger than the size of @code{void *}. The
3724 default is to store register values as @code{void *} type.
3728 @defmac ASSUME_EXTENDED_UNWIND_CONTEXT
3730 Define this macro to be 1 if the target always uses extended unwind
3731 context with version, args_size and by_value fields. If it is undefined,
3732 it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is
3733 defined and 0 otherwise.
3738 @subsection Eliminating Frame Pointer and Arg Pointer
3740 @c prevent bad page break with this line
3741 This is about eliminating the frame pointer and arg pointer.
3743 @deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void)
3744 This target hook should return @code{true} if a function must have and use
3745 a frame pointer. This target hook is called in the reload pass. If its return
3746 value is @code{true} the function will have a frame pointer.
3748 This target hook can in principle examine the current function and decide
3749 according to the facts, but on most machines the constant @code{false} or the
3750 constant @code{true} suffices. Use @code{false} when the machine allows code
3751 to be generated with no frame pointer, and doing so saves some time or space.
3752 Use @code{true} when there is no possible advantage to avoiding a frame
3755 In certain cases, the compiler does not know how to produce valid code
3756 without a frame pointer. The compiler recognizes those cases and
3757 automatically gives the function a frame pointer regardless of what
3758 @code{TARGET_FRAME_POINTER_REQUIRED} returns. You don't need to worry about
3761 In a function that does not require a frame pointer, the frame pointer
3762 register can be allocated for ordinary usage, unless you mark it as a
3763 fixed register. See @code{FIXED_REGISTERS} for more information.
3765 Default return value is @code{false}.
3768 @findex get_frame_size
3769 @defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var})
3770 A C statement to store in the variable @var{depth-var} the difference
3771 between the frame pointer and the stack pointer values immediately after
3772 the function prologue. The value would be computed from information
3773 such as the result of @code{get_frame_size ()} and the tables of
3774 registers @code{regs_ever_live} and @code{call_used_regs}.
3776 If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and
3777 need not be defined. Otherwise, it must be defined even if
3778 @code{TARGET_FRAME_POINTER_REQUIRED} always returns true; in that
3779 case, you may set @var{depth-var} to anything.
3782 @defmac ELIMINABLE_REGS
3783 If defined, this macro specifies a table of register pairs used to
3784 eliminate unneeded registers that point into the stack frame. If it is not
3785 defined, the only elimination attempted by the compiler is to replace
3786 references to the frame pointer with references to the stack pointer.
3788 The definition of this macro is a list of structure initializations, each
3789 of which specifies an original and replacement register.
3791 On some machines, the position of the argument pointer is not known until
3792 the compilation is completed. In such a case, a separate hard register
3793 must be used for the argument pointer. This register can be eliminated by
3794 replacing it with either the frame pointer or the argument pointer,
3795 depending on whether or not the frame pointer has been eliminated.
3797 In this case, you might specify:
3799 #define ELIMINABLE_REGS \
3800 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3801 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3802 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3805 Note that the elimination of the argument pointer with the stack pointer is
3806 specified first since that is the preferred elimination.
3809 @deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg})
3810 This target hook should returns @code{true} if the compiler is allowed to
3811 try to replace register number @var{from_reg} with register number
3812 @var{to_reg}. This target hook need only be defined if @code{ELIMINABLE_REGS}
3813 is defined, and will usually be @code{true}, since most of the cases
3814 preventing register elimination are things that the compiler already
3817 Default return value is @code{true}.
3820 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3821 This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It
3822 specifies the initial difference between the specified pair of
3823 registers. This macro must be defined if @code{ELIMINABLE_REGS} is
3827 @node Stack Arguments
3828 @subsection Passing Function Arguments on the Stack
3829 @cindex arguments on stack
3830 @cindex stack arguments
3832 The macros in this section control how arguments are passed
3833 on the stack. See the following section for other macros that
3834 control passing certain arguments in registers.
3836 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype})
3837 This target hook returns @code{true} if an argument declared in a
3838 prototype as an integral type smaller than @code{int} should actually be
3839 passed as an @code{int}. In addition to avoiding errors in certain
3840 cases of mismatch, it also makes for better code on certain machines.
3841 The default is to not promote prototypes.
3845 A C expression. If nonzero, push insns will be used to pass
3847 If the target machine does not have a push instruction, set it to zero.
3848 That directs GCC to use an alternate strategy: to
3849 allocate the entire argument block and then store the arguments into
3850 it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too.
3853 @defmac PUSH_ARGS_REVERSED
3854 A C expression. If nonzero, function arguments will be evaluated from
3855 last to first, rather than from first to last. If this macro is not
3856 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3857 and args grow in opposite directions, and 0 otherwise.
3860 @defmac PUSH_ROUNDING (@var{npushed})
3861 A C expression that is the number of bytes actually pushed onto the
3862 stack when an instruction attempts to push @var{npushed} bytes.
3864 On some machines, the definition
3867 #define PUSH_ROUNDING(BYTES) (BYTES)
3871 will suffice. But on other machines, instructions that appear
3872 to push one byte actually push two bytes in an attempt to maintain
3873 alignment. Then the definition should be
3876 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3879 If the value of this macro has a type, it should be an unsigned type.
3882 @findex current_function_outgoing_args_size
3883 @defmac ACCUMULATE_OUTGOING_ARGS
3884 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3885 will be computed and placed into the variable
3886 @code{current_function_outgoing_args_size}. No space will be pushed
3887 onto the stack for each call; instead, the function prologue should
3888 increase the stack frame size by this amount.
3890 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3894 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3895 Define this macro if functions should assume that stack space has been
3896 allocated for arguments even when their values are passed in
3899 The value of this macro is the size, in bytes, of the area reserved for
3900 arguments passed in registers for the function represented by @var{fndecl},
3901 which can be zero if GCC is calling a library function.
3902 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3905 This space can be allocated by the caller, or be a part of the
3906 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3909 @c above is overfull. not sure what to do. --mew 5feb93 did
3910 @c something, not sure if it looks good. --mew 10feb93
3912 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
3913 Define this to a nonzero value if it is the responsibility of the
3914 caller to allocate the area reserved for arguments passed in registers
3915 when calling a function of @var{fntype}. @var{fntype} may be NULL
3916 if the function called is a library function.
3918 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
3919 whether the space for these arguments counts in the value of
3920 @code{current_function_outgoing_args_size}.
3923 @defmac STACK_PARMS_IN_REG_PARM_AREA
3924 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
3925 stack parameters don't skip the area specified by it.
3926 @c i changed this, makes more sens and it should have taken care of the
3927 @c overfull.. not as specific, tho. --mew 5feb93
3929 Normally, when a parameter is not passed in registers, it is placed on the
3930 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
3931 suppresses this behavior and causes the parameter to be passed on the
3932 stack in its natural location.
3935 @deftypefn {Target Hook} int TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, int @var{size})
3936 This target hook returns the number of bytes of its own arguments that
3937 a function pops on returning, or 0 if the function pops no arguments
3938 and the caller must therefore pop them all after the function returns.
3940 @var{fundecl} is a C variable whose value is a tree node that describes
3941 the function in question. Normally it is a node of type
3942 @code{FUNCTION_DECL} that describes the declaration of the function.
3943 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
3945 @var{funtype} is a C variable whose value is a tree node that
3946 describes the function in question. Normally it is a node of type
3947 @code{FUNCTION_TYPE} that describes the data type of the function.
3948 From this it is possible to obtain the data types of the value and
3949 arguments (if known).
3951 When a call to a library function is being considered, @var{fundecl}
3952 will contain an identifier node for the library function. Thus, if
3953 you need to distinguish among various library functions, you can do so
3954 by their names. Note that ``library function'' in this context means
3955 a function used to perform arithmetic, whose name is known specially
3956 in the compiler and was not mentioned in the C code being compiled.
3958 @var{size} is the number of bytes of arguments passed on the
3959 stack. If a variable number of bytes is passed, it is zero, and
3960 argument popping will always be the responsibility of the calling function.
3962 On the VAX, all functions always pop their arguments, so the definition
3963 of this macro is @var{size}. On the 68000, using the standard
3964 calling convention, no functions pop their arguments, so the value of
3965 the macro is always 0 in this case. But an alternative calling
3966 convention is available in which functions that take a fixed number of
3967 arguments pop them but other functions (such as @code{printf}) pop
3968 nothing (the caller pops all). When this convention is in use,
3969 @var{funtype} is examined to determine whether a function takes a fixed
3970 number of arguments.
3973 @defmac CALL_POPS_ARGS (@var{cum})
3974 A C expression that should indicate the number of bytes a call sequence
3975 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
3976 when compiling a function call.
3978 @var{cum} is the variable in which all arguments to the called function
3979 have been accumulated.
3981 On certain architectures, such as the SH5, a call trampoline is used
3982 that pops certain registers off the stack, depending on the arguments
3983 that have been passed to the function. Since this is a property of the
3984 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
3988 @node Register Arguments
3989 @subsection Passing Arguments in Registers
3990 @cindex arguments in registers
3991 @cindex registers arguments
3993 This section describes the macros which let you control how various
3994 types of arguments are passed in registers or how they are arranged in
3997 @deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
3998 Return an RTX indicating whether a function argument is passed in a
3999 register and if so, which register.
4001 The arguments are @var{ca}, which summarizes all the previous
4002 arguments; @var{mode}, the machine mode of the argument; @var{type},
4003 the data type of the argument as a tree node or 0 if that is not known
4004 (which happens for C support library functions); and @var{named},
4005 which is @code{true} for an ordinary argument and @code{false} for
4006 nameless arguments that correspond to @samp{@dots{}} in the called
4007 function's prototype. @var{type} can be an incomplete type if a
4008 syntax error has previously occurred.
4010 The return value is usually either a @code{reg} RTX for the hard
4011 register in which to pass the argument, or zero to pass the argument
4014 The value of the expression can also be a @code{parallel} RTX@. This is
4015 used when an argument is passed in multiple locations. The mode of the
4016 @code{parallel} should be the mode of the entire argument. The
4017 @code{parallel} holds any number of @code{expr_list} pairs; each one
4018 describes where part of the argument is passed. In each
4019 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
4020 register in which to pass this part of the argument, and the mode of the
4021 register RTX indicates how large this part of the argument is. The
4022 second operand of the @code{expr_list} is a @code{const_int} which gives
4023 the offset in bytes into the entire argument of where this part starts.
4024 As a special exception the first @code{expr_list} in the @code{parallel}
4025 RTX may have a first operand of zero. This indicates that the entire
4026 argument is also stored on the stack.
4028 The last time this hook is called, it is called with @code{MODE ==
4029 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4030 pattern as operands 2 and 3 respectively.
4032 @cindex @file{stdarg.h} and register arguments
4033 The usual way to make the ISO library @file{stdarg.h} work on a
4034 machine where some arguments are usually passed in registers, is to
4035 cause nameless arguments to be passed on the stack instead. This is
4036 done by making @code{TARGET_FUNCTION_ARG} return 0 whenever
4037 @var{named} is @code{false}.
4039 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG}
4040 @cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG}
4041 You may use the hook @code{targetm.calls.must_pass_in_stack}
4042 in the definition of this macro to determine if this argument is of a
4043 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4044 is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an
4045 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4046 defined, the argument will be computed in the stack and then loaded into
4050 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (enum machine_mode @var{mode}, const_tree @var{type})
4051 This target hook should return @code{true} if we should not pass @var{type}
4052 solely in registers. The file @file{expr.h} defines a
4053 definition that is usually appropriate, refer to @file{expr.h} for additional
4057 @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})
4058 Define this hook if the target machine has ``register windows'', so
4059 that the register in which a function sees an arguments is not
4060 necessarily the same as the one in which the caller passed the
4063 For such machines, @code{TARGET_FUNCTION_ARG} computes the register in
4064 which the caller passes the value, and
4065 @code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar
4066 fashion to tell the function being called where the arguments will
4069 If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined,
4070 @code{TARGET_FUNCTION_ARG} serves both purposes.
4073 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, enum machine_mode @var{mode}, tree @var{type}, bool @var{named})
4074 This target hook returns the number of bytes at the beginning of an
4075 argument that must be put in registers. The value must be zero for
4076 arguments that are passed entirely in registers or that are entirely
4077 pushed on the stack.
4079 On some machines, certain arguments must be passed partially in
4080 registers and partially in memory. On these machines, typically the
4081 first few words of arguments are passed in registers, and the rest
4082 on the stack. If a multi-word argument (a @code{double} or a
4083 structure) crosses that boundary, its first few words must be passed
4084 in registers and the rest must be pushed. This macro tells the
4085 compiler when this occurs, and how many bytes should go in registers.
4087 @code{TARGET_FUNCTION_ARG} for these arguments should return the first
4088 register to be used by the caller for this argument; likewise
4089 @code{TARGET_FUNCTION_INCOMING_ARG}, for the called function.
4092 @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})
4093 This target hook should return @code{true} if an argument at the
4094 position indicated by @var{cum} should be passed by reference. This
4095 predicate is queried after target independent reasons for being
4096 passed by reference, such as @code{TREE_ADDRESSABLE (type)}.
4098 If the hook returns true, a copy of that argument is made in memory and a
4099 pointer to the argument is passed instead of the argument itself.
4100 The pointer is passed in whatever way is appropriate for passing a pointer
4104 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, enum machine_mode @var{mode}, const_tree @var{type}, bool @var{named})
4105 The function argument described by the parameters to this hook is
4106 known to be passed by reference. The hook should return true if the
4107 function argument should be copied by the callee instead of copied
4110 For any argument for which the hook returns true, if it can be
4111 determined that the argument is not modified, then a copy need
4114 The default version of this hook always returns false.
4117 @defmac CUMULATIVE_ARGS
4118 A C type for declaring a variable that is used as the first argument
4119 of @code{TARGET_FUNCTION_ARG} and other related values. For some
4120 target machines, the type @code{int} suffices and can hold the number
4121 of bytes of argument so far.
4123 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4124 arguments that have been passed on the stack. The compiler has other
4125 variables to keep track of that. For target machines on which all
4126 arguments are passed on the stack, there is no need to store anything in
4127 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4128 should not be empty, so use @code{int}.
4131 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4132 If defined, this macro is called before generating any code for a
4133 function, but after the @var{cfun} descriptor for the function has been
4134 created. The back end may use this macro to update @var{cfun} to
4135 reflect an ABI other than that which would normally be used by default.
4136 If the compiler is generating code for a compiler-generated function,
4137 @var{fndecl} may be @code{NULL}.
4140 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4141 A C statement (sans semicolon) for initializing the variable
4142 @var{cum} for the state at the beginning of the argument list. The
4143 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4144 is the tree node for the data type of the function which will receive
4145 the args, or 0 if the args are to a compiler support library function.
4146 For direct calls that are not libcalls, @var{fndecl} contain the
4147 declaration node of the function. @var{fndecl} is also set when
4148 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4149 being compiled. @var{n_named_args} is set to the number of named
4150 arguments, including a structure return address if it is passed as a
4151 parameter, when making a call. When processing incoming arguments,
4152 @var{n_named_args} is set to @minus{}1.
4154 When processing a call to a compiler support library function,
4155 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4156 contains the name of the function, as a string. @var{libname} is 0 when
4157 an ordinary C function call is being processed. Thus, each time this
4158 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4159 never both of them at once.
4162 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4163 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4164 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4165 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4166 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4167 0)} is used instead.
4170 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4171 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4172 finding the arguments for the function being compiled. If this macro is
4173 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4175 The value passed for @var{libname} is always 0, since library routines
4176 with special calling conventions are never compiled with GCC@. The
4177 argument @var{libname} exists for symmetry with
4178 @code{INIT_CUMULATIVE_ARGS}.
4179 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4180 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4183 @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})
4184 This hook updates the summarizer variable pointed to by @var{ca} to
4185 advance past an argument in the argument list. The values @var{mode},
4186 @var{type} and @var{named} describe that argument. Once this is done,
4187 the variable @var{cum} is suitable for analyzing the @emph{following}
4188 argument with @code{TARGET_FUNCTION_ARG}, etc.
4190 This hook need not do anything if the argument in question was passed
4191 on the stack. The compiler knows how to track the amount of stack space
4192 used for arguments without any special help.
4195 @defmac FUNCTION_ARG_OFFSET (@var{mode}, @var{type})
4196 If defined, a C expression that is the number of bytes to add to the
4197 offset of the argument passed in memory. This is needed for the SPU,
4198 which passes @code{char} and @code{short} arguments in the preferred
4199 slot that is in the middle of the quad word instead of starting at the
4203 @defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type})
4204 If defined, a C expression which determines whether, and in which direction,
4205 to pad out an argument with extra space. The value should be of type
4206 @code{enum direction}: either @code{upward} to pad above the argument,
4207 @code{downward} to pad below, or @code{none} to inhibit padding.
4209 The @emph{amount} of padding is not controlled by this macro, but by the
4210 target hook @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is
4211 always just enough to reach the next multiple of that boundary.
4213 This macro has a default definition which is right for most systems.
4214 For little-endian machines, the default is to pad upward. For
4215 big-endian machines, the default is to pad downward for an argument of
4216 constant size shorter than an @code{int}, and upward otherwise.
4219 @defmac PAD_VARARGS_DOWN
4220 If defined, a C expression which determines whether the default
4221 implementation of va_arg will attempt to pad down before reading the
4222 next argument, if that argument is smaller than its aligned space as
4223 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4224 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4227 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4228 Specify padding for the last element of a block move between registers and
4229 memory. @var{first} is nonzero if this is the only element. Defining this
4230 macro allows better control of register function parameters on big-endian
4231 machines, without using @code{PARALLEL} rtl. In particular,
4232 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4233 registers, as there is no longer a "wrong" part of a register; For example,
4234 a three byte aggregate may be passed in the high part of a register if so
4238 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (enum machine_mode @var{mode}, const_tree @var{type})
4239 This hook returns the alignment boundary, in bits, of an argument
4240 with the specified mode and type. The default hook returns
4241 @code{PARM_BOUNDARY} for all arguments.
4244 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (enum machine_mode @var{mode}, const_tree @var{type})
4245 Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY},
4246 which is the default value for this hook. You can define this hook to
4247 return a different value if an argument size must be rounded to a larger
4251 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4252 A C expression that is nonzero if @var{regno} is the number of a hard
4253 register in which function arguments are sometimes passed. This does
4254 @emph{not} include implicit arguments such as the static chain and
4255 the structure-value address. On many machines, no registers can be
4256 used for this purpose since all function arguments are pushed on the
4260 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type})
4261 This hook should return true if parameter of type @var{type} are passed
4262 as two scalar parameters. By default, GCC will attempt to pack complex
4263 arguments into the target's word size. Some ABIs require complex arguments
4264 to be split and treated as their individual components. For example, on
4265 AIX64, complex floats should be passed in a pair of floating point
4266 registers, even though a complex float would fit in one 64-bit floating
4269 The default value of this hook is @code{NULL}, which is treated as always
4273 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4274 This hook returns a type node for @code{va_list} for the target.
4275 The default version of the hook returns @code{void*}.
4278 @deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree})
4279 This target hook is used in function @code{c_common_nodes_and_builtins}
4280 to iterate through the target specific builtin types for va_list. The
4281 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4282 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4284 The arguments @var{pname} and @var{ptree} are used to store the result of
4285 this macro and are set to the name of the va_list builtin type and its
4287 If the return value of this macro is zero, then there is no more element.
4288 Otherwise the @var{IDX} should be increased for the next call of this
4289 macro to iterate through all types.
4292 @deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl})
4293 This hook returns the va_list type of the calling convention specified by
4295 The default version of this hook returns @code{va_list_type_node}.
4298 @deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type})
4299 This hook returns the va_list type of the calling convention specified by the
4300 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4304 @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})
4305 This hook performs target-specific gimplification of
4306 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4307 arguments to @code{va_arg}; the latter two are as in
4308 @code{gimplify.c:gimplify_expr}.
4311 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (enum machine_mode @var{mode})
4312 Define this to return nonzero if the port can handle pointers
4313 with machine mode @var{mode}. The default version of this
4314 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4317 @deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (struct ao_ref_s *@var{ref})
4318 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.
4321 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4322 Define this to return nonzero if the port is prepared to handle
4323 insns involving scalar mode @var{mode}. For a scalar mode to be
4324 considered supported, all the basic arithmetic and comparisons
4327 The default version of this hook returns true for any mode
4328 required to handle the basic C types (as defined by the port).
4329 Included here are the double-word arithmetic supported by the
4330 code in @file{optabs.c}.
4333 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (enum machine_mode @var{mode})
4334 Define this to return nonzero if the port is prepared to handle
4335 insns involving vector mode @var{mode}. At the very least, it
4336 must have move patterns for this mode.
4339 @deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (enum machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems})
4340 Return true if GCC should try to use a scalar mode to store an array
4341 of @var{nelems} elements, given that each element has mode @var{mode}.
4342 Returning true here overrides the usual @code{MAX_FIXED_MODE} limit
4343 and allows GCC to use any defined integer mode.
4345 One use of this hook is to support vector load and store operations
4346 that operate on several homogeneous vectors. For example, ARM NEON
4347 has operations like:
4350 int8x8x3_t vld3_s8 (const int8_t *)
4353 where the return type is defined as:
4356 typedef struct int8x8x3_t
4362 If this hook allows @code{val} to have a scalar mode, then
4363 @code{int8x8x3_t} can have the same mode. GCC can then store
4364 @code{int8x8x3_t}s in registers rather than forcing them onto the stack.
4367 @deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (enum machine_mode @var{mode})
4368 Define this to return nonzero for machine modes for which the port has
4369 small register classes. If this target hook returns nonzero for a given
4370 @var{mode}, the compiler will try to minimize the lifetime of registers
4371 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4372 In this case, the hook is expected to return nonzero if it returns nonzero
4375 On some machines, it is risky to let hard registers live across arbitrary
4376 insns. Typically, these machines have instructions that require values
4377 to be in specific registers (like an accumulator), and reload will fail
4378 if the required hard register is used for another purpose across such an
4381 Passes before reload do not know which hard registers will be used
4382 in an instruction, but the machine modes of the registers set or used in
4383 the instruction are already known. And for some machines, register
4384 classes are small for, say, integer registers but not for floating point
4385 registers. For example, the AMD x86-64 architecture requires specific
4386 registers for the legacy x86 integer instructions, but there are many
4387 SSE registers for floating point operations. On such targets, a good
4388 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4389 machine modes but zero for the SSE register classes.
4391 The default version of this hook returns false for any mode. It is always
4392 safe to redefine this hook to return with a nonzero value. But if you
4393 unnecessarily define it, you will reduce the amount of optimizations
4394 that can be performed in some cases. If you do not define this hook
4395 to return a nonzero value when it is required, the compiler will run out
4396 of spill registers and print a fatal error message.
4399 @deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM
4400 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.
4404 @subsection How Scalar Function Values Are Returned
4405 @cindex return values in registers
4406 @cindex values, returned by functions
4407 @cindex scalars, returned as values
4409 This section discusses the macros that control returning scalars as
4410 values---values that can fit in registers.
4412 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
4414 Define this to return an RTX representing the place where a function
4415 returns or receives a value of data type @var{ret_type}, a tree node
4416 representing a data type. @var{fn_decl_or_type} is a tree node
4417 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4418 function being called. If @var{outgoing} is false, the hook should
4419 compute the register in which the caller will see the return value.
4420 Otherwise, the hook should return an RTX representing the place where
4421 a function returns a value.
4423 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4424 (Actually, on most machines, scalar values are returned in the same
4425 place regardless of mode.) The value of the expression is usually a
4426 @code{reg} RTX for the hard register where the return value is stored.
4427 The value can also be a @code{parallel} RTX, if the return value is in
4428 multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the
4429 @code{parallel} form. Note that the callee will populate every
4430 location specified in the @code{parallel}, but if the first element of
4431 the @code{parallel} contains the whole return value, callers will use
4432 that element as the canonical location and ignore the others. The m68k
4433 port uses this type of @code{parallel} to return pointers in both
4434 @samp{%a0} (the canonical location) and @samp{%d0}.
4436 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4437 the same promotion rules specified in @code{PROMOTE_MODE} if
4438 @var{valtype} is a scalar type.
4440 If the precise function being called is known, @var{func} is a tree
4441 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4442 pointer. This makes it possible to use a different value-returning
4443 convention for specific functions when all their calls are
4446 Some target machines have ``register windows'' so that the register in
4447 which a function returns its value is not the same as the one in which
4448 the caller sees the value. For such machines, you should return
4449 different RTX depending on @var{outgoing}.
4451 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4452 aggregate data types, because these are returned in another way. See
4453 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4456 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4457 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4458 a new target instead.
4461 @defmac LIBCALL_VALUE (@var{mode})
4462 A C expression to create an RTX representing the place where a library
4463 function returns a value of mode @var{mode}.
4465 Note that ``library function'' in this context means a compiler
4466 support routine, used to perform arithmetic, whose name is known
4467 specially by the compiler and was not mentioned in the C code being
4471 @deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (enum machine_mode @var{mode}, const_rtx @var{fun})
4472 Define this hook if the back-end needs to know the name of the libcall
4473 function in order to determine where the result should be returned.
4475 The mode of the result is given by @var{mode} and the name of the called
4476 library function is given by @var{fun}. The hook should return an RTX
4477 representing the place where the library function result will be returned.
4479 If this hook is not defined, then LIBCALL_VALUE will be used.
4482 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4483 A C expression that is nonzero if @var{regno} is the number of a hard
4484 register in which the values of called function may come back.
4486 A register whose use for returning values is limited to serving as the
4487 second of a pair (for a value of type @code{double}, say) need not be
4488 recognized by this macro. So for most machines, this definition
4492 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4495 If the machine has register windows, so that the caller and the called
4496 function use different registers for the return value, this macro
4497 should recognize only the caller's register numbers.
4499 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4500 for a new target instead.
4503 @deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno})
4504 A target hook that return @code{true} if @var{regno} is the number of a hard
4505 register in which the values of called function may come back.
4507 A register whose use for returning values is limited to serving as the
4508 second of a pair (for a value of type @code{double}, say) need not be
4509 recognized by this target hook.
4511 If the machine has register windows, so that the caller and the called
4512 function use different registers for the return value, this target hook
4513 should recognize only the caller's register numbers.
4515 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4518 @defmac APPLY_RESULT_SIZE
4519 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4520 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4521 saving and restoring an arbitrary return value.
4524 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type})
4525 This hook should return true if values of type @var{type} are returned
4526 at the most significant end of a register (in other words, if they are
4527 padded at the least significant end). You can assume that @var{type}
4528 is returned in a register; the caller is required to check this.
4530 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4531 be able to hold the complete return value. For example, if a 1-, 2-
4532 or 3-byte structure is returned at the most significant end of a
4533 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4537 @node Aggregate Return
4538 @subsection How Large Values Are Returned
4539 @cindex aggregates as return values
4540 @cindex large return values
4541 @cindex returning aggregate values
4542 @cindex structure value address
4544 When a function value's mode is @code{BLKmode} (and in some other
4545 cases), the value is not returned according to
4546 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4547 caller passes the address of a block of memory in which the value
4548 should be stored. This address is called the @dfn{structure value
4551 This section describes how to control returning structure values in
4554 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype})
4555 This target hook should return a nonzero value to say to return the
4556 function value in memory, just as large structures are always returned.
4557 Here @var{type} will be the data type of the value, and @var{fntype}
4558 will be the type of the function doing the returning, or @code{NULL} for
4561 Note that values of mode @code{BLKmode} must be explicitly handled
4562 by this function. Also, the option @option{-fpcc-struct-return}
4563 takes effect regardless of this macro. On most systems, it is
4564 possible to leave the hook undefined; this causes a default
4565 definition to be used, whose value is the constant 1 for @code{BLKmode}
4566 values, and 0 otherwise.
4568 Do not use this hook to indicate that structures and unions should always
4569 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4573 @defmac DEFAULT_PCC_STRUCT_RETURN
4574 Define this macro to be 1 if all structure and union return values must be
4575 in memory. Since this results in slower code, this should be defined
4576 only if needed for compatibility with other compilers or with an ABI@.
4577 If you define this macro to be 0, then the conventions used for structure
4578 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4581 If not defined, this defaults to the value 1.
4584 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4585 This target hook should return the location of the structure value
4586 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4587 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4588 be @code{NULL}, for libcalls. You do not need to define this target
4589 hook if the address is always passed as an ``invisible'' first
4592 On some architectures the place where the structure value address
4593 is found by the called function is not the same place that the
4594 caller put it. This can be due to register windows, or it could
4595 be because the function prologue moves it to a different place.
4596 @var{incoming} is @code{1} or @code{2} when the location is needed in
4597 the context of the called function, and @code{0} in the context of
4600 If @var{incoming} is nonzero and the address is to be found on the
4601 stack, return a @code{mem} which refers to the frame pointer. If
4602 @var{incoming} is @code{2}, the result is being used to fetch the
4603 structure value address at the beginning of a function. If you need
4604 to emit adjusting code, you should do it at this point.
4607 @defmac PCC_STATIC_STRUCT_RETURN
4608 Define this macro if the usual system convention on the target machine
4609 for returning structures and unions is for the called function to return
4610 the address of a static variable containing the value.
4612 Do not define this if the usual system convention is for the caller to
4613 pass an address to the subroutine.
4615 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4616 nothing when you use @option{-freg-struct-return} mode.
4619 @deftypefn {Target Hook} {enum machine_mode} TARGET_GET_RAW_RESULT_MODE (int @var{regno})
4620 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.
4623 @deftypefn {Target Hook} {enum machine_mode} TARGET_GET_RAW_ARG_MODE (int @var{regno})
4624 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.
4628 @subsection Caller-Saves Register Allocation
4630 If you enable it, GCC can save registers around function calls. This
4631 makes it possible to use call-clobbered registers to hold variables that
4632 must live across calls.
4634 @defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls})
4635 A C expression to determine whether it is worthwhile to consider placing
4636 a pseudo-register in a call-clobbered hard register and saving and
4637 restoring it around each function call. The expression should be 1 when
4638 this is worth doing, and 0 otherwise.
4640 If you don't define this macro, a default is used which is good on most
4641 machines: @code{4 * @var{calls} < @var{refs}}.
4644 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4645 A C expression specifying which mode is required for saving @var{nregs}
4646 of a pseudo-register in call-clobbered hard register @var{regno}. If
4647 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4648 returned. For most machines this macro need not be defined since GCC
4649 will select the smallest suitable mode.
4652 @node Function Entry
4653 @subsection Function Entry and Exit
4654 @cindex function entry and exit
4658 This section describes the macros that output function entry
4659 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4661 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4662 If defined, a function that outputs the assembler code for entry to a
4663 function. The prologue is responsible for setting up the stack frame,
4664 initializing the frame pointer register, saving registers that must be
4665 saved, and allocating @var{size} additional bytes of storage for the
4666 local variables. @var{size} is an integer. @var{file} is a stdio
4667 stream to which the assembler code should be output.
4669 The label for the beginning of the function need not be output by this
4670 macro. That has already been done when the macro is run.
4672 @findex regs_ever_live
4673 To determine which registers to save, the macro can refer to the array
4674 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4675 @var{r} is used anywhere within the function. This implies the function
4676 prologue should save register @var{r}, provided it is not one of the
4677 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4678 @code{regs_ever_live}.)
4680 On machines that have ``register windows'', the function entry code does
4681 not save on the stack the registers that are in the windows, even if
4682 they are supposed to be preserved by function calls; instead it takes
4683 appropriate steps to ``push'' the register stack, if any non-call-used
4684 registers are used in the function.
4686 @findex frame_pointer_needed
4687 On machines where functions may or may not have frame-pointers, the
4688 function entry code must vary accordingly; it must set up the frame
4689 pointer if one is wanted, and not otherwise. To determine whether a
4690 frame pointer is in wanted, the macro can refer to the variable
4691 @code{frame_pointer_needed}. The variable's value will be 1 at run
4692 time in a function that needs a frame pointer. @xref{Elimination}.
4694 The function entry code is responsible for allocating any stack space
4695 required for the function. This stack space consists of the regions
4696 listed below. In most cases, these regions are allocated in the
4697 order listed, with the last listed region closest to the top of the
4698 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4699 the highest address if it is not defined). You can use a different order
4700 for a machine if doing so is more convenient or required for
4701 compatibility reasons. Except in cases where required by standard
4702 or by a debugger, there is no reason why the stack layout used by GCC
4703 need agree with that used by other compilers for a machine.
4706 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4707 If defined, a function that outputs assembler code at the end of a
4708 prologue. This should be used when the function prologue is being
4709 emitted as RTL, and you have some extra assembler that needs to be
4710 emitted. @xref{prologue instruction pattern}.
4713 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4714 If defined, a function that outputs assembler code at the start of an
4715 epilogue. This should be used when the function epilogue is being
4716 emitted as RTL, and you have some extra assembler that needs to be
4717 emitted. @xref{epilogue instruction pattern}.
4720 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size})
4721 If defined, a function that outputs the assembler code for exit from a
4722 function. The epilogue is responsible for restoring the saved
4723 registers and stack pointer to their values when the function was
4724 called, and returning control to the caller. This macro takes the
4725 same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4726 registers to restore are determined from @code{regs_ever_live} and
4727 @code{CALL_USED_REGISTERS} in the same way.
4729 On some machines, there is a single instruction that does all the work
4730 of returning from the function. On these machines, give that
4731 instruction the name @samp{return} and do not define the macro
4732 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4734 Do not define a pattern named @samp{return} if you want the
4735 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4736 switches to control whether return instructions or epilogues are used,
4737 define a @samp{return} pattern with a validity condition that tests the
4738 target switches appropriately. If the @samp{return} pattern's validity
4739 condition is false, epilogues will be used.
4741 On machines where functions may or may not have frame-pointers, the
4742 function exit code must vary accordingly. Sometimes the code for these
4743 two cases is completely different. To determine whether a frame pointer
4744 is wanted, the macro can refer to the variable
4745 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4746 a function that needs a frame pointer.
4748 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4749 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4750 The C variable @code{current_function_is_leaf} is nonzero for such a
4751 function. @xref{Leaf Functions}.
4753 On some machines, some functions pop their arguments on exit while
4754 others leave that for the caller to do. For example, the 68020 when
4755 given @option{-mrtd} pops arguments in functions that take a fixed
4756 number of arguments.
4758 @findex current_function_pops_args
4759 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4760 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4761 needs to know what was decided. The number of bytes of the current
4762 function's arguments that this function should pop is available in
4763 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4768 @findex current_function_pretend_args_size
4769 A region of @code{current_function_pretend_args_size} bytes of
4770 uninitialized space just underneath the first argument arriving on the
4771 stack. (This may not be at the very start of the allocated stack region
4772 if the calling sequence has pushed anything else since pushing the stack
4773 arguments. But usually, on such machines, nothing else has been pushed
4774 yet, because the function prologue itself does all the pushing.) This
4775 region is used on machines where an argument may be passed partly in
4776 registers and partly in memory, and, in some cases to support the
4777 features in @code{<stdarg.h>}.
4780 An area of memory used to save certain registers used by the function.
4781 The size of this area, which may also include space for such things as
4782 the return address and pointers to previous stack frames, is
4783 machine-specific and usually depends on which registers have been used
4784 in the function. Machines with register windows often do not require
4788 A region of at least @var{size} bytes, possibly rounded up to an allocation
4789 boundary, to contain the local variables of the function. On some machines,
4790 this region and the save area may occur in the opposite order, with the
4791 save area closer to the top of the stack.
4794 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4795 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4796 @code{current_function_outgoing_args_size} bytes to be used for outgoing
4797 argument lists of the function. @xref{Stack Arguments}.
4800 @defmac EXIT_IGNORE_STACK
4801 Define this macro as a C expression that is nonzero if the return
4802 instruction or the function epilogue ignores the value of the stack
4803 pointer; in other words, if it is safe to delete an instruction to
4804 adjust the stack pointer before a return from the function. The
4807 Note that this macro's value is relevant only for functions for which
4808 frame pointers are maintained. It is never safe to delete a final
4809 stack adjustment in a function that has no frame pointer, and the
4810 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
4813 @defmac EPILOGUE_USES (@var{regno})
4814 Define this macro as a C expression that is nonzero for registers that are
4815 used by the epilogue or the @samp{return} pattern. The stack and frame
4816 pointer registers are already assumed to be used as needed.
4819 @defmac EH_USES (@var{regno})
4820 Define this macro as a C expression that is nonzero for registers that are
4821 used by the exception handling mechanism, and so should be considered live
4822 on entry to an exception edge.
4825 @defmac DELAY_SLOTS_FOR_EPILOGUE
4826 Define this macro if the function epilogue contains delay slots to which
4827 instructions from the rest of the function can be ``moved''. The
4828 definition should be a C expression whose value is an integer
4829 representing the number of delay slots there.
4832 @defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n})
4833 A C expression that returns 1 if @var{insn} can be placed in delay
4834 slot number @var{n} of the epilogue.
4836 The argument @var{n} is an integer which identifies the delay slot now
4837 being considered (since different slots may have different rules of
4838 eligibility). It is never negative and is always less than the number
4839 of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns).
4840 If you reject a particular insn for a given delay slot, in principle, it
4841 may be reconsidered for a subsequent delay slot. Also, other insns may
4842 (at least in principle) be considered for the so far unfilled delay
4845 @findex current_function_epilogue_delay_list
4846 @findex final_scan_insn
4847 The insns accepted to fill the epilogue delay slots are put in an RTL
4848 list made with @code{insn_list} objects, stored in the variable
4849 @code{current_function_epilogue_delay_list}. The insn for the first
4850 delay slot comes first in the list. Your definition of the macro
4851 @code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by
4852 outputting the insns in this list, usually by calling
4853 @code{final_scan_insn}.
4855 You need not define this macro if you did not define
4856 @code{DELAY_SLOTS_FOR_EPILOGUE}.
4859 @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})
4860 A function that outputs the assembler code for a thunk
4861 function, used to implement C++ virtual function calls with multiple
4862 inheritance. The thunk acts as a wrapper around a virtual function,
4863 adjusting the implicit object parameter before handing control off to
4866 First, emit code to add the integer @var{delta} to the location that
4867 contains the incoming first argument. Assume that this argument
4868 contains a pointer, and is the one used to pass the @code{this} pointer
4869 in C++. This is the incoming argument @emph{before} the function prologue,
4870 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
4871 all other incoming arguments.
4873 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
4874 made after adding @code{delta}. In particular, if @var{p} is the
4875 adjusted pointer, the following adjustment should be made:
4878 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
4881 After the additions, emit code to jump to @var{function}, which is a
4882 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
4883 not touch the return address. Hence returning from @var{FUNCTION} will
4884 return to whoever called the current @samp{thunk}.
4886 The effect must be as if @var{function} had been called directly with
4887 the adjusted first argument. This macro is responsible for emitting all
4888 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
4889 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
4891 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
4892 have already been extracted from it.) It might possibly be useful on
4893 some targets, but probably not.
4895 If you do not define this macro, the target-independent code in the C++
4896 front end will generate a less efficient heavyweight thunk that calls
4897 @var{function} instead of jumping to it. The generic approach does
4898 not support varargs.
4901 @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})
4902 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
4903 to output the assembler code for the thunk function specified by the
4904 arguments it is passed, and false otherwise. In the latter case, the
4905 generic approach will be used by the C++ front end, with the limitations
4910 @subsection Generating Code for Profiling
4911 @cindex profiling, code generation
4913 These macros will help you generate code for profiling.
4915 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
4916 A C statement or compound statement to output to @var{file} some
4917 assembler code to call the profiling subroutine @code{mcount}.
4920 The details of how @code{mcount} expects to be called are determined by
4921 your operating system environment, not by GCC@. To figure them out,
4922 compile a small program for profiling using the system's installed C
4923 compiler and look at the assembler code that results.
4925 Older implementations of @code{mcount} expect the address of a counter
4926 variable to be loaded into some register. The name of this variable is
4927 @samp{LP} followed by the number @var{labelno}, so you would generate
4928 the name using @samp{LP%d} in a @code{fprintf}.
4931 @defmac PROFILE_HOOK
4932 A C statement or compound statement to output to @var{file} some assembly
4933 code to call the profiling subroutine @code{mcount} even the target does
4934 not support profiling.
4937 @defmac NO_PROFILE_COUNTERS
4938 Define this macro to be an expression with a nonzero value if the
4939 @code{mcount} subroutine on your system does not need a counter variable
4940 allocated for each function. This is true for almost all modern
4941 implementations. If you define this macro, you must not use the
4942 @var{labelno} argument to @code{FUNCTION_PROFILER}.
4945 @defmac PROFILE_BEFORE_PROLOGUE
4946 Define this macro if the code for function profiling should come before
4947 the function prologue. Normally, the profiling code comes after.
4951 @subsection Permitting tail calls
4954 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
4955 True if it is ok to do sibling call optimization for the specified
4956 call expression @var{exp}. @var{decl} will be the called function,
4957 or @code{NULL} if this is an indirect call.
4959 It is not uncommon for limitations of calling conventions to prevent
4960 tail calls to functions outside the current unit of translation, or
4961 during PIC compilation. The hook is used to enforce these restrictions,
4962 as the @code{sibcall} md pattern can not fail, or fall over to a
4963 ``normal'' call. The criteria for successful sibling call optimization
4964 may vary greatly between different architectures.
4967 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs})
4968 Add any hard registers to @var{regs} that are live on entry to the
4969 function. This hook only needs to be defined to provide registers that
4970 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
4971 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
4972 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
4973 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
4976 @deftypefn {Target Hook} void TARGET_SET_UP_BY_PROLOGUE (struct hard_reg_set_container *@var{})
4977 This hook should add additional registers that are computed by the prologue to the hard regset for shrink-wrapping optimization purposes.
4980 @node Stack Smashing Protection
4981 @subsection Stack smashing protection
4982 @cindex stack smashing protection
4984 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
4985 This hook returns a @code{DECL} node for the external variable to use
4986 for the stack protection guard. This variable is initialized by the
4987 runtime to some random value and is used to initialize the guard value
4988 that is placed at the top of the local stack frame. The type of this
4989 variable must be @code{ptr_type_node}.
4991 The default version of this hook creates a variable called
4992 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
4995 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
4996 This hook returns a @code{CALL_EXPR} that alerts the runtime that the
4997 stack protect guard variable has been modified. This expression should
4998 involve a call to a @code{noreturn} function.
5000 The default version of this hook invokes a function called
5001 @samp{__stack_chk_fail}, taking no arguments. This function is
5002 normally defined in @file{libgcc2.c}.
5005 @deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts})
5006 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
5010 @section Implementing the Varargs Macros
5011 @cindex varargs implementation
5013 GCC comes with an implementation of @code{<varargs.h>} and
5014 @code{<stdarg.h>} that work without change on machines that pass arguments
5015 on the stack. Other machines require their own implementations of
5016 varargs, and the two machine independent header files must have
5017 conditionals to include it.
5019 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
5020 the calling convention for @code{va_start}. The traditional
5021 implementation takes just one argument, which is the variable in which
5022 to store the argument pointer. The ISO implementation of
5023 @code{va_start} takes an additional second argument. The user is
5024 supposed to write the last named argument of the function here.
5026 However, @code{va_start} should not use this argument. The way to find
5027 the end of the named arguments is with the built-in functions described
5030 @defmac __builtin_saveregs ()
5031 Use this built-in function to save the argument registers in memory so
5032 that the varargs mechanism can access them. Both ISO and traditional
5033 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5034 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5036 On some machines, @code{__builtin_saveregs} is open-coded under the
5037 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5038 other machines, it calls a routine written in assembler language,
5039 found in @file{libgcc2.c}.
5041 Code generated for the call to @code{__builtin_saveregs} appears at the
5042 beginning of the function, as opposed to where the call to
5043 @code{__builtin_saveregs} is written, regardless of what the code is.
5044 This is because the registers must be saved before the function starts
5045 to use them for its own purposes.
5046 @c i rewrote the first sentence above to fix an overfull hbox. --mew
5050 @defmac __builtin_next_arg (@var{lastarg})
5051 This builtin returns the address of the first anonymous stack
5052 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5053 returns the address of the location above the first anonymous stack
5054 argument. Use it in @code{va_start} to initialize the pointer for
5055 fetching arguments from the stack. Also use it in @code{va_start} to
5056 verify that the second parameter @var{lastarg} is the last named argument
5057 of the current function.
5060 @defmac __builtin_classify_type (@var{object})
5061 Since each machine has its own conventions for which data types are
5062 passed in which kind of register, your implementation of @code{va_arg}
5063 has to embody these conventions. The easiest way to categorize the
5064 specified data type is to use @code{__builtin_classify_type} together
5065 with @code{sizeof} and @code{__alignof__}.
5067 @code{__builtin_classify_type} ignores the value of @var{object},
5068 considering only its data type. It returns an integer describing what
5069 kind of type that is---integer, floating, pointer, structure, and so on.
5071 The file @file{typeclass.h} defines an enumeration that you can use to
5072 interpret the values of @code{__builtin_classify_type}.
5075 These machine description macros help implement varargs:
5077 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
5078 If defined, this hook produces the machine-specific code for a call to
5079 @code{__builtin_saveregs}. This code will be moved to the very
5080 beginning of the function, before any parameter access are made. The
5081 return value of this function should be an RTX that contains the value
5082 to use as the return of @code{__builtin_saveregs}.
5085 @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})
5086 This target hook offers an alternative to using
5087 @code{__builtin_saveregs} and defining the hook
5088 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5089 register arguments into the stack so that all the arguments appear to
5090 have been passed consecutively on the stack. Once this is done, you can
5091 use the standard implementation of varargs that works for machines that
5092 pass all their arguments on the stack.
5094 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5095 structure, containing the values that are obtained after processing the
5096 named arguments. The arguments @var{mode} and @var{type} describe the
5097 last named argument---its machine mode and its data type as a tree node.
5099 The target hook should do two things: first, push onto the stack all the
5100 argument registers @emph{not} used for the named arguments, and second,
5101 store the size of the data thus pushed into the @code{int}-valued
5102 variable pointed to by @var{pretend_args_size}. The value that you
5103 store here will serve as additional offset for setting up the stack
5106 Because you must generate code to push the anonymous arguments at
5107 compile time without knowing their data types,
5108 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5109 have just a single category of argument register and use it uniformly
5112 If the argument @var{second_time} is nonzero, it means that the
5113 arguments of the function are being analyzed for the second time. This
5114 happens for an inline function, which is not actually compiled until the
5115 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5116 not generate any instructions in this case.
5119 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca})
5120 Define this hook to return @code{true} if the location where a function
5121 argument is passed depends on whether or not it is a named argument.
5123 This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG}
5124 is set for varargs and stdarg functions. If this hook returns
5125 @code{true}, the @var{named} argument is always true for named
5126 arguments, and false for unnamed arguments. If it returns @code{false},
5127 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5128 then all arguments are treated as named. Otherwise, all named arguments
5129 except the last are treated as named.
5131 You need not define this hook if it always returns @code{false}.
5134 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca})
5135 If you need to conditionally change ABIs so that one works with
5136 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5137 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5138 defined, then define this hook to return @code{true} if
5139 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5140 Otherwise, you should not define this hook.
5144 @section Trampolines for Nested Functions
5145 @cindex trampolines for nested functions
5146 @cindex nested functions, trampolines for
5148 A @dfn{trampoline} is a small piece of code that is created at run time
5149 when the address of a nested function is taken. It normally resides on
5150 the stack, in the stack frame of the containing function. These macros
5151 tell GCC how to generate code to allocate and initialize a
5154 The instructions in the trampoline must do two things: load a constant
5155 address into the static chain register, and jump to the real address of
5156 the nested function. On CISC machines such as the m68k, this requires
5157 two instructions, a move immediate and a jump. Then the two addresses
5158 exist in the trampoline as word-long immediate operands. On RISC
5159 machines, it is often necessary to load each address into a register in
5160 two parts. Then pieces of each address form separate immediate
5163 The code generated to initialize the trampoline must store the variable
5164 parts---the static chain value and the function address---into the
5165 immediate operands of the instructions. On a CISC machine, this is
5166 simply a matter of copying each address to a memory reference at the
5167 proper offset from the start of the trampoline. On a RISC machine, it
5168 may be necessary to take out pieces of the address and store them
5171 @deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f})
5172 This hook is called by @code{assemble_trampoline_template} to output,
5173 on the stream @var{f}, assembler code for a block of data that contains
5174 the constant parts of a trampoline. This code should not include a
5175 label---the label is taken care of automatically.
5177 If you do not define this hook, it means no template is needed
5178 for the target. Do not define this hook on systems where the block move
5179 code to copy the trampoline into place would be larger than the code
5180 to generate it on the spot.
5183 @defmac TRAMPOLINE_SECTION
5184 Return the section into which the trampoline template is to be placed
5185 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5188 @defmac TRAMPOLINE_SIZE
5189 A C expression for the size in bytes of the trampoline, as an integer.
5192 @defmac TRAMPOLINE_ALIGNMENT
5193 Alignment required for trampolines, in bits.
5195 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5196 is used for aligning trampolines.
5199 @deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain})
5200 This hook is called to initialize a trampoline.
5201 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5202 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5203 RTX for the static chain value that should be passed to the function
5206 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5207 first thing this hook should do is emit a block move into @var{m_tramp}
5208 from the memory block returned by @code{assemble_trampoline_template}.
5209 Note that the block move need only cover the constant parts of the
5210 trampoline. If the target isolates the variable parts of the trampoline
5211 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5213 If the target requires any other actions, such as flushing caches or
5214 enabling stack execution, these actions should be performed after
5215 initializing the trampoline proper.
5218 @deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr})
5219 This hook should perform any machine-specific adjustment in
5220 the address of the trampoline. Its argument contains the address of the
5221 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5222 the address to be used for a function call should be different from the
5223 address at which the template was stored, the different address should
5224 be returned; otherwise @var{addr} should be returned unchanged.
5225 If this hook is not defined, @var{addr} will be used for function calls.
5228 Implementing trampolines is difficult on many machines because they have
5229 separate instruction and data caches. Writing into a stack location
5230 fails to clear the memory in the instruction cache, so when the program
5231 jumps to that location, it executes the old contents.
5233 Here are two possible solutions. One is to clear the relevant parts of
5234 the instruction cache whenever a trampoline is set up. The other is to
5235 make all trampolines identical, by having them jump to a standard
5236 subroutine. The former technique makes trampoline execution faster; the
5237 latter makes initialization faster.
5239 To clear the instruction cache when a trampoline is initialized, define
5240 the following macro.
5242 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5243 If defined, expands to a C expression clearing the @emph{instruction
5244 cache} in the specified interval. The definition of this macro would
5245 typically be a series of @code{asm} statements. Both @var{beg} and
5246 @var{end} are both pointer expressions.
5249 To use a standard subroutine, define the following macro. In addition,
5250 you must make sure that the instructions in a trampoline fill an entire
5251 cache line with identical instructions, or else ensure that the
5252 beginning of the trampoline code is always aligned at the same point in
5253 its cache line. Look in @file{m68k.h} as a guide.
5255 @defmac TRANSFER_FROM_TRAMPOLINE
5256 Define this macro if trampolines need a special subroutine to do their
5257 work. The macro should expand to a series of @code{asm} statements
5258 which will be compiled with GCC@. They go in a library function named
5259 @code{__transfer_from_trampoline}.
5261 If you need to avoid executing the ordinary prologue code of a compiled
5262 C function when you jump to the subroutine, you can do so by placing a
5263 special label of your own in the assembler code. Use one @code{asm}
5264 statement to generate an assembler label, and another to make the label
5265 global. Then trampolines can use that label to jump directly to your
5266 special assembler code.
5270 @section Implicit Calls to Library Routines
5271 @cindex library subroutine names
5272 @cindex @file{libgcc.a}
5274 @c prevent bad page break with this line
5275 Here is an explanation of implicit calls to library routines.
5277 @defmac DECLARE_LIBRARY_RENAMES
5278 This macro, if defined, should expand to a piece of C code that will get
5279 expanded when compiling functions for libgcc.a. It can be used to
5280 provide alternate names for GCC's internal library functions if there
5281 are ABI-mandated names that the compiler should provide.
5284 @findex set_optab_libfunc
5285 @findex init_one_libfunc
5286 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5287 This hook should declare additional library routines or rename
5288 existing ones, using the functions @code{set_optab_libfunc} and
5289 @code{init_one_libfunc} defined in @file{optabs.c}.
5290 @code{init_optabs} calls this macro after initializing all the normal
5293 The default is to do nothing. Most ports don't need to define this hook.
5296 @deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX
5297 If false (the default), internal library routines start with two
5298 underscores. If set to true, these routines start with @code{__gnu_}
5299 instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This
5300 currently only affects functions defined in @file{libgcc2.c}. If this
5301 is set to true, the @file{tm.h} file must also
5302 @code{#define LIBGCC2_GNU_PREFIX}.
5305 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5306 This macro should return @code{true} if the library routine that
5307 implements the floating point comparison operator @var{comparison} in
5308 mode @var{mode} will return a boolean, and @var{false} if it will
5311 GCC's own floating point libraries return tristates from the
5312 comparison operators, so the default returns false always. Most ports
5313 don't need to define this macro.
5316 @defmac TARGET_LIB_INT_CMP_BIASED
5317 This macro should evaluate to @code{true} if the integer comparison
5318 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5319 operand is smaller than the second, 1 to indicate that they are equal,
5320 and 2 to indicate that the first operand is greater than the second.
5321 If this macro evaluates to @code{false} the comparison functions return
5322 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5323 in @file{libgcc.a}, you do not need to define this macro.
5326 @cindex @code{EDOM}, implicit usage
5329 The value of @code{EDOM} on the target machine, as a C integer constant
5330 expression. If you don't define this macro, GCC does not attempt to
5331 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5332 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5335 If you do not define @code{TARGET_EDOM}, then compiled code reports
5336 domain errors by calling the library function and letting it report the
5337 error. If mathematical functions on your system use @code{matherr} when
5338 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5339 that @code{matherr} is used normally.
5342 @cindex @code{errno}, implicit usage
5343 @defmac GEN_ERRNO_RTX
5344 Define this macro as a C expression to create an rtl expression that
5345 refers to the global ``variable'' @code{errno}. (On certain systems,
5346 @code{errno} may not actually be a variable.) If you don't define this
5347 macro, a reasonable default is used.
5350 @cindex C99 math functions, implicit usage
5351 @defmac TARGET_C99_FUNCTIONS
5352 When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into
5353 @code{sinf} and similarly for other functions defined by C99 standard. The
5354 default is zero because a number of existing systems lack support for these
5355 functions in their runtime so this macro needs to be redefined to one on
5356 systems that do support the C99 runtime.
5359 @cindex sincos math function, implicit usage
5360 @defmac TARGET_HAS_SINCOS
5361 When this macro is nonzero, GCC will implicitly optimize calls to @code{sin}
5362 and @code{cos} with the same argument to a call to @code{sincos}. The
5363 default is zero. The target has to provide the following functions:
5365 void sincos(double x, double *sin, double *cos);
5366 void sincosf(float x, float *sin, float *cos);
5367 void sincosl(long double x, long double *sin, long double *cos);
5371 @defmac NEXT_OBJC_RUNTIME
5372 Set this macro to 1 to use the "NeXT" Objective-C message sending conventions
5373 by default. This calling convention involves passing the object, the selector
5374 and the method arguments all at once to the method-lookup library function.
5375 This is the usual setting when targeting Darwin/Mac OS X systems, which have
5376 the NeXT runtime installed.
5378 If the macro is set to 0, the "GNU" Objective-C message sending convention
5379 will be used by default. This convention passes just the object and the
5380 selector to the method-lookup function, which returns a pointer to the method.
5382 In either case, it remains possible to select code-generation for the alternate
5383 scheme, by means of compiler command line switches.
5386 @node Addressing Modes
5387 @section Addressing Modes
5388 @cindex addressing modes
5390 @c prevent bad page break with this line
5391 This is about addressing modes.
5393 @defmac HAVE_PRE_INCREMENT
5394 @defmacx HAVE_PRE_DECREMENT
5395 @defmacx HAVE_POST_INCREMENT
5396 @defmacx HAVE_POST_DECREMENT
5397 A C expression that is nonzero if the machine supports pre-increment,
5398 pre-decrement, post-increment, or post-decrement addressing respectively.
5401 @defmac HAVE_PRE_MODIFY_DISP
5402 @defmacx HAVE_POST_MODIFY_DISP
5403 A C expression that is nonzero if the machine supports pre- or
5404 post-address side-effect generation involving constants other than
5405 the size of the memory operand.
5408 @defmac HAVE_PRE_MODIFY_REG
5409 @defmacx HAVE_POST_MODIFY_REG
5410 A C expression that is nonzero if the machine supports pre- or
5411 post-address side-effect generation involving a register displacement.
5414 @defmac CONSTANT_ADDRESS_P (@var{x})
5415 A C expression that is 1 if the RTX @var{x} is a constant which
5416 is a valid address. On most machines the default definition of
5417 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5418 is acceptable, but a few machines are more restrictive as to which
5419 constant addresses are supported.
5422 @defmac CONSTANT_P (@var{x})
5423 @code{CONSTANT_P}, which is defined by target-independent code,
5424 accepts integer-values expressions whose values are not explicitly
5425 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5426 expressions and @code{const} arithmetic expressions, in addition to
5427 @code{const_int} and @code{const_double} expressions.
5430 @defmac MAX_REGS_PER_ADDRESS
5431 A number, the maximum number of registers that can appear in a valid
5432 memory address. Note that it is up to you to specify a value equal to
5433 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5437 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (enum machine_mode @var{mode}, rtx @var{x}, bool @var{strict})
5438 A function that returns whether @var{x} (an RTX) is a legitimate memory
5439 address on the target machine for a memory operand of mode @var{mode}.
5441 Legitimate addresses are defined in two variants: a strict variant and a
5442 non-strict one. The @var{strict} parameter chooses which variant is
5443 desired by the caller.
5445 The strict variant is used in the reload pass. It must be defined so
5446 that any pseudo-register that has not been allocated a hard register is
5447 considered a memory reference. This is because in contexts where some
5448 kind of register is required, a pseudo-register with no hard register
5449 must be rejected. For non-hard registers, the strict variant should look
5450 up the @code{reg_renumber} array; it should then proceed using the hard
5451 register number in the array, or treat the pseudo as a memory reference
5452 if the array holds @code{-1}.
5454 The non-strict variant is used in other passes. It must be defined to
5455 accept all pseudo-registers in every context where some kind of
5456 register is required.
5458 Normally, constant addresses which are the sum of a @code{symbol_ref}
5459 and an integer are stored inside a @code{const} RTX to mark them as
5460 constant. Therefore, there is no need to recognize such sums
5461 specifically as legitimate addresses. Normally you would simply
5462 recognize any @code{const} as legitimate.
5464 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5465 sums that are not marked with @code{const}. It assumes that a naked
5466 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5467 naked constant sums as illegitimate addresses, so that none of them will
5468 be given to @code{PRINT_OPERAND_ADDRESS}.
5470 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5471 On some machines, whether a symbolic address is legitimate depends on
5472 the section that the address refers to. On these machines, define the
5473 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5474 into the @code{symbol_ref}, and then check for it here. When you see a
5475 @code{const}, you will have to look inside it to find the
5476 @code{symbol_ref} in order to determine the section. @xref{Assembler
5479 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5480 Some ports are still using a deprecated legacy substitute for
5481 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5485 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5489 and should @code{goto @var{label}} if the address @var{x} is a valid
5490 address on the target machine for a memory operand of mode @var{mode}.
5492 @findex REG_OK_STRICT
5493 Compiler source files that want to use the strict variant of this
5494 macro define the macro @code{REG_OK_STRICT}. You should use an
5495 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
5496 that case and the non-strict variant otherwise.
5498 Using the hook is usually simpler because it limits the number of
5499 files that are recompiled when changes are made.
5502 @defmac TARGET_MEM_CONSTRAINT
5503 A single character to be used instead of the default @code{'m'}
5504 character for general memory addresses. This defines the constraint
5505 letter which matches the memory addresses accepted by
5506 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5507 support new address formats in your back end without changing the
5508 semantics of the @code{'m'} constraint. This is necessary in order to
5509 preserve functionality of inline assembly constructs using the
5510 @code{'m'} constraint.
5513 @defmac FIND_BASE_TERM (@var{x})
5514 A C expression to determine the base term of address @var{x},
5515 or to provide a simplified version of @var{x} from which @file{alias.c}
5516 can easily find the base term. This macro is used in only two places:
5517 @code{find_base_value} and @code{find_base_term} in @file{alias.c}.
5519 It is always safe for this macro to not be defined. It exists so
5520 that alias analysis can understand machine-dependent addresses.
5522 The typical use of this macro is to handle addresses containing
5523 a label_ref or symbol_ref within an UNSPEC@.
5526 @deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, enum machine_mode @var{mode})
5527 This hook is given an invalid memory address @var{x} for an
5528 operand of mode @var{mode} and should try to return a valid memory
5531 @findex break_out_memory_refs
5532 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5533 and @var{oldx} will be the operand that was given to that function to produce
5536 The code of the hook should not alter the substructure of
5537 @var{x}. If it transforms @var{x} into a more legitimate form, it
5538 should return the new @var{x}.
5540 It is not necessary for this hook to come up with a legitimate address,
5541 with the exception of native TLS addresses (@pxref{Emulated TLS}).
5542 The compiler has standard ways of doing so in all cases. In fact, if
5543 the target supports only emulated TLS, it
5544 is safe to omit this hook or make it return @var{x} if it cannot find
5545 a valid way to legitimize the address. But often a machine-dependent
5546 strategy can generate better code.
5549 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5550 A C compound statement that attempts to replace @var{x}, which is an address
5551 that needs reloading, with a valid memory address for an operand of mode
5552 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5553 It is not necessary to define this macro, but it might be useful for
5554 performance reasons.
5556 For example, on the i386, it is sometimes possible to use a single
5557 reload register instead of two by reloading a sum of two pseudo
5558 registers into a register. On the other hand, for number of RISC
5559 processors offsets are limited so that often an intermediate address
5560 needs to be generated in order to address a stack slot. By defining
5561 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5562 generated for adjacent some stack slots can be made identical, and thus
5565 @emph{Note}: This macro should be used with caution. It is necessary
5566 to know something of how reload works in order to effectively use this,
5567 and it is quite easy to produce macros that build in too much knowledge
5568 of reload internals.
5570 @emph{Note}: This macro must be able to reload an address created by a
5571 previous invocation of this macro. If it fails to handle such addresses
5572 then the compiler may generate incorrect code or abort.
5575 The macro definition should use @code{push_reload} to indicate parts that
5576 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5577 suitable to be passed unaltered to @code{push_reload}.
5579 The code generated by this macro must not alter the substructure of
5580 @var{x}. If it transforms @var{x} into a more legitimate form, it
5581 should assign @var{x} (which will always be a C variable) a new value.
5582 This also applies to parts that you change indirectly by calling
5585 @findex strict_memory_address_p
5586 The macro definition may use @code{strict_memory_address_p} to test if
5587 the address has become legitimate.
5590 If you want to change only a part of @var{x}, one standard way of doing
5591 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5592 single level of rtl. Thus, if the part to be changed is not at the
5593 top level, you'll need to replace first the top level.
5594 It is not necessary for this macro to come up with a legitimate
5595 address; but often a machine-dependent strategy can generate better code.
5598 @deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr})
5599 This hook returns @code{true} if memory address @var{addr} can have
5600 different meanings depending on the machine mode of the memory
5601 reference it is used for or if the address is valid for some modes
5604 Autoincrement and autodecrement addresses typically have mode-dependent
5605 effects because the amount of the increment or decrement is the size
5606 of the operand being addressed. Some machines have other mode-dependent
5607 addresses. Many RISC machines have no mode-dependent addresses.
5609 You may assume that @var{addr} is a valid address for the machine.
5611 The default version of this hook returns @code{false}.
5614 @defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label})
5615 A C statement or compound statement with a conditional @code{goto
5616 @var{label};} executed if memory address @var{x} (an RTX) can have
5617 different meanings depending on the machine mode of the memory
5618 reference it is used for or if the address is valid for some modes
5621 Autoincrement and autodecrement addresses typically have mode-dependent
5622 effects because the amount of the increment or decrement is the size
5623 of the operand being addressed. Some machines have other mode-dependent
5624 addresses. Many RISC machines have no mode-dependent addresses.
5626 You may assume that @var{addr} is a valid address for the machine.
5628 These are obsolete macros, replaced by the
5629 @code{TARGET_MODE_DEPENDENT_ADDRESS_P} target hook.
5632 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (enum machine_mode @var{mode}, rtx @var{x})
5633 This hook returns true if @var{x} is a legitimate constant for a
5634 @var{mode}-mode immediate operand on the target machine. You can assume that
5635 @var{x} satisfies @code{CONSTANT_P}, so you need not check this.
5637 The default definition returns true.
5640 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
5641 This hook is used to undo the possibly obfuscating effects of the
5642 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
5643 macros. Some backend implementations of these macros wrap symbol
5644 references inside an @code{UNSPEC} rtx to represent PIC or similar
5645 addressing modes. This target hook allows GCC's optimizers to understand
5646 the semantics of these opaque @code{UNSPEC}s by converting them back
5647 into their original form.
5650 @deftypefn {Target Hook} bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx @var{x})
5651 This hook should return true if @var{x} should not be emitted into
5655 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (enum machine_mode @var{mode}, rtx @var{x})
5656 This hook should return true if @var{x} is of a form that cannot (or
5657 should not) be spilled to the constant pool. @var{mode} is the mode
5660 The default version of this hook returns false.
5662 The primary reason to define this hook is to prevent reload from
5663 deciding that a non-legitimate constant would be better reloaded
5664 from the constant pool instead of spilling and reloading a register
5665 holding the constant. This restriction is often true of addresses
5666 of TLS symbols for various targets.
5669 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (enum machine_mode @var{mode}, const_rtx @var{x})
5670 This hook should return true if pool entries for constant @var{x} can
5671 be placed in an @code{object_block} structure. @var{mode} is the mode
5674 The default version returns false for all constants.
5677 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (unsigned @var{fn}, bool @var{md_fn}, bool @var{sqrt})
5678 This hook should return the DECL of a function that implements reciprocal of
5679 the builtin function with builtin function code @var{fn}, or
5680 @code{NULL_TREE} if such a function is not available. @var{md_fn} is true
5681 when @var{fn} is a code of a machine-dependent builtin function. When
5682 @var{sqrt} is true, additional optimizations that apply only to the reciprocal
5683 of a square root function are performed, and only reciprocals of @code{sqrt}
5687 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
5688 This hook should return the DECL of a function @var{f} that given an
5689 address @var{addr} as an argument returns a mask @var{m} that can be
5690 used to extract from two vectors the relevant data that resides in
5691 @var{addr} in case @var{addr} is not properly aligned.
5693 The autovectorizer, when vectorizing a load operation from an address
5694 @var{addr} that may be unaligned, will generate two vector loads from
5695 the two aligned addresses around @var{addr}. It then generates a
5696 @code{REALIGN_LOAD} operation to extract the relevant data from the
5697 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
5698 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
5699 the third argument, @var{OFF}, defines how the data will be extracted
5700 from these two vectors: if @var{OFF} is 0, then the returned vector is
5701 @var{v2}; otherwise, the returned vector is composed from the last
5702 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
5703 @var{OFF} elements of @var{v2}.
5705 If this hook is defined, the autovectorizer will generate a call
5706 to @var{f} (using the DECL tree that this hook returns) and will
5707 use the return value of @var{f} as the argument @var{OFF} to
5708 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
5709 should comply with the semantics expected by @code{REALIGN_LOAD}
5711 If this hook is not defined, then @var{addr} will be used as
5712 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
5713 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
5716 @deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign})
5717 Returns cost of different scalar or vector statements for vectorization cost model.
5718 For vector memory operations the cost may depend on type (@var{vectype}) and
5719 misalignment value (@var{misalign}).
5722 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed})
5723 Return true if vector alignment is reachable (by peeling N iterations) for the given type.
5726 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST_OK (enum @var{machine_mode}, const unsigned char *@var{sel})
5727 Return true if a vector created for @code{vec_perm_const} is valid.
5730 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (unsigned @var{code}, tree @var{dest_type}, tree @var{src_type})
5731 This hook should return the DECL of a function that implements conversion of the
5732 input vector of type @var{src_type} to type @var{dest_type}.
5733 The value of @var{code} is one of the enumerators in @code{enum tree_code} and
5734 specifies how the conversion is to be applied
5735 (truncation, rounding, etc.).
5737 If this hook is defined, the autovectorizer will use the
5738 @code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing
5739 conversion. Otherwise, it will return @code{NULL_TREE}.
5742 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in})
5743 This hook should return the decl of a function that implements the
5744 vectorized variant of the builtin function with builtin function code
5745 @var{code} or @code{NULL_TREE} if such a function is not available.
5746 The value of @var{fndecl} is the builtin function declaration. The
5747 return type of the vectorized function shall be of vector type
5748 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
5751 @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})
5752 This hook should return true if the target supports misaligned vector
5753 store/load of a specific factor denoted in the @var{misalignment}
5754 parameter. The vector store/load should be of machine mode @var{mode} and
5755 the elements in the vectors should be of type @var{type}. @var{is_packed}
5756 parameter is true if the memory access is defined in a packed struct.
5759 @deftypefn {Target Hook} {enum machine_mode} TARGET_VECTORIZE_PREFERRED_SIMD_MODE (enum machine_mode @var{mode})
5760 This hook should return the preferred mode for vectorizing scalar
5761 mode @var{mode}. The default is
5762 equal to @code{word_mode}, because the vectorizer can do some
5763 transformations even in absence of specialized @acronym{SIMD} hardware.
5766 @deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES (void)
5767 This hook should return a mask of sizes that should be iterated over
5768 after trying to autovectorize using the vector size derived from the
5769 mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}.
5770 The default is zero which means to not iterate over other vector sizes.
5773 @deftypefn {Target Hook} {void *} TARGET_VECTORIZE_INIT_COST (struct loop *@var{loop_info})
5774 This hook should initialize target-specific data structures in preparation for modeling the costs of vectorizing a loop or basic block. The default allocates an unsigned integer for accumulating a single cost. If @var{loop_info} is non-NULL, it identifies the loop being vectorized; otherwise a single block is being vectorized.
5777 @deftypefn {Target Hook} unsigned TARGET_VECTORIZE_ADD_STMT_COST (void *@var{data}, int @var{count}, enum vect_cost_for_stmt @var{kind}, struct _stmt_vec_info *@var{stmt_info}, int @var{misalign})
5778 This hook should update the target-specific @var{data} in response to adding @var{count} copies of the given @var{kind} of statement to the body of a loop or basic block. The default adds the builtin vectorizer cost for the copies of the statement to the accumulator, and returns the amount added. The return value should be viewed as a tentative cost that may later be overridden.
5781 @deftypefn {Target Hook} unsigned TARGET_VECTORIZE_FINISH_COST (void *@var{data})
5782 This hook should complete calculations of the cost of vectorizing a loop or basic block based on @var{data}, and return that cost as an unsigned integer. The default returns the value of the accumulator.
5785 @deftypefn {Target Hook} void TARGET_VECTORIZE_DESTROY_COST_DATA (void *@var{data})
5786 This hook should release @var{data} and any related data structures allocated by TARGET_VECTORIZE_INIT_COST. The default releases the accumulator.
5789 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_LOAD (tree)
5790 This hook should return the built-in decl needed to load a vector of the given type within a transaction.
5793 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_TM_STORE (tree)
5794 This hook should return the built-in decl needed to store a vector of the given type within a transaction.
5797 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale})
5798 Target builtin that implements vector gather operation. @var{mem_vectype}
5799 is the vector type of the load and @var{index_type} is scalar type of
5800 the index, scaled by @var{scale}.
5801 The default is @code{NULL_TREE} which means to not vectorize gather
5805 @node Anchored Addresses
5806 @section Anchored Addresses
5807 @cindex anchored addresses
5808 @cindex @option{-fsection-anchors}
5810 GCC usually addresses every static object as a separate entity.
5811 For example, if we have:
5815 int foo (void) @{ return a + b + c; @}
5818 the code for @code{foo} will usually calculate three separate symbolic
5819 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
5820 it would be better to calculate just one symbolic address and access
5821 the three variables relative to it. The equivalent pseudocode would
5827 register int *xr = &x;
5828 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5832 (which isn't valid C). We refer to shared addresses like @code{x} as
5833 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
5835 The hooks below describe the target properties that GCC needs to know
5836 in order to make effective use of section anchors. It won't use
5837 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
5838 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
5840 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
5841 The minimum offset that should be applied to a section anchor.
5842 On most targets, it should be the smallest offset that can be
5843 applied to a base register while still giving a legitimate address
5844 for every mode. The default value is 0.
5847 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
5848 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
5849 offset that should be applied to section anchors. The default
5853 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
5854 Write the assembly code to define section anchor @var{x}, which is a
5855 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
5856 The hook is called with the assembly output position set to the beginning
5857 of @code{SYMBOL_REF_BLOCK (@var{x})}.
5859 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
5860 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
5861 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
5862 is @code{NULL}, which disables the use of section anchors altogether.
5865 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x})
5866 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
5867 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
5868 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
5870 The default version is correct for most targets, but you might need to
5871 intercept this hook to handle things like target-specific attributes
5872 or target-specific sections.
5875 @node Condition Code
5876 @section Condition Code Status
5877 @cindex condition code status
5879 The macros in this section can be split in two families, according to the
5880 two ways of representing condition codes in GCC.
5882 The first representation is the so called @code{(cc0)} representation
5883 (@pxref{Jump Patterns}), where all instructions can have an implicit
5884 clobber of the condition codes. The second is the condition code
5885 register representation, which provides better schedulability for
5886 architectures that do have a condition code register, but on which
5887 most instructions do not affect it. The latter category includes
5890 The implicit clobbering poses a strong restriction on the placement of
5891 the definition and use of the condition code, which need to be in adjacent
5892 insns for machines using @code{(cc0)}. This can prevent important
5893 optimizations on some machines. For example, on the IBM RS/6000, there
5894 is a delay for taken branches unless the condition code register is set
5895 three instructions earlier than the conditional branch. The instruction
5896 scheduler cannot perform this optimization if it is not permitted to
5897 separate the definition and use of the condition code register.
5899 For this reason, it is possible and suggested to use a register to
5900 represent the condition code for new ports. If there is a specific
5901 condition code register in the machine, use a hard register. If the
5902 condition code or comparison result can be placed in any general register,
5903 or if there are multiple condition registers, use a pseudo register.
5904 Registers used to store the condition code value will usually have a mode
5905 that is in class @code{MODE_CC}.
5907 Alternatively, you can use @code{BImode} if the comparison operator is
5908 specified already in the compare instruction. In this case, you are not
5909 interested in most macros in this section.
5912 * CC0 Condition Codes:: Old style representation of condition codes.
5913 * MODE_CC Condition Codes:: Modern representation of condition codes.
5914 * Cond Exec Macros:: Macros to control conditional execution.
5917 @node CC0 Condition Codes
5918 @subsection Representation of condition codes using @code{(cc0)}
5922 The file @file{conditions.h} defines a variable @code{cc_status} to
5923 describe how the condition code was computed (in case the interpretation of
5924 the condition code depends on the instruction that it was set by). This
5925 variable contains the RTL expressions on which the condition code is
5926 currently based, and several standard flags.
5928 Sometimes additional machine-specific flags must be defined in the machine
5929 description header file. It can also add additional machine-specific
5930 information by defining @code{CC_STATUS_MDEP}.
5932 @defmac CC_STATUS_MDEP
5933 C code for a data type which is used for declaring the @code{mdep}
5934 component of @code{cc_status}. It defaults to @code{int}.
5936 This macro is not used on machines that do not use @code{cc0}.
5939 @defmac CC_STATUS_MDEP_INIT
5940 A C expression to initialize the @code{mdep} field to ``empty''.
5941 The default definition does nothing, since most machines don't use
5942 the field anyway. If you want to use the field, you should probably
5943 define this macro to initialize it.
5945 This macro is not used on machines that do not use @code{cc0}.
5948 @defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn})
5949 A C compound statement to set the components of @code{cc_status}
5950 appropriately for an insn @var{insn} whose body is @var{exp}. It is
5951 this macro's responsibility to recognize insns that set the condition
5952 code as a byproduct of other activity as well as those that explicitly
5955 This macro is not used on machines that do not use @code{cc0}.
5957 If there are insns that do not set the condition code but do alter
5958 other machine registers, this macro must check to see whether they
5959 invalidate the expressions that the condition code is recorded as
5960 reflecting. For example, on the 68000, insns that store in address
5961 registers do not set the condition code, which means that usually
5962 @code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such
5963 insns. But suppose that the previous insn set the condition code
5964 based on location @samp{a4@@(102)} and the current insn stores a new
5965 value in @samp{a4}. Although the condition code is not changed by
5966 this, it will no longer be true that it reflects the contents of
5967 @samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter
5968 @code{cc_status} in this case to say that nothing is known about the
5969 condition code value.
5971 The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal
5972 with the results of peephole optimization: insns whose patterns are
5973 @code{parallel} RTXs containing various @code{reg}, @code{mem} or
5974 constants which are just the operands. The RTL structure of these
5975 insns is not sufficient to indicate what the insns actually do. What
5976 @code{NOTICE_UPDATE_CC} should do when it sees one is just to run
5977 @code{CC_STATUS_INIT}.
5979 A possible definition of @code{NOTICE_UPDATE_CC} is to call a function
5980 that looks at an attribute (@pxref{Insn Attributes}) named, for example,
5981 @samp{cc}. This avoids having detailed information about patterns in
5982 two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}.
5985 @node MODE_CC Condition Codes
5986 @subsection Representation of condition codes using registers
5990 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
5991 On many machines, the condition code may be produced by other instructions
5992 than compares, for example the branch can use directly the condition
5993 code set by a subtract instruction. However, on some machines
5994 when the condition code is set this way some bits (such as the overflow
5995 bit) are not set in the same way as a test instruction, so that a different
5996 branch instruction must be used for some conditional branches. When
5997 this happens, use the machine mode of the condition code register to
5998 record different formats of the condition code register. Modes can
5999 also be used to record which compare instruction (e.g. a signed or an
6000 unsigned comparison) produced the condition codes.
6002 If other modes than @code{CCmode} are required, add them to
6003 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
6004 a mode given an operand of a compare. This is needed because the modes
6005 have to be chosen not only during RTL generation but also, for example,
6006 by instruction combination. The result of @code{SELECT_CC_MODE} should
6007 be consistent with the mode used in the patterns; for example to support
6008 the case of the add on the SPARC discussed above, we have the pattern
6012 [(set (reg:CC_NOOV 0)
6014 (plus:SI (match_operand:SI 0 "register_operand" "%r")
6015 (match_operand:SI 1 "arith_operand" "rI"))
6022 together with a @code{SELECT_CC_MODE} that returns @code{CC_NOOVmode}
6023 for comparisons whose argument is a @code{plus}:
6026 #define SELECT_CC_MODE(OP,X,Y) \
6027 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
6028 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \
6029 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
6030 || GET_CODE (X) == NEG) \
6031 ? CC_NOOVmode : CCmode))
6034 Another reason to use modes is to retain information on which operands
6035 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6038 You should define this macro if and only if you define extra CC modes
6039 in @file{@var{machine}-modes.def}.
6042 @defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1})
6043 On some machines not all possible comparisons are defined, but you can
6044 convert an invalid comparison into a valid one. For example, the Alpha
6045 does not have a @code{GT} comparison, but you can use an @code{LT}
6046 comparison instead and swap the order of the operands.
6048 On such machines, define this macro to be a C statement to do any
6049 required conversions. @var{code} is the initial comparison code
6050 and @var{op0} and @var{op1} are the left and right operands of the
6051 comparison, respectively. You should modify @var{code}, @var{op0}, and
6052 @var{op1} as required.
6054 GCC will not assume that the comparison resulting from this macro is
6055 valid but will see if the resulting insn matches a pattern in the
6058 You need not define this macro if it would never change the comparison
6062 @defmac REVERSIBLE_CC_MODE (@var{mode})
6063 A C expression whose value is one if it is always safe to reverse a
6064 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6065 can ever return @var{mode} for a floating-point inequality comparison,
6066 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6068 You need not define this macro if it would always returns zero or if the
6069 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6070 For example, here is the definition used on the SPARC, where floating-point
6071 inequality comparisons are always given @code{CCFPEmode}:
6074 #define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode)
6078 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6079 A C expression whose value is reversed condition code of the @var{code} for
6080 comparison done in CC_MODE @var{mode}. The macro is used only in case
6081 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6082 machine has some non-standard way how to reverse certain conditionals. For
6083 instance in case all floating point conditions are non-trapping, compiler may
6084 freely convert unordered compares to ordered one. Then definition may look
6088 #define REVERSE_CONDITION(CODE, MODE) \
6089 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6090 : reverse_condition_maybe_unordered (CODE))
6094 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2})
6095 On targets which do not use @code{(cc0)}, and which use a hard
6096 register rather than a pseudo-register to hold condition codes, the
6097 regular CSE passes are often not able to identify cases in which the
6098 hard register is set to a common value. Use this hook to enable a
6099 small pass which optimizes such cases. This hook should return true
6100 to enable this pass, and it should set the integers to which its
6101 arguments point to the hard register numbers used for condition codes.
6102 When there is only one such register, as is true on most systems, the
6103 integer pointed to by @var{p2} should be set to
6104 @code{INVALID_REGNUM}.
6106 The default version of this hook returns false.
6109 @deftypefn {Target Hook} {enum machine_mode} TARGET_CC_MODES_COMPATIBLE (enum machine_mode @var{m1}, enum machine_mode @var{m2})
6110 On targets which use multiple condition code modes in class
6111 @code{MODE_CC}, it is sometimes the case that a comparison can be
6112 validly done in more than one mode. On such a system, define this
6113 target hook to take two mode arguments and to return a mode in which
6114 both comparisons may be validly done. If there is no such mode,
6115 return @code{VOIDmode}.
6117 The default version of this hook checks whether the modes are the
6118 same. If they are, it returns that mode. If they are different, it
6119 returns @code{VOIDmode}.
6122 @node Cond Exec Macros
6123 @subsection Macros to control conditional execution
6124 @findex conditional execution
6127 There is one macro that may need to be defined for targets
6128 supporting conditional execution, independent of how they
6129 represent conditional branches.
6132 @section Describing Relative Costs of Operations
6133 @cindex costs of instructions
6134 @cindex relative costs
6135 @cindex speed of instructions
6137 These macros let you describe the relative speed of various operations
6138 on the target machine.
6140 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6141 A C expression for the cost of moving data of mode @var{mode} from a
6142 register in class @var{from} to one in class @var{to}. The classes are
6143 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6144 value of 2 is the default; other values are interpreted relative to
6147 It is not required that the cost always equal 2 when @var{from} is the
6148 same as @var{to}; on some machines it is expensive to move between
6149 registers if they are not general registers.
6151 If reload sees an insn consisting of a single @code{set} between two
6152 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6153 classes returns a value of 2, reload does not check to ensure that the
6154 constraints of the insn are met. Setting a cost of other than 2 will
6155 allow reload to verify that the constraints are met. You should do this
6156 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6158 These macros are obsolete, new ports should use the target hook
6159 @code{TARGET_REGISTER_MOVE_COST} instead.
6162 @deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (enum machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to})
6163 This target hook should return the cost of moving data of mode @var{mode}
6164 from a register in class @var{from} to one in class @var{to}. The classes
6165 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6166 A value of 2 is the default; other values are interpreted relative to
6169 It is not required that the cost always equal 2 when @var{from} is the
6170 same as @var{to}; on some machines it is expensive to move between
6171 registers if they are not general registers.
6173 If reload sees an insn consisting of a single @code{set} between two
6174 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6175 classes returns a value of 2, reload does not check to ensure that the
6176 constraints of the insn are met. Setting a cost of other than 2 will
6177 allow reload to verify that the constraints are met. You should do this
6178 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6180 The default version of this function returns 2.
6183 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6184 A C expression for the cost of moving data of mode @var{mode} between a
6185 register of class @var{class} and memory; @var{in} is zero if the value
6186 is to be written to memory, nonzero if it is to be read in. This cost
6187 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6188 registers and memory is more expensive than between two registers, you
6189 should define this macro to express the relative cost.
6191 If you do not define this macro, GCC uses a default cost of 4 plus
6192 the cost of copying via a secondary reload register, if one is
6193 needed. If your machine requires a secondary reload register to copy
6194 between memory and a register of @var{class} but the reload mechanism is
6195 more complex than copying via an intermediate, define this macro to
6196 reflect the actual cost of the move.
6198 GCC defines the function @code{memory_move_secondary_cost} if
6199 secondary reloads are needed. It computes the costs due to copying via
6200 a secondary register. If your machine copies from memory using a
6201 secondary register in the conventional way but the default base value of
6202 4 is not correct for your machine, define this macro to add some other
6203 value to the result of that function. The arguments to that function
6204 are the same as to this macro.
6206 These macros are obsolete, new ports should use the target hook
6207 @code{TARGET_MEMORY_MOVE_COST} instead.
6210 @deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (enum machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in})
6211 This target hook should return the cost of moving data of mode @var{mode}
6212 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6213 if the value is to be written to memory, @code{true} if it is to be read in.
6214 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6215 If moving between registers and memory is more expensive than between two
6216 registers, you should add this target hook to express the relative cost.
6218 If you do not add this target hook, GCC uses a default cost of 4 plus
6219 the cost of copying via a secondary reload register, if one is
6220 needed. If your machine requires a secondary reload register to copy
6221 between memory and a register of @var{rclass} but the reload mechanism is
6222 more complex than copying via an intermediate, use this target hook to
6223 reflect the actual cost of the move.
6225 GCC defines the function @code{memory_move_secondary_cost} if
6226 secondary reloads are needed. It computes the costs due to copying via
6227 a secondary register. If your machine copies from memory using a
6228 secondary register in the conventional way but the default base value of
6229 4 is not correct for your machine, use this target hook to add some other
6230 value to the result of that function. The arguments to that function
6231 are the same as to this target hook.
6234 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6235 A C expression for the cost of a branch instruction. A value of 1 is
6236 the default; other values are interpreted relative to that. Parameter
6237 @var{speed_p} is true when the branch in question should be optimized
6238 for speed. When it is false, @code{BRANCH_COST} should return a value
6239 optimal for code size rather than performance. @var{predictable_p} is
6240 true for well-predicted branches. On many architectures the
6241 @code{BRANCH_COST} can be reduced then.
6244 Here are additional macros which do not specify precise relative costs,
6245 but only that certain actions are more expensive than GCC would
6248 @defmac SLOW_BYTE_ACCESS
6249 Define this macro as a C expression which is nonzero if accessing less
6250 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6251 faster than accessing a word of memory, i.e., if such access
6252 require more than one instruction or if there is no difference in cost
6253 between byte and (aligned) word loads.
6255 When this macro is not defined, the compiler will access a field by
6256 finding the smallest containing object; when it is defined, a fullword
6257 load will be used if alignment permits. Unless bytes accesses are
6258 faster than word accesses, using word accesses is preferable since it
6259 may eliminate subsequent memory access if subsequent accesses occur to
6260 other fields in the same word of the structure, but to different bytes.
6263 @defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment})
6264 Define this macro to be the value 1 if memory accesses described by the
6265 @var{mode} and @var{alignment} parameters have a cost many times greater
6266 than aligned accesses, for example if they are emulated in a trap
6269 When this macro is nonzero, the compiler will act as if
6270 @code{STRICT_ALIGNMENT} were nonzero when generating code for block
6271 moves. This can cause significantly more instructions to be produced.
6272 Therefore, do not set this macro nonzero if unaligned accesses only add a
6273 cycle or two to the time for a memory access.
6275 If the value of this macro is always zero, it need not be defined. If
6276 this macro is defined, it should produce a nonzero value when
6277 @code{STRICT_ALIGNMENT} is nonzero.
6280 @defmac MOVE_RATIO (@var{speed})
6281 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6282 which a sequence of insns should be generated instead of a
6283 string move insn or a library call. Increasing the value will always
6284 make code faster, but eventually incurs high cost in increased code size.
6286 Note that on machines where the corresponding move insn is a
6287 @code{define_expand} that emits a sequence of insns, this macro counts
6288 the number of such sequences.
6290 The parameter @var{speed} is true if the code is currently being
6291 optimized for speed rather than size.
6293 If you don't define this, a reasonable default is used.
6296 @defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment})
6297 A C expression used to determine whether @code{move_by_pieces} will be used to
6298 copy a chunk of memory, or whether some other block move mechanism
6299 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6300 than @code{MOVE_RATIO}.
6303 @defmac MOVE_MAX_PIECES
6304 A C expression used by @code{move_by_pieces} to determine the largest unit
6305 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6308 @defmac CLEAR_RATIO (@var{speed})
6309 The threshold of number of scalar move insns, @emph{below} which a sequence
6310 of insns should be generated to clear memory instead of a string clear insn
6311 or a library call. Increasing the value will always make code faster, but
6312 eventually incurs high cost in increased code size.
6314 The parameter @var{speed} is true if the code is currently being
6315 optimized for speed rather than size.
6317 If you don't define this, a reasonable default is used.
6320 @defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment})
6321 A C expression used to determine whether @code{clear_by_pieces} will be used
6322 to clear a chunk of memory, or whether some other block clear mechanism
6323 will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6324 than @code{CLEAR_RATIO}.
6327 @defmac SET_RATIO (@var{speed})
6328 The threshold of number of scalar move insns, @emph{below} which a sequence
6329 of insns should be generated to set memory to a constant value, instead of
6330 a block set insn or a library call.
6331 Increasing the value will always make code faster, but
6332 eventually incurs high cost in increased code size.
6334 The parameter @var{speed} is true if the code is currently being
6335 optimized for speed rather than size.
6337 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6340 @defmac SET_BY_PIECES_P (@var{size}, @var{alignment})
6341 A C expression used to determine whether @code{store_by_pieces} will be
6342 used to set a chunk of memory to a constant value, or whether some
6343 other mechanism will be used. Used by @code{__builtin_memset} when
6344 storing values other than constant zero.
6345 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6346 than @code{SET_RATIO}.
6349 @defmac STORE_BY_PIECES_P (@var{size}, @var{alignment})
6350 A C expression used to determine whether @code{store_by_pieces} will be
6351 used to set a chunk of memory to a constant string value, or whether some
6352 other mechanism will be used. Used by @code{__builtin_strcpy} when
6353 called with a constant source string.
6354 Defaults to 1 if @code{move_by_pieces_ninsns} returns less
6355 than @code{MOVE_RATIO}.
6358 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6359 A C expression used to determine whether a load postincrement is a good
6360 thing to use for a given mode. Defaults to the value of
6361 @code{HAVE_POST_INCREMENT}.
6364 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6365 A C expression used to determine whether a load postdecrement is a good
6366 thing to use for a given mode. Defaults to the value of
6367 @code{HAVE_POST_DECREMENT}.
6370 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6371 A C expression used to determine whether a load preincrement is a good
6372 thing to use for a given mode. Defaults to the value of
6373 @code{HAVE_PRE_INCREMENT}.
6376 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6377 A C expression used to determine whether a load predecrement is a good
6378 thing to use for a given mode. Defaults to the value of
6379 @code{HAVE_PRE_DECREMENT}.
6382 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6383 A C expression used to determine whether a store postincrement is a good
6384 thing to use for a given mode. Defaults to the value of
6385 @code{HAVE_POST_INCREMENT}.
6388 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6389 A C expression used to determine whether a store postdecrement is a good
6390 thing to use for a given mode. Defaults to the value of
6391 @code{HAVE_POST_DECREMENT}.
6394 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6395 This macro is used to determine whether a store preincrement is a good
6396 thing to use for a given mode. Defaults to the value of
6397 @code{HAVE_PRE_INCREMENT}.
6400 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6401 This macro is used to determine whether a store predecrement is a good
6402 thing to use for a given mode. Defaults to the value of
6403 @code{HAVE_PRE_DECREMENT}.
6406 @defmac NO_FUNCTION_CSE
6407 Define this macro if it is as good or better to call a constant
6408 function address than to call an address kept in a register.
6411 @defmac LOGICAL_OP_NON_SHORT_CIRCUIT
6412 Define this macro if a non-short-circuit operation produced by
6413 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6414 @code{BRANCH_COST} is greater than or equal to the value 2.
6417 @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})
6418 This target hook describes the relative costs of RTL expressions.
6420 The cost may depend on the precise form of the expression, which is
6421 available for examination in @var{x}, and the fact that @var{x} appears
6422 as operand @var{opno} of an expression with rtx code @var{outer_code}.
6423 That is, the hook can assume that there is some rtx @var{y} such
6424 that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that
6425 either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or
6426 (b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}.
6428 @var{code} is @var{x}'s expression code---redundant, since it can be
6429 obtained with @code{GET_CODE (@var{x})}.
6431 In implementing this hook, you can use the construct
6432 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
6435 On entry to the hook, @code{*@var{total}} contains a default estimate
6436 for the cost of the expression. The hook should modify this value as
6437 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
6438 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
6439 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
6441 When optimizing for code size, i.e.@: when @code{speed} is
6442 false, this target hook should be used to estimate the relative
6443 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
6445 The hook returns true when all subexpressions of @var{x} have been
6446 processed, and false when @code{rtx_cost} should recurse.
6449 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, bool @var{speed})
6450 This hook computes the cost of an addressing mode that contains
6451 @var{address}. If not defined, the cost is computed from
6452 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
6454 For most CISC machines, the default cost is a good approximation of the
6455 true cost of the addressing mode. However, on RISC machines, all
6456 instructions normally have the same length and execution time. Hence
6457 all addresses will have equal costs.
6459 In cases where more than one form of an address is known, the form with
6460 the lowest cost will be used. If multiple forms have the same, lowest,
6461 cost, the one that is the most complex will be used.
6463 For example, suppose an address that is equal to the sum of a register
6464 and a constant is used twice in the same basic block. When this macro
6465 is not defined, the address will be computed in a register and memory
6466 references will be indirect through that register. On machines where
6467 the cost of the addressing mode containing the sum is no higher than
6468 that of a simple indirect reference, this will produce an additional
6469 instruction and possibly require an additional register. Proper
6470 specification of this macro eliminates this overhead for such machines.
6472 This hook is never called with an invalid address.
6474 On machines where an address involving more than one register is as
6475 cheap as an address computation involving only one register, defining
6476 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
6477 be live over a region of code where only one would have been if
6478 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
6479 should be considered in the definition of this macro. Equivalent costs
6480 should probably only be given to addresses with different numbers of
6481 registers on machines with lots of registers.
6485 @section Adjusting the Instruction Scheduler
6487 The instruction scheduler may need a fair amount of machine-specific
6488 adjustment in order to produce good code. GCC provides several target
6489 hooks for this purpose. It is usually enough to define just a few of
6490 them: try the first ones in this list first.
6492 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
6493 This hook returns the maximum number of instructions that can ever
6494 issue at the same time on the target machine. The default is one.
6495 Although the insn scheduler can define itself the possibility of issue
6496 an insn on the same cycle, the value can serve as an additional
6497 constraint to issue insns on the same simulated processor cycle (see
6498 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
6499 This value must be constant over the entire compilation. If you need
6500 it to vary depending on what the instructions are, you must use
6501 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
6504 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more})
6505 This hook is executed by the scheduler after it has scheduled an insn
6506 from the ready list. It should return the number of insns which can
6507 still be issued in the current cycle. The default is
6508 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
6509 @code{USE}, which normally are not counted against the issue rate.
6510 You should define this hook if some insns take more machine resources
6511 than others, so that fewer insns can follow them in the same cycle.
6512 @var{file} is either a null pointer, or a stdio stream to write any
6513 debug output to. @var{verbose} is the verbose level provided by
6514 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
6518 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost})
6519 This function corrects the value of @var{cost} based on the
6520 relationship between @var{insn} and @var{dep_insn} through the
6521 dependence @var{link}. It should return the new value. The default
6522 is to make no adjustment to @var{cost}. This can be used for example
6523 to specify to the scheduler using the traditional pipeline description
6524 that an output- or anti-dependence does not incur the same cost as a
6525 data-dependence. If the scheduler using the automaton based pipeline
6526 description, the cost of anti-dependence is zero and the cost of
6527 output-dependence is maximum of one and the difference of latency
6528 times of the first and the second insns. If these values are not
6529 acceptable, you could use the hook to modify them too. See also
6530 @pxref{Processor pipeline description}.
6533 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority})
6534 This hook adjusts the integer scheduling priority @var{priority} of
6535 @var{insn}. It should return the new priority. Increase the priority to
6536 execute @var{insn} earlier, reduce the priority to execute @var{insn}
6537 later. Do not define this hook if you do not need to adjust the
6538 scheduling priorities of insns.
6541 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6542 This hook is executed by the scheduler after it has scheduled the ready
6543 list, to allow the machine description to reorder it (for example to
6544 combine two small instructions together on @samp{VLIW} machines).
6545 @var{file} is either a null pointer, or a stdio stream to write any
6546 debug output to. @var{verbose} is the verbose level provided by
6547 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
6548 list of instructions that are ready to be scheduled. @var{n_readyp} is
6549 a pointer to the number of elements in the ready list. The scheduler
6550 reads the ready list in reverse order, starting with
6551 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
6552 is the timer tick of the scheduler. You may modify the ready list and
6553 the number of ready insns. The return value is the number of insns that
6554 can issue this cycle; normally this is just @code{issue_rate}. See also
6555 @samp{TARGET_SCHED_REORDER2}.
6558 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock})
6559 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
6560 function is called whenever the scheduler starts a new cycle. This one
6561 is called once per iteration over a cycle, immediately after
6562 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
6563 return the number of insns to be scheduled in the same cycle. Defining
6564 this hook can be useful if there are frequent situations where
6565 scheduling one insn causes other insns to become ready in the same
6566 cycle. These other insns can then be taken into account properly.
6569 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail})
6570 This hook is called after evaluation forward dependencies of insns in
6571 chain given by two parameter values (@var{head} and @var{tail}
6572 correspondingly) but before insns scheduling of the insn chain. For
6573 example, it can be used for better insn classification if it requires
6574 analysis of dependencies. This hook can use backward and forward
6575 dependencies of the insn scheduler because they are already
6579 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
6580 This hook is executed by the scheduler at the beginning of each block of
6581 instructions that are to be scheduled. @var{file} is either a null
6582 pointer, or a stdio stream to write any debug output to. @var{verbose}
6583 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6584 @var{max_ready} is the maximum number of insns in the current scheduling
6585 region that can be live at the same time. This can be used to allocate
6586 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
6589 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
6590 This hook is executed by the scheduler at the end of each block of
6591 instructions that are to be scheduled. It can be used to perform
6592 cleanup of any actions done by the other scheduling hooks. @var{file}
6593 is either a null pointer, or a stdio stream to write any debug output
6594 to. @var{verbose} is the verbose level provided by
6595 @option{-fsched-verbose-@var{n}}.
6598 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
6599 This hook is executed by the scheduler after function level initializations.
6600 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6601 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6602 @var{old_max_uid} is the maximum insn uid when scheduling begins.
6605 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
6606 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
6607 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
6608 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
6611 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
6612 The hook returns an RTL insn. The automaton state used in the
6613 pipeline hazard recognizer is changed as if the insn were scheduled
6614 when the new simulated processor cycle starts. Usage of the hook may
6615 simplify the automaton pipeline description for some @acronym{VLIW}
6616 processors. If the hook is defined, it is used only for the automaton
6617 based pipeline description. The default is not to change the state
6618 when the new simulated processor cycle starts.
6621 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
6622 The hook can be used to initialize data used by the previous hook.
6625 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
6626 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6627 to changed the state as if the insn were scheduled when the new
6628 simulated processor cycle finishes.
6631 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
6632 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
6633 used to initialize data used by the previous hook.
6636 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)
6637 The hook to notify target that the current simulated cycle is about to finish.
6638 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
6639 to change the state in more complicated situations - e.g., when advancing
6640 state on a single insn is not enough.
6643 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)
6644 The hook to notify target that new simulated cycle has just started.
6645 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
6646 to change the state in more complicated situations - e.g., when advancing
6647 state on a single insn is not enough.
6650 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
6651 This hook controls better choosing an insn from the ready insn queue
6652 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
6653 chooses the first insn from the queue. If the hook returns a positive
6654 value, an additional scheduler code tries all permutations of
6655 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
6656 subsequent ready insns to choose an insn whose issue will result in
6657 maximal number of issued insns on the same cycle. For the
6658 @acronym{VLIW} processor, the code could actually solve the problem of
6659 packing simple insns into the @acronym{VLIW} insn. Of course, if the
6660 rules of @acronym{VLIW} packing are described in the automaton.
6662 This code also could be used for superscalar @acronym{RISC}
6663 processors. Let us consider a superscalar @acronym{RISC} processor
6664 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
6665 @var{B}, some insns can be executed only in pipelines @var{B} or
6666 @var{C}, and one insn can be executed in pipeline @var{B}. The
6667 processor may issue the 1st insn into @var{A} and the 2nd one into
6668 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
6669 until the next cycle. If the scheduler issues the 3rd insn the first,
6670 the processor could issue all 3 insns per cycle.
6672 Actually this code demonstrates advantages of the automaton based
6673 pipeline hazard recognizer. We try quickly and easy many insn
6674 schedules to choose the best one.
6676 The default is no multipass scheduling.
6679 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx @var{insn})
6681 This hook controls what insns from the ready insn queue will be
6682 considered for the multipass insn scheduling. If the hook returns
6683 zero for @var{insn}, the insn will be not chosen to
6686 The default is that any ready insns can be chosen to be issued.
6689 @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})
6690 This hook prepares the target backend for a new round of multipass
6694 @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})
6695 This hook is called when multipass scheduling evaluates instruction INSN.
6698 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, char *@var{ready_try}, int @var{n_ready})
6699 This is called when multipass scheduling backtracks from evaluation of
6703 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data})
6704 This hook notifies the target about the result of the concluded current
6705 round of multipass scheduling.
6708 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data})
6709 This hook initializes target-specific data used in multipass scheduling.
6712 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data})
6713 This hook finalizes target-specific data used in multipass scheduling.
6716 @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})
6717 This hook is called by the insn scheduler before issuing @var{insn}
6718 on cycle @var{clock}. If the hook returns nonzero,
6719 @var{insn} is not issued on this processor cycle. Instead,
6720 the processor cycle is advanced. If *@var{sort_p}
6721 is zero, the insn ready queue is not sorted on the new cycle
6722 start as usually. @var{dump} and @var{verbose} specify the file and
6723 verbosity level to use for debugging output.
6724 @var{last_clock} and @var{clock} are, respectively, the
6725 processor cycle on which the previous insn has been issued,
6726 and the current processor cycle.
6729 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance})
6730 This hook is used to define which dependences are considered costly by
6731 the target, so costly that it is not advisable to schedule the insns that
6732 are involved in the dependence too close to one another. The parameters
6733 to this hook are as follows: The first parameter @var{_dep} is the dependence
6734 being evaluated. The second parameter @var{cost} is the cost of the
6735 dependence as estimated by the scheduler, and the third
6736 parameter @var{distance} is the distance in cycles between the two insns.
6737 The hook returns @code{true} if considering the distance between the two
6738 insns the dependence between them is considered costly by the target,
6739 and @code{false} otherwise.
6741 Defining this hook can be useful in multiple-issue out-of-order machines,
6742 where (a) it's practically hopeless to predict the actual data/resource
6743 delays, however: (b) there's a better chance to predict the actual grouping
6744 that will be formed, and (c) correctly emulating the grouping can be very
6745 important. In such targets one may want to allow issuing dependent insns
6746 closer to one another---i.e., closer than the dependence distance; however,
6747 not in cases of ``costly dependences'', which this hooks allows to define.
6750 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
6751 This hook is called by the insn scheduler after emitting a new instruction to
6752 the instruction stream. The hook notifies a target backend to extend its
6753 per instruction data structures.
6756 @deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
6757 Return a pointer to a store large enough to hold target scheduling context.
6760 @deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p})
6761 Initialize store pointed to by @var{tc} to hold target scheduling context.
6762 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
6763 beginning of the block. Otherwise, copy the current context into @var{tc}.
6766 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc})
6767 Copy target scheduling context pointed to by @var{tc} to the current context.
6770 @deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc})
6771 Deallocate internal data in target scheduling context pointed to by @var{tc}.
6774 @deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc})
6775 Deallocate a store for target scheduling context pointed to by @var{tc}.
6778 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx @var{insn}, int @var{request}, rtx *@var{new_pat})
6779 This hook is called by the insn scheduler when @var{insn} has only
6780 speculative dependencies and therefore can be scheduled speculatively.
6781 The hook is used to check if the pattern of @var{insn} has a speculative
6782 version and, in case of successful check, to generate that speculative
6783 pattern. The hook should return 1, if the instruction has a speculative form,
6784 or @minus{}1, if it doesn't. @var{request} describes the type of requested
6785 speculation. If the return value equals 1 then @var{new_pat} is assigned
6786 the generated speculative pattern.
6789 @deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (int @var{dep_status})
6790 This hook is called by the insn scheduler during generation of recovery code
6791 for @var{insn}. It should return @code{true}, if the corresponding check
6792 instruction should branch to recovery code, or @code{false} otherwise.
6795 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx @var{insn}, rtx @var{label}, int @var{mutate_p})
6796 This hook is called by the insn scheduler to generate a pattern for recovery
6797 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
6798 speculative instruction for which the check should be generated.
6799 @var{label} is either a label of a basic block, where recovery code should
6800 be emitted, or a null pointer, when requested check doesn't branch to
6801 recovery code (a simple check). If @var{mutate_p} is nonzero, then
6802 a pattern for a branchy check corresponding to a simple check denoted by
6803 @var{insn} should be generated. In this case @var{label} can't be null.
6806 @deftypefn {Target Hook} bool TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD_SPEC (const_rtx @var{insn})
6807 This hook is used as a workaround for
6808 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD} not being
6809 called on the first instruction of the ready list. The hook is used to
6810 discard speculative instructions that stand first in the ready list from
6811 being scheduled on the current cycle. If the hook returns @code{false},
6812 @var{insn} will not be chosen to be issued.
6813 For non-speculative instructions,
6814 the hook should always return @code{true}. For example, in the ia64 backend
6815 the hook is used to cancel data speculative insns when the ALAT table
6819 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info})
6820 This hook is used by the insn scheduler to find out what features should be
6822 The structure *@var{spec_info} should be filled in by the target.
6823 The structure describes speculation types that can be used in the scheduler.
6826 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
6827 This hook is called by the swing modulo scheduler to calculate a
6828 resource-based lower bound which is based on the resources available in
6829 the machine and the resources required by each instruction. The target
6830 backend can use @var{g} to calculate such bound. A very simple lower
6831 bound will be used in case this hook is not implemented: the total number
6832 of instructions divided by the issue rate.
6835 @deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx @var{insn}, int @var{x})
6836 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
6837 is supported in hardware and the condition specified in the parameter is true.
6840 @deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx @var{insn}, int @var{x})
6841 This hook is called by Haifa Scheduler. It performs the operation specified
6842 in its second parameter.
6845 @deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE
6846 True if the processor has an exposed pipeline, which means that not just
6847 the order of instructions is important for correctness when scheduling, but
6848 also the latencies of operations.
6851 @deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, enum machine_mode @var{mode})
6852 This hook is called by tree reassociator to determine a level of
6853 parallelism required in output calculations chain.
6857 @section Dividing the Output into Sections (Texts, Data, @dots{})
6858 @c the above section title is WAY too long. maybe cut the part between
6859 @c the (...)? --mew 10feb93
6861 An object file is divided into sections containing different types of
6862 data. In the most common case, there are three sections: the @dfn{text
6863 section}, which holds instructions and read-only data; the @dfn{data
6864 section}, which holds initialized writable data; and the @dfn{bss
6865 section}, which holds uninitialized data. Some systems have other kinds
6868 @file{varasm.c} provides several well-known sections, such as
6869 @code{text_section}, @code{data_section} and @code{bss_section}.
6870 The normal way of controlling a @code{@var{foo}_section} variable
6871 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
6872 as described below. The macros are only read once, when @file{varasm.c}
6873 initializes itself, so their values must be run-time constants.
6874 They may however depend on command-line flags.
6876 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
6877 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
6878 to be string literals.
6880 Some assemblers require a different string to be written every time a
6881 section is selected. If your assembler falls into this category, you
6882 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
6883 @code{get_unnamed_section} to set up the sections.
6885 You must always create a @code{text_section}, either by defining
6886 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
6887 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
6888 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
6889 create a distinct @code{readonly_data_section}, the default is to
6890 reuse @code{text_section}.
6892 All the other @file{varasm.c} sections are optional, and are null
6893 if the target does not provide them.
6895 @defmac TEXT_SECTION_ASM_OP
6896 A C expression whose value is a string, including spacing, containing the
6897 assembler operation that should precede instructions and read-only data.
6898 Normally @code{"\t.text"} is right.
6901 @defmac HOT_TEXT_SECTION_NAME
6902 If defined, a C string constant for the name of the section containing most
6903 frequently executed functions of the program. If not defined, GCC will provide
6904 a default definition if the target supports named sections.
6907 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
6908 If defined, a C string constant for the name of the section containing unlikely
6909 executed functions in the program.
6912 @defmac DATA_SECTION_ASM_OP
6913 A C expression whose value is a string, including spacing, containing the
6914 assembler operation to identify the following data as writable initialized
6915 data. Normally @code{"\t.data"} is right.
6918 @defmac SDATA_SECTION_ASM_OP
6919 If defined, a C expression whose value is a string, including spacing,
6920 containing the assembler operation to identify the following data as
6921 initialized, writable small data.
6924 @defmac READONLY_DATA_SECTION_ASM_OP
6925 A C expression whose value is a string, including spacing, containing the
6926 assembler operation to identify the following data as read-only initialized
6930 @defmac BSS_SECTION_ASM_OP
6931 If defined, a C expression whose value is a string, including spacing,
6932 containing the assembler operation to identify the following data as
6933 uninitialized global data. If not defined, and
6934 @code{ASM_OUTPUT_ALIGNED_BSS} not defined,
6935 uninitialized global data will be output in the data section if
6936 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
6940 @defmac SBSS_SECTION_ASM_OP
6941 If defined, a C expression whose value is a string, including spacing,
6942 containing the assembler operation to identify the following data as
6943 uninitialized, writable small data.
6946 @defmac TLS_COMMON_ASM_OP
6947 If defined, a C expression whose value is a string containing the
6948 assembler operation to identify the following data as thread-local
6949 common data. The default is @code{".tls_common"}.
6952 @defmac TLS_SECTION_ASM_FLAG
6953 If defined, a C expression whose value is a character constant
6954 containing the flag used to mark a section as a TLS section. The
6955 default is @code{'T'}.
6958 @defmac INIT_SECTION_ASM_OP
6959 If defined, a C expression whose value is a string, including spacing,
6960 containing the assembler operation to identify the following data as
6961 initialization code. If not defined, GCC will assume such a section does
6962 not exist. This section has no corresponding @code{init_section}
6963 variable; it is used entirely in runtime code.
6966 @defmac FINI_SECTION_ASM_OP
6967 If defined, a C expression whose value is a string, including spacing,
6968 containing the assembler operation to identify the following data as
6969 finalization code. If not defined, GCC will assume such a section does
6970 not exist. This section has no corresponding @code{fini_section}
6971 variable; it is used entirely in runtime code.
6974 @defmac INIT_ARRAY_SECTION_ASM_OP
6975 If defined, a C expression whose value is a string, including spacing,
6976 containing the assembler operation to identify the following data as
6977 part of the @code{.init_array} (or equivalent) section. If not
6978 defined, GCC will assume such a section does not exist. Do not define
6979 both this macro and @code{INIT_SECTION_ASM_OP}.
6982 @defmac FINI_ARRAY_SECTION_ASM_OP
6983 If defined, a C expression whose value is a string, including spacing,
6984 containing the assembler operation to identify the following data as
6985 part of the @code{.fini_array} (or equivalent) section. If not
6986 defined, GCC will assume such a section does not exist. Do not define
6987 both this macro and @code{FINI_SECTION_ASM_OP}.
6990 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
6991 If defined, an ASM statement that switches to a different section
6992 via @var{section_op}, calls @var{function}, and switches back to
6993 the text section. This is used in @file{crtstuff.c} if
6994 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
6995 to initialization and finalization functions from the init and fini
6996 sections. By default, this macro uses a simple function call. Some
6997 ports need hand-crafted assembly code to avoid dependencies on
6998 registers initialized in the function prologue or to ensure that
6999 constant pools don't end up too far way in the text section.
7002 @defmac TARGET_LIBGCC_SDATA_SECTION
7003 If defined, a string which names the section into which small
7004 variables defined in crtstuff and libgcc should go. This is useful
7005 when the target has options for optimizing access to small data, and
7006 you want the crtstuff and libgcc routines to be conservative in what
7007 they expect of your application yet liberal in what your application
7008 expects. For example, for targets with a @code{.sdata} section (like
7009 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
7010 require small data support from your application, but use this macro
7011 to put small data into @code{.sdata} so that your application can
7012 access these variables whether it uses small data or not.
7015 @defmac FORCE_CODE_SECTION_ALIGN
7016 If defined, an ASM statement that aligns a code section to some
7017 arbitrary boundary. This is used to force all fragments of the
7018 @code{.init} and @code{.fini} sections to have to same alignment
7019 and thus prevent the linker from having to add any padding.
7022 @defmac JUMP_TABLES_IN_TEXT_SECTION
7023 Define this macro to be an expression with a nonzero value if jump
7024 tables (for @code{tablejump} insns) should be output in the text
7025 section, along with the assembler instructions. Otherwise, the
7026 readonly data section is used.
7028 This macro is irrelevant if there is no separate readonly data section.
7031 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
7032 Define this hook if you need to do something special to set up the
7033 @file{varasm.c} sections, or if your target has some special sections
7034 of its own that you need to create.
7036 GCC calls this hook after processing the command line, but before writing
7037 any assembly code, and before calling any of the section-returning hooks
7041 @deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void)
7042 Return a mask describing how relocations should be treated when
7043 selecting sections. Bit 1 should be set if global relocations
7044 should be placed in a read-write section; bit 0 should be set if
7045 local relocations should be placed in a read-write section.
7047 The default version of this function returns 3 when @option{-fpic}
7048 is in effect, and 0 otherwise. The hook is typically redefined
7049 when the target cannot support (some kinds of) dynamic relocations
7050 in read-only sections even in executables.
7053 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
7054 Return the section into which @var{exp} should be placed. You can
7055 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
7056 some sort. @var{reloc} indicates whether the initial value of @var{exp}
7057 requires link-time relocations. Bit 0 is set when variable contains
7058 local relocations only, while bit 1 is set for global relocations.
7059 @var{align} is the constant alignment in bits.
7061 The default version of this function takes care of putting read-only
7062 variables in @code{readonly_data_section}.
7064 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7067 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7068 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7069 for @code{FUNCTION_DECL}s as well as for variables and constants.
7071 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7072 function has been determined to be likely to be called, and nonzero if
7073 it is unlikely to be called.
7076 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
7077 Build up a unique section name, expressed as a @code{STRING_CST} node,
7078 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7079 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7080 the initial value of @var{exp} requires link-time relocations.
7082 The default version of this function appends the symbol name to the
7083 ELF section name that would normally be used for the symbol. For
7084 example, the function @code{foo} would be placed in @code{.text.foo}.
7085 Whatever the actual target object format, this is often good enough.
7088 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl})
7089 Return the readonly data section associated with
7090 @samp{DECL_SECTION_NAME (@var{decl})}.
7091 The default version of this function selects @code{.gnu.linkonce.r.name} if
7092 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7093 if function is in @code{.text.name}, and the normal readonly-data section
7097 @deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX
7098 Usually, the compiler uses the prefix @code{".rodata"} to construct
7099 section names for mergeable constant data. Define this macro to override
7100 the string if a different section name should be used.
7103 @deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void)
7104 Return the section that should be used for transactional memory clone tables.
7107 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
7108 Return the section into which a constant @var{x}, of mode @var{mode},
7109 should be placed. You can assume that @var{x} is some kind of
7110 constant in RTL@. The argument @var{mode} is redundant except in the
7111 case of a @code{const_int} rtx. @var{align} is the constant alignment
7114 The default version of this function takes care of putting symbolic
7115 constants in @code{flag_pic} mode in @code{data_section} and everything
7116 else in @code{readonly_data_section}.
7119 @deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
7120 Define this hook if you need to postprocess the assembler name generated
7121 by target-independent code. The @var{id} provided to this hook will be
7122 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7123 or the mangled name of the @var{decl} in C++). The return value of the
7124 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7125 your target system. The default implementation of this hook just
7126 returns the @var{id} provided.
7129 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
7130 Define this hook if references to a symbol or a constant must be
7131 treated differently depending on something about the variable or
7132 function named by the symbol (such as what section it is in).
7134 The hook is executed immediately after rtl has been created for
7135 @var{decl}, which may be a variable or function declaration or
7136 an entry in the constant pool. In either case, @var{rtl} is the
7137 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7138 in this hook; that field may not have been initialized yet.
7140 In the case of a constant, it is safe to assume that the rtl is
7141 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7142 will also have this form, but that is not guaranteed. Global
7143 register variables, for instance, will have a @code{reg} for their
7144 rtl. (Normally the right thing to do with such unusual rtl is
7147 The @var{new_decl_p} argument will be true if this is the first time
7148 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7149 be false for subsequent invocations, which will happen for duplicate
7150 declarations. Whether or not anything must be done for the duplicate
7151 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7152 @var{new_decl_p} is always true when the hook is called for a constant.
7154 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7155 The usual thing for this hook to do is to record flags in the
7156 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7157 Historically, the name string was modified if it was necessary to
7158 encode more than one bit of information, but this practice is now
7159 discouraged; use @code{SYMBOL_REF_FLAGS}.
7161 The default definition of this hook, @code{default_encode_section_info}
7162 in @file{varasm.c}, sets a number of commonly-useful bits in
7163 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7164 before overriding it.
7167 @deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name})
7168 Decode @var{name} and return the real name part, sans
7169 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7173 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp})
7174 Returns true if @var{exp} should be placed into a ``small data'' section.
7175 The default version of this hook always returns false.
7178 @deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
7179 Contains the value true if the target places read-only
7180 ``small data'' into a separate section. The default value is false.
7183 @deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void)
7184 It returns true if target wants profile code emitted before prologue.
7186 The default version of this hook use the target macro
7187 @code{PROFILE_BEFORE_PROLOGUE}.
7190 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp})
7191 Returns true if @var{exp} names an object for which name resolution
7192 rules must resolve to the current ``module'' (dynamic shared library
7193 or executable image).
7195 The default version of this hook implements the name resolution rules
7196 for ELF, which has a looser model of global name binding than other
7197 currently supported object file formats.
7200 @deftypevr {Target Hook} bool TARGET_HAVE_TLS
7201 Contains the value true if the target supports thread-local storage.
7202 The default value is false.
7207 @section Position Independent Code
7208 @cindex position independent code
7211 This section describes macros that help implement generation of position
7212 independent code. Simply defining these macros is not enough to
7213 generate valid PIC; you must also add support to the hook
7214 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7215 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7216 must modify the definition of @samp{movsi} to do something appropriate
7217 when the source operand contains a symbolic address. You may also
7218 need to alter the handling of switch statements so that they use
7220 @c i rearranged the order of the macros above to try to force one of
7221 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7223 @defmac PIC_OFFSET_TABLE_REGNUM
7224 The register number of the register used to address a table of static
7225 data addresses in memory. In some cases this register is defined by a
7226 processor's ``application binary interface'' (ABI)@. When this macro
7227 is defined, RTL is generated for this register once, as with the stack
7228 pointer and frame pointer registers. If this macro is not defined, it
7229 is up to the machine-dependent files to allocate such a register (if
7230 necessary). Note that this register must be fixed when in use (e.g.@:
7231 when @code{flag_pic} is true).
7234 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7235 A C expression that is nonzero if the register defined by
7236 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7237 the default is zero. Do not define
7238 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7241 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7242 A C expression that is nonzero if @var{x} is a legitimate immediate
7243 operand on the target machine when generating position independent code.
7244 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7245 check this. You can also assume @var{flag_pic} is true, so you need not
7246 check it either. You need not define this macro if all constants
7247 (including @code{SYMBOL_REF}) can be immediate operands when generating
7248 position independent code.
7251 @node Assembler Format
7252 @section Defining the Output Assembler Language
7254 This section describes macros whose principal purpose is to describe how
7255 to write instructions in assembler language---rather than what the
7259 * File Framework:: Structural information for the assembler file.
7260 * Data Output:: Output of constants (numbers, strings, addresses).
7261 * Uninitialized Data:: Output of uninitialized variables.
7262 * Label Output:: Output and generation of labels.
7263 * Initialization:: General principles of initialization
7264 and termination routines.
7265 * Macros for Initialization::
7266 Specific macros that control the handling of
7267 initialization and termination routines.
7268 * Instruction Output:: Output of actual instructions.
7269 * Dispatch Tables:: Output of jump tables.
7270 * Exception Region Output:: Output of exception region code.
7271 * Alignment Output:: Pseudo ops for alignment and skipping data.
7274 @node File Framework
7275 @subsection The Overall Framework of an Assembler File
7276 @cindex assembler format
7277 @cindex output of assembler code
7279 @c prevent bad page break with this line
7280 This describes the overall framework of an assembly file.
7282 @findex default_file_start
7283 @deftypefn {Target Hook} void TARGET_ASM_FILE_START (void)
7284 Output to @code{asm_out_file} any text which the assembler expects to
7285 find at the beginning of a file. The default behavior is controlled
7286 by two flags, documented below. Unless your target's assembler is
7287 quite unusual, if you override the default, you should call
7288 @code{default_file_start} at some point in your target hook. This
7289 lets other target files rely on these variables.
7292 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
7293 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
7294 printed as the very first line in the assembly file, unless
7295 @option{-fverbose-asm} is in effect. (If that macro has been defined
7296 to the empty string, this variable has no effect.) With the normal
7297 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
7298 assembler that it need not bother stripping comments or extra
7299 whitespace from its input. This allows it to work a bit faster.
7301 The default is false. You should not set it to true unless you have
7302 verified that your port does not generate any extra whitespace or
7303 comments that will cause GAS to issue errors in NO_APP mode.
7306 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
7307 If this flag is true, @code{output_file_directive} will be called
7308 for the primary source file, immediately after printing
7309 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
7310 this to be done. The default is false.
7313 @deftypefn {Target Hook} void TARGET_ASM_FILE_END (void)
7314 Output to @code{asm_out_file} any text which the assembler expects
7315 to find at the end of a file. The default is to output nothing.
7318 @deftypefun void file_end_indicate_exec_stack ()
7319 Some systems use a common convention, the @samp{.note.GNU-stack}
7320 special section, to indicate whether or not an object file relies on
7321 the stack being executable. If your system uses this convention, you
7322 should define @code{TARGET_ASM_FILE_END} to this function. If you
7323 need to do other things in that hook, have your hook function call
7327 @deftypefn {Target Hook} void TARGET_ASM_LTO_START (void)
7328 Output to @code{asm_out_file} any text which the assembler expects
7329 to find at the start of an LTO section. The default is to output
7333 @deftypefn {Target Hook} void TARGET_ASM_LTO_END (void)
7334 Output to @code{asm_out_file} any text which the assembler expects
7335 to find at the end of an LTO section. The default is to output
7339 @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void)
7340 Output to @code{asm_out_file} any text which is needed before emitting
7341 unwind info and debug info at the end of a file. Some targets emit
7342 here PIC setup thunks that cannot be emitted at the end of file,
7343 because they couldn't have unwind info then. The default is to output
7347 @defmac ASM_COMMENT_START
7348 A C string constant describing how to begin a comment in the target
7349 assembler language. The compiler assumes that the comment will end at
7350 the end of the line.
7354 A C string constant for text to be output before each @code{asm}
7355 statement or group of consecutive ones. Normally this is
7356 @code{"#APP"}, which is a comment that has no effect on most
7357 assemblers but tells the GNU assembler that it must check the lines
7358 that follow for all valid assembler constructs.
7362 A C string constant for text to be output after each @code{asm}
7363 statement or group of consecutive ones. Normally this is
7364 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
7365 time-saving assumptions that are valid for ordinary compiler output.
7368 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
7369 A C statement to output COFF information or DWARF debugging information
7370 which indicates that filename @var{name} is the current source file to
7371 the stdio stream @var{stream}.
7373 This macro need not be defined if the standard form of output
7374 for the file format in use is appropriate.
7377 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name})
7378 Output COFF information or DWARF debugging information which indicates that filename @var{name} is the current source file to the stdio stream @var{file}.
7380 This target hook need not be defined if the standard form of output for the file format in use is appropriate.
7383 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_IDENT (const char *@var{name})
7384 Output a string based on @var{name}, suitable for the @samp{#ident} directive, or the equivalent directive or pragma in non-C-family languages. If this hook is not defined, nothing is output for the @samp{#ident} directive.
7387 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
7388 A C statement to output the string @var{string} to the stdio stream
7389 @var{stream}. If you do not call the function @code{output_quoted_string}
7390 in your config files, GCC will only call it to output filenames to
7391 the assembler source. So you can use it to canonicalize the format
7392 of the filename using this macro.
7395 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl})
7396 Output assembly directives to switch to section @var{name}. The section
7397 should have attributes as specified by @var{flags}, which is a bit mask
7398 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
7399 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
7400 this section is associated.
7403 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit})
7404 Return preferred text (sub)section for function @var{decl}.
7405 Main purpose of this function is to separate cold, normal and hot
7406 functions. @var{startup} is true when function is known to be used only
7407 at startup (from static constructors or it is @code{main()}).
7408 @var{exit} is true when function is known to be used only at exit
7409 (from static destructors).
7410 Return NULL if function should go to default text section.
7413 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold})
7414 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}.
7417 @deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
7418 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
7419 It must not be modified by command-line option processing.
7422 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
7423 @deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
7424 This flag is true if we can create zeroed data by switching to a BSS
7425 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
7426 This is true on most ELF targets.
7429 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
7430 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
7431 based on a variable or function decl, a section name, and whether or not the
7432 declaration's initializer may contain runtime relocations. @var{decl} may be
7433 null, in which case read-write data should be assumed.
7435 The default version of this function handles choosing code vs data,
7436 read-only vs read-write data, and @code{flag_pic}. You should only
7437 need to override this if your target has special flags that might be
7438 set via @code{__attribute__}.
7441 @deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text})
7442 Provides the target with the ability to record the gcc command line
7443 switches that have been passed to the compiler, and options that are
7444 enabled. The @var{type} argument specifies what is being recorded.
7445 It can take the following values:
7448 @item SWITCH_TYPE_PASSED
7449 @var{text} is a command line switch that has been set by the user.
7451 @item SWITCH_TYPE_ENABLED
7452 @var{text} is an option which has been enabled. This might be as a
7453 direct result of a command line switch, or because it is enabled by
7454 default or because it has been enabled as a side effect of a different
7455 command line switch. For example, the @option{-O2} switch enables
7456 various different individual optimization passes.
7458 @item SWITCH_TYPE_DESCRIPTIVE
7459 @var{text} is either NULL or some descriptive text which should be
7460 ignored. If @var{text} is NULL then it is being used to warn the
7461 target hook that either recording is starting or ending. The first
7462 time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the
7463 warning is for start up and the second time the warning is for
7464 wind down. This feature is to allow the target hook to make any
7465 necessary preparations before it starts to record switches and to
7466 perform any necessary tidying up after it has finished recording
7469 @item SWITCH_TYPE_LINE_START
7470 This option can be ignored by this target hook.
7472 @item SWITCH_TYPE_LINE_END
7473 This option can be ignored by this target hook.
7476 The hook's return value must be zero. Other return values may be
7477 supported in the future.
7479 By default this hook is set to NULL, but an example implementation is
7480 provided for ELF based targets. Called @var{elf_record_gcc_switches},
7481 it records the switches as ASCII text inside a new, string mergeable
7482 section in the assembler output file. The name of the new section is
7483 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
7487 @deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
7488 This is the name of the section that will be created by the example
7489 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
7495 @subsection Output of Data
7498 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
7499 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
7500 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
7501 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
7502 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
7503 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
7504 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
7505 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
7506 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
7507 These hooks specify assembly directives for creating certain kinds
7508 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
7509 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
7510 aligned two-byte object, and so on. Any of the hooks may be
7511 @code{NULL}, indicating that no suitable directive is available.
7513 The compiler will print these strings at the start of a new line,
7514 followed immediately by the object's initial value. In most cases,
7515 the string should contain a tab, a pseudo-op, and then another tab.
7518 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
7519 The @code{assemble_integer} function uses this hook to output an
7520 integer object. @var{x} is the object's value, @var{size} is its size
7521 in bytes and @var{aligned_p} indicates whether it is aligned. The
7522 function should return @code{true} if it was able to output the
7523 object. If it returns false, @code{assemble_integer} will try to
7524 split the object into smaller parts.
7526 The default implementation of this hook will use the
7527 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
7528 when the relevant string is @code{NULL}.
7531 @deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x})
7532 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
7533 can't deal with, and output assembly code to @var{file} corresponding to
7534 the pattern @var{x}. This may be used to allow machine-dependent
7535 @code{UNSPEC}s to appear within constants.
7537 If target hook fails to recognize a pattern, it must return @code{false},
7538 so that a standard error message is printed. If it prints an error message
7539 itself, by calling, for example, @code{output_operand_lossage}, it may just
7543 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
7544 A C statement to output to the stdio stream @var{stream} an assembler
7545 instruction to assemble a string constant containing the @var{len}
7546 bytes at @var{ptr}. @var{ptr} will be a C expression of type
7547 @code{char *} and @var{len} a C expression of type @code{int}.
7549 If the assembler has a @code{.ascii} pseudo-op as found in the
7550 Berkeley Unix assembler, do not define the macro
7551 @code{ASM_OUTPUT_ASCII}.
7554 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
7555 A C statement to output word @var{n} of a function descriptor for
7556 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
7557 is defined, and is otherwise unused.
7560 @defmac CONSTANT_POOL_BEFORE_FUNCTION
7561 You may define this macro as a C expression. You should define the
7562 expression to have a nonzero value if GCC should output the constant
7563 pool for a function before the code for the function, or a zero value if
7564 GCC should output the constant pool after the function. If you do
7565 not define this macro, the usual case, GCC will output the constant
7566 pool before the function.
7569 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
7570 A C statement to output assembler commands to define the start of the
7571 constant pool for a function. @var{funname} is a string giving
7572 the name of the function. Should the return type of the function
7573 be required, it can be obtained via @var{fundecl}. @var{size}
7574 is the size, in bytes, of the constant pool that will be written
7575 immediately after this call.
7577 If no constant-pool prefix is required, the usual case, this macro need
7581 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
7582 A C statement (with or without semicolon) to output a constant in the
7583 constant pool, if it needs special treatment. (This macro need not do
7584 anything for RTL expressions that can be output normally.)
7586 The argument @var{file} is the standard I/O stream to output the
7587 assembler code on. @var{x} is the RTL expression for the constant to
7588 output, and @var{mode} is the machine mode (in case @var{x} is a
7589 @samp{const_int}). @var{align} is the required alignment for the value
7590 @var{x}; you should output an assembler directive to force this much
7593 The argument @var{labelno} is a number to use in an internal label for
7594 the address of this pool entry. The definition of this macro is
7595 responsible for outputting the label definition at the proper place.
7596 Here is how to do this:
7599 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
7602 When you output a pool entry specially, you should end with a
7603 @code{goto} to the label @var{jumpto}. This will prevent the same pool
7604 entry from being output a second time in the usual manner.
7606 You need not define this macro if it would do nothing.
7609 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
7610 A C statement to output assembler commands to at the end of the constant
7611 pool for a function. @var{funname} is a string giving the name of the
7612 function. Should the return type of the function be required, you can
7613 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
7614 constant pool that GCC wrote immediately before this call.
7616 If no constant-pool epilogue is required, the usual case, you need not
7620 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
7621 Define this macro as a C expression which is nonzero if @var{C} is
7622 used as a logical line separator by the assembler. @var{STR} points
7623 to the position in the string where @var{C} was found; this can be used if
7624 a line separator uses multiple characters.
7626 If you do not define this macro, the default is that only
7627 the character @samp{;} is treated as a logical line separator.
7630 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
7631 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
7632 These target hooks are C string constants, describing the syntax in the
7633 assembler for grouping arithmetic expressions. If not overridden, they
7634 default to normal parentheses, which is correct for most assemblers.
7637 These macros are provided by @file{real.h} for writing the definitions
7638 of @code{ASM_OUTPUT_DOUBLE} and the like:
7640 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
7641 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
7642 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
7643 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
7644 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
7645 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
7646 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
7647 target's floating point representation, and store its bit pattern in
7648 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
7649 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
7650 simple @code{long int}. For the others, it should be an array of
7651 @code{long int}. The number of elements in this array is determined
7652 by the size of the desired target floating point data type: 32 bits of
7653 it go in each @code{long int} array element. Each array element holds
7654 32 bits of the result, even if @code{long int} is wider than 32 bits
7655 on the host machine.
7657 The array element values are designed so that you can print them out
7658 using @code{fprintf} in the order they should appear in the target
7662 @node Uninitialized Data
7663 @subsection Output of Uninitialized Variables
7665 Each of the macros in this section is used to do the whole job of
7666 outputting a single uninitialized variable.
7668 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
7669 A C statement (sans semicolon) to output to the stdio stream
7670 @var{stream} the assembler definition of a common-label named
7671 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7672 is the size rounded up to whatever alignment the caller wants. It is
7673 possible that @var{size} may be zero, for instance if a struct with no
7674 other member than a zero-length array is defined. In this case, the
7675 backend must output a symbol definition that allocates at least one
7676 byte, both so that the address of the resulting object does not compare
7677 equal to any other, and because some object formats cannot even express
7678 the concept of a zero-sized common symbol, as that is how they represent
7679 an ordinary undefined external.
7681 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7682 output the name itself; before and after that, output the additional
7683 assembler syntax for defining the name, and a newline.
7685 This macro controls how the assembler definitions of uninitialized
7686 common global variables are output.
7689 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
7690 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
7691 separate, explicit argument. If you define this macro, it is used in
7692 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
7693 handling the required alignment of the variable. The alignment is specified
7694 as the number of bits.
7697 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7698 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
7699 variable to be output, if there is one, or @code{NULL_TREE} if there
7700 is no corresponding variable. If you define this macro, GCC will use it
7701 in place of both @code{ASM_OUTPUT_COMMON} and
7702 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
7703 the variable's decl in order to chose what to output.
7706 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7707 A C statement (sans semicolon) to output to the stdio stream
7708 @var{stream} the assembler definition of uninitialized global @var{decl} named
7709 @var{name} whose size is @var{size} bytes. The variable @var{alignment}
7710 is the alignment specified as the number of bits.
7712 Try to use function @code{asm_output_aligned_bss} defined in file
7713 @file{varasm.c} when defining this macro. If unable, use the expression
7714 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
7715 before and after that, output the additional assembler syntax for defining
7716 the name, and a newline.
7718 There are two ways of handling global BSS@. One is to define this macro.
7719 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
7720 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
7721 You do not need to do both.
7723 Some languages do not have @code{common} data, and require a
7724 non-common form of global BSS in order to handle uninitialized globals
7725 efficiently. C++ is one example of this. However, if the target does
7726 not support global BSS, the front end may choose to make globals
7727 common in order to save space in the object file.
7730 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
7731 A C statement (sans semicolon) to output to the stdio stream
7732 @var{stream} the assembler definition of a local-common-label named
7733 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
7734 is the size rounded up to whatever alignment the caller wants.
7736 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7737 output the name itself; before and after that, output the additional
7738 assembler syntax for defining the name, and a newline.
7740 This macro controls how the assembler definitions of uninitialized
7741 static variables are output.
7744 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
7745 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
7746 separate, explicit argument. If you define this macro, it is used in
7747 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
7748 handling the required alignment of the variable. The alignment is specified
7749 as the number of bits.
7752 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
7753 Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the
7754 variable to be output, if there is one, or @code{NULL_TREE} if there
7755 is no corresponding variable. If you define this macro, GCC will use it
7756 in place of both @code{ASM_OUTPUT_DECL} and
7757 @code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see
7758 the variable's decl in order to chose what to output.
7762 @subsection Output and Generation of Labels
7764 @c prevent bad page break with this line
7765 This is about outputting labels.
7767 @findex assemble_name
7768 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
7769 A C statement (sans semicolon) to output to the stdio stream
7770 @var{stream} the assembler definition of a label named @var{name}.
7771 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7772 output the name itself; before and after that, output the additional
7773 assembler syntax for defining the name, and a newline. A default
7774 definition of this macro is provided which is correct for most systems.
7777 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
7778 A C statement (sans semicolon) to output to the stdio stream
7779 @var{stream} the assembler definition of a label named @var{name} of
7781 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
7782 output the name itself; before and after that, output the additional
7783 assembler syntax for defining the name, and a newline. A default
7784 definition of this macro is provided which is correct for most systems.
7786 If this macro is not defined, then the function name is defined in the
7787 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7790 @findex assemble_name_raw
7791 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
7792 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
7793 to refer to a compiler-generated label. The default definition uses
7794 @code{assemble_name_raw}, which is like @code{assemble_name} except
7795 that it is more efficient.
7799 A C string containing the appropriate assembler directive to specify the
7800 size of a symbol, without any arguments. On systems that use ELF, the
7801 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
7802 systems, the default is not to define this macro.
7804 Define this macro only if it is correct to use the default definitions
7805 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
7806 for your system. If you need your own custom definitions of those
7807 macros, or if you do not need explicit symbol sizes at all, do not
7811 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
7812 A C statement (sans semicolon) to output to the stdio stream
7813 @var{stream} a directive telling the assembler that the size of the
7814 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
7815 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7819 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
7820 A C statement (sans semicolon) to output to the stdio stream
7821 @var{stream} a directive telling the assembler to calculate the size of
7822 the symbol @var{name} by subtracting its address from the current
7825 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
7826 provided. The default assumes that the assembler recognizes a special
7827 @samp{.} symbol as referring to the current address, and can calculate
7828 the difference between this and another symbol. If your assembler does
7829 not recognize @samp{.} or cannot do calculations with it, you will need
7830 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
7834 A C string containing the appropriate assembler directive to specify the
7835 type of a symbol, without any arguments. On systems that use ELF, the
7836 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
7837 systems, the default is not to define this macro.
7839 Define this macro only if it is correct to use the default definition of
7840 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7841 custom definition of this macro, or if you do not need explicit symbol
7842 types at all, do not define this macro.
7845 @defmac TYPE_OPERAND_FMT
7846 A C string which specifies (using @code{printf} syntax) the format of
7847 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
7848 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
7849 the default is not to define this macro.
7851 Define this macro only if it is correct to use the default definition of
7852 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
7853 custom definition of this macro, or if you do not need explicit symbol
7854 types at all, do not define this macro.
7857 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
7858 A C statement (sans semicolon) to output to the stdio stream
7859 @var{stream} a directive telling the assembler that the type of the
7860 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
7861 that string is always either @samp{"function"} or @samp{"object"}, but
7862 you should not count on this.
7864 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
7865 definition of this macro is provided.
7868 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
7869 A C statement (sans semicolon) to output to the stdio stream
7870 @var{stream} any text necessary for declaring the name @var{name} of a
7871 function which is being defined. This macro is responsible for
7872 outputting the label definition (perhaps using
7873 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
7874 @code{FUNCTION_DECL} tree node representing the function.
7876 If this macro is not defined, then the function name is defined in the
7877 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
7879 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
7883 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
7884 A C statement (sans semicolon) to output to the stdio stream
7885 @var{stream} any text necessary for declaring the size of a function
7886 which is being defined. The argument @var{name} is the name of the
7887 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
7888 representing the function.
7890 If this macro is not defined, then the function size is not defined.
7892 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
7896 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
7897 A C statement (sans semicolon) to output to the stdio stream
7898 @var{stream} any text necessary for declaring the name @var{name} of an
7899 initialized variable which is being defined. This macro must output the
7900 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
7901 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
7903 If this macro is not defined, then the variable name is defined in the
7904 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
7906 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
7907 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
7910 @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})
7911 A target hook to output to the stdio stream @var{file} any text necessary
7912 for declaring the name @var{name} of a constant which is being defined. This
7913 target hook is responsible for outputting the label definition (perhaps using
7914 @code{assemble_label}). The argument @var{exp} is the value of the constant,
7915 and @var{size} is the size of the constant in bytes. The @var{name}
7916 will be an internal label.
7918 The default version of this target hook, define the @var{name} in the
7919 usual manner as a label (by means of @code{assemble_label}).
7921 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
7924 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
7925 A C statement (sans semicolon) to output to the stdio stream
7926 @var{stream} any text necessary for claiming a register @var{regno}
7927 for a global variable @var{decl} with name @var{name}.
7929 If you don't define this macro, that is equivalent to defining it to do
7933 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
7934 A C statement (sans semicolon) to finish up declaring a variable name
7935 once the compiler has processed its initializer fully and thus has had a
7936 chance to determine the size of an array when controlled by an
7937 initializer. This is used on systems where it's necessary to declare
7938 something about the size of the object.
7940 If you don't define this macro, that is equivalent to defining it to do
7943 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
7944 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
7947 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
7948 This target hook is a function to output to the stdio stream
7949 @var{stream} some commands that will make the label @var{name} global;
7950 that is, available for reference from other files.
7952 The default implementation relies on a proper definition of
7953 @code{GLOBAL_ASM_OP}.
7956 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
7957 This target hook is a function to output to the stdio stream
7958 @var{stream} some commands that will make the name associated with @var{decl}
7959 global; that is, available for reference from other files.
7961 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
7964 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
7965 A C statement (sans semicolon) to output to the stdio stream
7966 @var{stream} some commands that will make the label @var{name} weak;
7967 that is, available for reference from other files but only used if
7968 no other definition is available. Use the expression
7969 @code{assemble_name (@var{stream}, @var{name})} to output the name
7970 itself; before and after that, output the additional assembler syntax
7971 for making that name weak, and a newline.
7973 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
7974 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
7978 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
7979 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
7980 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
7981 or variable decl. If @var{value} is not @code{NULL}, this C statement
7982 should output to the stdio stream @var{stream} assembler code which
7983 defines (equates) the weak symbol @var{name} to have the value
7984 @var{value}. If @var{value} is @code{NULL}, it should output commands
7985 to make @var{name} weak.
7988 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
7989 Outputs a directive that enables @var{name} to be used to refer to
7990 symbol @var{value} with weak-symbol semantics. @code{decl} is the
7991 declaration of @code{name}.
7994 @defmac SUPPORTS_WEAK
7995 A preprocessor constant expression which evaluates to true if the target
7996 supports weak symbols.
7998 If you don't define this macro, @file{defaults.h} provides a default
7999 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
8000 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
8003 @defmac TARGET_SUPPORTS_WEAK
8004 A C expression which evaluates to true if the target supports weak symbols.
8006 If you don't define this macro, @file{defaults.h} provides a default
8007 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
8008 this macro if you want to control weak symbol support with a compiler
8009 flag such as @option{-melf}.
8012 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
8013 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
8014 public symbol such that extra copies in multiple translation units will
8015 be discarded by the linker. Define this macro if your object file
8016 format provides support for this concept, such as the @samp{COMDAT}
8017 section flags in the Microsoft Windows PE/COFF format, and this support
8018 requires changes to @var{decl}, such as putting it in a separate section.
8021 @defmac SUPPORTS_ONE_ONLY
8022 A C expression which evaluates to true if the target supports one-only
8025 If you don't define this macro, @file{varasm.c} provides a default
8026 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
8027 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
8028 you want to control one-only symbol support with a compiler flag, or if
8029 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
8030 be emitted as one-only.
8033 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility})
8034 This target hook is a function to output to @var{asm_out_file} some
8035 commands that will make the symbol(s) associated with @var{decl} have
8036 hidden, protected or internal visibility as specified by @var{visibility}.
8039 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
8040 A C expression that evaluates to true if the target's linker expects
8041 that weak symbols do not appear in a static archive's table of contents.
8042 The default is @code{0}.
8044 Leaving weak symbols out of an archive's table of contents means that,
8045 if a symbol will only have a definition in one translation unit and
8046 will have undefined references from other translation units, that
8047 symbol should not be weak. Defining this macro to be nonzero will
8048 thus have the effect that certain symbols that would normally be weak
8049 (explicit template instantiations, and vtables for polymorphic classes
8050 with noninline key methods) will instead be nonweak.
8052 The C++ ABI requires this macro to be zero. Define this macro for
8053 targets where full C++ ABI compliance is impossible and where linker
8054 restrictions require weak symbols to be left out of a static archive's
8058 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
8059 A C statement (sans semicolon) to output to the stdio stream
8060 @var{stream} any text necessary for declaring the name of an external
8061 symbol named @var{name} which is referenced in this compilation but
8062 not defined. The value of @var{decl} is the tree node for the
8065 This macro need not be defined if it does not need to output anything.
8066 The GNU assembler and most Unix assemblers don't require anything.
8069 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
8070 This target hook is a function to output to @var{asm_out_file} an assembler
8071 pseudo-op to declare a library function name external. The name of the
8072 library function is given by @var{symref}, which is a @code{symbol_ref}.
8075 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol})
8076 This target hook is a function to output to @var{asm_out_file} an assembler
8077 directive to annotate @var{symbol} as used. The Darwin target uses the
8078 .no_dead_code_strip directive.
8081 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8082 A C statement (sans semicolon) to output to the stdio stream
8083 @var{stream} a reference in assembler syntax to a label named
8084 @var{name}. This should add @samp{_} to the front of the name, if that
8085 is customary on your operating system, as it is in most Berkeley Unix
8086 systems. This macro is used in @code{assemble_name}.
8089 @deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name})
8090 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.
8093 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8094 A C statement (sans semicolon) to output a reference to
8095 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8096 will be used to output the name of the symbol. This macro may be used
8097 to modify the way a symbol is referenced depending on information
8098 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8101 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8102 A C statement (sans semicolon) to output a reference to @var{buf}, the
8103 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8104 @code{assemble_name} will be used to output the name of the symbol.
8105 This macro is not used by @code{output_asm_label}, or the @code{%l}
8106 specifier that calls it; the intention is that this macro should be set
8107 when it is necessary to output a label differently when its address is
8111 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
8112 A function to output to the stdio stream @var{stream} a label whose
8113 name is made from the string @var{prefix} and the number @var{labelno}.
8115 It is absolutely essential that these labels be distinct from the labels
8116 used for user-level functions and variables. Otherwise, certain programs
8117 will have name conflicts with internal labels.
8119 It is desirable to exclude internal labels from the symbol table of the
8120 object file. Most assemblers have a naming convention for labels that
8121 should be excluded; on many systems, the letter @samp{L} at the
8122 beginning of a label has this effect. You should find out what
8123 convention your system uses, and follow it.
8125 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8128 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8129 A C statement to output to the stdio stream @var{stream} a debug info
8130 label whose name is made from the string @var{prefix} and the number
8131 @var{num}. This is useful for VLIW targets, where debug info labels
8132 may need to be treated differently than branch target labels. On some
8133 systems, branch target labels must be at the beginning of instruction
8134 bundles, but debug info labels can occur in the middle of instruction
8137 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8141 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8142 A C statement to store into the string @var{string} a label whose name
8143 is made from the string @var{prefix} and the number @var{num}.
8145 This string, when output subsequently by @code{assemble_name}, should
8146 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8147 with the same @var{prefix} and @var{num}.
8149 If the string begins with @samp{*}, then @code{assemble_name} will
8150 output the rest of the string unchanged. It is often convenient for
8151 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8152 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8153 to output the string, and may change it. (Of course,
8154 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8155 you should know what it does on your machine.)
8158 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8159 A C expression to assign to @var{outvar} (which is a variable of type
8160 @code{char *}) a newly allocated string made from the string
8161 @var{name} and the number @var{number}, with some suitable punctuation
8162 added. Use @code{alloca} to get space for the string.
8164 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8165 produce an assembler label for an internal static variable whose name is
8166 @var{name}. Therefore, the string must be such as to result in valid
8167 assembler code. The argument @var{number} is different each time this
8168 macro is executed; it prevents conflicts between similarly-named
8169 internal static variables in different scopes.
8171 Ideally this string should not be a valid C identifier, to prevent any
8172 conflict with the user's own symbols. Most assemblers allow periods
8173 or percent signs in assembler symbols; putting at least one of these
8174 between the name and the number will suffice.
8176 If this macro is not defined, a default definition will be provided
8177 which is correct for most systems.
8180 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8181 A C statement to output to the stdio stream @var{stream} assembler code
8182 which defines (equates) the symbol @var{name} to have the value @var{value}.
8185 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8186 correct for most systems.
8189 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8190 A C statement to output to the stdio stream @var{stream} assembler code
8191 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8192 to have the value of the tree node @var{decl_of_value}. This macro will
8193 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8194 the tree nodes are available.
8197 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8198 correct for most systems.
8201 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8202 A C statement that evaluates to true if the assembler code which defines
8203 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8204 of the tree node @var{decl_of_value} should be emitted near the end of the
8205 current compilation unit. The default is to not defer output of defines.
8206 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
8207 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
8210 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
8211 A C statement to output to the stdio stream @var{stream} assembler code
8212 which defines (equates) the weak symbol @var{name} to have the value
8213 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
8214 an undefined weak symbol.
8216 Define this macro if the target only supports weak aliases; define
8217 @code{ASM_OUTPUT_DEF} instead if possible.
8220 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
8221 Define this macro to override the default assembler names used for
8222 Objective-C methods.
8224 The default name is a unique method number followed by the name of the
8225 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
8226 the category is also included in the assembler name (e.g.@:
8229 These names are safe on most systems, but make debugging difficult since
8230 the method's selector is not present in the name. Therefore, particular
8231 systems define other ways of computing names.
8233 @var{buf} is an expression of type @code{char *} which gives you a
8234 buffer in which to store the name; its length is as long as
8235 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
8236 50 characters extra.
8238 The argument @var{is_inst} specifies whether the method is an instance
8239 method or a class method; @var{class_name} is the name of the class;
8240 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
8241 in a category); and @var{sel_name} is the name of the selector.
8243 On systems where the assembler can handle quoted names, you can use this
8244 macro to provide more human-readable names.
8247 @node Initialization
8248 @subsection How Initialization Functions Are Handled
8249 @cindex initialization routines
8250 @cindex termination routines
8251 @cindex constructors, output of
8252 @cindex destructors, output of
8254 The compiled code for certain languages includes @dfn{constructors}
8255 (also called @dfn{initialization routines})---functions to initialize
8256 data in the program when the program is started. These functions need
8257 to be called before the program is ``started''---that is to say, before
8258 @code{main} is called.
8260 Compiling some languages generates @dfn{destructors} (also called
8261 @dfn{termination routines}) that should be called when the program
8264 To make the initialization and termination functions work, the compiler
8265 must output something in the assembler code to cause those functions to
8266 be called at the appropriate time. When you port the compiler to a new
8267 system, you need to specify how to do this.
8269 There are two major ways that GCC currently supports the execution of
8270 initialization and termination functions. Each way has two variants.
8271 Much of the structure is common to all four variations.
8273 @findex __CTOR_LIST__
8274 @findex __DTOR_LIST__
8275 The linker must build two lists of these functions---a list of
8276 initialization functions, called @code{__CTOR_LIST__}, and a list of
8277 termination functions, called @code{__DTOR_LIST__}.
8279 Each list always begins with an ignored function pointer (which may hold
8280 0, @minus{}1, or a count of the function pointers after it, depending on
8281 the environment). This is followed by a series of zero or more function
8282 pointers to constructors (or destructors), followed by a function
8283 pointer containing zero.
8285 Depending on the operating system and its executable file format, either
8286 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
8287 time and exit time. Constructors are called in reverse order of the
8288 list; destructors in forward order.
8290 The best way to handle static constructors works only for object file
8291 formats which provide arbitrarily-named sections. A section is set
8292 aside for a list of constructors, and another for a list of destructors.
8293 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
8294 object file that defines an initialization function also puts a word in
8295 the constructor section to point to that function. The linker
8296 accumulates all these words into one contiguous @samp{.ctors} section.
8297 Termination functions are handled similarly.
8299 This method will be chosen as the default by @file{target-def.h} if
8300 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
8301 support arbitrary sections, but does support special designated
8302 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
8303 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
8305 When arbitrary sections are available, there are two variants, depending
8306 upon how the code in @file{crtstuff.c} is called. On systems that
8307 support a @dfn{.init} section which is executed at program startup,
8308 parts of @file{crtstuff.c} are compiled into that section. The
8309 program is linked by the @command{gcc} driver like this:
8312 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
8315 The prologue of a function (@code{__init}) appears in the @code{.init}
8316 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
8317 for the function @code{__fini} in the @dfn{.fini} section. Normally these
8318 files are provided by the operating system or by the GNU C library, but
8319 are provided by GCC for a few targets.
8321 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
8322 compiled from @file{crtstuff.c}. They contain, among other things, code
8323 fragments within the @code{.init} and @code{.fini} sections that branch
8324 to routines in the @code{.text} section. The linker will pull all parts
8325 of a section together, which results in a complete @code{__init} function
8326 that invokes the routines we need at startup.
8328 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
8331 If no init section is available, when GCC compiles any function called
8332 @code{main} (or more accurately, any function designated as a program
8333 entry point by the language front end calling @code{expand_main_function}),
8334 it inserts a procedure call to @code{__main} as the first executable code
8335 after the function prologue. The @code{__main} function is defined
8336 in @file{libgcc2.c} and runs the global constructors.
8338 In file formats that don't support arbitrary sections, there are again
8339 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
8340 and an `a.out' format must be used. In this case,
8341 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
8342 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
8343 and with the address of the void function containing the initialization
8344 code as its value. The GNU linker recognizes this as a request to add
8345 the value to a @dfn{set}; the values are accumulated, and are eventually
8346 placed in the executable as a vector in the format described above, with
8347 a leading (ignored) count and a trailing zero element.
8348 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
8349 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
8350 the compilation of @code{main} to call @code{__main} as above, starting
8351 the initialization process.
8353 The last variant uses neither arbitrary sections nor the GNU linker.
8354 This is preferable when you want to do dynamic linking and when using
8355 file formats which the GNU linker does not support, such as `ECOFF'@. In
8356 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
8357 termination functions are recognized simply by their names. This requires
8358 an extra program in the linkage step, called @command{collect2}. This program
8359 pretends to be the linker, for use with GCC; it does its job by running
8360 the ordinary linker, but also arranges to include the vectors of
8361 initialization and termination functions. These functions are called
8362 via @code{__main} as described above. In order to use this method,
8363 @code{use_collect2} must be defined in the target in @file{config.gcc}.
8366 The following section describes the specific macros that control and
8367 customize the handling of initialization and termination functions.
8370 @node Macros for Initialization
8371 @subsection Macros Controlling Initialization Routines
8373 Here are the macros that control how the compiler handles initialization
8374 and termination functions:
8376 @defmac INIT_SECTION_ASM_OP
8377 If defined, a C string constant, including spacing, for the assembler
8378 operation to identify the following data as initialization code. If not
8379 defined, GCC will assume such a section does not exist. When you are
8380 using special sections for initialization and termination functions, this
8381 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
8382 run the initialization functions.
8385 @defmac HAS_INIT_SECTION
8386 If defined, @code{main} will not call @code{__main} as described above.
8387 This macro should be defined for systems that control start-up code
8388 on a symbol-by-symbol basis, such as OSF/1, and should not
8389 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
8392 @defmac LD_INIT_SWITCH
8393 If defined, a C string constant for a switch that tells the linker that
8394 the following symbol is an initialization routine.
8397 @defmac LD_FINI_SWITCH
8398 If defined, a C string constant for a switch that tells the linker that
8399 the following symbol is a finalization routine.
8402 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
8403 If defined, a C statement that will write a function that can be
8404 automatically called when a shared library is loaded. The function
8405 should call @var{func}, which takes no arguments. If not defined, and
8406 the object format requires an explicit initialization function, then a
8407 function called @code{_GLOBAL__DI} will be generated.
8409 This function and the following one are used by collect2 when linking a
8410 shared library that needs constructors or destructors, or has DWARF2
8411 exception tables embedded in the code.
8414 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
8415 If defined, a C statement that will write a function that can be
8416 automatically called when a shared library is unloaded. The function
8417 should call @var{func}, which takes no arguments. If not defined, and
8418 the object format requires an explicit finalization function, then a
8419 function called @code{_GLOBAL__DD} will be generated.
8422 @defmac INVOKE__main
8423 If defined, @code{main} will call @code{__main} despite the presence of
8424 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
8425 where the init section is not actually run automatically, but is still
8426 useful for collecting the lists of constructors and destructors.
8429 @defmac SUPPORTS_INIT_PRIORITY
8430 If nonzero, the C++ @code{init_priority} attribute is supported and the
8431 compiler should emit instructions to control the order of initialization
8432 of objects. If zero, the compiler will issue an error message upon
8433 encountering an @code{init_priority} attribute.
8436 @deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS
8437 This value is true if the target supports some ``native'' method of
8438 collecting constructors and destructors to be run at startup and exit.
8439 It is false if we must use @command{collect2}.
8442 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
8443 If defined, a function that outputs assembler code to arrange to call
8444 the function referenced by @var{symbol} at initialization time.
8446 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
8447 no arguments and with no return value. If the target supports initialization
8448 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
8449 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
8451 If this macro is not defined by the target, a suitable default will
8452 be chosen if (1) the target supports arbitrary section names, (2) the
8453 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
8457 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
8458 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
8459 functions rather than initialization functions.
8462 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
8463 generated for the generated object file will have static linkage.
8465 If your system uses @command{collect2} as the means of processing
8466 constructors, then that program normally uses @command{nm} to scan
8467 an object file for constructor functions to be called.
8469 On certain kinds of systems, you can define this macro to make
8470 @command{collect2} work faster (and, in some cases, make it work at all):
8472 @defmac OBJECT_FORMAT_COFF
8473 Define this macro if the system uses COFF (Common Object File Format)
8474 object files, so that @command{collect2} can assume this format and scan
8475 object files directly for dynamic constructor/destructor functions.
8477 This macro is effective only in a native compiler; @command{collect2} as
8478 part of a cross compiler always uses @command{nm} for the target machine.
8481 @defmac REAL_NM_FILE_NAME
8482 Define this macro as a C string constant containing the file name to use
8483 to execute @command{nm}. The default is to search the path normally for
8488 @command{collect2} calls @command{nm} to scan object files for static
8489 constructors and destructors and LTO info. By default, @option{-n} is
8490 passed. Define @code{NM_FLAGS} to a C string constant if other options
8491 are needed to get the same output format as GNU @command{nm -n}
8495 If your system supports shared libraries and has a program to list the
8496 dynamic dependencies of a given library or executable, you can define
8497 these macros to enable support for running initialization and
8498 termination functions in shared libraries:
8501 Define this macro to a C string constant containing the name of the program
8502 which lists dynamic dependencies, like @command{ldd} under SunOS 4.
8505 @defmac PARSE_LDD_OUTPUT (@var{ptr})
8506 Define this macro to be C code that extracts filenames from the output
8507 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
8508 of type @code{char *} that points to the beginning of a line of output
8509 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
8510 code must advance @var{ptr} to the beginning of the filename on that
8511 line. Otherwise, it must set @var{ptr} to @code{NULL}.
8514 @defmac SHLIB_SUFFIX
8515 Define this macro to a C string constant containing the default shared
8516 library extension of the target (e.g., @samp{".so"}). @command{collect2}
8517 strips version information after this suffix when generating global
8518 constructor and destructor names. This define is only needed on targets
8519 that use @command{collect2} to process constructors and destructors.
8522 @node Instruction Output
8523 @subsection Output of Assembler Instructions
8525 @c prevent bad page break with this line
8526 This describes assembler instruction output.
8528 @defmac REGISTER_NAMES
8529 A C initializer containing the assembler's names for the machine
8530 registers, each one as a C string constant. This is what translates
8531 register numbers in the compiler into assembler language.
8534 @defmac ADDITIONAL_REGISTER_NAMES
8535 If defined, a C initializer for an array of structures containing a name
8536 and a register number. This macro defines additional names for hard
8537 registers, thus allowing the @code{asm} option in declarations to refer
8538 to registers using alternate names.
8541 @defmac OVERLAPPING_REGISTER_NAMES
8542 If defined, a C initializer for an array of structures containing a
8543 name, a register number and a count of the number of consecutive
8544 machine registers the name overlaps. This macro defines additional
8545 names for hard registers, thus allowing the @code{asm} option in
8546 declarations to refer to registers using alternate names. Unlike
8547 @code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the
8548 register name implies multiple underlying registers.
8550 This macro should be used when it is important that a clobber in an
8551 @code{asm} statement clobbers all the underlying values implied by the
8552 register name. For example, on ARM, clobbering the double-precision
8553 VFP register ``d0'' implies clobbering both single-precision registers
8557 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
8558 Define this macro if you are using an unusual assembler that
8559 requires different names for the machine instructions.
8561 The definition is a C statement or statements which output an
8562 assembler instruction opcode to the stdio stream @var{stream}. The
8563 macro-operand @var{ptr} is a variable of type @code{char *} which
8564 points to the opcode name in its ``internal'' form---the form that is
8565 written in the machine description. The definition should output the
8566 opcode name to @var{stream}, performing any translation you desire, and
8567 increment the variable @var{ptr} to point at the end of the opcode
8568 so that it will not be output twice.
8570 In fact, your macro definition may process less than the entire opcode
8571 name, or more than the opcode name; but if you want to process text
8572 that includes @samp{%}-sequences to substitute operands, you must take
8573 care of the substitution yourself. Just be sure to increment
8574 @var{ptr} over whatever text should not be output normally.
8576 @findex recog_data.operand
8577 If you need to look at the operand values, they can be found as the
8578 elements of @code{recog_data.operand}.
8580 If the macro definition does nothing, the instruction is output
8584 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
8585 If defined, a C statement to be executed just prior to the output of
8586 assembler code for @var{insn}, to modify the extracted operands so
8587 they will be output differently.
8589 Here the argument @var{opvec} is the vector containing the operands
8590 extracted from @var{insn}, and @var{noperands} is the number of
8591 elements of the vector which contain meaningful data for this insn.
8592 The contents of this vector are what will be used to convert the insn
8593 template into assembler code, so you can change the assembler output
8594 by changing the contents of the vector.
8596 This macro is useful when various assembler syntaxes share a single
8597 file of instruction patterns; by defining this macro differently, you
8598 can cause a large class of instructions to be output differently (such
8599 as with rearranged operands). Naturally, variations in assembler
8600 syntax affecting individual insn patterns ought to be handled by
8601 writing conditional output routines in those patterns.
8603 If this macro is not defined, it is equivalent to a null statement.
8606 @deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx @var{insn}, rtx *@var{opvec}, int @var{noperands})
8607 If defined, this target hook is a function which is executed just after the
8608 output of assembler code for @var{insn}, to change the mode of the assembler
8611 Here the argument @var{opvec} is the vector containing the operands
8612 extracted from @var{insn}, and @var{noperands} is the number of
8613 elements of the vector which contain meaningful data for this insn.
8614 The contents of this vector are what was used to convert the insn
8615 template into assembler code, so you can change the assembler mode
8616 by checking the contents of the vector.
8619 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
8620 A C compound statement to output to stdio stream @var{stream} the
8621 assembler syntax for an instruction operand @var{x}. @var{x} is an
8624 @var{code} is a value that can be used to specify one of several ways
8625 of printing the operand. It is used when identical operands must be
8626 printed differently depending on the context. @var{code} comes from
8627 the @samp{%} specification that was used to request printing of the
8628 operand. If the specification was just @samp{%@var{digit}} then
8629 @var{code} is 0; if the specification was @samp{%@var{ltr}
8630 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
8633 If @var{x} is a register, this macro should print the register's name.
8634 The names can be found in an array @code{reg_names} whose type is
8635 @code{char *[]}. @code{reg_names} is initialized from
8636 @code{REGISTER_NAMES}.
8638 When the machine description has a specification @samp{%@var{punct}}
8639 (a @samp{%} followed by a punctuation character), this macro is called
8640 with a null pointer for @var{x} and the punctuation character for
8644 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
8645 A C expression which evaluates to true if @var{code} is a valid
8646 punctuation character for use in the @code{PRINT_OPERAND} macro. If
8647 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
8648 punctuation characters (except for the standard one, @samp{%}) are used
8652 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
8653 A C compound statement to output to stdio stream @var{stream} the
8654 assembler syntax for an instruction operand that is a memory reference
8655 whose address is @var{x}. @var{x} is an RTL expression.
8657 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
8658 On some machines, the syntax for a symbolic address depends on the
8659 section that the address refers to. On these machines, define the hook
8660 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
8661 @code{symbol_ref}, and then check for it here. @xref{Assembler
8665 @findex dbr_sequence_length
8666 @defmac DBR_OUTPUT_SEQEND (@var{file})
8667 A C statement, to be executed after all slot-filler instructions have
8668 been output. If necessary, call @code{dbr_sequence_length} to
8669 determine the number of slots filled in a sequence (zero if not
8670 currently outputting a sequence), to decide how many no-ops to output,
8673 Don't define this macro if it has nothing to do, but it is helpful in
8674 reading assembly output if the extent of the delay sequence is made
8675 explicit (e.g.@: with white space).
8678 @findex final_sequence
8679 Note that output routines for instructions with delay slots must be
8680 prepared to deal with not being output as part of a sequence
8681 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
8682 found.) The variable @code{final_sequence} is null when not
8683 processing a sequence, otherwise it contains the @code{sequence} rtx
8687 @defmac REGISTER_PREFIX
8688 @defmacx LOCAL_LABEL_PREFIX
8689 @defmacx USER_LABEL_PREFIX
8690 @defmacx IMMEDIATE_PREFIX
8691 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
8692 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
8693 @file{final.c}). These are useful when a single @file{md} file must
8694 support multiple assembler formats. In that case, the various @file{tm.h}
8695 files can define these macros differently.
8698 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
8699 If defined this macro should expand to a series of @code{case}
8700 statements which will be parsed inside the @code{switch} statement of
8701 the @code{asm_fprintf} function. This allows targets to define extra
8702 printf formats which may useful when generating their assembler
8703 statements. Note that uppercase letters are reserved for future
8704 generic extensions to asm_fprintf, and so are not available to target
8705 specific code. The output file is given by the parameter @var{file}.
8706 The varargs input pointer is @var{argptr} and the rest of the format
8707 string, starting the character after the one that is being switched
8708 upon, is pointed to by @var{format}.
8711 @defmac ASSEMBLER_DIALECT
8712 If your target supports multiple dialects of assembler language (such as
8713 different opcodes), define this macro as a C expression that gives the
8714 numeric index of the assembler language dialect to use, with zero as the
8717 If this macro is defined, you may use constructs of the form
8719 @samp{@{option0|option1|option2@dots{}@}}
8722 in the output templates of patterns (@pxref{Output Template}) or in the
8723 first argument of @code{asm_fprintf}. This construct outputs
8724 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
8725 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
8726 within these strings retain their usual meaning. If there are fewer
8727 alternatives within the braces than the value of
8728 @code{ASSEMBLER_DIALECT}, the construct outputs nothing.
8730 If you do not define this macro, the characters @samp{@{}, @samp{|} and
8731 @samp{@}} do not have any special meaning when used in templates or
8732 operands to @code{asm_fprintf}.
8734 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
8735 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
8736 the variations in assembler language syntax with that mechanism. Define
8737 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
8738 if the syntax variant are larger and involve such things as different
8739 opcodes or operand order.
8742 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
8743 A C expression to output to @var{stream} some assembler code
8744 which will push hard register number @var{regno} onto the stack.
8745 The code need not be optimal, since this macro is used only when
8749 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
8750 A C expression to output to @var{stream} some assembler code
8751 which will pop hard register number @var{regno} off of the stack.
8752 The code need not be optimal, since this macro is used only when
8756 @node Dispatch Tables
8757 @subsection Output of Dispatch Tables
8759 @c prevent bad page break with this line
8760 This concerns dispatch tables.
8762 @cindex dispatch table
8763 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
8764 A C statement to output to the stdio stream @var{stream} an assembler
8765 pseudo-instruction to generate a difference between two labels.
8766 @var{value} and @var{rel} are the numbers of two internal labels. The
8767 definitions of these labels are output using
8768 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
8769 way here. For example,
8772 fprintf (@var{stream}, "\t.word L%d-L%d\n",
8773 @var{value}, @var{rel})
8776 You must provide this macro on machines where the addresses in a
8777 dispatch table are relative to the table's own address. If defined, GCC
8778 will also use this macro on all machines when producing PIC@.
8779 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
8780 mode and flags can be read.
8783 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
8784 This macro should be provided on machines where the addresses
8785 in a dispatch table are absolute.
8787 The definition should be a C statement to output to the stdio stream
8788 @var{stream} an assembler pseudo-instruction to generate a reference to
8789 a label. @var{value} is the number of an internal label whose
8790 definition is output using @code{(*targetm.asm_out.internal_label)}.
8794 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
8798 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
8799 Define this if the label before a jump-table needs to be output
8800 specially. The first three arguments are the same as for
8801 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
8802 jump-table which follows (a @code{jump_insn} containing an
8803 @code{addr_vec} or @code{addr_diff_vec}).
8805 This feature is used on system V to output a @code{swbeg} statement
8808 If this macro is not defined, these labels are output with
8809 @code{(*targetm.asm_out.internal_label)}.
8812 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
8813 Define this if something special must be output at the end of a
8814 jump-table. The definition should be a C statement to be executed
8815 after the assembler code for the table is written. It should write
8816 the appropriate code to stdio stream @var{stream}. The argument
8817 @var{table} is the jump-table insn, and @var{num} is the label-number
8818 of the preceding label.
8820 If this macro is not defined, nothing special is output at the end of
8824 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty})
8825 This target hook emits a label at the beginning of each FDE@. It
8826 should be defined on targets where FDEs need special labels, and it
8827 should write the appropriate label, for the FDE associated with the
8828 function declaration @var{decl}, to the stdio stream @var{stream}.
8829 The third argument, @var{for_eh}, is a boolean: true if this is for an
8830 exception table. The fourth argument, @var{empty}, is a boolean:
8831 true if this is a placeholder label for an omitted FDE@.
8833 The default is that FDEs are not given nonlocal labels.
8836 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream})
8837 This target hook emits a label at the beginning of the exception table.
8838 It should be defined on targets where it is desirable for the table
8839 to be broken up according to function.
8841 The default is that no label is emitted.
8844 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality})
8845 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.
8848 @deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx @var{insn})
8849 This target hook emits assembly directives required to unwind the
8850 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
8851 returns @code{UI_TARGET}.
8854 @deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
8855 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.
8858 @node Exception Region Output
8859 @subsection Assembler Commands for Exception Regions
8861 @c prevent bad page break with this line
8863 This describes commands marking the start and the end of an exception
8866 @defmac EH_FRAME_SECTION_NAME
8867 If defined, a C string constant for the name of the section containing
8868 exception handling frame unwind information. If not defined, GCC will
8869 provide a default definition if the target supports named sections.
8870 @file{crtstuff.c} uses this macro to switch to the appropriate section.
8872 You should define this symbol if your target supports DWARF 2 frame
8873 unwind information and the default definition does not work.
8876 @defmac EH_FRAME_IN_DATA_SECTION
8877 If defined, DWARF 2 frame unwind information will be placed in the
8878 data section even though the target supports named sections. This
8879 might be necessary, for instance, if the system linker does garbage
8880 collection and sections cannot be marked as not to be collected.
8882 Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is
8886 @defmac EH_TABLES_CAN_BE_READ_ONLY
8887 Define this macro to 1 if your target is such that no frame unwind
8888 information encoding used with non-PIC code will ever require a
8889 runtime relocation, but the linker may not support merging read-only
8890 and read-write sections into a single read-write section.
8893 @defmac MASK_RETURN_ADDR
8894 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
8895 that it does not contain any extraneous set bits in it.
8898 @defmac DWARF2_UNWIND_INFO
8899 Define this macro to 0 if your target supports DWARF 2 frame unwind
8900 information, but it does not yet work with exception handling.
8901 Otherwise, if your target supports this information (if it defines
8902 @code{INCOMING_RETURN_ADDR_RTX} and @code{OBJECT_FORMAT_ELF}),
8903 GCC will provide a default definition of 1.
8906 @deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts})
8907 This hook defines the mechanism that will be used for exception handling
8908 by the target. If the target has ABI specified unwind tables, the hook
8909 should return @code{UI_TARGET}. If the target is to use the
8910 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
8911 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
8912 information, the hook should return @code{UI_DWARF2}.
8914 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
8915 This may end up simplifying other parts of target-specific code. The
8916 default implementation of this hook never returns @code{UI_NONE}.
8918 Note that the value returned by this hook should be constant. It should
8919 not depend on anything except the command-line switches described by
8920 @var{opts}. In particular, the
8921 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
8922 macros and builtin functions related to exception handling are set up
8923 depending on this setting.
8925 The default implementation of the hook first honors the
8926 @option{--enable-sjlj-exceptions} configure option, then
8927 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If
8928 @code{DWARF2_UNWIND_INFO} depends on command-line options, the target
8929 must define this hook so that @var{opts} is used correctly.
8932 @deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
8933 This variable should be set to @code{true} if the target ABI requires unwinding
8934 tables even when exceptions are not used. It must not be modified by
8935 command-line option processing.
8938 @defmac DONT_USE_BUILTIN_SETJMP
8939 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
8940 should use the @code{setjmp}/@code{longjmp} functions from the C library
8941 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
8944 @defmac JMP_BUF_SIZE
8945 This macro has no effect unless @code{DONT_USE_BUILTIN_SETJMP} is also
8946 defined. Define this macro if the default size of @code{jmp_buf} buffer
8947 for the @code{setjmp}/@code{longjmp}-based exception handling mechanism
8948 is not large enough, or if it is much too large.
8949 The default size is @code{FIRST_PSEUDO_REGISTER * sizeof(void *)}.
8952 @defmac DWARF_CIE_DATA_ALIGNMENT
8953 This macro need only be defined if the target might save registers in the
8954 function prologue at an offset to the stack pointer that is not aligned to
8955 @code{UNITS_PER_WORD}. The definition should be the negative minimum
8956 alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive
8957 minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if
8958 the target supports DWARF 2 frame unwind information.
8961 @deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
8962 Contains the value true if the target should add a zero word onto the
8963 end of a Dwarf-2 frame info section when used for exception handling.
8964 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
8968 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
8969 Given a register, this hook should return a parallel of registers to
8970 represent where to find the register pieces. Define this hook if the
8971 register and its mode are represented in Dwarf in non-contiguous
8972 locations, or if the register should be represented in more than one
8973 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
8974 If not defined, the default is to return @code{NULL_RTX}.
8977 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
8978 If some registers are represented in Dwarf-2 unwind information in
8979 multiple pieces, define this hook to fill in information about the
8980 sizes of those pieces in the table used by the unwinder at runtime.
8981 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
8982 filling in a single size corresponding to each hard register;
8983 @var{address} is the address of the table.
8986 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
8987 This hook is used to output a reference from a frame unwinding table to
8988 the type_info object identified by @var{sym}. It should return @code{true}
8989 if the reference was output. Returning @code{false} will cause the
8990 reference to be output using the normal Dwarf2 routines.
8993 @deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER
8994 This flag should be set to @code{true} on targets that use an ARM EABI
8995 based unwinding library, and @code{false} on other targets. This effects
8996 the format of unwinding tables, and how the unwinder in entered after
8997 running a cleanup. The default is @code{false}.
9000 @node Alignment Output
9001 @subsection Assembler Commands for Alignment
9003 @c prevent bad page break with this line
9004 This describes commands for alignment.
9006 @defmac JUMP_ALIGN (@var{label})
9007 The alignment (log base 2) to put in front of @var{label}, which is
9008 a common destination of jumps and has no fallthru incoming edge.
9010 This macro need not be defined if you don't want any special alignment
9011 to be done at such a time. Most machine descriptions do not currently
9014 Unless it's necessary to inspect the @var{label} parameter, it is better
9015 to set the variable @var{align_jumps} in the target's
9016 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9017 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
9020 @deftypefn {Target Hook} int TARGET_ASM_JUMP_ALIGN_MAX_SKIP (rtx @var{label})
9021 The maximum number of bytes to skip before @var{label} when applying
9022 @code{JUMP_ALIGN}. This works only if
9023 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9026 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
9027 The alignment (log base 2) to put in front of @var{label}, which follows
9030 This macro need not be defined if you don't want any special alignment
9031 to be done at such a time. Most machine descriptions do not currently
9035 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP (rtx @var{label})
9036 The maximum number of bytes to skip before @var{label} when applying
9037 @code{LABEL_ALIGN_AFTER_BARRIER}. This works only if
9038 @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined.
9041 @defmac LOOP_ALIGN (@var{label})
9042 The alignment (log base 2) to put in front of @var{label}, which follows
9043 a @code{NOTE_INSN_LOOP_BEG} note.
9045 This macro need not be defined if you don't want any special alignment
9046 to be done at such a time. Most machine descriptions do not currently
9049 Unless it's necessary to inspect the @var{label} parameter, it is better
9050 to set the variable @code{align_loops} in the target's
9051 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9052 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
9055 @deftypefn {Target Hook} int TARGET_ASM_LOOP_ALIGN_MAX_SKIP (rtx @var{label})
9056 The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to
9057 @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is
9061 @defmac LABEL_ALIGN (@var{label})
9062 The alignment (log base 2) to put in front of @var{label}.
9063 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
9064 the maximum of the specified values is used.
9066 Unless it's necessary to inspect the @var{label} parameter, it is better
9067 to set the variable @code{align_labels} in the target's
9068 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9069 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
9072 @deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_MAX_SKIP (rtx @var{label})
9073 The maximum number of bytes to skip when applying @code{LABEL_ALIGN}
9074 to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN}
9078 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
9079 A C statement to output to the stdio stream @var{stream} an assembler
9080 instruction to advance the location counter by @var{nbytes} bytes.
9081 Those bytes should be zero when loaded. @var{nbytes} will be a C
9082 expression of type @code{unsigned HOST_WIDE_INT}.
9085 @defmac ASM_NO_SKIP_IN_TEXT
9086 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
9087 text section because it fails to put zeros in the bytes that are skipped.
9088 This is true on many Unix systems, where the pseudo--op to skip bytes
9089 produces no-op instructions rather than zeros when used in the text
9093 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9094 A C statement to output to the stdio stream @var{stream} an assembler
9095 command to advance the location counter to a multiple of 2 to the
9096 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9099 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9100 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9101 for padding, if necessary.
9104 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9105 A C statement to output to the stdio stream @var{stream} an assembler
9106 command to advance the location counter to a multiple of 2 to the
9107 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9108 satisfy the alignment request. @var{power} and @var{max_skip} will be
9109 a C expression of type @code{int}.
9113 @node Debugging Info
9114 @section Controlling Debugging Information Format
9116 @c prevent bad page break with this line
9117 This describes how to specify debugging information.
9120 * All Debuggers:: Macros that affect all debugging formats uniformly.
9121 * DBX Options:: Macros enabling specific options in DBX format.
9122 * DBX Hooks:: Hook macros for varying DBX format.
9123 * File Names and DBX:: Macros controlling output of file names in DBX format.
9124 * SDB and DWARF:: Macros for SDB (COFF) and DWARF formats.
9125 * VMS Debug:: Macros for VMS debug format.
9129 @subsection Macros Affecting All Debugging Formats
9131 @c prevent bad page break with this line
9132 These macros affect all debugging formats.
9134 @defmac DBX_REGISTER_NUMBER (@var{regno})
9135 A C expression that returns the DBX register number for the compiler
9136 register number @var{regno}. In the default macro provided, the value
9137 of this expression will be @var{regno} itself. But sometimes there are
9138 some registers that the compiler knows about and DBX does not, or vice
9139 versa. In such cases, some register may need to have one number in the
9140 compiler and another for DBX@.
9142 If two registers have consecutive numbers inside GCC, and they can be
9143 used as a pair to hold a multiword value, then they @emph{must} have
9144 consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}.
9145 Otherwise, debuggers will be unable to access such a pair, because they
9146 expect register pairs to be consecutive in their own numbering scheme.
9148 If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that
9149 does not preserve register pairs, then what you must do instead is
9150 redefine the actual register numbering scheme.
9153 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9154 A C expression that returns the integer offset value for an automatic
9155 variable having address @var{x} (an RTL expression). The default
9156 computation assumes that @var{x} is based on the frame-pointer and
9157 gives the offset from the frame-pointer. This is required for targets
9158 that produce debugging output for DBX or COFF-style debugging output
9159 for SDB and allow the frame-pointer to be eliminated when the
9160 @option{-g} options is used.
9163 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9164 A C expression that returns the integer offset value for an argument
9165 having address @var{x} (an RTL expression). The nominal offset is
9169 @defmac PREFERRED_DEBUGGING_TYPE
9170 A C expression that returns the type of debugging output GCC should
9171 produce when the user specifies just @option{-g}. Define
9172 this if you have arranged for GCC to support more than one format of
9173 debugging output. Currently, the allowable values are @code{DBX_DEBUG},
9174 @code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG},
9175 @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}.
9177 When the user specifies @option{-ggdb}, GCC normally also uses the
9178 value of this macro to select the debugging output format, but with two
9179 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9180 value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is
9181 defined, GCC uses @code{DBX_DEBUG}.
9183 The value of this macro only affects the default debugging output; the
9184 user can always get a specific type of output by using @option{-gstabs},
9185 @option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}.
9189 @subsection Specific Options for DBX Output
9191 @c prevent bad page break with this line
9192 These are specific options for DBX output.
9194 @defmac DBX_DEBUGGING_INFO
9195 Define this macro if GCC should produce debugging output for DBX
9196 in response to the @option{-g} option.
9199 @defmac XCOFF_DEBUGGING_INFO
9200 Define this macro if GCC should produce XCOFF format debugging output
9201 in response to the @option{-g} option. This is a variant of DBX format.
9204 @defmac DEFAULT_GDB_EXTENSIONS
9205 Define this macro to control whether GCC should by default generate
9206 GDB's extended version of DBX debugging information (assuming DBX-format
9207 debugging information is enabled at all). If you don't define the
9208 macro, the default is 1: always generate the extended information
9209 if there is any occasion to.
9212 @defmac DEBUG_SYMS_TEXT
9213 Define this macro if all @code{.stabs} commands should be output while
9214 in the text section.
9217 @defmac ASM_STABS_OP
9218 A C string constant, including spacing, naming the assembler pseudo op to
9219 use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol.
9220 If you don't define this macro, @code{"\t.stabs\t"} is used. This macro
9221 applies only to DBX debugging information format.
9224 @defmac ASM_STABD_OP
9225 A C string constant, including spacing, naming the assembler pseudo op to
9226 use instead of @code{"\t.stabd\t"} to define a debugging symbol whose
9227 value is the current location. If you don't define this macro,
9228 @code{"\t.stabd\t"} is used. This macro applies only to DBX debugging
9232 @defmac ASM_STABN_OP
9233 A C string constant, including spacing, naming the assembler pseudo op to
9234 use instead of @code{"\t.stabn\t"} to define a debugging symbol with no
9235 name. If you don't define this macro, @code{"\t.stabn\t"} is used. This
9236 macro applies only to DBX debugging information format.
9239 @defmac DBX_NO_XREFS
9240 Define this macro if DBX on your system does not support the construct
9241 @samp{xs@var{tagname}}. On some systems, this construct is used to
9242 describe a forward reference to a structure named @var{tagname}.
9243 On other systems, this construct is not supported at all.
9246 @defmac DBX_CONTIN_LENGTH
9247 A symbol name in DBX-format debugging information is normally
9248 continued (split into two separate @code{.stabs} directives) when it
9249 exceeds a certain length (by default, 80 characters). On some
9250 operating systems, DBX requires this splitting; on others, splitting
9251 must not be done. You can inhibit splitting by defining this macro
9252 with the value zero. You can override the default splitting-length by
9253 defining this macro as an expression for the length you desire.
9256 @defmac DBX_CONTIN_CHAR
9257 Normally continuation is indicated by adding a @samp{\} character to
9258 the end of a @code{.stabs} string when a continuation follows. To use
9259 a different character instead, define this macro as a character
9260 constant for the character you want to use. Do not define this macro
9261 if backslash is correct for your system.
9264 @defmac DBX_STATIC_STAB_DATA_SECTION
9265 Define this macro if it is necessary to go to the data section before
9266 outputting the @samp{.stabs} pseudo-op for a non-global static
9270 @defmac DBX_TYPE_DECL_STABS_CODE
9271 The value to use in the ``code'' field of the @code{.stabs} directive
9272 for a typedef. The default is @code{N_LSYM}.
9275 @defmac DBX_STATIC_CONST_VAR_CODE
9276 The value to use in the ``code'' field of the @code{.stabs} directive
9277 for a static variable located in the text section. DBX format does not
9278 provide any ``right'' way to do this. The default is @code{N_FUN}.
9281 @defmac DBX_REGPARM_STABS_CODE
9282 The value to use in the ``code'' field of the @code{.stabs} directive
9283 for a parameter passed in registers. DBX format does not provide any
9284 ``right'' way to do this. The default is @code{N_RSYM}.
9287 @defmac DBX_REGPARM_STABS_LETTER
9288 The letter to use in DBX symbol data to identify a symbol as a parameter
9289 passed in registers. DBX format does not customarily provide any way to
9290 do this. The default is @code{'P'}.
9293 @defmac DBX_FUNCTION_FIRST
9294 Define this macro if the DBX information for a function and its
9295 arguments should precede the assembler code for the function. Normally,
9296 in DBX format, the debugging information entirely follows the assembler
9300 @defmac DBX_BLOCKS_FUNCTION_RELATIVE
9301 Define this macro, with value 1, if the value of a symbol describing
9302 the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be
9303 relative to the start of the enclosing function. Normally, GCC uses
9304 an absolute address.
9307 @defmac DBX_LINES_FUNCTION_RELATIVE
9308 Define this macro, with value 1, if the value of a symbol indicating
9309 the current line number (@code{N_SLINE}) should be relative to the
9310 start of the enclosing function. Normally, GCC uses an absolute address.
9313 @defmac DBX_USE_BINCL
9314 Define this macro if GCC should generate @code{N_BINCL} and
9315 @code{N_EINCL} stabs for included header files, as on Sun systems. This
9316 macro also directs GCC to output a type number as a pair of a file
9317 number and a type number within the file. Normally, GCC does not
9318 generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single
9319 number for a type number.
9323 @subsection Open-Ended Hooks for DBX Format
9325 @c prevent bad page break with this line
9326 These are hooks for DBX format.
9328 @defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter})
9329 A C statement to output DBX debugging information before code for line
9330 number @var{line} of the current source file to the stdio stream
9331 @var{stream}. @var{counter} is the number of time the macro was
9332 invoked, including the current invocation; it is intended to generate
9333 unique labels in the assembly output.
9335 This macro should not be defined if the default output is correct, or
9336 if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}.
9339 @defmac NO_DBX_FUNCTION_END
9340 Some stabs encapsulation formats (in particular ECOFF), cannot handle the
9341 @code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct.
9342 On those machines, define this macro to turn this feature off without
9343 disturbing the rest of the gdb extensions.
9346 @defmac NO_DBX_BNSYM_ENSYM
9347 Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx
9348 extension construct. On those machines, define this macro to turn this
9349 feature off without disturbing the rest of the gdb extensions.
9352 @node File Names and DBX
9353 @subsection File Names in DBX Format
9355 @c prevent bad page break with this line
9356 This describes file names in DBX format.
9358 @defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name})
9359 A C statement to output DBX debugging information to the stdio stream
9360 @var{stream}, which indicates that file @var{name} is the main source
9361 file---the file specified as the input file for compilation.
9362 This macro is called only once, at the beginning of compilation.
9364 This macro need not be defined if the standard form of output
9365 for DBX debugging information is appropriate.
9367 It may be necessary to refer to a label equal to the beginning of the
9368 text section. You can use @samp{assemble_name (stream, ltext_label_name)}
9369 to do so. If you do this, you must also set the variable
9370 @var{used_ltext_label_name} to @code{true}.
9373 @defmac NO_DBX_MAIN_SOURCE_DIRECTORY
9374 Define this macro, with value 1, if GCC should not emit an indication
9375 of the current directory for compilation and current source language at
9376 the beginning of the file.
9379 @defmac NO_DBX_GCC_MARKER
9380 Define this macro, with value 1, if GCC should not emit an indication
9381 that this object file was compiled by GCC@. The default is to emit
9382 an @code{N_OPT} stab at the beginning of every source file, with
9383 @samp{gcc2_compiled.} for the string and value 0.
9386 @defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name})
9387 A C statement to output DBX debugging information at the end of
9388 compilation of the main source file @var{name}. Output should be
9389 written to the stdio stream @var{stream}.
9391 If you don't define this macro, nothing special is output at the end
9392 of compilation, which is correct for most machines.
9395 @defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END
9396 Define this macro @emph{instead of} defining
9397 @code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at
9398 the end of compilation is an @code{N_SO} stab with an empty string,
9399 whose value is the highest absolute text address in the file.
9404 @subsection Macros for SDB and DWARF Output
9406 @c prevent bad page break with this line
9407 Here are macros for SDB and DWARF output.
9409 @defmac SDB_DEBUGGING_INFO
9410 Define this macro if GCC should produce COFF-style debugging output
9411 for SDB in response to the @option{-g} option.
9414 @defmac DWARF2_DEBUGGING_INFO
9415 Define this macro if GCC should produce dwarf version 2 format
9416 debugging output in response to the @option{-g} option.
9418 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function})
9419 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
9420 be emitted for each function. Instead of an integer return the enum
9421 value for the @code{DW_CC_} tag.
9424 To support optional call frame debugging information, you must also
9425 define @code{INCOMING_RETURN_ADDR_RTX} and either set
9426 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
9427 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
9428 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
9431 @defmac DWARF2_FRAME_INFO
9432 Define this macro to a nonzero value if GCC should always output
9433 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
9434 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
9435 exceptions are enabled, GCC will output this information not matter
9436 how you define @code{DWARF2_FRAME_INFO}.
9439 @deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void)
9440 This hook defines the mechanism that will be used for describing frame
9441 unwind information to the debugger. Normally the hook will return
9442 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
9443 return @code{UI_NONE} otherwise.
9445 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
9446 is disabled in order to always output DWARF 2 frame information.
9448 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
9449 This will suppress generation of the normal debug frame unwind information.
9452 @defmac DWARF2_ASM_LINE_DEBUG_INFO
9453 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
9454 line debug info sections. This will result in much more compact line number
9455 tables, and hence is desirable if it works.
9458 @deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS
9459 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.
9462 @deftypevr {Target Hook} bool TARGET_FORCE_AT_COMP_DIR
9463 True if the @code{DW_AT_comp_dir} attribute should be emitted for each compilation unit. This attribute is required for the darwin linker to emit debug information.
9466 @deftypevr {Target Hook} bool TARGET_DELAY_SCHED2
9467 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.
9470 @deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK
9471 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.
9474 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9475 A C statement to issue assembly directives that create a difference
9476 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
9479 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
9480 A C statement to issue assembly directives that create a difference
9481 between the two given labels in system defined units, e.g. instruction
9482 slots on IA64 VMS, using an integer of the given size.
9485 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{section})
9486 A C statement to issue assembly directives that create a
9487 section-relative reference to the given @var{label}, using an integer of the
9488 given @var{size}. The label is known to be defined in the given @var{section}.
9491 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
9492 A C statement to issue assembly directives that create a self-relative
9493 reference to the given @var{label}, using an integer of the given @var{size}.
9496 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
9497 A C statement to issue assembly directives that create a reference to
9498 the DWARF table identifier @var{label} from the current section. This
9499 is used on some systems to avoid garbage collecting a DWARF table which
9500 is referenced by a function.
9503 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x})
9504 If defined, this target hook is a function which outputs a DTP-relative
9505 reference to the given TLS symbol of the specified size.
9508 @defmac PUT_SDB_@dots{}
9509 Define these macros to override the assembler syntax for the special
9510 SDB assembler directives. See @file{sdbout.c} for a list of these
9511 macros and their arguments. If the standard syntax is used, you need
9512 not define them yourself.
9516 Some assemblers do not support a semicolon as a delimiter, even between
9517 SDB assembler directives. In that case, define this macro to be the
9518 delimiter to use (usually @samp{\n}). It is not necessary to define
9519 a new set of @code{PUT_SDB_@var{op}} macros if this is the only change
9523 @defmac SDB_ALLOW_UNKNOWN_REFERENCES
9524 Define this macro to allow references to unknown structure,
9525 union, or enumeration tags to be emitted. Standard COFF does not
9526 allow handling of unknown references, MIPS ECOFF has support for
9530 @defmac SDB_ALLOW_FORWARD_REFERENCES
9531 Define this macro to allow references to structure, union, or
9532 enumeration tags that have not yet been seen to be handled. Some
9533 assemblers choke if forward tags are used, while some require it.
9536 @defmac SDB_OUTPUT_SOURCE_LINE (@var{stream}, @var{line})
9537 A C statement to output SDB debugging information before code for line
9538 number @var{line} of the current source file to the stdio stream
9539 @var{stream}. The default is to emit an @code{.ln} directive.
9544 @subsection Macros for VMS Debug Format
9546 @c prevent bad page break with this line
9547 Here are macros for VMS debug format.
9549 @defmac VMS_DEBUGGING_INFO
9550 Define this macro if GCC should produce debugging output for VMS
9551 in response to the @option{-g} option. The default behavior for VMS
9552 is to generate minimal debug info for a traceback in the absence of
9553 @option{-g} unless explicitly overridden with @option{-g0}. This
9554 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
9555 @code{TARGET_OPTION_OVERRIDE}.
9558 @node Floating Point
9559 @section Cross Compilation and Floating Point
9560 @cindex cross compilation and floating point
9561 @cindex floating point and cross compilation
9563 While all modern machines use twos-complement representation for integers,
9564 there are a variety of representations for floating point numbers. This
9565 means that in a cross-compiler the representation of floating point numbers
9566 in the compiled program may be different from that used in the machine
9567 doing the compilation.
9569 Because different representation systems may offer different amounts of
9570 range and precision, all floating point constants must be represented in
9571 the target machine's format. Therefore, the cross compiler cannot
9572 safely use the host machine's floating point arithmetic; it must emulate
9573 the target's arithmetic. To ensure consistency, GCC always uses
9574 emulation to work with floating point values, even when the host and
9575 target floating point formats are identical.
9577 The following macros are provided by @file{real.h} for the compiler to
9578 use. All parts of the compiler which generate or optimize
9579 floating-point calculations must use these macros. They may evaluate
9580 their operands more than once, so operands must not have side effects.
9582 @defmac REAL_VALUE_TYPE
9583 The C data type to be used to hold a floating point value in the target
9584 machine's format. Typically this is a @code{struct} containing an
9585 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
9589 @deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9590 Compares for equality the two values, @var{x} and @var{y}. If the target
9591 floating point format supports negative zeroes and/or NaNs,
9592 @samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and
9593 @samp{REAL_VALUES_EQUAL (NaN, NaN)} is false.
9596 @deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y})
9597 Tests whether @var{x} is less than @var{y}.
9600 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
9601 Truncates @var{x} to a signed integer, rounding toward zero.
9604 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
9605 Truncates @var{x} to an unsigned integer, rounding toward zero. If
9606 @var{x} is negative, returns zero.
9609 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode})
9610 Converts @var{string} into a floating point number in the target machine's
9611 representation for mode @var{mode}. This routine can handle both
9612 decimal and hexadecimal floating point constants, using the syntax
9613 defined by the C language for both.
9616 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
9617 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
9620 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
9621 Determines whether @var{x} represents infinity (positive or negative).
9624 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
9625 Determines whether @var{x} represents a ``NaN'' (not-a-number).
9628 @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})
9629 Calculates an arithmetic operation on the two floating point values
9630 @var{x} and @var{y}, storing the result in @var{output} (which must be a
9633 The operation to be performed is specified by @var{code}. Only the
9634 following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR},
9635 @code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}.
9637 If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the
9638 target's floating point format cannot represent infinity, it will call
9639 @code{abort}. Callers should check for this situation first, using
9640 @code{MODE_HAS_INFINITIES}. @xref{Storage Layout}.
9643 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
9644 Returns the negative of the floating point value @var{x}.
9647 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
9648 Returns the absolute value of @var{x}.
9651 @deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x})
9652 Converts a floating point value @var{x} into a double-precision integer
9653 which is then stored into @var{low} and @var{high}. If the value is not
9654 integral, it is truncated.
9657 @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})
9658 Converts a double-precision integer found in @var{low} and @var{high},
9659 into a floating point value which is then stored into @var{x}. The
9660 value is truncated to fit in mode @var{mode}.
9663 @node Mode Switching
9664 @section Mode Switching Instructions
9665 @cindex mode switching
9666 The following macros control mode switching optimizations:
9668 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
9669 Define this macro if the port needs extra instructions inserted for mode
9670 switching in an optimizing compilation.
9672 For an example, the SH4 can perform both single and double precision
9673 floating point operations, but to perform a single precision operation,
9674 the FPSCR PR bit has to be cleared, while for a double precision
9675 operation, this bit has to be set. Changing the PR bit requires a general
9676 purpose register as a scratch register, hence these FPSCR sets have to
9677 be inserted before reload, i.e.@: you can't put this into instruction emitting
9678 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
9680 You can have multiple entities that are mode-switched, and select at run time
9681 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
9682 return nonzero for any @var{entity} that needs mode-switching.
9683 If you define this macro, you also have to define
9684 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED},
9685 @code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}.
9686 @code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT}
9690 @defmac NUM_MODES_FOR_MODE_SWITCHING
9691 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
9692 initializer for an array of integers. Each initializer element
9693 N refers to an entity that needs mode switching, and specifies the number
9694 of different modes that might need to be set for this entity.
9695 The position of the initializer in the initializer---starting counting at
9696 zero---determines the integer that is used to refer to the mode-switched
9698 In macros that take mode arguments / yield a mode result, modes are
9699 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
9700 switch is needed / supplied.
9703 @defmac MODE_NEEDED (@var{entity}, @var{insn})
9704 @var{entity} is an integer specifying a mode-switched entity. If
9705 @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to
9706 return an integer value not larger than the corresponding element in
9707 @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must
9708 be switched into prior to the execution of @var{insn}.
9711 @defmac MODE_AFTER (@var{entity}, @var{mode}, @var{insn})
9712 @var{entity} is an integer specifying a mode-switched entity. If
9713 this macro is defined, it is evaluated for every @var{insn} during
9714 mode switching. It determines the mode that an insn results in (if
9715 different from the incoming mode).
9718 @defmac MODE_ENTRY (@var{entity})
9719 If this macro is defined, it is evaluated for every @var{entity} that needs
9720 mode switching. It should evaluate to an integer, which is a mode that
9721 @var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY}
9722 is defined then @code{MODE_EXIT} must be defined.
9725 @defmac MODE_EXIT (@var{entity})
9726 If this macro is defined, it is evaluated for every @var{entity} that needs
9727 mode switching. It should evaluate to an integer, which is a mode that
9728 @var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT}
9729 is defined then @code{MODE_ENTRY} must be defined.
9732 @defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n})
9733 This macro specifies the order in which modes for @var{entity} are processed.
9734 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the
9735 lowest. The value of the macro should be an integer designating a mode
9736 for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode}
9737 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
9738 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
9741 @defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live})
9742 Generate one or more insns to set @var{entity} to @var{mode}.
9743 @var{hard_reg_live} is the set of hard registers live at the point where
9744 the insn(s) are to be inserted.
9747 @node Target Attributes
9748 @section Defining target-specific uses of @code{__attribute__}
9749 @cindex target attributes
9750 @cindex machine attributes
9751 @cindex attributes, target-specific
9753 Target-specific attributes may be defined for functions, data and types.
9754 These are described using the following target hooks; they also need to
9755 be documented in @file{extend.texi}.
9757 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
9758 If defined, this target hook points to an array of @samp{struct
9759 attribute_spec} (defined in @file{tree.h}) specifying the machine
9760 specific attributes for this target and some of the restrictions on the
9761 entities to which these attributes are applied and the arguments they
9765 @deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name})
9766 If defined, this target hook is a function which returns true if the
9767 machine-specific attribute named @var{name} expects an identifier
9768 given as its first argument to be passed on as a plain identifier, not
9769 subjected to name lookup. If this is not defined, the default is
9770 false for all machine-specific attributes.
9773 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2})
9774 If defined, this target hook is a function which returns zero if the attributes on
9775 @var{type1} and @var{type2} are incompatible, one if they are compatible,
9776 and two if they are nearly compatible (which causes a warning to be
9777 generated). If this is not defined, machine-specific attributes are
9778 supposed always to be compatible.
9781 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
9782 If defined, this target hook is a function which assigns default attributes to
9783 the newly defined @var{type}.
9786 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
9787 Define this target hook if the merging of type attributes needs special
9788 handling. If defined, the result is a list of the combined
9789 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
9790 that @code{comptypes} has already been called and returned 1. This
9791 function may call @code{merge_attributes} to handle machine-independent
9795 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
9796 Define this target hook if the merging of decl attributes needs special
9797 handling. If defined, the result is a list of the combined
9798 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
9799 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
9800 when this is needed are when one attribute overrides another, or when an
9801 attribute is nullified by a subsequent definition. This function may
9802 call @code{merge_attributes} to handle machine-independent merging.
9804 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
9805 If the only target-specific handling you require is @samp{dllimport}
9806 for Microsoft Windows targets, you should define the macro
9807 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
9808 will then define a function called
9809 @code{merge_dllimport_decl_attributes} which can then be defined as
9810 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
9811 add @code{handle_dll_attribute} in the attribute table for your port
9812 to perform initial processing of the @samp{dllimport} and
9813 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
9814 @file{i386/i386.c}, for example.
9817 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl})
9818 @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}.
9821 @defmac TARGET_DECLSPEC
9822 Define this macro to a nonzero value if you want to treat
9823 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
9824 default, this behavior is enabled only for targets that define
9825 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
9826 of @code{__declspec} is via a built-in macro, but you should not rely
9827 on this implementation detail.
9830 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
9831 Define this target hook if you want to be able to add attributes to a decl
9832 when it is being created. This is normally useful for back ends which
9833 wish to implement a pragma by using the attributes which correspond to
9834 the pragma's effect. The @var{node} argument is the decl which is being
9835 created. The @var{attr_ptr} argument is a pointer to the attribute list
9836 for this decl. The list itself should not be modified, since it may be
9837 shared with other decls, but attributes may be chained on the head of
9838 the list and @code{*@var{attr_ptr}} modified to point to the new
9839 attributes, or a copy of the list may be made if further changes are
9843 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl})
9845 This target hook returns @code{true} if it is ok to inline @var{fndecl}
9846 into the current function, despite its having target-specific
9847 attributes, @code{false} otherwise. By default, if a function has a
9848 target specific attribute attached to it, it will not be inlined.
9851 @deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags})
9852 This hook is called to parse the @code{attribute(option("..."))}, and
9853 it allows the function to set different target machine compile time
9854 options for the current function that might be different than the
9855 options specified on the command line. The hook should return
9856 @code{true} if the options are valid.
9858 The hook should set the @var{DECL_FUNCTION_SPECIFIC_TARGET} field in
9859 the function declaration to hold a pointer to a target specific
9860 @var{struct cl_target_option} structure.
9863 @deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr})
9864 This hook is called to save any additional target specific information
9865 in the @var{struct cl_target_option} structure for function specific
9867 @xref{Option file format}.
9870 @deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct cl_target_option *@var{ptr})
9871 This hook is called to restore any additional target specific
9872 information in the @var{struct cl_target_option} structure for
9873 function specific options.
9876 @deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr})
9877 This hook is called to print any additional target specific
9878 information in the @var{struct cl_target_option} structure for
9879 function specific options.
9882 @deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target})
9883 This target hook parses the options for @code{#pragma GCC option} to
9884 set the machine specific options for functions that occur later in the
9885 input stream. The options should be the same as handled by the
9886 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
9889 @deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void)
9890 Sometimes certain combinations of command options do not make sense on
9891 a particular target machine. You can override the hook
9892 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
9893 once just after all the command options have been parsed.
9895 Don't use this hook to turn on various extra optimizations for
9896 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
9898 If you need to do something whenever the optimization level is
9899 changed via the optimize attribute or pragma, see
9900 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
9903 @deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee})
9904 This target hook returns @code{false} if the @var{caller} function
9905 cannot inline @var{callee}, based on target specific information. By
9906 default, inlining is not allowed if the callee function has function
9907 specific target options and the caller does not use the same options.
9911 @section Emulating TLS
9912 @cindex Emulated TLS
9914 For targets whose psABI does not provide Thread Local Storage via
9915 specific relocations and instruction sequences, an emulation layer is
9916 used. A set of target hooks allows this emulation layer to be
9917 configured for the requirements of a particular target. For instance
9918 the psABI may in fact specify TLS support in terms of an emulation
9921 The emulation layer works by creating a control object for every TLS
9922 object. To access the TLS object, a lookup function is provided
9923 which, when given the address of the control object, will return the
9924 address of the current thread's instance of the TLS object.
9926 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
9927 Contains the name of the helper function that uses a TLS control
9928 object to locate a TLS instance. The default causes libgcc's
9929 emulated TLS helper function to be used.
9932 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
9933 Contains the name of the helper function that should be used at
9934 program startup to register TLS objects that are implicitly
9935 initialized to zero. If this is @code{NULL}, all TLS objects will
9936 have explicit initializers. The default causes libgcc's emulated TLS
9937 registration function to be used.
9940 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
9941 Contains the name of the section in which TLS control variables should
9942 be placed. The default of @code{NULL} allows these to be placed in
9946 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
9947 Contains the name of the section in which TLS initializers should be
9948 placed. The default of @code{NULL} allows these to be placed in any
9952 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
9953 Contains the prefix to be prepended to TLS control variable names.
9954 The default of @code{NULL} uses a target-specific prefix.
9957 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
9958 Contains the prefix to be prepended to TLS initializer objects. The
9959 default of @code{NULL} uses a target-specific prefix.
9962 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
9963 Specifies a function that generates the FIELD_DECLs for a TLS control
9964 object type. @var{type} is the RECORD_TYPE the fields are for and
9965 @var{name} should be filled with the structure tag, if the default of
9966 @code{__emutls_object} is unsuitable. The default creates a type suitable
9967 for libgcc's emulated TLS function.
9970 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
9971 Specifies a function that generates the CONSTRUCTOR to initialize a
9972 TLS control object. @var{var} is the TLS control object, @var{decl}
9973 is the TLS object and @var{tmpl_addr} is the address of the
9974 initializer. The default initializes libgcc's emulated TLS control object.
9977 @deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED
9978 Specifies whether the alignment of TLS control variable objects is
9979 fixed and should not be increased as some backends may do to optimize
9980 single objects. The default is false.
9983 @deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
9984 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
9985 may be used to describe emulated TLS control objects.
9988 @node MIPS Coprocessors
9989 @section Defining coprocessor specifics for MIPS targets.
9990 @cindex MIPS coprocessor-definition macros
9992 The MIPS specification allows MIPS implementations to have as many as 4
9993 coprocessors, each with as many as 32 private registers. GCC supports
9994 accessing these registers and transferring values between the registers
9995 and memory using asm-ized variables. For example:
9998 register unsigned int cp0count asm ("c0r1");
10004 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
10005 names may be added as described below, or the default names may be
10006 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
10008 Coprocessor registers are assumed to be epilogue-used; sets to them will
10009 be preserved even if it does not appear that the register is used again
10010 later in the function.
10012 Another note: according to the MIPS spec, coprocessor 1 (if present) is
10013 the FPU@. One accesses COP1 registers through standard mips
10014 floating-point support; they are not included in this mechanism.
10016 There is one macro used in defining the MIPS coprocessor interface which
10017 you may want to override in subtargets; it is described below.
10020 @section Parameters for Precompiled Header Validity Checking
10021 @cindex parameters, precompiled headers
10023 @deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
10024 This hook returns a pointer to the data needed by
10025 @code{TARGET_PCH_VALID_P} and sets
10026 @samp{*@var{sz}} to the size of the data in bytes.
10029 @deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
10030 This hook checks whether the options used to create a PCH file are
10031 compatible with the current settings. It returns @code{NULL}
10032 if so and a suitable error message if not. Error messages will
10033 be presented to the user and must be localized using @samp{_(@var{msg})}.
10035 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
10036 when the PCH file was created and @var{sz} is the size of that data in bytes.
10037 It's safe to assume that the data was created by the same version of the
10038 compiler, so no format checking is needed.
10040 The default definition of @code{default_pch_valid_p} should be
10041 suitable for most targets.
10044 @deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
10045 If this hook is nonnull, the default implementation of
10046 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
10047 of @code{target_flags}. @var{pch_flags} specifies the value that
10048 @code{target_flags} had when the PCH file was created. The return
10049 value is the same as for @code{TARGET_PCH_VALID_P}.
10052 @deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void)
10053 Called before writing out a PCH file. If the target has some
10054 garbage-collected data that needs to be in a particular state on PCH loads,
10055 it can use this hook to enforce that state. Very few targets need
10056 to do anything here.
10060 @section C++ ABI parameters
10061 @cindex parameters, c++ abi
10063 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
10064 Define this hook to override the integer type used for guard variables.
10065 These are used to implement one-time construction of static objects. The
10066 default is long_long_integer_type_node.
10069 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
10070 This hook determines how guard variables are used. It should return
10071 @code{false} (the default) if the first byte should be used. A return value of
10072 @code{true} indicates that only the least significant bit should be used.
10075 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
10076 This hook returns the size of the cookie to use when allocating an array
10077 whose elements have the indicated @var{type}. Assumes that it is already
10078 known that a cookie is needed. The default is
10079 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
10080 IA64/Generic C++ ABI@.
10083 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
10084 This hook should return @code{true} if the element size should be stored in
10085 array cookies. The default is to return @code{false}.
10088 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
10089 If defined by a backend this hook allows the decision made to export
10090 class @var{type} to be overruled. Upon entry @var{import_export}
10091 will contain 1 if the class is going to be exported, @minus{}1 if it is going
10092 to be imported and 0 otherwise. This function should return the
10093 modified value and perform any other actions necessary to support the
10094 backend's targeted operating system.
10097 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
10098 This hook should return @code{true} if constructors and destructors return
10099 the address of the object created/destroyed. The default is to return
10103 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
10104 This hook returns true if the key method for a class (i.e., the method
10105 which, if defined in the current translation unit, causes the virtual
10106 table to be emitted) may be an inline function. Under the standard
10107 Itanium C++ ABI the key method may be an inline function so long as
10108 the function is not declared inline in the class definition. Under
10109 some variants of the ABI, an inline function can never be the key
10110 method. The default is to return @code{true}.
10113 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
10114 @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}.
10117 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
10118 This hook returns true (the default) if virtual tables and other
10119 similar implicit class data objects are always COMDAT if they have
10120 external linkage. If this hook returns false, then class data for
10121 classes whose virtual table will be emitted in only one translation
10122 unit will not be COMDAT.
10125 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
10126 This hook returns true (the default) if the RTTI information for
10127 the basic types which is defined in the C++ runtime should always
10128 be COMDAT, false if it should not be COMDAT.
10131 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
10132 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10133 should be used to register static destructors when @option{-fuse-cxa-atexit}
10134 is in effect. The default is to return false to use @code{__cxa_atexit}.
10137 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
10138 This hook returns true if the target @code{atexit} function can be used
10139 in the same manner as @code{__cxa_atexit} to register C++ static
10140 destructors. This requires that @code{atexit}-registered functions in
10141 shared libraries are run in the correct order when the libraries are
10142 unloaded. The default is to return false.
10145 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
10146 @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).
10149 @deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl})
10150 Return target-specific mangling context of @var{decl} or @code{NULL_TREE}.
10153 @node Named Address Spaces
10154 @section Adding support for named address spaces
10155 @cindex named address spaces
10157 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10158 standards committee, @cite{Programming Languages - C - Extensions to
10159 support embedded processors}, specifies a syntax for embedded
10160 processors to specify alternate address spaces. You can configure a
10161 GCC port to support section 5.1 of the draft report to add support for
10162 address spaces other than the default address space. These address
10163 spaces are new keywords that are similar to the @code{volatile} and
10164 @code{const} type attributes.
10166 Pointers to named address spaces can have a different size than
10167 pointers to the generic address space.
10169 For example, the SPU port uses the @code{__ea} address space to refer
10170 to memory in the host processor, rather than memory local to the SPU
10171 processor. Access to memory in the @code{__ea} address space involves
10172 issuing DMA operations to move data between the host processor and the
10173 local processor memory address space. Pointers in the @code{__ea}
10174 address space are either 32 bits or 64 bits based on the
10175 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10178 Internally, address spaces are represented as a small integer in the
10179 range 0 to 15 with address space 0 being reserved for the generic
10182 To register a named address space qualifier keyword with the C front end,
10183 the target may call the @code{c_register_addr_space} routine. For example,
10184 the SPU port uses the following to declare @code{__ea} as the keyword for
10185 named address space #1:
10187 #define ADDR_SPACE_EA 1
10188 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10191 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space})
10192 Define this to return the machine mode to use for pointers to
10193 @var{address_space} if the target supports named address spaces.
10194 The default version of this hook returns @code{ptr_mode} for the
10195 generic address space only.
10198 @deftypefn {Target Hook} {enum machine_mode} TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space})
10199 Define this to return the machine mode to use for addresses in
10200 @var{address_space} if the target supports named address spaces.
10201 The default version of this hook returns @code{Pmode} for the
10202 generic address space only.
10205 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (enum machine_mode @var{mode}, addr_space_t @var{as})
10206 Define this to return nonzero if the port can handle pointers
10207 with machine mode @var{mode} to address space @var{as}. This target
10208 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10209 except that it includes explicit named address space support. The default
10210 version of this hook returns true for the modes returned by either the
10211 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10212 target hooks for the given address space.
10215 @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})
10216 Define this to return true if @var{exp} is a valid address for mode
10217 @var{mode} in the named address space @var{as}. The @var{strict}
10218 parameter says whether strict addressing is in effect after reload has
10219 finished. This target hook is the same as the
10220 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10221 explicit named address space support.
10224 @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})
10225 Define this to modify an invalid address @var{x} to be a valid address
10226 with mode @var{mode} in the named address space @var{as}. This target
10227 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10228 except that it includes explicit named address space support.
10231 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset})
10232 Define this to return whether the @var{subset} named address space is
10233 contained within the @var{superset} named address space. Pointers to
10234 a named address space that is a subset of another named address space
10235 will be converted automatically without a cast if used together in
10236 arithmetic operations. Pointers to a superset address space can be
10237 converted to pointers to a subset address space via explicit casts.
10240 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type})
10241 Define this to convert the pointer expression represented by the RTL
10242 @var{op} with type @var{from_type} that points to a named address
10243 space to a new pointer expression with type @var{to_type} that points
10244 to a different named address space. When this hook it called, it is
10245 guaranteed that one of the two address spaces is a subset of the other,
10246 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10250 @section Miscellaneous Parameters
10251 @cindex parameters, miscellaneous
10253 @c prevent bad page break with this line
10254 Here are several miscellaneous parameters.
10256 @defmac HAS_LONG_COND_BRANCH
10257 Define this boolean macro to indicate whether or not your architecture
10258 has conditional branches that can span all of memory. It is used in
10259 conjunction with an optimization that partitions hot and cold basic
10260 blocks into separate sections of the executable. If this macro is
10261 set to false, gcc will convert any conditional branches that attempt
10262 to cross between sections into unconditional branches or indirect jumps.
10265 @defmac HAS_LONG_UNCOND_BRANCH
10266 Define this boolean macro to indicate whether or not your architecture
10267 has unconditional branches that can span all of memory. It is used in
10268 conjunction with an optimization that partitions hot and cold basic
10269 blocks into separate sections of the executable. If this macro is
10270 set to false, gcc will convert any unconditional branches that attempt
10271 to cross between sections into indirect jumps.
10274 @defmac CASE_VECTOR_MODE
10275 An alias for a machine mode name. This is the machine mode that
10276 elements of a jump-table should have.
10279 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
10280 Optional: return the preferred mode for an @code{addr_diff_vec}
10281 when the minimum and maximum offset are known. If you define this,
10282 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
10283 To make this work, you also have to define @code{INSN_ALIGN} and
10284 make the alignment for @code{addr_diff_vec} explicit.
10285 The @var{body} argument is provided so that the offset_unsigned and scale
10286 flags can be updated.
10289 @defmac CASE_VECTOR_PC_RELATIVE
10290 Define this macro to be a C expression to indicate when jump-tables
10291 should contain relative addresses. You need not define this macro if
10292 jump-tables never contain relative addresses, or jump-tables should
10293 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
10297 @deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void)
10298 This function return the smallest number of different values for which it
10299 is best to use a jump-table instead of a tree of conditional branches.
10300 The default is four for machines with a @code{casesi} instruction and
10301 five otherwise. This is best for most machines.
10304 @defmac WORD_REGISTER_OPERATIONS
10305 Define this macro if operations between registers with integral mode
10306 smaller than a word are always performed on the entire register.
10307 Most RISC machines have this property and most CISC machines do not.
10310 @defmac LOAD_EXTEND_OP (@var{mem_mode})
10311 Define this macro to be a C expression indicating when insns that read
10312 memory in @var{mem_mode}, an integral mode narrower than a word, set the
10313 bits outside of @var{mem_mode} to be either the sign-extension or the
10314 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
10315 of @var{mem_mode} for which the
10316 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
10317 @code{UNKNOWN} for other modes.
10319 This macro is not called with @var{mem_mode} non-integral or with a width
10320 greater than or equal to @code{BITS_PER_WORD}, so you may return any
10321 value in this case. Do not define this macro if it would always return
10322 @code{UNKNOWN}. On machines where this macro is defined, you will normally
10323 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
10325 You may return a non-@code{UNKNOWN} value even if for some hard registers
10326 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
10327 of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero
10328 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
10329 integral mode larger than this but not larger than @code{word_mode}.
10331 You must return @code{UNKNOWN} if for some hard registers that allow this
10332 mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to
10333 @code{word_mode}, but that they can change to another integral mode that
10334 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
10337 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
10338 Define this macro if loading short immediate values into registers sign
10342 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (enum machine_mode @var{mode})
10343 When @option{-ffast-math} is in effect, GCC tries to optimize
10344 divisions by the same divisor, by turning them into multiplications by
10345 the reciprocal. This target hook specifies the minimum number of divisions
10346 that should be there for GCC to perform the optimization for a variable
10347 of mode @var{mode}. The default implementation returns 3 if the machine
10348 has an instruction for the division, and 2 if it does not.
10352 The maximum number of bytes that a single instruction can move quickly
10353 between memory and registers or between two memory locations.
10356 @defmac MAX_MOVE_MAX
10357 The maximum number of bytes that a single instruction can move quickly
10358 between memory and registers or between two memory locations. If this
10359 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
10360 constant value that is the largest value that @code{MOVE_MAX} can have
10364 @defmac SHIFT_COUNT_TRUNCATED
10365 A C expression that is nonzero if on this machine the number of bits
10366 actually used for the count of a shift operation is equal to the number
10367 of bits needed to represent the size of the object being shifted. When
10368 this macro is nonzero, the compiler will assume that it is safe to omit
10369 a sign-extend, zero-extend, and certain bitwise `and' instructions that
10370 truncates the count of a shift operation. On machines that have
10371 instructions that act on bit-fields at variable positions, which may
10372 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
10373 also enables deletion of truncations of the values that serve as
10374 arguments to bit-field instructions.
10376 If both types of instructions truncate the count (for shifts) and
10377 position (for bit-field operations), or if no variable-position bit-field
10378 instructions exist, you should define this macro.
10380 However, on some machines, such as the 80386 and the 680x0, truncation
10381 only applies to shift operations and not the (real or pretended)
10382 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
10383 such machines. Instead, add patterns to the @file{md} file that include
10384 the implied truncation of the shift instructions.
10386 You need not define this macro if it would always have the value of zero.
10389 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
10390 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (enum machine_mode @var{mode})
10391 This function describes how the standard shift patterns for @var{mode}
10392 deal with shifts by negative amounts or by more than the width of the mode.
10393 @xref{shift patterns}.
10395 On many machines, the shift patterns will apply a mask @var{m} to the
10396 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
10397 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
10398 this is true for mode @var{mode}, the function should return @var{m},
10399 otherwise it should return 0. A return value of 0 indicates that no
10400 particular behavior is guaranteed.
10402 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
10403 @emph{not} apply to general shift rtxes; it applies only to instructions
10404 that are generated by the named shift patterns.
10406 The default implementation of this function returns
10407 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
10408 and 0 otherwise. This definition is always safe, but if
10409 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
10410 nevertheless truncate the shift count, you may get better code
10414 @defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec})
10415 A C expression which is nonzero if on this machine it is safe to
10416 ``convert'' an integer of @var{inprec} bits to one of @var{outprec}
10417 bits (where @var{outprec} is smaller than @var{inprec}) by merely
10418 operating on it as if it had only @var{outprec} bits.
10420 On many machines, this expression can be 1.
10422 @c rearranged this, removed the phrase "it is reported that". this was
10423 @c to fix an overfull hbox. --mew 10feb93
10424 When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for
10425 modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result.
10426 If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in
10427 such cases may improve things.
10430 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (enum machine_mode @var{mode}, enum machine_mode @var{rep_mode})
10431 The representation of an integral mode can be such that the values
10432 are always extended to a wider integral mode. Return
10433 @code{SIGN_EXTEND} if values of @var{mode} are represented in
10434 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
10435 otherwise. (Currently, none of the targets use zero-extended
10436 representation this way so unlike @code{LOAD_EXTEND_OP},
10437 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
10438 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
10439 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
10440 widest integral mode and currently we take advantage of this fact.)
10442 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
10443 value even if the extension is not performed on certain hard registers
10444 as long as for the @code{REGNO_REG_CLASS} of these hard registers
10445 @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero.
10447 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
10448 describe two related properties. If you define
10449 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
10450 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
10453 In order to enforce the representation of @code{mode},
10454 @code{TRULY_NOOP_TRUNCATION} should return false when truncating to
10458 @defmac STORE_FLAG_VALUE
10459 A C expression describing the value returned by a comparison operator
10460 with an integral mode and stored by a store-flag instruction
10461 (@samp{cstore@var{mode}4}) when the condition is true. This description must
10462 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
10463 comparison operators whose results have a @code{MODE_INT} mode.
10465 A value of 1 or @minus{}1 means that the instruction implementing the
10466 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
10467 and 0 when the comparison is false. Otherwise, the value indicates
10468 which bits of the result are guaranteed to be 1 when the comparison is
10469 true. This value is interpreted in the mode of the comparison
10470 operation, which is given by the mode of the first operand in the
10471 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
10472 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
10475 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
10476 generate code that depends only on the specified bits. It can also
10477 replace comparison operators with equivalent operations if they cause
10478 the required bits to be set, even if the remaining bits are undefined.
10479 For example, on a machine whose comparison operators return an
10480 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
10481 @samp{0x80000000}, saying that just the sign bit is relevant, the
10485 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
10489 can be converted to
10492 (ashift:SI @var{x} (const_int @var{n}))
10496 where @var{n} is the appropriate shift count to move the bit being
10497 tested into the sign bit.
10499 There is no way to describe a machine that always sets the low-order bit
10500 for a true value, but does not guarantee the value of any other bits,
10501 but we do not know of any machine that has such an instruction. If you
10502 are trying to port GCC to such a machine, include an instruction to
10503 perform a logical-and of the result with 1 in the pattern for the
10504 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
10506 Often, a machine will have multiple instructions that obtain a value
10507 from a comparison (or the condition codes). Here are rules to guide the
10508 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
10513 Use the shortest sequence that yields a valid definition for
10514 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
10515 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
10516 comparison operators to do so because there may be opportunities to
10517 combine the normalization with other operations.
10520 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
10521 slightly preferred on machines with expensive jumps and 1 preferred on
10525 As a second choice, choose a value of @samp{0x80000001} if instructions
10526 exist that set both the sign and low-order bits but do not define the
10530 Otherwise, use a value of @samp{0x80000000}.
10533 Many machines can produce both the value chosen for
10534 @code{STORE_FLAG_VALUE} and its negation in the same number of
10535 instructions. On those machines, you should also define a pattern for
10536 those cases, e.g., one matching
10539 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
10542 Some machines can also perform @code{and} or @code{plus} operations on
10543 condition code values with less instructions than the corresponding
10544 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
10545 machines, define the appropriate patterns. Use the names @code{incscc}
10546 and @code{decscc}, respectively, for the patterns which perform
10547 @code{plus} or @code{minus} operations on condition code values. See
10548 @file{rs6000.md} for some examples. The GNU Superoptimizer can be used to
10549 find such instruction sequences on other machines.
10551 If this macro is not defined, the default value, 1, is used. You need
10552 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
10553 instructions, or if the value generated by these instructions is 1.
10556 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
10557 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
10558 returned when comparison operators with floating-point results are true.
10559 Define this macro on machines that have comparison operations that return
10560 floating-point values. If there are no such operations, do not define
10564 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
10565 A C expression that gives a rtx representing the nonzero true element
10566 for vector comparisons. The returned rtx should be valid for the inner
10567 mode of @var{mode} which is guaranteed to be a vector mode. Define
10568 this macro on machines that have vector comparison operations that
10569 return a vector result. If there are no such operations, do not define
10570 this macro. Typically, this macro is defined as @code{const1_rtx} or
10571 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
10572 the compiler optimizing such vector comparison operations for the
10576 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10577 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
10578 A C expression that indicates whether the architecture defines a value
10579 for @code{clz} or @code{ctz} with a zero operand.
10580 A result of @code{0} indicates the value is undefined.
10581 If the value is defined for only the RTL expression, the macro should
10582 evaluate to @code{1}; if the value applies also to the corresponding optab
10583 entry (which is normally the case if it expands directly into
10584 the corresponding RTL), then the macro should evaluate to @code{2}.
10585 In the cases where the value is defined, @var{value} should be set to
10588 If this macro is not defined, the value of @code{clz} or
10589 @code{ctz} at zero is assumed to be undefined.
10591 This macro must be defined if the target's expansion for @code{ffs}
10592 relies on a particular value to get correct results. Otherwise it
10593 is not necessary, though it may be used to optimize some corner cases, and
10594 to provide a default expansion for the @code{ffs} optab.
10596 Note that regardless of this macro the ``definedness'' of @code{clz}
10597 and @code{ctz} at zero do @emph{not} extend to the builtin functions
10598 visible to the user. Thus one may be free to adjust the value at will
10599 to match the target expansion of these operations without fear of
10604 An alias for the machine mode for pointers. On most machines, define
10605 this to be the integer mode corresponding to the width of a hardware
10606 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
10607 On some machines you must define this to be one of the partial integer
10608 modes, such as @code{PSImode}.
10610 The width of @code{Pmode} must be at least as large as the value of
10611 @code{POINTER_SIZE}. If it is not equal, you must define the macro
10612 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
10616 @defmac FUNCTION_MODE
10617 An alias for the machine mode used for memory references to functions
10618 being called, in @code{call} RTL expressions. On most CISC machines,
10619 where an instruction can begin at any byte address, this should be
10620 @code{QImode}. On most RISC machines, where all instructions have fixed
10621 size and alignment, this should be a mode with the same size and alignment
10622 as the machine instruction words - typically @code{SImode} or @code{HImode}.
10625 @defmac STDC_0_IN_SYSTEM_HEADERS
10626 In normal operation, the preprocessor expands @code{__STDC__} to the
10627 constant 1, to signify that GCC conforms to ISO Standard C@. On some
10628 hosts, like Solaris, the system compiler uses a different convention,
10629 where @code{__STDC__} is normally 0, but is 1 if the user specifies
10630 strict conformance to the C Standard.
10632 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
10633 convention when processing system header files, but when processing user
10634 files @code{__STDC__} will always expand to 1.
10637 @defmac NO_IMPLICIT_EXTERN_C
10638 Define this macro if the system header files support C++ as well as C@.
10639 This macro inhibits the usual method of using system header files in
10640 C++, which is to pretend that the file's contents are enclosed in
10641 @samp{extern "C" @{@dots{}@}}.
10646 @defmac REGISTER_TARGET_PRAGMAS ()
10647 Define this macro if you want to implement any target-specific pragmas.
10648 If defined, it is a C expression which makes a series of calls to
10649 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
10650 for each pragma. The macro may also do any
10651 setup required for the pragmas.
10653 The primary reason to define this macro is to provide compatibility with
10654 other compilers for the same target. In general, we discourage
10655 definition of target-specific pragmas for GCC@.
10657 If the pragma can be implemented by attributes then you should consider
10658 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
10660 Preprocessor macros that appear on pragma lines are not expanded. All
10661 @samp{#pragma} directives that do not match any registered pragma are
10662 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
10665 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10666 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
10668 Each call to @code{c_register_pragma} or
10669 @code{c_register_pragma_with_expansion} establishes one pragma. The
10670 @var{callback} routine will be called when the preprocessor encounters a
10674 #pragma [@var{space}] @var{name} @dots{}
10677 @var{space} is the case-sensitive namespace of the pragma, or
10678 @code{NULL} to put the pragma in the global namespace. The callback
10679 routine receives @var{pfile} as its first argument, which can be passed
10680 on to cpplib's functions if necessary. You can lex tokens after the
10681 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
10682 callback will be silently ignored. The end of the line is indicated by
10683 a token of type @code{CPP_EOF}. Macro expansion occurs on the
10684 arguments of pragmas registered with
10685 @code{c_register_pragma_with_expansion} but not on the arguments of
10686 pragmas registered with @code{c_register_pragma}.
10688 Note that the use of @code{pragma_lex} is specific to the C and C++
10689 compilers. It will not work in the Java or Fortran compilers, or any
10690 other language compilers for that matter. Thus if @code{pragma_lex} is going
10691 to be called from target-specific code, it must only be done so when
10692 building the C and C++ compilers. This can be done by defining the
10693 variables @code{c_target_objs} and @code{cxx_target_objs} in the
10694 target entry in the @file{config.gcc} file. These variables should name
10695 the target-specific, language-specific object file which contains the
10696 code that uses @code{pragma_lex}. Note it will also be necessary to add a
10697 rule to the makefile fragment pointed to by @code{tmake_file} that shows
10698 how to build this object file.
10701 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
10702 Define this macro if macros should be expanded in the
10703 arguments of @samp{#pragma pack}.
10706 @defmac TARGET_DEFAULT_PACK_STRUCT
10707 If your target requires a structure packing default other than 0 (meaning
10708 the machine default), define this macro to the necessary value (in bytes).
10709 This must be a value that would also be valid to use with
10710 @samp{#pragma pack()} (that is, a small power of two).
10713 @defmac DOLLARS_IN_IDENTIFIERS
10714 Define this macro to control use of the character @samp{$} in
10715 identifier names for the C family of languages. 0 means @samp{$} is
10716 not allowed by default; 1 means it is allowed. 1 is the default;
10717 there is no need to define this macro in that case.
10720 @defmac NO_DOLLAR_IN_LABEL
10721 Define this macro if the assembler does not accept the character
10722 @samp{$} in label names. By default constructors and destructors in
10723 G++ have @samp{$} in the identifiers. If this macro is defined,
10724 @samp{.} is used instead.
10727 @defmac NO_DOT_IN_LABEL
10728 Define this macro if the assembler does not accept the character
10729 @samp{.} in label names. By default constructors and destructors in G++
10730 have names that use @samp{.}. If this macro is defined, these names
10731 are rewritten to avoid @samp{.}.
10734 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
10735 Define this macro as a C expression that is nonzero if it is safe for the
10736 delay slot scheduler to place instructions in the delay slot of @var{insn},
10737 even if they appear to use a resource set or clobbered in @var{insn}.
10738 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
10739 every @code{call_insn} has this behavior. On machines where some @code{insn}
10740 or @code{jump_insn} is really a function call and hence has this behavior,
10741 you should define this macro.
10743 You need not define this macro if it would always return zero.
10746 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
10747 Define this macro as a C expression that is nonzero if it is safe for the
10748 delay slot scheduler to place instructions in the delay slot of @var{insn},
10749 even if they appear to set or clobber a resource referenced in @var{insn}.
10750 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
10751 some @code{insn} or @code{jump_insn} is really a function call and its operands
10752 are registers whose use is actually in the subroutine it calls, you should
10753 define this macro. Doing so allows the delay slot scheduler to move
10754 instructions which copy arguments into the argument registers into the delay
10755 slot of @var{insn}.
10757 You need not define this macro if it would always return zero.
10760 @defmac MULTIPLE_SYMBOL_SPACES
10761 Define this macro as a C expression that is nonzero if, in some cases,
10762 global symbols from one translation unit may not be bound to undefined
10763 symbols in another translation unit without user intervention. For
10764 instance, under Microsoft Windows symbols must be explicitly imported
10765 from shared libraries (DLLs).
10767 You need not define this macro if it would always evaluate to zero.
10770 @deftypefn {Target Hook} tree TARGET_MD_ASM_CLOBBERS (tree @var{outputs}, tree @var{inputs}, tree @var{clobbers})
10771 This target hook should add to @var{clobbers} @code{STRING_CST} trees for
10772 any hard regs the port wishes to automatically clobber for an asm.
10773 It should return the result of the last @code{tree_cons} used to add a
10774 clobber. The @var{outputs}, @var{inputs} and @var{clobber} lists are the
10775 corresponding parameters to the asm and may be inspected to avoid
10776 clobbering a register that is an input or output of the asm. You can use
10777 @code{tree_overlaps_hard_reg_set}, declared in @file{tree.h}, to test
10778 for overlap with regards to asm-declared registers.
10781 @defmac MATH_LIBRARY
10782 Define this macro as a C string constant for the linker argument to link
10783 in the system math library, minus the initial @samp{"-l"}, or
10784 @samp{""} if the target does not have a
10785 separate math library.
10787 You need only define this macro if the default of @samp{"m"} is wrong.
10790 @defmac LIBRARY_PATH_ENV
10791 Define this macro as a C string constant for the environment variable that
10792 specifies where the linker should look for libraries.
10794 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
10798 @defmac TARGET_POSIX_IO
10799 Define this macro if the target supports the following POSIX@ file
10800 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
10801 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
10802 to use file locking when exiting a program, which avoids race conditions
10803 if the program has forked. It will also create directories at run-time
10804 for cross-profiling.
10807 @defmac MAX_CONDITIONAL_EXECUTE
10809 A C expression for the maximum number of instructions to execute via
10810 conditional execution instructions instead of a branch. A value of
10811 @code{BRANCH_COST}+1 is the default if the machine does not use cc0, and
10812 1 if it does use cc0.
10815 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
10816 Used if the target needs to perform machine-dependent modifications on the
10817 conditionals used for turning basic blocks into conditionally executed code.
10818 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
10819 contains information about the currently processed blocks. @var{true_expr}
10820 and @var{false_expr} are the tests that are used for converting the
10821 then-block and the else-block, respectively. Set either @var{true_expr} or
10822 @var{false_expr} to a null pointer if the tests cannot be converted.
10825 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
10826 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
10827 if-statements into conditions combined by @code{and} and @code{or} operations.
10828 @var{bb} contains the basic block that contains the test that is currently
10829 being processed and about to be turned into a condition.
10832 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
10833 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
10834 be converted to conditional execution format. @var{ce_info} points to
10835 a data structure, @code{struct ce_if_block}, which contains information
10836 about the currently processed blocks.
10839 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
10840 A C expression to perform any final machine dependent modifications in
10841 converting code to conditional execution. The involved basic blocks
10842 can be found in the @code{struct ce_if_block} structure that is pointed
10843 to by @var{ce_info}.
10846 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
10847 A C expression to cancel any machine dependent modifications in
10848 converting code to conditional execution. The involved basic blocks
10849 can be found in the @code{struct ce_if_block} structure that is pointed
10850 to by @var{ce_info}.
10853 @defmac IFCVT_MACHDEP_INIT (@var{ce_info})
10854 A C expression to initialize any machine specific data for if-conversion
10855 of the if-block in the @code{struct ce_if_block} structure that is pointed
10856 to by @var{ce_info}.
10859 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void)
10860 If non-null, this hook performs a target-specific pass over the
10861 instruction stream. The compiler will run it at all optimization levels,
10862 just before the point at which it normally does delayed-branch scheduling.
10864 The exact purpose of the hook varies from target to target. Some use
10865 it to do transformations that are necessary for correctness, such as
10866 laying out in-function constant pools or avoiding hardware hazards.
10867 Others use it as an opportunity to do some machine-dependent optimizations.
10869 You need not implement the hook if it has nothing to do. The default
10870 definition is null.
10873 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void)
10874 Define this hook if you have any machine-specific built-in functions
10875 that need to be defined. It should be a function that performs the
10878 Machine specific built-in functions can be useful to expand special machine
10879 instructions that would otherwise not normally be generated because
10880 they have no equivalent in the source language (for example, SIMD vector
10881 instructions or prefetch instructions).
10883 To create a built-in function, call the function
10884 @code{lang_hooks.builtin_function}
10885 which is defined by the language front end. You can use any type nodes set
10886 up by @code{build_common_tree_nodes};
10887 only language front ends that use those two functions will call
10888 @samp{TARGET_INIT_BUILTINS}.
10891 @deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p})
10892 Define this hook if you have any machine-specific built-in functions
10893 that need to be defined. It should be a function that returns the
10894 builtin function declaration for the builtin function code @var{code}.
10895 If there is no such builtin and it cannot be initialized at this time
10896 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
10897 If @var{code} is out of range the function should return
10898 @code{error_mark_node}.
10901 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore})
10903 Expand a call to a machine specific built-in function that was set up by
10904 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
10905 function call; the result should go to @var{target} if that is
10906 convenient, and have mode @var{mode} if that is convenient.
10907 @var{subtarget} may be used as the target for computing one of
10908 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
10909 ignored. This function should return the result of the call to the
10913 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist})
10914 Select a replacement for a machine specific built-in function that
10915 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
10916 @emph{before} regular type checking, and so allows the target to
10917 implement a crude form of function overloading. @var{fndecl} is the
10918 declaration of the built-in function. @var{arglist} is the list of
10919 arguments passed to the built-in function. The result is a
10920 complete expression that implements the operation, usually
10921 another @code{CALL_EXPR}.
10922 @var{arglist} really has type @samp{VEC(tree,gc)*}
10925 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore})
10926 Fold a call to a machine specific built-in function that was set up by
10927 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
10928 built-in function. @var{n_args} is the number of arguments passed to
10929 the function; the arguments themselves are pointed to by @var{argp}.
10930 The result is another tree containing a simplified expression for the
10931 call's result. If @var{ignore} is true the value will be ignored.
10934 @deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const_rtx @var{insn})
10936 Take an instruction in @var{insn} and return NULL if it is valid within a
10937 low-overhead loop, otherwise return a string explaining why doloop
10938 could not be applied.
10940 Many targets use special registers for low-overhead looping. For any
10941 instruction that clobbers these this function should return a string indicating
10942 the reason why the doloop could not be applied.
10943 By default, the RTL loop optimizer does not use a present doloop pattern for
10944 loops containing function calls or branch on table instructions.
10947 @defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2})
10949 Take a branch insn in @var{branch1} and another in @var{branch2}.
10950 Return true if redirecting @var{branch1} to the destination of
10951 @var{branch2} is possible.
10953 On some targets, branches may have a limited range. Optimizing the
10954 filling of delay slots can result in branches being redirected, and this
10955 may in turn cause a branch offset to overflow.
10958 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code})
10959 This target hook returns @code{true} if @var{x} is considered to be commutative.
10960 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
10961 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
10962 of the enclosing rtl, if known, otherwise it is UNKNOWN.
10965 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
10967 When the initial value of a hard register has been copied in a pseudo
10968 register, it is often not necessary to actually allocate another register
10969 to this pseudo register, because the original hard register or a stack slot
10970 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
10971 is called at the start of register allocation once for each hard register
10972 that had its initial value copied by using
10973 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
10974 Possible values are @code{NULL_RTX}, if you don't want
10975 to do any special allocation, a @code{REG} rtx---that would typically be
10976 the hard register itself, if it is known not to be clobbered---or a
10978 If you are returning a @code{MEM}, this is only a hint for the allocator;
10979 it might decide to use another register anyways.
10980 You may use @code{current_function_leaf_function} in the hook, functions
10981 that use @code{REG_N_SETS}, to determine if the hard
10982 register in question will not be clobbered.
10983 The default value of this hook is @code{NULL}, which disables any special
10987 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
10988 This target hook returns nonzero if @var{x}, an @code{unspec} or
10989 @code{unspec_volatile} operation, might cause a trap. Targets can use
10990 this hook to enhance precision of analysis for @code{unspec} and
10991 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
10992 to analyze inner elements of @var{x} in which case @var{flags} should be
10996 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
10997 The compiler invokes this hook whenever it changes its current function
10998 context (@code{cfun}). You can define this function if
10999 the back end needs to perform any initialization or reset actions on a
11000 per-function basis. For example, it may be used to implement function
11001 attributes that affect register usage or code generation patterns.
11002 The argument @var{decl} is the declaration for the new function context,
11003 and may be null to indicate that the compiler has left a function context
11004 and is returning to processing at the top level.
11005 The default hook function does nothing.
11007 GCC sets @code{cfun} to a dummy function context during initialization of
11008 some parts of the back end. The hook function is not invoked in this
11009 situation; you need not worry about the hook being invoked recursively,
11010 or when the back end is in a partially-initialized state.
11011 @code{cfun} might be @code{NULL} to indicate processing at top level,
11012 outside of any function scope.
11015 @defmac TARGET_OBJECT_SUFFIX
11016 Define this macro to be a C string representing the suffix for object
11017 files on your target machine. If you do not define this macro, GCC will
11018 use @samp{.o} as the suffix for object files.
11021 @defmac TARGET_EXECUTABLE_SUFFIX
11022 Define this macro to be a C string representing the suffix to be
11023 automatically added to executable files on your target machine. If you
11024 do not define this macro, GCC will use the null string as the suffix for
11028 @defmac COLLECT_EXPORT_LIST
11029 If defined, @code{collect2} will scan the individual object files
11030 specified on its command line and create an export list for the linker.
11031 Define this macro for systems like AIX, where the linker discards
11032 object files that are not referenced from @code{main} and uses export
11036 @defmac MODIFY_JNI_METHOD_CALL (@var{mdecl})
11037 Define this macro to a C expression representing a variant of the
11038 method call @var{mdecl}, if Java Native Interface (JNI) methods
11039 must be invoked differently from other methods on your target.
11040 For example, on 32-bit Microsoft Windows, JNI methods must be invoked using
11041 the @code{stdcall} calling convention and this macro is then
11042 defined as this expression:
11045 build_type_attribute_variant (@var{mdecl},
11047 (get_identifier ("stdcall"),
11052 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
11053 This target hook returns @code{true} past the point in which new jump
11054 instructions could be created. On machines that require a register for
11055 every jump such as the SHmedia ISA of SH5, this point would typically be
11056 reload, so this target hook should be defined to a function such as:
11060 cannot_modify_jumps_past_reload_p ()
11062 return (reload_completed || reload_in_progress);
11067 @deftypefn {Target Hook} reg_class_t TARGET_BRANCH_TARGET_REGISTER_CLASS (void)
11068 This target hook returns a register class for which branch target register
11069 optimizations should be applied. All registers in this class should be
11070 usable interchangeably. After reload, registers in this class will be
11071 re-allocated and loads will be hoisted out of loops and be subjected
11072 to inter-block scheduling.
11075 @deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen})
11076 Branch target register optimization will by default exclude callee-saved
11078 that are not already live during the current function; if this target hook
11079 returns true, they will be included. The target code must than make sure
11080 that all target registers in the class returned by
11081 @samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are
11082 saved. @var{after_prologue_epilogue_gen} indicates if prologues and
11083 epilogues have already been generated. Note, even if you only return
11084 true when @var{after_prologue_epilogue_gen} is false, you still are likely
11085 to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET}
11086 to reserve space for caller-saved target registers.
11089 @deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
11090 This target hook returns true if the target supports conditional execution.
11091 This target hook is required only when the target has several different
11092 modes and they have different conditional execution capability, such as ARM.
11095 @deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, struct loop *@var{loop})
11096 This target hook returns a new value for the number of times @var{loop}
11097 should be unrolled. The parameter @var{nunroll} is the number of times
11098 the loop is to be unrolled. The parameter @var{loop} is a pointer to
11099 the loop, which is going to be checked for unrolling. This target hook
11100 is required only when the target has special constraints like maximum
11101 number of memory accesses.
11104 @defmac POWI_MAX_MULTS
11105 If defined, this macro is interpreted as a signed integer C expression
11106 that specifies the maximum number of floating point multiplications
11107 that should be emitted when expanding exponentiation by an integer
11108 constant inline. When this value is defined, exponentiation requiring
11109 more than this number of multiplications is implemented by calling the
11110 system library's @code{pow}, @code{powf} or @code{powl} routines.
11111 The default value places no upper bound on the multiplication count.
11114 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11115 This target hook should register any extra include files for the
11116 target. The parameter @var{stdinc} indicates if normal include files
11117 are present. The parameter @var{sysroot} is the system root directory.
11118 The parameter @var{iprefix} is the prefix for the gcc directory.
11121 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
11122 This target hook should register any extra include files for the
11123 target before any standard headers. The parameter @var{stdinc}
11124 indicates if normal include files are present. The parameter
11125 @var{sysroot} is the system root directory. The parameter
11126 @var{iprefix} is the prefix for the gcc directory.
11129 @deftypefn Macro void TARGET_OPTF (char *@var{path})
11130 This target hook should register special include paths for the target.
11131 The parameter @var{path} is the include to register. On Darwin
11132 systems, this is used for Framework includes, which have semantics
11133 that are different from @option{-I}.
11136 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
11137 This target macro returns @code{true} if it is safe to use a local alias
11138 for a virtual function @var{fndecl} when constructing thunks,
11139 @code{false} otherwise. By default, the macro returns @code{true} for all
11140 functions, if a target supports aliases (i.e.@: defines
11141 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
11144 @defmac TARGET_FORMAT_TYPES
11145 If defined, this macro is the name of a global variable containing
11146 target-specific format checking information for the @option{-Wformat}
11147 option. The default is to have no target-specific format checks.
11150 @defmac TARGET_N_FORMAT_TYPES
11151 If defined, this macro is the number of entries in
11152 @code{TARGET_FORMAT_TYPES}.
11155 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
11156 If defined, this macro is the name of a global variable containing
11157 target-specific format overrides for the @option{-Wformat} option. The
11158 default is to have no target-specific format overrides. If defined,
11159 @code{TARGET_FORMAT_TYPES} must be defined, too.
11162 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
11163 If defined, this macro specifies the number of entries in
11164 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
11167 @defmac TARGET_OVERRIDES_FORMAT_INIT
11168 If defined, this macro specifies the optional initialization
11169 routine for target specific customizations of the system printf
11170 and scanf formatter settings.
11173 @deftypevr {Target Hook} bool TARGET_RELAXED_ORDERING
11174 If set to @code{true}, means that the target's memory model does not
11175 guarantee that loads which do not depend on one another will access
11176 main memory in the order of the instruction stream; if ordering is
11177 important, an explicit memory barrier must be used. This is true of
11178 many recent processors which implement a policy of ``relaxed,''
11179 ``weak,'' or ``release'' memory consistency, such as Alpha, PowerPC,
11180 and ia64. The default is @code{false}.
11183 @deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val})
11184 If defined, this macro returns the diagnostic message when it is
11185 illegal to pass argument @var{val} to function @var{funcdecl}
11186 with prototype @var{typelist}.
11189 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype})
11190 If defined, this macro returns the diagnostic message when it is
11191 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
11192 if validity should be determined by the front end.
11195 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type})
11196 If defined, this macro returns the diagnostic message when it is
11197 invalid to apply operation @var{op} (where unary plus is denoted by
11198 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
11199 if validity should be determined by the front end.
11202 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2})
11203 If defined, this macro returns the diagnostic message when it is
11204 invalid to apply operation @var{op} to operands of types @var{type1}
11205 and @var{type2}, or @code{NULL} if validity should be determined by
11209 @deftypefn {Target Hook} {const char *} TARGET_INVALID_PARAMETER_TYPE (const_tree @var{type})
11210 If defined, this macro returns the diagnostic message when it is
11211 invalid for functions to include parameters of type @var{type},
11212 or @code{NULL} if validity should be determined by
11213 the front end. This is currently used only by the C and C++ front ends.
11216 @deftypefn {Target Hook} {const char *} TARGET_INVALID_RETURN_TYPE (const_tree @var{type})
11217 If defined, this macro returns the diagnostic message when it is
11218 invalid for functions to have return type @var{type},
11219 or @code{NULL} if validity should be determined by
11220 the front end. This is currently used only by the C and C++ front ends.
11223 @deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type})
11224 If defined, this target hook returns the type to which values of
11225 @var{type} should be promoted when they appear in expressions,
11226 analogous to the integer promotions, or @code{NULL_TREE} to use the
11227 front end's normal promotion rules. This hook is useful when there are
11228 target-specific types with special promotion rules.
11229 This is currently used only by the C and C++ front ends.
11232 @deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr})
11233 If defined, this hook returns the result of converting @var{expr} to
11234 @var{type}. It should return the converted expression,
11235 or @code{NULL_TREE} to apply the front end's normal conversion rules.
11236 This hook is useful when there are target-specific types with special
11238 This is currently used only by the C and C++ front ends.
11241 @defmac TARGET_USE_JCR_SECTION
11242 This macro determines whether to use the JCR section to register Java
11243 classes. By default, TARGET_USE_JCR_SECTION is defined to 1 if both
11244 SUPPORTS_WEAK and TARGET_HAVE_NAMED_SECTIONS are true, else 0.
11248 This macro determines the size of the objective C jump buffer for the
11249 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
11252 @defmac LIBGCC2_UNWIND_ATTRIBUTE
11253 Define this macro if any target-specific attributes need to be attached
11254 to the functions in @file{libgcc} that provide low-level support for
11255 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
11256 and the associated definitions of those functions.
11259 @deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void)
11260 Define this macro to update the current function stack boundary if
11264 @deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void)
11265 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
11266 different argument pointer register is needed to access the function's
11267 argument list due to stack realignment. Return @code{NULL} if no DRAP
11271 @deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
11272 When optimization is disabled, this hook indicates whether or not
11273 arguments should be allocated to stack slots. Normally, GCC allocates
11274 stacks slots for arguments when not optimizing in order to make
11275 debugging easier. However, when a function is declared with
11276 @code{__attribute__((naked))}, there is no stack frame, and the compiler
11277 cannot safely move arguments from the registers in which they are passed
11278 to the stack. Therefore, this hook should return true in general, but
11279 false for naked functions. The default implementation always returns true.
11282 @deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR
11283 On some architectures it can take multiple instructions to synthesize
11284 a constant. If there is another constant already in a register that
11285 is close enough in value then it is preferable that the new constant
11286 is computed from this register using immediate addition or
11287 subtraction. We accomplish this through CSE. Besides the value of
11288 the constant we also add a lower and an upper constant anchor to the
11289 available expressions. These are then queried when encountering new
11290 constants. The anchors are computed by rounding the constant up and
11291 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
11292 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
11293 accepted by immediate-add plus one. We currently assume that the
11294 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
11295 MIPS, where add-immediate takes a 16-bit signed value,
11296 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
11297 is zero, which disables this optimization. @end deftypevr
11299 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_MEMMODEL_CHECK (unsigned HOST_WIDE_INT @var{val})
11300 Validate target specific memory model mask bits. When NULL no target specific
11301 memory model bits are allowed.
11304 @deftypevr {Target Hook} {unsigned char} TARGET_ATOMIC_TEST_AND_SET_TRUEVAL
11305 This value should be set if the result written by @code{atomic_test_and_set} is not exactly 1, i.e. the @code{bool} @code{true}.