1 @c Copyright (C) 1988-2022 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter Target Description Macros and Functions
7 @cindex machine description macros
8 @cindex target description macros
9 @cindex macros, target description
10 @cindex @file{tm.h} macros
12 In addition to the file @file{@var{machine}.md}, a machine description
13 includes a C header file conventionally given the name
14 @file{@var{machine}.h} and a C source file named @file{@var{machine}.c}.
15 The header file defines numerous macros that convey the information
16 about the target machine that does not fit into the scheme of the
17 @file{.md} file. The file @file{tm.h} should be a link to
18 @file{@var{machine}.h}. The header file @file{config.h} includes
19 @file{tm.h} and most compiler source files include @file{config.h}. The
20 source file defines a variable @code{targetm}, which is a structure
21 containing pointers to functions and data relating to the target
22 machine. @file{@var{machine}.c} should also contain their definitions,
23 if they are not defined elsewhere in GCC, and other functions called
24 through the macros defined in the @file{.h} file.
27 * Target Structure:: The @code{targetm} variable.
28 * Driver:: Controlling how the driver runs the compilation passes.
29 * Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}.
30 * Per-Function Data:: Defining data structures for per-function information.
31 * Storage Layout:: Defining sizes and alignments of data.
32 * Type Layout:: Defining sizes and properties of basic user data types.
33 * Registers:: Naming and describing the hardware registers.
34 * Register Classes:: Defining the classes of hardware registers.
35 * Stack and Calling:: Defining which way the stack grows and by how much.
36 * Varargs:: Defining the varargs macros.
37 * Trampolines:: Code set up at run time to enter a nested function.
38 * Library Calls:: Controlling how library routines are implicitly called.
39 * Addressing Modes:: Defining addressing modes valid for memory operands.
40 * Anchored Addresses:: Defining how @option{-fsection-anchors} should work.
41 * Condition Code:: Defining how insns update the condition code.
42 * Costs:: Defining relative costs of different operations.
43 * Scheduling:: Adjusting the behavior of the instruction scheduler.
44 * Sections:: Dividing storage into text, data, and other sections.
45 * PIC:: Macros for position independent code.
46 * Assembler Format:: Defining how to write insns and pseudo-ops to output.
47 * Debugging Info:: Defining the format of debugging output.
48 * Floating Point:: Handling floating point for cross-compilers.
49 * Mode Switching:: Insertion of mode-switching instructions.
50 * Target Attributes:: Defining target-specific uses of @code{__attribute__}.
51 * Emulated TLS:: Emulated TLS support.
52 * MIPS Coprocessors:: MIPS coprocessor support and how to customize it.
53 * PCH Target:: Validity checking for precompiled headers.
54 * C++ ABI:: Controlling C++ ABI changes.
55 * D Language and ABI:: Controlling D ABI changes.
56 * Named Address Spaces:: Adding support for named address spaces
57 * Misc:: Everything else.
60 @node Target Structure
61 @section The Global @code{targetm} Variable
63 @cindex target functions
65 @deftypevar {struct gcc_target} targetm
66 The target @file{.c} file must define the global @code{targetm} variable
67 which contains pointers to functions and data relating to the target
68 machine. The variable is declared in @file{target.h};
69 @file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is
70 used to initialize the variable, and macros for the default initializers
71 for elements of the structure. The @file{.c} file should override those
72 macros for which the default definition is inappropriate. For example:
75 #include "target-def.h"
77 /* @r{Initialize the GCC target structure.} */
79 #undef TARGET_COMP_TYPE_ATTRIBUTES
80 #define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes
82 struct gcc_target targetm = TARGET_INITIALIZER;
86 Where a macro should be defined in the @file{.c} file in this manner to
87 form part of the @code{targetm} structure, it is documented below as a
88 ``Target Hook'' with a prototype. Many macros will change in future
89 from being defined in the @file{.h} file to being part of the
90 @code{targetm} structure.
92 Similarly, there is a @code{targetcm} variable for hooks that are
93 specific to front ends for C-family languages, documented as ``C
94 Target Hook''. This is declared in @file{c-family/c-target.h}, the
95 initializer @code{TARGETCM_INITIALIZER} in
96 @file{c-family/c-target-def.h}. If targets initialize @code{targetcm}
97 themselves, they should set @code{target_has_targetcm=yes} in
98 @file{config.gcc}; otherwise a default definition is used.
100 Similarly, there is a @code{targetm_common} variable for hooks that
101 are shared between the compiler driver and the compilers proper,
102 documented as ``Common Target Hook''. This is declared in
103 @file{common/common-target.h}, the initializer
104 @code{TARGETM_COMMON_INITIALIZER} in
105 @file{common/common-target-def.h}. If targets initialize
106 @code{targetm_common} themselves, they should set
107 @code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a
108 default definition is used.
110 Similarly, there is a @code{targetdm} variable for hooks that are
111 specific to the D language front end, documented as ``D Target Hook''.
112 This is declared in @file{d/d-target.h}, the initializer
113 @code{TARGETDM_INITIALIZER} in @file{d/d-target-def.h}. If targets
114 initialize @code{targetdm} themselves, they should set
115 @code{target_has_targetdm=yes} in @file{config.gcc}; otherwise a default
119 @section Controlling the Compilation Driver, @file{gcc}
121 @cindex controlling the compilation driver
123 @c prevent bad page break with this line
124 You can control the compilation driver.
126 @defmac DRIVER_SELF_SPECS
127 A list of specs for the driver itself. It should be a suitable
128 initializer for an array of strings, with no surrounding braces.
130 The driver applies these specs to its own command line between loading
131 default @file{specs} files (but not command-line specified ones) and
132 choosing the multilib directory or running any subcommands. It
133 applies them in the order given, so each spec can depend on the
134 options added by earlier ones. It is also possible to remove options
135 using @samp{%<@var{option}} in the usual way.
137 This macro can be useful when a port has several interdependent target
138 options. It provides a way of standardizing the command line so
139 that the other specs are easier to write.
141 Do not define this macro if it does not need to do anything.
144 @defmac OPTION_DEFAULT_SPECS
145 A list of specs used to support configure-time default options (i.e.@:
146 @option{--with} options) in the driver. It should be a suitable initializer
147 for an array of structures, each containing two strings, without the
148 outermost pair of surrounding braces.
150 The first item in the pair is the name of the default. This must match
151 the code in @file{config.gcc} for the target. The second item is a spec
152 to apply if a default with this name was specified. The string
153 @samp{%(VALUE)} in the spec will be replaced by the value of the default
154 everywhere it occurs.
156 The driver will apply these specs to its own command line between loading
157 default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using
158 the same mechanism as @code{DRIVER_SELF_SPECS}.
160 Do not define this macro if it does not need to do anything.
164 A C string constant that tells the GCC driver program options to
165 pass to CPP@. It can also specify how to translate options you
166 give to GCC into options for GCC to pass to the CPP@.
168 Do not define this macro if it does not need to do anything.
171 @defmac CPLUSPLUS_CPP_SPEC
172 This macro is just like @code{CPP_SPEC}, but is used for C++, rather
173 than C@. If you do not define this macro, then the value of
174 @code{CPP_SPEC} (if any) will be used instead.
178 A C string constant that tells the GCC driver program options to
179 pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language
181 It can also specify how to translate options you give to GCC into options
182 for GCC to pass to front ends.
184 Do not define this macro if it does not need to do anything.
188 A C string constant that tells the GCC driver program options to
189 pass to @code{cc1plus}. It can also specify how to translate options you
190 give to GCC into options for GCC to pass to the @code{cc1plus}.
192 Do not define this macro if it does not need to do anything.
193 Note that everything defined in CC1_SPEC is already passed to
194 @code{cc1plus} so there is no need to duplicate the contents of
195 CC1_SPEC in CC1PLUS_SPEC@.
199 A C string constant that tells the GCC driver program options to
200 pass to the assembler. It can also specify how to translate options
201 you give to GCC into options for GCC to pass to the assembler.
202 See the file @file{sun3.h} for an example of this.
204 Do not define this macro if it does not need to do anything.
207 @defmac ASM_FINAL_SPEC
208 A C string constant that tells the GCC driver program how to
209 run any programs which cleanup after the normal assembler.
210 Normally, this is not needed. See the file @file{mips.h} for
213 Do not define this macro if it does not need to do anything.
216 @defmac AS_NEEDS_DASH_FOR_PIPED_INPUT
217 Define this macro, with no value, if the driver should give the assembler
218 an argument consisting of a single dash, @option{-}, to instruct it to
219 read from its standard input (which will be a pipe connected to the
220 output of the compiler proper). This argument is given after any
221 @option{-o} option specifying the name of the output file.
223 If you do not define this macro, the assembler is assumed to read its
224 standard input if given no non-option arguments. If your assembler
225 cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct;
226 see @file{mips.h} for instance.
230 A C string constant that tells the GCC driver program options to
231 pass to the linker. It can also specify how to translate options you
232 give to GCC into options for GCC to pass to the linker.
234 Do not define this macro if it does not need to do anything.
238 Another C string constant used much like @code{LINK_SPEC}. The difference
239 between the two is that @code{LIB_SPEC} is used at the end of the
240 command given to the linker.
242 If this macro is not defined, a default is provided that
243 loads the standard C library from the usual place. See @file{gcc.cc}.
247 Another C string constant that tells the GCC driver program
248 how and when to place a reference to @file{libgcc.a} into the
249 linker command line. This constant is placed both before and after
250 the value of @code{LIB_SPEC}.
252 If this macro is not defined, the GCC driver provides a default that
253 passes the string @option{-lgcc} to the linker.
256 @defmac REAL_LIBGCC_SPEC
257 By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the
258 @code{LIBGCC_SPEC} is not directly used by the driver program but is
259 instead modified to refer to different versions of @file{libgcc.a}
260 depending on the values of the command line flags @option{-static},
261 @option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On
262 targets where these modifications are inappropriate, define
263 @code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the
264 driver how to place a reference to @file{libgcc} on the link command
265 line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified.
268 @defmac USE_LD_AS_NEEDED
269 A macro that controls the modifications to @code{LIBGCC_SPEC}
270 mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be
271 generated that uses @option{--as-needed} or equivalent options and the
272 shared @file{libgcc} in place of the
273 static exception handler library, when linking without any of
274 @code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}.
278 If defined, this C string constant is added to @code{LINK_SPEC}.
279 When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects
280 the modifications to @code{LIBGCC_SPEC} mentioned in
281 @code{REAL_LIBGCC_SPEC}.
284 @defmac STARTFILE_SPEC
285 Another C string constant used much like @code{LINK_SPEC}. The
286 difference between the two is that @code{STARTFILE_SPEC} is used at
287 the very beginning of the command given to the linker.
289 If this macro is not defined, a default is provided that loads the
290 standard C startup file from the usual place. See @file{gcc.cc}.
294 Another C string constant used much like @code{LINK_SPEC}. The
295 difference between the two is that @code{ENDFILE_SPEC} is used at
296 the very end of the command given to the linker.
298 Do not define this macro if it does not need to do anything.
301 @defmac THREAD_MODEL_SPEC
302 GCC @code{-v} will print the thread model GCC was configured to use.
303 However, this doesn't work on platforms that are multilibbed on thread
304 models, such as AIX 4.3. On such platforms, define
305 @code{THREAD_MODEL_SPEC} such that it evaluates to a string without
306 blanks that names one of the recognized thread models. @code{%*}, the
307 default value of this macro, will expand to the value of
308 @code{thread_file} set in @file{config.gcc}.
311 @defmac SYSROOT_SUFFIX_SPEC
312 Define this macro to add a suffix to the target sysroot when GCC is
313 configured with a sysroot. This will cause GCC to search for usr/lib,
314 et al, within sysroot+suffix.
317 @defmac SYSROOT_HEADERS_SUFFIX_SPEC
318 Define this macro to add a headers_suffix to the target sysroot when
319 GCC is configured with a sysroot. This will cause GCC to pass the
320 updated sysroot+headers_suffix to CPP, causing it to search for
321 usr/include, et al, within sysroot+headers_suffix.
325 Define this macro to provide additional specifications to put in the
326 @file{specs} file that can be used in various specifications like
329 The definition should be an initializer for an array of structures,
330 containing a string constant, that defines the specification name, and a
331 string constant that provides the specification.
333 Do not define this macro if it does not need to do anything.
335 @code{EXTRA_SPECS} is useful when an architecture contains several
336 related targets, which have various @code{@dots{}_SPECS} which are similar
337 to each other, and the maintainer would like one central place to keep
340 For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to
341 define either @code{_CALL_SYSV} when the System V calling sequence is
342 used or @code{_CALL_AIX} when the older AIX-based calling sequence is
345 The @file{config/rs6000/rs6000.h} target file defines:
348 #define EXTRA_SPECS \
349 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @},
351 #define CPP_SYS_DEFAULT ""
354 The @file{config/rs6000/sysv.h} target file defines:
358 "%@{posix: -D_POSIX_SOURCE @} \
359 %@{mcall-sysv: -D_CALL_SYSV @} \
360 %@{!mcall-sysv: %(cpp_sysv_default) @} \
361 %@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}"
363 #undef CPP_SYSV_DEFAULT
364 #define CPP_SYSV_DEFAULT "-D_CALL_SYSV"
367 while the @file{config/rs6000/eabiaix.h} target file defines
368 @code{CPP_SYSV_DEFAULT} as:
371 #undef CPP_SYSV_DEFAULT
372 #define CPP_SYSV_DEFAULT "-D_CALL_AIX"
376 @defmac LINK_LIBGCC_SPECIAL_1
377 Define this macro if the driver program should find the library
378 @file{libgcc.a}. If you do not define this macro, the driver program will pass
379 the argument @option{-lgcc} to tell the linker to do the search.
382 @defmac LINK_GCC_C_SEQUENCE_SPEC
383 The sequence in which libgcc and libc are specified to the linker.
384 By default this is @code{%G %L %G}.
387 @defmac POST_LINK_SPEC
388 Define this macro to add additional steps to be executed after linker.
389 The default value of this macro is empty string.
392 @defmac LINK_COMMAND_SPEC
393 A C string constant giving the complete command line need to execute the
394 linker. When you do this, you will need to update your port each time a
395 change is made to the link command line within @file{gcc.cc}. Therefore,
396 define this macro only if you need to completely redefine the command
397 line for invoking the linker and there is no other way to accomplish
398 the effect you need. Overriding this macro may be avoidable by overriding
399 @code{LINK_GCC_C_SEQUENCE_SPEC} instead.
402 @deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT
403 True if @file{..} components should always be removed from directory names
404 computed relative to GCC's internal directories, false (default) if such
405 components should be preserved and directory names containing them passed
406 to other tools such as the linker.
409 @defmac MULTILIB_DEFAULTS
410 Define this macro as a C expression for the initializer of an array of
411 string to tell the driver program which options are defaults for this
412 target and thus do not need to be handled specially when using
413 @code{MULTILIB_OPTIONS}.
415 Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in
416 the target makefile fragment or if none of the options listed in
417 @code{MULTILIB_OPTIONS} are set by default.
418 @xref{Target Fragment}.
421 @defmac RELATIVE_PREFIX_NOT_LINKDIR
422 Define this macro to tell @command{gcc} that it should only translate
423 a @option{-B} prefix into a @option{-L} linker option if the prefix
424 indicates an absolute file name.
427 @defmac MD_EXEC_PREFIX
428 If defined, this macro is an additional prefix to try after
429 @code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched
430 when the compiler is built as a cross
431 compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it
432 to the list of directories used to find the assembler in @file{configure.ac}.
435 @defmac STANDARD_STARTFILE_PREFIX
436 Define this macro as a C string constant if you wish to override the
437 standard choice of @code{libdir} as the default prefix to
438 try when searching for startup files such as @file{crt0.o}.
439 @code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler
440 is built as a cross compiler.
443 @defmac STANDARD_STARTFILE_PREFIX_1
444 Define this macro as a C string constant if you wish to override the
445 standard choice of @code{/lib} as a prefix to try after the default prefix
446 when searching for startup files such as @file{crt0.o}.
447 @code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler
448 is built as a cross compiler.
451 @defmac STANDARD_STARTFILE_PREFIX_2
452 Define this macro as a C string constant if you wish to override the
453 standard choice of @code{/lib} as yet another prefix to try after the
454 default prefix when searching for startup files such as @file{crt0.o}.
455 @code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler
456 is built as a cross compiler.
459 @defmac MD_STARTFILE_PREFIX
460 If defined, this macro supplies an additional prefix to try after the
461 standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the
462 compiler is built as a cross compiler.
465 @defmac MD_STARTFILE_PREFIX_1
466 If defined, this macro supplies yet another prefix to try after the
467 standard prefixes. It is not searched when the compiler is built as a
471 @defmac INIT_ENVIRONMENT
472 Define this macro as a C string constant if you wish to set environment
473 variables for programs called by the driver, such as the assembler and
474 loader. The driver passes the value of this macro to @code{putenv} to
475 initialize the necessary environment variables.
478 @defmac LOCAL_INCLUDE_DIR
479 Define this macro as a C string constant if you wish to override the
480 standard choice of @file{/usr/local/include} as the default prefix to
481 try when searching for local header files. @code{LOCAL_INCLUDE_DIR}
482 comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in
483 @file{config.gcc}, normally @file{/usr/include}) in the search order.
485 Cross compilers do not search either @file{/usr/local/include} or its
489 @defmac NATIVE_SYSTEM_HEADER_COMPONENT
490 The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}.
491 See @code{INCLUDE_DEFAULTS}, below, for the description of components.
492 If you do not define this macro, no component is used.
495 @defmac INCLUDE_DEFAULTS
496 Define this macro if you wish to override the entire default search path
497 for include files. For a native compiler, the default search path
498 usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR},
499 @code{GPLUSPLUS_INCLUDE_DIR}, and
500 @code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR}
501 and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile},
502 and specify private search areas for GCC@. The directory
503 @code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs.
505 The definition should be an initializer for an array of structures.
506 Each array element should have four elements: the directory name (a
507 string constant), the component name (also a string constant), a flag
508 for C++-only directories,
509 and a flag showing that the includes in the directory don't need to be
510 wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of
511 the array with a null element.
513 The component name denotes what GNU package the include file is part of,
514 if any, in all uppercase letters. For example, it might be @samp{GCC}
515 or @samp{BINUTILS}. If the package is part of a vendor-supplied
516 operating system, code the component name as @samp{0}.
518 For example, here is the definition used for VAX/VMS:
521 #define INCLUDE_DEFAULTS \
523 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \
524 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \
525 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \
532 Here is the order of prefixes tried for exec files:
536 Any prefixes specified by the user with @option{-B}.
539 The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX}
540 is not set and the compiler has not been installed in the configure-time
541 @var{prefix}, the location in which the compiler has actually been installed.
544 The directories specified by the environment variable @code{COMPILER_PATH}.
547 The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed
548 in the configured-time @var{prefix}.
551 The location @file{/usr/libexec/gcc/}, but only if this is a native compiler.
554 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
557 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
561 Here is the order of prefixes tried for startfiles:
565 Any prefixes specified by the user with @option{-B}.
568 The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined
569 value based on the installed toolchain location.
572 The directories specified by the environment variable @code{LIBRARY_PATH}
573 (or port-specific name; native only, cross compilers do not use this).
576 The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed
577 in the configured @var{prefix} or this is a native compiler.
580 The location @file{/usr/lib/gcc/}, but only if this is a native compiler.
583 The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native
587 The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a
588 native compiler, or we have a target system root.
591 The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a
592 native compiler, or we have a target system root.
595 The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications.
596 If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and
597 the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix.
600 The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native
601 compiler, or we have a target system root. The default for this macro is
605 The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native
606 compiler, or we have a target system root. The default for this macro is
610 @node Run-time Target
611 @section Run-time Target Specification
612 @cindex run-time target specification
613 @cindex predefined macros
614 @cindex target specifications
616 @c prevent bad page break with this line
617 Here are run-time target specifications.
619 @defmac TARGET_CPU_CPP_BUILTINS ()
620 This function-like macro expands to a block of code that defines
621 built-in preprocessor macros and assertions for the target CPU, using
622 the functions @code{builtin_define}, @code{builtin_define_std} and
623 @code{builtin_assert}. When the front end
624 calls this macro it provides a trailing semicolon, and since it has
625 finished command line option processing your code can use those
628 @code{builtin_assert} takes a string in the form you pass to the
629 command-line option @option{-A}, such as @code{cpu=mips}, and creates
630 the assertion. @code{builtin_define} takes a string in the form
631 accepted by option @option{-D} and unconditionally defines the macro.
633 @code{builtin_define_std} takes a string representing the name of an
634 object-like macro. If it doesn't lie in the user's namespace,
635 @code{builtin_define_std} defines it unconditionally. Otherwise, it
636 defines a version with two leading underscores, and another version
637 with two leading and trailing underscores, and defines the original
638 only if an ISO standard was not requested on the command line. For
639 example, passing @code{unix} defines @code{__unix}, @code{__unix__}
640 and possibly @code{unix}; passing @code{_mips} defines @code{__mips},
641 @code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64}
642 defines only @code{_ABI64}.
644 You can also test for the C dialect being compiled. The variable
645 @code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus}
646 or @code{clk_objective_c}. Note that if we are preprocessing
647 assembler, this variable will be @code{clk_c} but the function-like
648 macro @code{preprocessing_asm_p()} will return true, so you might want
649 to check for that first. If you need to check for strict ANSI, the
650 variable @code{flag_iso} can be used. The function-like macro
651 @code{preprocessing_trad_p()} can be used to check for traditional
655 @defmac TARGET_OS_CPP_BUILTINS ()
656 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
657 and is used for the target operating system instead.
660 @defmac TARGET_OBJFMT_CPP_BUILTINS ()
661 Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional
662 and is used for the target object format. @file{elfos.h} uses this
663 macro to define @code{__ELF__}, so you probably do not need to define
667 @deftypevar {extern int} target_flags
668 This variable is declared in @file{options.h}, which is included before
669 any target-specific headers.
672 @deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS
673 This variable specifies the initial value of @code{target_flags}.
674 Its default setting is 0.
677 @cindex optional hardware or system features
678 @cindex features, optional, in system conventions
680 @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})
681 This hook is called whenever the user specifies one of the
682 target-specific options described by the @file{.opt} definition files
683 (@pxref{Options}). It has the opportunity to do some option-specific
684 processing and should return true if the option is valid. The default
685 definition does nothing but return true.
687 @var{decoded} specifies the option and its arguments. @var{opts} and
688 @var{opts_set} are the @code{gcc_options} structures to be used for
689 storing option state, and @var{loc} is the location at which the
690 option was passed (@code{UNKNOWN_LOCATION} except for options passed
694 @deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value})
695 This target hook is called whenever the user specifies one of the
696 target-specific C language family options described by the @file{.opt}
697 definition files(@pxref{Options}). It has the opportunity to do some
698 option-specific processing and should return true if the option is
699 valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The
700 default definition does nothing but return false.
702 In general, you should use @code{TARGET_HANDLE_OPTION} to handle
703 options. However, if processing an option requires routines that are
704 only available in the C (and related language) front ends, then you
705 should use @code{TARGET_HANDLE_C_OPTION} instead.
708 @deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string})
709 Targets may provide a string object type that can be used within
710 and between C, C++ and their respective Objective-C dialects.
711 A string object might, for example, embed encoding and length information.
712 These objects are considered opaque to the compiler and handled as references.
713 An ideal implementation makes the composition of the string object
714 match that of the Objective-C @code{NSString} (@code{NXString} for GNUStep),
715 allowing efficient interworking between C-only and Objective-C code.
716 If a target implements string objects then this hook should return a
717 reference to such an object constructed from the normal `C' string
718 representation provided in @var{string}.
719 At present, the hook is used by Objective-C only, to obtain a
720 common-format string object when the target provides one.
723 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE (const char *@var{classname})
724 Declare that Objective C class @var{classname} is referenced
728 @deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char *@var{classname})
729 Declare that Objective C class @var{classname} is defined
733 @deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref})
734 If a target implements string objects then this hook should return
735 @code{true} if @var{stringref} is a valid reference to such an object.
738 @deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list})
739 If a target implements string objects then this hook should
740 provide a facility to check the function arguments in @var{args_list}
741 against the format specifiers in @var{format_arg} where the type of
742 @var{format_arg} is one recognized as a valid string reference type.
745 @deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void)
746 This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE}
747 but is called when the optimize level is changed via an attribute or
748 pragma or when it is reset at the end of the code affected by the
749 attribute or pragma. It is not called at the beginning of compilation
750 when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these
751 actions then, you should have @code{TARGET_OPTION_OVERRIDE} call
752 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}.
755 @defmac C_COMMON_OVERRIDE_OPTIONS
756 This is similar to the @code{TARGET_OPTION_OVERRIDE} hook
757 but is only used in the C
758 language frontends (C, Objective-C, C++, Objective-C++) and so can be
759 used to alter option flag variables which only exist in those
763 @deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE
764 Some machines may desire to change what optimizations are performed for
765 various optimization levels. This variable, if defined, describes
766 options to enable at particular sets of optimization levels. These
767 options are processed once
768 just after the optimization level is determined and before the remainder
769 of the command options have been parsed, so may be overridden by other
770 options passed explicitly.
772 This processing is run once at program startup and when the optimization
773 options are changed via @code{#pragma GCC optimize} or by using the
774 @code{optimize} attribute.
777 @deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts})
778 Set target-dependent initial values of fields in @var{opts}.
781 @deftypefn {Common Target Hook} {const char *} TARGET_COMPUTE_MULTILIB (const struct switchstr *@var{switches}, int @var{n_switches}, const char *@var{multilib_dir}, const char *@var{multilib_defaults}, const char *@var{multilib_select}, const char *@var{multilib_matches}, const char *@var{multilib_exclusions}, const char *@var{multilib_reuse})
782 Some targets like RISC-V might have complicated multilib reuse rules which
783 are hard to implement with the current multilib scheme. This hook allows
784 targets to override the result from the built-in multilib mechanism.
785 @var{switches} is the raw option list with @var{n_switches} items;
786 @var{multilib_dir} is the multi-lib result which is computed by the built-in
788 @var{multilib_defaults} is the default options list for multi-lib;
789 @var{multilib_select} is the string containing the list of supported
790 multi-libs, and the option checking list.
791 @var{multilib_matches}, @var{multilib_exclusions}, and @var{multilib_reuse}
792 are corresponding to @var{MULTILIB_MATCHES}, @var{MULTILIB_EXCLUSIONS},
793 and @var{MULTILIB_REUSE}.
794 The default definition does nothing but return @var{multilib_dir} directly.
798 @defmac SWITCHABLE_TARGET
799 Some targets need to switch between substantially different subtargets
800 during compilation. For example, the MIPS target has one subtarget for
801 the traditional MIPS architecture and another for MIPS16. Source code
802 can switch between these two subarchitectures using the @code{mips16}
803 and @code{nomips16} attributes.
805 Such subtargets can differ in things like the set of available
806 registers, the set of available instructions, the costs of various
807 operations, and so on. GCC caches a lot of this type of information
808 in global variables, and recomputing them for each subtarget takes a
809 significant amount of time. The compiler therefore provides a facility
810 for maintaining several versions of the global variables and quickly
811 switching between them; see @file{target-globals.h} for details.
813 Define this macro to 1 if your target needs this facility. The default
817 @deftypefn {Target Hook} bool TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P (void)
818 Returns true if the target supports IEEE 754 floating-point exceptions
819 and rounding modes, false otherwise. This is intended to relate to the
820 @code{float} and @code{double} types, but not necessarily @code{long double}.
821 By default, returns true if the @code{adddf3} instruction pattern is
822 available and false otherwise, on the assumption that hardware floating
823 point supports exceptions and rounding modes but software floating point
827 @node Per-Function Data
828 @section Defining data structures for per-function information.
829 @cindex per-function data
830 @cindex data structures
832 If the target needs to store information on a per-function basis, GCC
833 provides a macro and a couple of variables to allow this. Note, just
834 using statics to store the information is a bad idea, since GCC supports
835 nested functions, so you can be halfway through encoding one function
836 when another one comes along.
838 GCC defines a data structure called @code{struct function} which
839 contains all of the data specific to an individual function. This
840 structure contains a field called @code{machine} whose type is
841 @code{struct machine_function *}, which can be used by targets to point
842 to their own specific data.
844 If a target needs per-function specific data it should define the type
845 @code{struct machine_function} and also the macro @code{INIT_EXPANDERS}.
846 This macro should be used to initialize the function pointer
847 @code{init_machine_status}. This pointer is explained below.
849 One typical use of per-function, target specific data is to create an
850 RTX to hold the register containing the function's return address. This
851 RTX can then be used to implement the @code{__builtin_return_address}
852 function, for level 0.
854 Note---earlier implementations of GCC used a single data area to hold
855 all of the per-function information. Thus when processing of a nested
856 function began the old per-function data had to be pushed onto a
857 stack, and when the processing was finished, it had to be popped off the
858 stack. GCC used to provide function pointers called
859 @code{save_machine_status} and @code{restore_machine_status} to handle
860 the saving and restoring of the target specific information. Since the
861 single data area approach is no longer used, these pointers are no
864 @defmac INIT_EXPANDERS
865 Macro called to initialize any target specific information. This macro
866 is called once per function, before generation of any RTL has begun.
867 The intention of this macro is to allow the initialization of the
868 function pointer @code{init_machine_status}.
871 @deftypevar {void (*)(struct function *)} init_machine_status
872 If this function pointer is non-@code{NULL} it will be called once per
873 function, before function compilation starts, in order to allow the
874 target to perform any target specific initialization of the
875 @code{struct function} structure. It is intended that this would be
876 used to initialize the @code{machine} of that structure.
878 @code{struct machine_function} structures are expected to be freed by GC@.
879 Generally, any memory that they reference must be allocated by using
880 GC allocation, including the structure itself.
884 @section Storage Layout
885 @cindex storage layout
887 Note that the definitions of the macros in this table which are sizes or
888 alignments measured in bits do not need to be constant. They can be C
889 expressions that refer to static variables, such as the @code{target_flags}.
890 @xref{Run-time Target}.
892 @defmac BITS_BIG_ENDIAN
893 Define this macro to have the value 1 if the most significant bit in a
894 byte has the lowest number; otherwise define it to have the value zero.
895 This means that bit-field instructions count from the most significant
896 bit. If the machine has no bit-field instructions, then this must still
897 be defined, but it doesn't matter which value it is defined to. This
898 macro need not be a constant.
900 This macro does not affect the way structure fields are packed into
901 bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}.
904 @defmac BYTES_BIG_ENDIAN
905 Define this macro to have the value 1 if the most significant byte in a
906 word has the lowest number. This macro need not be a constant.
909 @defmac WORDS_BIG_ENDIAN
910 Define this macro to have the value 1 if, in a multiword object, the
911 most significant word has the lowest number. This applies to both
912 memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the
913 order of words in memory is not the same as the order in registers. This
914 macro need not be a constant.
917 @defmac REG_WORDS_BIG_ENDIAN
918 On some machines, the order of words in a multiword object differs between
919 registers in memory. In such a situation, define this macro to describe
920 the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls
921 the order of words in memory.
924 @defmac FLOAT_WORDS_BIG_ENDIAN
925 Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or
926 @code{TFmode} floating point numbers are stored in memory with the word
927 containing the sign bit at the lowest address; otherwise define it to
928 have the value 0. This macro need not be a constant.
930 You need not define this macro if the ordering is the same as for
934 @defmac BITS_PER_WORD
935 Number of bits in a word. If you do not define this macro, the default
936 is @code{BITS_PER_UNIT * UNITS_PER_WORD}.
939 @defmac MAX_BITS_PER_WORD
940 Maximum number of bits in a word. If this is undefined, the default is
941 @code{BITS_PER_WORD}. Otherwise, it is the constant value that is the
942 largest value that @code{BITS_PER_WORD} can have at run-time.
945 @defmac UNITS_PER_WORD
946 Number of storage units in a word; normally the size of a general-purpose
947 register, a power of two from 1 or 8.
950 @defmac MIN_UNITS_PER_WORD
951 Minimum number of units in a word. If this is undefined, the default is
952 @code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the
953 smallest value that @code{UNITS_PER_WORD} can have at run-time.
957 Width of a pointer, in bits. You must specify a value no wider than the
958 width of @code{Pmode}. If it is not equal to the width of @code{Pmode},
959 you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify
960 a value the default is @code{BITS_PER_WORD}.
963 @defmac POINTERS_EXTEND_UNSIGNED
964 A C expression that determines how pointers should be extended from
965 @code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is
966 greater than zero if pointers should be zero-extended, zero if they
967 should be sign-extended, and negative if some other sort of conversion
968 is needed. In the last case, the extension is done by the target's
969 @code{ptr_extend} instruction.
971 You need not define this macro if the @code{ptr_mode}, @code{Pmode}
972 and @code{word_mode} are all the same width.
975 @defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type})
976 A macro to update @var{m} and @var{unsignedp} when an object whose type
977 is @var{type} and which has the specified mode and signedness is to be
978 stored in a register. This macro is only called when @var{type} is a
981 On most RISC machines, which only have operations that operate on a full
982 register, define this macro to set @var{m} to @code{word_mode} if
983 @var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most
984 cases, only integer modes should be widened because wider-precision
985 floating-point operations are usually more expensive than their narrower
988 For most machines, the macro definition does not change @var{unsignedp}.
989 However, some machines, have instructions that preferentially handle
990 either signed or unsigned quantities of certain modes. For example, on
991 the DEC Alpha, 32-bit loads from memory and 32-bit add instructions
992 sign-extend the result to 64 bits. On such machines, set
993 @var{unsignedp} according to which kind of extension is more efficient.
995 Do not define this macro if it would never modify @var{m}.
998 @deftypefn {Target Hook} {enum flt_eval_method} TARGET_C_EXCESS_PRECISION (enum excess_precision_type @var{type})
999 Return a value, with the same meaning as the C99 macro
1000 @code{FLT_EVAL_METHOD} that describes which excess precision should be
1001 applied. @var{type} is either @code{EXCESS_PRECISION_TYPE_IMPLICIT},
1002 @code{EXCESS_PRECISION_TYPE_FAST},
1003 @code{EXCESS_PRECISION_TYPE_STANDARD}, or
1004 @code{EXCESS_PRECISION_TYPE_FLOAT16}. For
1005 @code{EXCESS_PRECISION_TYPE_IMPLICIT}, the target should return which
1006 precision and range operations will be implictly evaluated in regardless
1007 of the excess precision explicitly added. For
1008 @code{EXCESS_PRECISION_TYPE_STANDARD},
1009 @code{EXCESS_PRECISION_TYPE_FLOAT16}, and
1010 @code{EXCESS_PRECISION_TYPE_FAST}, the target should return the
1011 explicit excess precision that should be added depending on the
1012 value set for @option{-fexcess-precision=@r{[}standard@r{|}fast@r{]}}.
1013 Note that unpredictable explicit excess precision does not make sense,
1014 so a target should never return @code{FLT_EVAL_METHOD_UNPREDICTABLE}
1015 when @var{type} is @code{EXCESS_PRECISION_TYPE_STANDARD},
1016 @code{EXCESS_PRECISION_TYPE_FLOAT16} or
1017 @code{EXCESS_PRECISION_TYPE_FAST}.
1019 Return a value, with the same meaning as the C99 macro
1020 @code{FLT_EVAL_METHOD} that describes which excess precision should be
1023 @deftypefn {Target Hook} machine_mode TARGET_PROMOTE_FUNCTION_MODE (const_tree @var{type}, machine_mode @var{mode}, int *@var{punsignedp}, const_tree @var{funtype}, int @var{for_return})
1024 Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or
1025 function return values. The target hook should return the new mode
1026 and possibly change @code{*@var{punsignedp}} if the promotion should
1027 change signedness. This function is called only for scalar @emph{or
1030 @var{for_return} allows to distinguish the promotion of arguments and
1031 return values. If it is @code{1}, a return value is being promoted and
1032 @code{TARGET_FUNCTION_VALUE} must perform the same promotions done here.
1033 If it is @code{2}, the returned mode should be that of the register in
1034 which an incoming parameter is copied, or the outgoing result is computed;
1035 then the hook should return the same mode as @code{promote_mode}, though
1036 the signedness may be different.
1038 @var{type} can be NULL when promoting function arguments of libcalls.
1040 The default is to not promote arguments and return values. You can
1041 also define the hook to @code{default_promote_function_mode_always_promote}
1042 if you would like to apply the same rules given by @code{PROMOTE_MODE}.
1045 @defmac PARM_BOUNDARY
1046 Normal alignment required for function parameters on the stack, in
1047 bits. All stack parameters receive at least this much alignment
1048 regardless of data type. On most machines, this is the same as the
1052 @defmac STACK_BOUNDARY
1053 Define this macro to the minimum alignment enforced by hardware for the
1054 stack pointer on this machine. The definition is a C expression for the
1055 desired alignment (measured in bits). This value is used as a default
1056 if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines,
1057 this should be the same as @code{PARM_BOUNDARY}.
1060 @defmac PREFERRED_STACK_BOUNDARY
1061 Define this macro if you wish to preserve a certain alignment for the
1062 stack pointer, greater than what the hardware enforces. The definition
1063 is a C expression for the desired alignment (measured in bits). This
1064 macro must evaluate to a value equal to or larger than
1065 @code{STACK_BOUNDARY}.
1068 @defmac INCOMING_STACK_BOUNDARY
1069 Define this macro if the incoming stack boundary may be different
1070 from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate
1071 to a value equal to or larger than @code{STACK_BOUNDARY}.
1074 @defmac FUNCTION_BOUNDARY
1075 Alignment required for a function entry point, in bits.
1078 @defmac BIGGEST_ALIGNMENT
1079 Biggest alignment that any data type can require on this machine, in
1080 bits. Note that this is not the biggest alignment that is supported,
1081 just the biggest alignment that, when violated, may cause a fault.
1084 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_ABSOLUTE_BIGGEST_ALIGNMENT
1085 If defined, this target hook specifies the absolute biggest alignment
1086 that a type or variable can have on this machine, otherwise,
1087 @code{BIGGEST_ALIGNMENT} is used.
1090 @defmac MALLOC_ABI_ALIGNMENT
1091 Alignment, in bits, a C conformant malloc implementation has to
1092 provide. If not defined, the default value is @code{BITS_PER_WORD}.
1095 @defmac ATTRIBUTE_ALIGNED_VALUE
1096 Alignment used by the @code{__attribute__ ((aligned))} construct. If
1097 not defined, the default value is @code{BIGGEST_ALIGNMENT}.
1100 @defmac MINIMUM_ATOMIC_ALIGNMENT
1101 If defined, the smallest alignment, in bits, that can be given to an
1102 object that can be referenced in one operation, without disturbing any
1103 nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger
1104 on machines that don't have byte or half-word store operations.
1107 @defmac BIGGEST_FIELD_ALIGNMENT
1108 Biggest alignment that any structure or union field can require on this
1109 machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for
1110 structure and union fields only, unless the field alignment has been set
1111 by the @code{__attribute__ ((aligned (@var{n})))} construct.
1114 @defmac ADJUST_FIELD_ALIGN (@var{field}, @var{type}, @var{computed})
1115 An expression for the alignment of a structure field @var{field} of
1116 type @var{type} if the alignment computed in the usual way (including
1117 applying of @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the
1118 alignment) is @var{computed}. It overrides alignment only if the
1119 field alignment has not been set by the
1120 @code{__attribute__ ((aligned (@var{n})))} construct. Note that @var{field}
1121 may be @code{NULL_TREE} in case we just query for the minimum alignment
1122 of a field of type @var{type} in structure context.
1125 @defmac MAX_STACK_ALIGNMENT
1126 Biggest stack alignment guaranteed by the backend. Use this macro
1127 to specify the maximum alignment of a variable on stack.
1129 If not defined, the default value is @code{STACK_BOUNDARY}.
1131 @c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}.
1132 @c But the fix for PR 32893 indicates that we can only guarantee
1133 @c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not
1134 @c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported.
1137 @defmac MAX_OFILE_ALIGNMENT
1138 Biggest alignment supported by the object file format of this machine.
1139 Use this macro to limit the alignment which can be specified using the
1140 @code{__attribute__ ((aligned (@var{n})))} construct for functions and
1141 objects with static storage duration. The alignment of automatic
1142 objects may exceed the object file format maximum up to the maximum
1143 supported by GCC. If not defined, the default value is
1144 @code{BIGGEST_ALIGNMENT}.
1146 On systems that use ELF, the default (in @file{config/elfos.h}) is
1147 the largest supported 32-bit ELF section alignment representable on
1148 a 32-bit host e.g.@: @samp{(((uint64_t) 1 << 28) * 8)}.
1149 On 32-bit ELF the largest supported section alignment in bits is
1150 @samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts.
1153 @deftypefn {Target Hook} void TARGET_LOWER_LOCAL_DECL_ALIGNMENT (tree @var{decl})
1154 Define this hook to lower alignment of local, parm or result
1155 decl @samp{(@var{decl})}.
1158 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_STATIC_RTX_ALIGNMENT (machine_mode @var{mode})
1159 This hook returns the preferred alignment in bits for a
1160 statically-allocated rtx, such as a constant pool entry. @var{mode}
1161 is the mode of the rtx. The default implementation returns
1162 @samp{GET_MODE_ALIGNMENT (@var{mode})}.
1165 @defmac DATA_ALIGNMENT (@var{type}, @var{basic-align})
1166 If defined, a C expression to compute the alignment for a variable in
1167 the static store. @var{type} is the data type, and @var{basic-align} is
1168 the alignment that the object would ordinarily have. The value of this
1169 macro is used instead of that alignment to align the object.
1171 If this macro is not defined, then @var{basic-align} is used.
1174 One use of this macro is to increase alignment of medium-size data to
1175 make it all fit in fewer cache lines. Another is to cause character
1176 arrays to be word-aligned so that @code{strcpy} calls that copy
1177 constants to character arrays can be done inline.
1180 @defmac DATA_ABI_ALIGNMENT (@var{type}, @var{basic-align})
1181 Similar to @code{DATA_ALIGNMENT}, but for the cases where the ABI mandates
1182 some alignment increase, instead of optimization only purposes. E.g.@
1183 AMD x86-64 psABI says that variables with array type larger than 15 bytes
1184 must be aligned to 16 byte boundaries.
1186 If this macro is not defined, then @var{basic-align} is used.
1189 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_CONSTANT_ALIGNMENT (const_tree @var{constant}, HOST_WIDE_INT @var{basic_align})
1190 This hook returns the alignment in bits of a constant that is being
1191 placed in memory. @var{constant} is the constant and @var{basic_align}
1192 is the alignment that the object would ordinarily have.
1194 The default definition just returns @var{basic_align}.
1196 The typical use of this hook is to increase alignment for string
1197 constants to be word aligned so that @code{strcpy} calls that copy
1198 constants can be done inline. The function
1199 @code{constant_alignment_word_strings} provides such a definition.
1202 @defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align})
1203 If defined, a C expression to compute the alignment for a variable in
1204 the local store. @var{type} is the data type, and @var{basic-align} is
1205 the alignment that the object would ordinarily have. The value of this
1206 macro is used instead of that alignment to align the object.
1208 If this macro is not defined, then @var{basic-align} is used.
1210 One use of this macro is to increase alignment of medium-size data to
1211 make it all fit in fewer cache lines.
1213 If the value of this macro has a type, it should be an unsigned type.
1216 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_VECTOR_ALIGNMENT (const_tree @var{type})
1217 This hook can be used to define the alignment for a vector of type
1218 @var{type}, in order to comply with a platform ABI. The default is to
1219 require natural alignment for vector types. The alignment returned by
1220 this hook must be a power-of-two multiple of the default alignment of
1221 the vector element type.
1224 @defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align})
1225 If defined, a C expression to compute the alignment for stack slot.
1226 @var{type} is the data type, @var{mode} is the widest mode available,
1227 and @var{basic-align} is the alignment that the slot would ordinarily
1228 have. The value of this macro is used instead of that alignment to
1231 If this macro is not defined, then @var{basic-align} is used when
1232 @var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will
1235 This macro is to set alignment of stack slot to the maximum alignment
1236 of all possible modes which the slot may have.
1238 If the value of this macro has a type, it should be an unsigned type.
1241 @defmac LOCAL_DECL_ALIGNMENT (@var{decl})
1242 If defined, a C expression to compute the alignment for a local
1243 variable @var{decl}.
1245 If this macro is not defined, then
1246 @code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))}
1249 One use of this macro is to increase alignment of medium-size data to
1250 make it all fit in fewer cache lines.
1252 If the value of this macro has a type, it should be an unsigned type.
1255 @defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align})
1256 If defined, a C expression to compute the minimum required alignment
1257 for dynamic stack realignment purposes for @var{exp} (a type or decl),
1258 @var{mode}, assuming normal alignment @var{align}.
1260 If this macro is not defined, then @var{align} will be used.
1263 @defmac EMPTY_FIELD_BOUNDARY
1264 Alignment in bits to be given to a structure bit-field that follows an
1265 empty field such as @code{int : 0;}.
1267 If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro.
1270 @defmac STRUCTURE_SIZE_BOUNDARY
1271 Number of bits which any structure or union's size must be a multiple of.
1272 Each structure or union's size is rounded up to a multiple of this.
1274 If you do not define this macro, the default is the same as
1275 @code{BITS_PER_UNIT}.
1278 @defmac STRICT_ALIGNMENT
1279 Define this macro to be the value 1 if instructions will fail to work
1280 if given data not on the nominal alignment. If instructions will merely
1281 go slower in that case, define this macro as 0.
1284 @defmac PCC_BITFIELD_TYPE_MATTERS
1285 Define this if you wish to imitate the way many other C compilers handle
1286 alignment of bit-fields and the structures that contain them.
1288 The behavior is that the type written for a named bit-field (@code{int},
1289 @code{short}, or other integer type) imposes an alignment for the entire
1290 structure, as if the structure really did contain an ordinary field of
1291 that type. In addition, the bit-field is placed within the structure so
1292 that it would fit within such a field, not crossing a boundary for it.
1294 Thus, on most machines, a named bit-field whose type is written as
1295 @code{int} would not cross a four-byte boundary, and would force
1296 four-byte alignment for the whole structure. (The alignment used may
1297 not be four bytes; it is controlled by the other alignment parameters.)
1299 An unnamed bit-field will not affect the alignment of the containing
1302 If the macro is defined, its definition should be a C expression;
1303 a nonzero value for the expression enables this behavior.
1305 Note that if this macro is not defined, or its value is zero, some
1306 bit-fields may cross more than one alignment boundary. The compiler can
1307 support such references if there are @samp{insv}, @samp{extv}, and
1308 @samp{extzv} insns that can directly reference memory.
1310 The other known way of making bit-fields work is to define
1311 @code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}.
1312 Then every structure can be accessed with fullwords.
1314 Unless the machine has bit-field instructions or you define
1315 @code{STRUCTURE_SIZE_BOUNDARY} that way, you must define
1316 @code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value.
1318 If your aim is to make GCC use the same conventions for laying out
1319 bit-fields as are used by another compiler, here is how to investigate
1320 what the other compiler does. Compile and run this program:
1339 printf ("Size of foo1 is %d\n",
1340 sizeof (struct foo1));
1341 printf ("Size of foo2 is %d\n",
1342 sizeof (struct foo2));
1347 If this prints 2 and 5, then the compiler's behavior is what you would
1348 get from @code{PCC_BITFIELD_TYPE_MATTERS}.
1351 @defmac BITFIELD_NBYTES_LIMITED
1352 Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited
1353 to aligning a bit-field within the structure.
1356 @deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void)
1357 When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine
1358 whether unnamed bitfields affect the alignment of the containing
1359 structure. The hook should return true if the structure should inherit
1360 the alignment requirements of an unnamed bitfield's type.
1363 @deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void)
1364 This target hook should return @code{true} if accesses to volatile bitfields
1365 should use the narrowest mode possible. It should return @code{false} if
1366 these accesses should use the bitfield container type.
1368 The default is @code{false}.
1371 @deftypefn {Target Hook} bool TARGET_MEMBER_TYPE_FORCES_BLK (const_tree @var{field}, machine_mode @var{mode})
1372 Return true if a structure, union or array containing @var{field} should
1373 be accessed using @code{BLKMODE}.
1375 If @var{field} is the only field in the structure, @var{mode} is its
1376 mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the
1377 case where structures of one field would require the structure's mode to
1378 retain the field's mode.
1380 Normally, this is not needed.
1383 @defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified})
1384 Define this macro as an expression for the alignment of a type (given
1385 by @var{type} as a tree node) if the alignment computed in the usual
1386 way is @var{computed} and the alignment explicitly specified was
1389 The default is to use @var{specified} if it is larger; otherwise, use
1390 the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT}
1393 @defmac MAX_FIXED_MODE_SIZE
1394 An integer expression for the size in bits of the largest integer
1395 machine mode that should actually be used. All integer machine modes of
1396 this size or smaller can be used for structures and unions with the
1397 appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE
1398 (DImode)} is assumed.
1401 @defmac STACK_SAVEAREA_MODE (@var{save_level})
1402 If defined, an expression of type @code{machine_mode} that
1403 specifies the mode of the save area operand of a
1404 @code{save_stack_@var{level}} named pattern (@pxref{Standard Names}).
1405 @var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or
1406 @code{SAVE_NONLOCAL} and selects which of the three named patterns is
1407 having its mode specified.
1409 You need not define this macro if it always returns @code{Pmode}. You
1410 would most commonly define this macro if the
1411 @code{save_stack_@var{level}} patterns need to support both a 32- and a
1415 @defmac STACK_SIZE_MODE
1416 If defined, an expression of type @code{machine_mode} that
1417 specifies the mode of the size increment operand of an
1418 @code{allocate_stack} named pattern (@pxref{Standard Names}).
1420 You need not define this macro if it always returns @code{word_mode}.
1421 You would most commonly define this macro if the @code{allocate_stack}
1422 pattern needs to support both a 32- and a 64-bit mode.
1425 @deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_CMP_RETURN_MODE (void)
1426 This target hook should return the mode to be used for the return value
1427 of compare instructions expanded to libgcc calls. If not defined
1428 @code{word_mode} is returned which is the right choice for a majority of
1432 @deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_SHIFT_COUNT_MODE (void)
1433 This target hook should return the mode to be used for the shift count operand
1434 of shift instructions expanded to libgcc calls. If not defined
1435 @code{word_mode} is returned which is the right choice for a majority of
1439 @deftypefn {Target Hook} scalar_int_mode TARGET_UNWIND_WORD_MODE (void)
1440 Return machine mode to be used for @code{_Unwind_Word} type.
1441 The default is to use @code{word_mode}.
1444 @deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type})
1445 This target hook returns @code{true} if bit-fields in the given
1446 @var{record_type} are to be laid out following the rules of Microsoft
1447 Visual C/C++, namely: (i) a bit-field won't share the same storage
1448 unit with the previous bit-field if their underlying types have
1449 different sizes, and the bit-field will be aligned to the highest
1450 alignment of the underlying types of itself and of the previous
1451 bit-field; (ii) a zero-sized bit-field will affect the alignment of
1452 the whole enclosing structure, even if it is unnamed; except that
1453 (iii) a zero-sized bit-field will be disregarded unless it follows
1454 another bit-field of nonzero size. If this hook returns @code{true},
1455 other macros that control bit-field layout are ignored.
1457 When a bit-field is inserted into a packed record, the whole size
1458 of the underlying type is used by one or more same-size adjacent
1459 bit-fields (that is, if its long:3, 32 bits is used in the record,
1460 and any additional adjacent long bit-fields are packed into the same
1461 chunk of 32 bits. However, if the size changes, a new field of that
1462 size is allocated). In an unpacked record, this is the same as using
1463 alignment, but not equivalent when packing.
1465 If both MS bit-fields and @samp{__attribute__((packed))} are used,
1466 the latter will take precedence. If @samp{__attribute__((packed))} is
1467 used on a single field when MS bit-fields are in use, it will take
1468 precedence for that field, but the alignment of the rest of the structure
1469 may affect its placement.
1472 @deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void)
1473 Returns true if the target supports decimal floating point.
1476 @deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void)
1477 Returns true if the target supports fixed-point arithmetic.
1480 @deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void)
1481 This hook is called just before expansion into rtl, allowing the target
1482 to perform additional initializations or analysis before the expansion.
1483 For example, the rs6000 port uses it to allocate a scratch stack slot
1484 for use in copying SDmode values between memory and floating point
1485 registers whenever the function being expanded has any SDmode
1489 @deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void)
1490 This hook allows the backend to perform additional instantiations on rtl
1491 that are not actually in any insns yet, but will be later.
1494 @deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type})
1495 If your target defines any fundamental types, or any types your target
1496 uses should be mangled differently from the default, define this hook
1497 to return the appropriate encoding for these types as part of a C++
1498 mangled name. The @var{type} argument is the tree structure representing
1499 the type to be mangled. The hook may be applied to trees which are
1500 not target-specific fundamental types; it should return @code{NULL}
1501 for all such types, as well as arguments it does not recognize. If the
1502 return value is not @code{NULL}, it must point to a statically-allocated
1505 Target-specific fundamental types might be new fundamental types or
1506 qualified versions of ordinary fundamental types. Encode new
1507 fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name}
1508 is the name used for the type in source code, and @var{n} is the
1509 length of @var{name} in decimal. Encode qualified versions of
1510 ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where
1511 @var{name} is the name used for the type qualifier in source code,
1512 @var{n} is the length of @var{name} as above, and @var{code} is the
1513 code used to represent the unqualified version of this type. (See
1514 @code{write_builtin_type} in @file{cp/mangle.cc} for the list of
1515 codes.) In both cases the spaces are for clarity; do not include any
1516 spaces in your string.
1518 This hook is applied to types prior to typedef resolution. If the mangled
1519 name for a particular type depends only on that type's main variant, you
1520 can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT}
1523 The default version of this hook always returns @code{NULL}, which is
1524 appropriate for a target that does not define any new fundamental
1529 @section Layout of Source Language Data Types
1531 These macros define the sizes and other characteristics of the standard
1532 basic data types used in programs being compiled. Unlike the macros in
1533 the previous section, these apply to specific features of C and related
1534 languages, rather than to fundamental aspects of storage layout.
1536 @defmac INT_TYPE_SIZE
1537 A C expression for the size in bits of the type @code{int} on the
1538 target machine. If you don't define this, the default is one word.
1541 @defmac SHORT_TYPE_SIZE
1542 A C expression for the size in bits of the type @code{short} on the
1543 target machine. If you don't define this, the default is half a word.
1544 (If this would be less than one storage unit, it is rounded up to one
1548 @defmac LONG_TYPE_SIZE
1549 A C expression for the size in bits of the type @code{long} on the
1550 target machine. If you don't define this, the default is one word.
1553 @defmac ADA_LONG_TYPE_SIZE
1554 On some machines, the size used for the Ada equivalent of the type
1555 @code{long} by a native Ada compiler differs from that used by C@. In
1556 that situation, define this macro to be a C expression to be used for
1557 the size of that type. If you don't define this, the default is the
1558 value of @code{LONG_TYPE_SIZE}.
1561 @defmac LONG_LONG_TYPE_SIZE
1562 A C expression for the size in bits of the type @code{long long} on the
1563 target machine. If you don't define this, the default is two
1564 words. If you want to support GNU Ada on your machine, the value of this
1565 macro must be at least 64.
1568 @defmac CHAR_TYPE_SIZE
1569 A C expression for the size in bits of the type @code{char} on the
1570 target machine. If you don't define this, the default is
1571 @code{BITS_PER_UNIT}.
1574 @defmac BOOL_TYPE_SIZE
1575 A C expression for the size in bits of the C++ type @code{bool} and
1576 C99 type @code{_Bool} on the target machine. If you don't define
1577 this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}.
1580 @defmac FLOAT_TYPE_SIZE
1581 A C expression for the size in bits of the type @code{float} on the
1582 target machine. If you don't define this, the default is one word.
1585 @defmac DOUBLE_TYPE_SIZE
1586 A C expression for the size in bits of the type @code{double} on the
1587 target machine. If you don't define this, the default is two
1591 @defmac LONG_DOUBLE_TYPE_SIZE
1592 A C expression for the size in bits of the type @code{long double} on
1593 the target machine. If you don't define this, the default is two
1597 @defmac SHORT_FRACT_TYPE_SIZE
1598 A C expression for the size in bits of the type @code{short _Fract} on
1599 the target machine. If you don't define this, the default is
1600 @code{BITS_PER_UNIT}.
1603 @defmac FRACT_TYPE_SIZE
1604 A C expression for the size in bits of the type @code{_Fract} on
1605 the target machine. If you don't define this, the default is
1606 @code{BITS_PER_UNIT * 2}.
1609 @defmac LONG_FRACT_TYPE_SIZE
1610 A C expression for the size in bits of the type @code{long _Fract} on
1611 the target machine. If you don't define this, the default is
1612 @code{BITS_PER_UNIT * 4}.
1615 @defmac LONG_LONG_FRACT_TYPE_SIZE
1616 A C expression for the size in bits of the type @code{long long _Fract} on
1617 the target machine. If you don't define this, the default is
1618 @code{BITS_PER_UNIT * 8}.
1621 @defmac SHORT_ACCUM_TYPE_SIZE
1622 A C expression for the size in bits of the type @code{short _Accum} on
1623 the target machine. If you don't define this, the default is
1624 @code{BITS_PER_UNIT * 2}.
1627 @defmac ACCUM_TYPE_SIZE
1628 A C expression for the size in bits of the type @code{_Accum} on
1629 the target machine. If you don't define this, the default is
1630 @code{BITS_PER_UNIT * 4}.
1633 @defmac LONG_ACCUM_TYPE_SIZE
1634 A C expression for the size in bits of the type @code{long _Accum} on
1635 the target machine. If you don't define this, the default is
1636 @code{BITS_PER_UNIT * 8}.
1639 @defmac LONG_LONG_ACCUM_TYPE_SIZE
1640 A C expression for the size in bits of the type @code{long long _Accum} on
1641 the target machine. If you don't define this, the default is
1642 @code{BITS_PER_UNIT * 16}.
1645 @defmac LIBGCC2_GNU_PREFIX
1646 This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target
1647 hook and should be defined if that hook is overriden to be true. It
1648 causes function names in libgcc to be changed to use a @code{__gnu_}
1649 prefix for their name rather than the default @code{__}. A port which
1650 uses this macro should also arrange to use @file{t-gnu-prefix} in
1651 the libgcc @file{config.host}.
1654 @defmac WIDEST_HARDWARE_FP_SIZE
1655 A C expression for the size in bits of the widest floating-point format
1656 supported by the hardware. If you define this macro, you must specify a
1657 value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}.
1658 If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE}
1662 @defmac DEFAULT_SIGNED_CHAR
1663 An expression whose value is 1 or 0, according to whether the type
1664 @code{char} should be signed or unsigned by default. The user can
1665 always override this default with the options @option{-fsigned-char}
1666 and @option{-funsigned-char}.
1669 @deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void)
1670 This target hook should return true if the compiler should give an
1671 @code{enum} type only as many bytes as it takes to represent the range
1672 of possible values of that type. It should return false if all
1673 @code{enum} types should be allocated like @code{int}.
1675 The default is to return false.
1679 A C expression for a string describing the name of the data type to use
1680 for size values. The typedef name @code{size_t} is defined using the
1681 contents of the string.
1683 The string can contain more than one keyword. If so, separate them with
1684 spaces, and write first any length keyword, then @code{unsigned} if
1685 appropriate, and finally @code{int}. The string must exactly match one
1686 of the data type names defined in the function
1687 @code{c_common_nodes_and_builtins} in the file @file{c-family/c-common.cc}.
1688 You may not omit @code{int} or change the order---that would cause the
1689 compiler to crash on startup.
1691 If you don't define this macro, the default is @code{"long unsigned
1696 GCC defines internal types (@code{sizetype}, @code{ssizetype},
1697 @code{bitsizetype} and @code{sbitsizetype}) for expressions
1698 dealing with size. This macro is a C expression for a string describing
1699 the name of the data type from which the precision of @code{sizetype}
1702 The string has the same restrictions as @code{SIZE_TYPE} string.
1704 If you don't define this macro, the default is @code{SIZE_TYPE}.
1707 @defmac PTRDIFF_TYPE
1708 A C expression for a string describing the name of the data type to use
1709 for the result of subtracting two pointers. The typedef name
1710 @code{ptrdiff_t} is defined using the contents of the string. See
1711 @code{SIZE_TYPE} above for more information.
1713 If you don't define this macro, the default is @code{"long int"}.
1717 A C expression for a string describing the name of the data type to use
1718 for wide characters. The typedef name @code{wchar_t} is defined using
1719 the contents of the string. See @code{SIZE_TYPE} above for more
1722 If you don't define this macro, the default is @code{"int"}.
1725 @defmac WCHAR_TYPE_SIZE
1726 A C expression for the size in bits of the data type for wide
1727 characters. This is used in @code{cpp}, which cannot make use of
1732 A C expression for a string describing the name of the data type to
1733 use for wide characters passed to @code{printf} and returned from
1734 @code{getwc}. The typedef name @code{wint_t} is defined using the
1735 contents of the string. See @code{SIZE_TYPE} above for more
1738 If you don't define this macro, the default is @code{"unsigned int"}.
1742 A C expression for a string describing the name of the data type that
1743 can represent any value of any standard or extended signed integer type.
1744 The typedef name @code{intmax_t} is defined using the contents of the
1745 string. See @code{SIZE_TYPE} above for more information.
1747 If you don't define this macro, the default is the first of
1748 @code{"int"}, @code{"long int"}, or @code{"long long int"} that has as
1749 much precision as @code{long long int}.
1752 @defmac UINTMAX_TYPE
1753 A C expression for a string describing the name of the data type that
1754 can represent any value of any standard or extended unsigned integer
1755 type. The typedef name @code{uintmax_t} is defined using the contents
1756 of the string. See @code{SIZE_TYPE} above for more information.
1758 If you don't define this macro, the default is the first of
1759 @code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long
1760 unsigned int"} that has as much precision as @code{long long unsigned
1764 @defmac SIG_ATOMIC_TYPE
1770 @defmacx UINT16_TYPE
1771 @defmacx UINT32_TYPE
1772 @defmacx UINT64_TYPE
1773 @defmacx INT_LEAST8_TYPE
1774 @defmacx INT_LEAST16_TYPE
1775 @defmacx INT_LEAST32_TYPE
1776 @defmacx INT_LEAST64_TYPE
1777 @defmacx UINT_LEAST8_TYPE
1778 @defmacx UINT_LEAST16_TYPE
1779 @defmacx UINT_LEAST32_TYPE
1780 @defmacx UINT_LEAST64_TYPE
1781 @defmacx INT_FAST8_TYPE
1782 @defmacx INT_FAST16_TYPE
1783 @defmacx INT_FAST32_TYPE
1784 @defmacx INT_FAST64_TYPE
1785 @defmacx UINT_FAST8_TYPE
1786 @defmacx UINT_FAST16_TYPE
1787 @defmacx UINT_FAST32_TYPE
1788 @defmacx UINT_FAST64_TYPE
1789 @defmacx INTPTR_TYPE
1790 @defmacx UINTPTR_TYPE
1791 C expressions for the standard types @code{sig_atomic_t},
1792 @code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t},
1793 @code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t},
1794 @code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t},
1795 @code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t},
1796 @code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t},
1797 @code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t},
1798 @code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t},
1799 @code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See
1800 @code{SIZE_TYPE} above for more information.
1802 If any of these macros evaluates to a null pointer, the corresponding
1803 type is not supported; if GCC is configured to provide
1804 @code{<stdint.h>} in such a case, the header provided may not conform
1805 to C99, depending on the type in question. The defaults for all of
1806 these macros are null pointers.
1809 @defmac TARGET_PTRMEMFUNC_VBIT_LOCATION
1810 The C++ compiler represents a pointer-to-member-function with a struct
1817 ptrdiff_t vtable_index;
1824 The C++ compiler must use one bit to indicate whether the function that
1825 will be called through a pointer-to-member-function is virtual.
1826 Normally, we assume that the low-order bit of a function pointer must
1827 always be zero. Then, by ensuring that the vtable_index is odd, we can
1828 distinguish which variant of the union is in use. But, on some
1829 platforms function pointers can be odd, and so this doesn't work. In
1830 that case, we use the low-order bit of the @code{delta} field, and shift
1831 the remainder of the @code{delta} field to the left.
1833 GCC will automatically make the right selection about where to store
1834 this bit using the @code{FUNCTION_BOUNDARY} setting for your platform.
1835 However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY}
1836 set such that functions always start at even addresses, but the lowest
1837 bit of pointers to functions indicate whether the function at that
1838 address is in ARM or Thumb mode. If this is the case of your
1839 architecture, you should define this macro to
1840 @code{ptrmemfunc_vbit_in_delta}.
1842 In general, you should not have to define this macro. On architectures
1843 in which function addresses are always even, according to
1844 @code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to
1845 @code{ptrmemfunc_vbit_in_pfn}.
1848 @defmac TARGET_VTABLE_USES_DESCRIPTORS
1849 Normally, the C++ compiler uses function pointers in vtables. This
1850 macro allows the target to change to use ``function descriptors''
1851 instead. Function descriptors are found on targets for whom a
1852 function pointer is actually a small data structure. Normally the
1853 data structure consists of the actual code address plus a data
1854 pointer to which the function's data is relative.
1856 If vtables are used, the value of this macro should be the number
1857 of words that the function descriptor occupies.
1860 @defmac TARGET_VTABLE_ENTRY_ALIGN
1861 By default, the vtable entries are void pointers, the so the alignment
1862 is the same as pointer alignment. The value of this macro specifies
1863 the alignment of the vtable entry in bits. It should be defined only
1864 when special alignment is necessary. */
1867 @defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE
1868 There are a few non-descriptor entries in the vtable at offsets below
1869 zero. If these entries must be padded (say, to preserve the alignment
1870 specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number
1871 of words in each data entry.
1875 @section Register Usage
1876 @cindex register usage
1878 This section explains how to describe what registers the target machine
1879 has, and how (in general) they can be used.
1881 The description of which registers a specific instruction can use is
1882 done with register classes; see @ref{Register Classes}. For information
1883 on using registers to access a stack frame, see @ref{Frame Registers}.
1884 For passing values in registers, see @ref{Register Arguments}.
1885 For returning values in registers, see @ref{Scalar Return}.
1888 * Register Basics:: Number and kinds of registers.
1889 * Allocation Order:: Order in which registers are allocated.
1890 * Values in Registers:: What kinds of values each reg can hold.
1891 * Leaf Functions:: Renumbering registers for leaf functions.
1892 * Stack Registers:: Handling a register stack such as 80387.
1895 @node Register Basics
1896 @subsection Basic Characteristics of Registers
1898 @c prevent bad page break with this line
1899 Registers have various characteristics.
1901 @defmac FIRST_PSEUDO_REGISTER
1902 Number of hardware registers known to the compiler. They receive
1903 numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first
1904 pseudo register's number really is assigned the number
1905 @code{FIRST_PSEUDO_REGISTER}.
1908 @defmac FIXED_REGISTERS
1909 @cindex fixed register
1910 An initializer that says which registers are used for fixed purposes
1911 all throughout the compiled code and are therefore not available for
1912 general allocation. These would include the stack pointer, the frame
1913 pointer (except on machines where that can be used as a general
1914 register when no frame pointer is needed), the program counter on
1915 machines where that is considered one of the addressable registers,
1916 and any other numbered register with a standard use.
1918 This information is expressed as a sequence of numbers, separated by
1919 commas and surrounded by braces. The @var{n}th number is 1 if
1920 register @var{n} is fixed, 0 otherwise.
1922 The table initialized from this macro, and the table initialized by
1923 the following one, may be overridden at run time either automatically,
1924 by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by
1925 the user with the command options @option{-ffixed-@var{reg}},
1926 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}.
1929 @defmac CALL_USED_REGISTERS
1930 @cindex call-used register
1931 @cindex call-clobbered register
1932 @cindex call-saved register
1933 Like @code{FIXED_REGISTERS} but has 1 for each register that is
1934 clobbered (in general) by function calls as well as for fixed
1935 registers. This macro therefore identifies the registers that are not
1936 available for general allocation of values that must live across
1939 If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler
1940 automatically saves it on function entry and restores it on function
1941 exit, if the register is used within the function.
1943 Exactly one of @code{CALL_USED_REGISTERS} and @code{CALL_REALLY_USED_REGISTERS}
1944 must be defined. Modern ports should define @code{CALL_REALLY_USED_REGISTERS}.
1947 @defmac CALL_REALLY_USED_REGISTERS
1948 @cindex call-used register
1949 @cindex call-clobbered register
1950 @cindex call-saved register
1951 Like @code{CALL_USED_REGISTERS} except this macro doesn't require
1952 that the entire set of @code{FIXED_REGISTERS} be included.
1953 (@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}).
1955 Exactly one of @code{CALL_USED_REGISTERS} and @code{CALL_REALLY_USED_REGISTERS}
1956 must be defined. Modern ports should define @code{CALL_REALLY_USED_REGISTERS}.
1959 @cindex call-used register
1960 @cindex call-clobbered register
1961 @cindex call-saved register
1962 @deftypefn {Target Hook} {const predefined_function_abi &} TARGET_FNTYPE_ABI (const_tree @var{type})
1963 Return the ABI used by a function with type @var{type}; see the
1964 definition of @code{predefined_function_abi} for details of the ABI
1965 descriptor. Targets only need to define this hook if they support
1966 interoperability between several ABIs in the same translation unit.
1969 @deftypefn {Target Hook} {const predefined_function_abi &} TARGET_INSN_CALLEE_ABI (const rtx_insn *@var{insn})
1970 This hook returns a description of the ABI used by the target of
1971 call instruction @var{insn}; see the definition of
1972 @code{predefined_function_abi} for details of the ABI descriptor.
1973 Only the global function @code{insn_callee_abi} should call this hook
1976 Targets only need to define this hook if they support
1977 interoperability between several ABIs in the same translation unit.
1980 @cindex call-used register
1981 @cindex call-clobbered register
1982 @cindex call-saved register
1983 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_CALL_PART_CLOBBERED (unsigned int @var{abi_id}, unsigned int @var{regno}, machine_mode @var{mode})
1984 ABIs usually specify that calls must preserve the full contents
1985 of a particular register, or that calls can alter any part of a
1986 particular register. This information is captured by the target macro
1987 @code{CALL_REALLY_USED_REGISTERS}. However, some ABIs specify that calls
1988 must preserve certain bits of a particular register but can alter others.
1989 This hook should return true if this applies to at least one of the
1990 registers in @samp{(reg:@var{mode} @var{regno})}, and if as a result the
1991 call would alter part of the @var{mode} value. For example, if a call
1992 preserves the low 32 bits of a 64-bit hard register @var{regno} but can
1993 clobber the upper 32 bits, this hook should return true for a 64-bit mode
1994 but false for a 32-bit mode.
1996 The value of @var{abi_id} comes from the @code{predefined_function_abi}
1997 structure that describes the ABI of the call; see the definition of the
1998 structure for more details. If (as is usual) the target uses the same ABI
1999 for all functions in a translation unit, @var{abi_id} is always 0.
2001 The default implementation returns false, which is correct
2002 for targets that don't have partly call-clobbered registers.
2005 @deftypefn {Target Hook} {const char *} TARGET_GET_MULTILIB_ABI_NAME (void)
2006 This hook returns name of multilib ABI name.
2010 @findex call_used_regs
2013 @findex reg_class_contents
2014 @deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void)
2015 This hook may conditionally modify five variables
2016 @code{fixed_regs}, @code{call_used_regs}, @code{global_regs},
2017 @code{reg_names}, and @code{reg_class_contents}, to take into account
2018 any dependence of these register sets on target flags. The first three
2019 of these are of type @code{char []} (interpreted as boolean vectors).
2020 @code{global_regs} is a @code{const char *[]}, and
2021 @code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is
2022 called, @code{fixed_regs}, @code{call_used_regs},
2023 @code{reg_class_contents}, and @code{reg_names} have been initialized
2024 from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS},
2025 @code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively.
2026 @code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}},
2027 @option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}
2028 command options have been applied.
2030 @cindex disabling certain registers
2031 @cindex controlling register usage
2032 If the usage of an entire class of registers depends on the target
2033 flags, you may indicate this to GCC by using this macro to modify
2034 @code{fixed_regs} and @code{call_used_regs} to 1 for each of the
2035 registers in the classes which should not be used by GCC@. Also make
2036 @code{define_register_constraint}s return @code{NO_REGS} for constraints
2037 that shouldn't be used.
2039 (However, if this class is not included in @code{GENERAL_REGS} and all
2040 of the insn patterns whose constraints permit this class are
2041 controlled by target switches, then GCC will automatically avoid using
2042 these registers when the target switches are opposed to them.)
2045 @defmac INCOMING_REGNO (@var{out})
2046 Define this macro if the target machine has register windows. This C
2047 expression returns the register number as seen by the called function
2048 corresponding to the register number @var{out} as seen by the calling
2049 function. Return @var{out} if register number @var{out} is not an
2053 @defmac OUTGOING_REGNO (@var{in})
2054 Define this macro if the target machine has register windows. This C
2055 expression returns the register number as seen by the calling function
2056 corresponding to the register number @var{in} as seen by the called
2057 function. Return @var{in} if register number @var{in} is not an inbound
2061 @defmac LOCAL_REGNO (@var{regno})
2062 Define this macro if the target machine has register windows. This C
2063 expression returns true if the register is call-saved but is in the
2064 register window. Unlike most call-saved registers, such registers
2065 need not be explicitly restored on function exit or during non-local
2070 If the program counter has a register number, define this as that
2071 register number. Otherwise, do not define it.
2074 @node Allocation Order
2075 @subsection Order of Allocation of Registers
2076 @cindex order of register allocation
2077 @cindex register allocation order
2079 @c prevent bad page break with this line
2080 Registers are allocated in order.
2082 @defmac REG_ALLOC_ORDER
2083 If defined, an initializer for a vector of integers, containing the
2084 numbers of hard registers in the order in which GCC should prefer
2085 to use them (from most preferred to least).
2087 If this macro is not defined, registers are used lowest numbered first
2088 (all else being equal).
2090 One use of this macro is on machines where the highest numbered
2091 registers must always be saved and the save-multiple-registers
2092 instruction supports only sequences of consecutive registers. On such
2093 machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists
2094 the highest numbered allocable register first.
2097 @defmac ADJUST_REG_ALLOC_ORDER
2098 A C statement (sans semicolon) to choose the order in which to allocate
2099 hard registers for pseudo-registers local to a basic block.
2101 Store the desired register order in the array @code{reg_alloc_order}.
2102 Element 0 should be the register to allocate first; element 1, the next
2103 register; and so on.
2105 The macro body should not assume anything about the contents of
2106 @code{reg_alloc_order} before execution of the macro.
2108 On most machines, it is not necessary to define this macro.
2111 @defmac HONOR_REG_ALLOC_ORDER
2112 Normally, IRA tries to estimate the costs for saving a register in the
2113 prologue and restoring it in the epilogue. This discourages it from
2114 using call-saved registers. If a machine wants to ensure that IRA
2115 allocates registers in the order given by REG_ALLOC_ORDER even if some
2116 call-saved registers appear earlier than call-used ones, then define this
2117 macro as a C expression to nonzero. Default is 0.
2120 @defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno})
2121 In some case register allocation order is not enough for the
2122 Integrated Register Allocator (@acronym{IRA}) to generate a good code.
2123 If this macro is defined, it should return a floating point value
2124 based on @var{regno}. The cost of using @var{regno} for a pseudo will
2125 be increased by approximately the pseudo's usage frequency times the
2126 value returned by this macro. Not defining this macro is equivalent
2127 to having it always return @code{0.0}.
2129 On most machines, it is not necessary to define this macro.
2132 @node Values in Registers
2133 @subsection How Values Fit in Registers
2135 This section discusses the macros that describe which kinds of values
2136 (specifically, which machine modes) each register can hold, and how many
2137 consecutive registers are needed for a given mode.
2139 @deftypefn {Target Hook} {unsigned int} TARGET_HARD_REGNO_NREGS (unsigned int @var{regno}, machine_mode @var{mode})
2140 This hook returns the number of consecutive hard registers, starting
2141 at register number @var{regno}, required to hold a value of mode
2142 @var{mode}. This hook must never return zero, even if a register
2143 cannot hold the requested mode - indicate that with
2144 @code{TARGET_HARD_REGNO_MODE_OK} and/or
2145 @code{TARGET_CAN_CHANGE_MODE_CLASS} instead.
2147 The default definition returns the number of words in @var{mode}.
2150 @defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode})
2151 A C expression that is nonzero if a value of mode @var{mode}, stored
2152 in memory, ends with padding that causes it to take up more space than
2153 in registers starting at register number @var{regno} (as determined by
2154 multiplying GCC's notion of the size of the register when containing
2155 this mode by the number of registers returned by
2156 @code{TARGET_HARD_REGNO_NREGS}). By default this is zero.
2158 For example, if a floating-point value is stored in three 32-bit
2159 registers but takes up 128 bits in memory, then this would be
2162 This macros only needs to be defined if there are cases where
2163 @code{subreg_get_info}
2164 would otherwise wrongly determine that a @code{subreg} can be
2165 represented by an offset to the register number, when in fact such a
2166 @code{subreg} would contain some of the padding not stored in
2167 registers and so not be representable.
2170 @defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode})
2171 For values of @var{regno} and @var{mode} for which
2172 @code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression
2173 returning the greater number of registers required to hold the value
2174 including any padding. In the example above, the value would be four.
2177 @defmac REGMODE_NATURAL_SIZE (@var{mode})
2178 Define this macro if the natural size of registers that hold values
2179 of mode @var{mode} is not the word size. It is a C expression that
2180 should give the natural size in bytes for the specified mode. It is
2181 used by the register allocator to try to optimize its results. This
2182 happens for example on SPARC 64-bit where the natural size of
2183 floating-point registers is still 32-bit.
2186 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_MODE_OK (unsigned int @var{regno}, machine_mode @var{mode})
2187 This hook returns true if it is permissible to store a value
2188 of mode @var{mode} in hard register number @var{regno} (or in several
2189 registers starting with that one). The default definition returns true
2192 You need not include code to check for the numbers of fixed registers,
2193 because the allocation mechanism considers them to be always occupied.
2195 @cindex register pairs
2196 On some machines, double-precision values must be kept in even/odd
2197 register pairs. You can implement that by defining this hook to reject
2198 odd register numbers for such modes.
2200 The minimum requirement for a mode to be OK in a register is that the
2201 @samp{mov@var{mode}} instruction pattern support moves between the
2202 register and other hard register in the same class and that moving a
2203 value into the register and back out not alter it.
2205 Since the same instruction used to move @code{word_mode} will work for
2206 all narrower integer modes, it is not necessary on any machine for
2207 this hook to distinguish between these modes, provided you define
2208 patterns @samp{movhi}, etc., to take advantage of this. This is
2209 useful because of the interaction between @code{TARGET_HARD_REGNO_MODE_OK}
2210 and @code{TARGET_MODES_TIEABLE_P}; it is very desirable for all integer
2211 modes to be tieable.
2213 Many machines have special registers for floating point arithmetic.
2214 Often people assume that floating point machine modes are allowed only
2215 in floating point registers. This is not true. Any registers that
2216 can hold integers can safely @emph{hold} a floating point machine
2217 mode, whether or not floating arithmetic can be done on it in those
2218 registers. Integer move instructions can be used to move the values.
2220 On some machines, though, the converse is true: fixed-point machine
2221 modes may not go in floating registers. This is true if the floating
2222 registers normalize any value stored in them, because storing a
2223 non-floating value there would garble it. In this case,
2224 @code{TARGET_HARD_REGNO_MODE_OK} should reject fixed-point machine modes in
2225 floating registers. But if the floating registers do not automatically
2226 normalize, if you can store any bit pattern in one and retrieve it
2227 unchanged without a trap, then any machine mode may go in a floating
2228 register, so you can define this hook to say so.
2230 The primary significance of special floating registers is rather that
2231 they are the registers acceptable in floating point arithmetic
2232 instructions. However, this is of no concern to
2233 @code{TARGET_HARD_REGNO_MODE_OK}. You handle it by writing the proper
2234 constraints for those instructions.
2236 On some machines, the floating registers are especially slow to access,
2237 so that it is better to store a value in a stack frame than in such a
2238 register if floating point arithmetic is not being done. As long as the
2239 floating registers are not in class @code{GENERAL_REGS}, they will not
2240 be used unless some pattern's constraint asks for one.
2243 @defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to})
2244 A C expression that is nonzero if it is OK to rename a hard register
2245 @var{from} to another hard register @var{to}.
2247 One common use of this macro is to prevent renaming of a register to
2248 another register that is not saved by a prologue in an interrupt
2251 The default is always nonzero.
2254 @deftypefn {Target Hook} bool TARGET_MODES_TIEABLE_P (machine_mode @var{mode1}, machine_mode @var{mode2})
2255 This hook returns true if a value of mode @var{mode1} is accessible
2256 in mode @var{mode2} without copying.
2258 If @code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and
2259 @code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always
2260 the same for any @var{r}, then
2261 @code{TARGET_MODES_TIEABLE_P (@var{mode1}, @var{mode2})}
2262 should be true. If they differ for any @var{r}, you should define
2263 this hook to return false unless some other mechanism ensures the
2264 accessibility of the value in a narrower mode.
2266 You should define this hook to return true in as many cases as
2267 possible since doing so will allow GCC to perform better register
2268 allocation. The default definition returns true unconditionally.
2271 @deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno})
2272 This target hook should return @code{true} if it is OK to use a hard register
2273 @var{regno} as scratch reg in peephole2.
2275 One common use of this macro is to prevent using of a register that
2276 is not saved by a prologue in an interrupt handler.
2278 The default version of this hook always returns @code{true}.
2281 @defmac AVOID_CCMODE_COPIES
2282 Define this macro if the compiler should avoid copies to/from @code{CCmode}
2283 registers. You should only define this macro if support for copying to/from
2284 @code{CCmode} is incomplete.
2287 @node Leaf Functions
2288 @subsection Handling Leaf Functions
2290 @cindex leaf functions
2291 @cindex functions, leaf
2292 On some machines, a leaf function (i.e., one which makes no calls) can run
2293 more efficiently if it does not make its own register window. Often this
2294 means it is required to receive its arguments in the registers where they
2295 are passed by the caller, instead of the registers where they would
2298 The special treatment for leaf functions generally applies only when
2299 other conditions are met; for example, often they may use only those
2300 registers for its own variables and temporaries. We use the term ``leaf
2301 function'' to mean a function that is suitable for this special
2302 handling, so that functions with no calls are not necessarily ``leaf
2305 GCC assigns register numbers before it knows whether the function is
2306 suitable for leaf function treatment. So it needs to renumber the
2307 registers in order to output a leaf function. The following macros
2310 @defmac LEAF_REGISTERS
2311 Name of a char vector, indexed by hard register number, which
2312 contains 1 for a register that is allowable in a candidate for leaf
2315 If leaf function treatment involves renumbering the registers, then the
2316 registers marked here should be the ones before renumbering---those that
2317 GCC would ordinarily allocate. The registers which will actually be
2318 used in the assembler code, after renumbering, should not be marked with 1
2321 Define this macro only if the target machine offers a way to optimize
2322 the treatment of leaf functions.
2325 @defmac LEAF_REG_REMAP (@var{regno})
2326 A C expression whose value is the register number to which @var{regno}
2327 should be renumbered, when a function is treated as a leaf function.
2329 If @var{regno} is a register number which should not appear in a leaf
2330 function before renumbering, then the expression should yield @minus{}1, which
2331 will cause the compiler to abort.
2333 Define this macro only if the target machine offers a way to optimize the
2334 treatment of leaf functions, and registers need to be renumbered to do
2338 @findex current_function_is_leaf
2339 @findex current_function_uses_only_leaf_regs
2340 @code{TARGET_ASM_FUNCTION_PROLOGUE} and
2341 @code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions
2342 specially. They can test the C variable @code{current_function_is_leaf}
2343 which is nonzero for leaf functions. @code{current_function_is_leaf} is
2344 set prior to local register allocation and is valid for the remaining
2345 compiler passes. They can also test the C variable
2346 @code{current_function_uses_only_leaf_regs} which is nonzero for leaf
2347 functions which only use leaf registers.
2348 @code{current_function_uses_only_leaf_regs} is valid after all passes
2349 that modify the instructions have been run and is only useful if
2350 @code{LEAF_REGISTERS} is defined.
2351 @c changed this to fix overfull. ALSO: why the "it" at the beginning
2352 @c of the next paragraph?! --mew 2feb93
2354 @node Stack Registers
2355 @subsection Registers That Form a Stack
2357 There are special features to handle computers where some of the
2358 ``registers'' form a stack. Stack registers are normally written by
2359 pushing onto the stack, and are numbered relative to the top of the
2362 Currently, GCC can only handle one group of stack-like registers, and
2363 they must be consecutively numbered. Furthermore, the existing
2364 support for stack-like registers is specific to the 80387 floating
2365 point coprocessor. If you have a new architecture that uses
2366 stack-like registers, you will need to do substantial work on
2367 @file{reg-stack.cc} and write your machine description to cooperate
2368 with it, as well as defining these macros.
2371 Define this if the machine has any stack-like registers.
2374 @defmac STACK_REG_COVER_CLASS
2375 This is a cover class containing the stack registers. Define this if
2376 the machine has any stack-like registers.
2379 @defmac FIRST_STACK_REG
2380 The number of the first stack-like register. This one is the top
2384 @defmac LAST_STACK_REG
2385 The number of the last stack-like register. This one is the bottom of
2389 @node Register Classes
2390 @section Register Classes
2391 @cindex register class definitions
2392 @cindex class definitions, register
2394 On many machines, the numbered registers are not all equivalent.
2395 For example, certain registers may not be allowed for indexed addressing;
2396 certain registers may not be allowed in some instructions. These machine
2397 restrictions are described to the compiler using @dfn{register classes}.
2399 You define a number of register classes, giving each one a name and saying
2400 which of the registers belong to it. Then you can specify register classes
2401 that are allowed as operands to particular instruction patterns.
2405 In general, each register will belong to several classes. In fact, one
2406 class must be named @code{ALL_REGS} and contain all the registers. Another
2407 class must be named @code{NO_REGS} and contain no registers. Often the
2408 union of two classes will be another class; however, this is not required.
2410 @findex GENERAL_REGS
2411 One of the classes must be named @code{GENERAL_REGS}. There is nothing
2412 terribly special about the name, but the operand constraint letters
2413 @samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is
2414 the same as @code{ALL_REGS}, just define it as a macro which expands
2417 Order the classes so that if class @var{x} is contained in class @var{y}
2418 then @var{x} has a lower class number than @var{y}.
2420 The way classes other than @code{GENERAL_REGS} are specified in operand
2421 constraints is through machine-dependent operand constraint letters.
2422 You can define such letters to correspond to various classes, then use
2423 them in operand constraints.
2425 You must define the narrowest register classes for allocatable
2426 registers, so that each class either has no subclasses, or that for
2427 some mode, the move cost between registers within the class is
2428 cheaper than moving a register in the class to or from memory
2431 You should define a class for the union of two classes whenever some
2432 instruction allows both classes. For example, if an instruction allows
2433 either a floating point (coprocessor) register or a general register for a
2434 certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS}
2435 which includes both of them. Otherwise you will get suboptimal code,
2436 or even internal compiler errors when reload cannot find a register in the
2437 class computed via @code{reg_class_subunion}.
2439 You must also specify certain redundant information about the register
2440 classes: for each class, which classes contain it and which ones are
2441 contained in it; for each pair of classes, the largest class contained
2444 When a value occupying several consecutive registers is expected in a
2445 certain class, all the registers used must belong to that class.
2446 Therefore, register classes cannot be used to enforce a requirement for
2447 a register pair to start with an even-numbered register. The way to
2448 specify this requirement is with @code{TARGET_HARD_REGNO_MODE_OK}.
2450 Register classes used for input-operands of bitwise-and or shift
2451 instructions have a special requirement: each such class must have, for
2452 each fixed-point machine mode, a subclass whose registers can transfer that
2453 mode to or from memory. For example, on some machines, the operations for
2454 single-byte values (@code{QImode}) are limited to certain registers. When
2455 this is so, each register class that is used in a bitwise-and or shift
2456 instruction must have a subclass consisting of registers from which
2457 single-byte values can be loaded or stored. This is so that
2458 @code{PREFERRED_RELOAD_CLASS} can always have a possible value to return.
2460 @deftp {Data type} {enum reg_class}
2461 An enumerated type that must be defined with all the register class names
2462 as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS}
2463 must be the last register class, followed by one more enumerated value,
2464 @code{LIM_REG_CLASSES}, which is not a register class but rather
2465 tells how many classes there are.
2467 Each register class has a number, which is the value of casting
2468 the class name to type @code{int}. The number serves as an index
2469 in many of the tables described below.
2472 @defmac N_REG_CLASSES
2473 The number of distinct register classes, defined as follows:
2476 #define N_REG_CLASSES (int) LIM_REG_CLASSES
2480 @defmac REG_CLASS_NAMES
2481 An initializer containing the names of the register classes as C string
2482 constants. These names are used in writing some of the debugging dumps.
2485 @defmac REG_CLASS_CONTENTS
2486 An initializer containing the contents of the register classes, as integers
2487 which are bit masks. The @var{n}th integer specifies the contents of class
2488 @var{n}. The way the integer @var{mask} is interpreted is that
2489 register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1.
2491 When the machine has more than 32 registers, an integer does not suffice.
2492 Then the integers are replaced by sub-initializers, braced groupings containing
2493 several integers. Each sub-initializer must be suitable as an initializer
2494 for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}.
2495 In this situation, the first integer in each sub-initializer corresponds to
2496 registers 0 through 31, the second integer to registers 32 through 63, and
2500 @defmac REGNO_REG_CLASS (@var{regno})
2501 A C expression whose value is a register class containing hard register
2502 @var{regno}. In general there is more than one such class; choose a class
2503 which is @dfn{minimal}, meaning that no smaller class also contains the
2507 @defmac BASE_REG_CLASS
2508 A macro whose definition is the name of the class to which a valid
2509 base register must belong. A base register is one used in an address
2510 which is the register value plus a displacement.
2513 @defmac MODE_BASE_REG_CLASS (@var{mode})
2514 This is a variation of the @code{BASE_REG_CLASS} macro which allows
2515 the selection of a base register in a mode dependent manner. If
2516 @var{mode} is VOIDmode then it should return the same value as
2517 @code{BASE_REG_CLASS}.
2520 @defmac MODE_BASE_REG_REG_CLASS (@var{mode})
2521 A C expression whose value is the register class to which a valid
2522 base register must belong in order to be used in a base plus index
2523 register address. You should define this macro if base plus index
2524 addresses have different requirements than other base register uses.
2527 @defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2528 A C expression whose value is the register class to which a valid
2529 base register for a memory reference in mode @var{mode} to address
2530 space @var{address_space} must belong. @var{outer_code} and @var{index_code}
2531 define the context in which the base register occurs. @var{outer_code} is
2532 the code of the immediately enclosing expression (@code{MEM} for the top level
2533 of an address, @code{ADDRESS} for something that occurs in an
2534 @code{address_operand}). @var{index_code} is the code of the corresponding
2535 index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise.
2538 @defmac INDEX_REG_CLASS
2539 A macro whose definition is the name of the class to which a valid
2540 index register must belong. An index register is one used in an
2541 address where its value is either multiplied by a scale factor or
2542 added to another register (as well as added to a displacement).
2545 @defmac REGNO_OK_FOR_BASE_P (@var{num})
2546 A C expression which is nonzero if register number @var{num} is
2547 suitable for use as a base register in operand addresses.
2550 @defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode})
2551 A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that
2552 that expression may examine the mode of the memory reference in
2553 @var{mode}. You should define this macro if the mode of the memory
2554 reference affects whether a register may be used as a base register. If
2555 you define this macro, the compiler will use it instead of
2556 @code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for
2557 addresses that appear outside a @code{MEM}, i.e., as an
2558 @code{address_operand}.
2561 @defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode})
2562 A C expression which is nonzero if register number @var{num} is suitable for
2563 use as a base register in base plus index operand addresses, accessing
2564 memory in mode @var{mode}. It may be either a suitable hard register or a
2565 pseudo register that has been allocated such a hard register. You should
2566 define this macro if base plus index addresses have different requirements
2567 than other base register uses.
2569 Use of this macro is deprecated; please use the more general
2570 @code{REGNO_MODE_CODE_OK_FOR_BASE_P}.
2573 @defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code})
2574 A C expression which is nonzero if register number @var{num} is
2575 suitable for use as a base register in operand addresses, accessing
2576 memory in mode @var{mode} in address space @var{address_space}.
2577 This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except
2578 that that expression may examine the context in which the register
2579 appears in the memory reference. @var{outer_code} is the code of the
2580 immediately enclosing expression (@code{MEM} if at the top level of the
2581 address, @code{ADDRESS} for something that occurs in an
2582 @code{address_operand}). @var{index_code} is the code of the
2583 corresponding index expression if @var{outer_code} is @code{PLUS};
2584 @code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses
2585 that appear outside a @code{MEM}, i.e., as an @code{address_operand}.
2588 @defmac REGNO_OK_FOR_INDEX_P (@var{num})
2589 A C expression which is nonzero if register number @var{num} is
2590 suitable for use as an index register in operand addresses. It may be
2591 either a suitable hard register or a pseudo register that has been
2592 allocated such a hard register.
2594 The difference between an index register and a base register is that
2595 the index register may be scaled. If an address involves the sum of
2596 two registers, neither one of them scaled, then either one may be
2597 labeled the ``base'' and the other the ``index''; but whichever
2598 labeling is used must fit the machine's constraints of which registers
2599 may serve in each capacity. The compiler will try both labelings,
2600 looking for one that is valid, and will reload one or both registers
2601 only if neither labeling works.
2604 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass})
2605 A target hook that places additional preference on the register
2606 class to use when it is necessary to rename a register in class
2607 @var{rclass} to another class, or perhaps @var{NO_REGS}, if no
2608 preferred register class is found or hook @code{preferred_rename_class}
2610 Sometimes returning a more restrictive class makes better code. For
2611 example, on ARM, thumb-2 instructions using @code{LO_REGS} may be
2612 smaller than instructions using @code{GENERIC_REGS}. By returning
2613 @code{LO_REGS} from @code{preferred_rename_class}, code size can
2617 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2618 A target hook that places additional restrictions on the register class
2619 to use when it is necessary to copy value @var{x} into a register in class
2620 @var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps
2621 another, smaller class.
2623 The default version of this hook always returns value of @code{rclass} argument.
2625 Sometimes returning a more restrictive class makes better code. For
2626 example, on the 68000, when @var{x} is an integer constant that is in range
2627 for a @samp{moveq} instruction, the value of this macro is always
2628 @code{DATA_REGS} as long as @var{rclass} includes the data registers.
2629 Requiring a data register guarantees that a @samp{moveq} will be used.
2631 One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return
2632 @var{rclass} is if @var{x} is a legitimate constant which cannot be
2633 loaded into some register class. By returning @code{NO_REGS} you can
2634 force @var{x} into a memory location. For example, rs6000 can load
2635 immediate values into general-purpose registers, but does not have an
2636 instruction for loading an immediate value into a floating-point
2637 register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2638 @var{x} is a floating-point constant. If the constant can't be loaded
2639 into any kind of register, code generation will be better if
2640 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2641 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2643 If an insn has pseudos in it after register allocation, reload will go
2644 through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS}
2645 to find the best one. Returning @code{NO_REGS}, in this case, makes
2646 reload add a @code{!} in front of the constraint: the x86 back-end uses
2647 this feature to discourage usage of 387 registers when math is done in
2648 the SSE registers (and vice versa).
2651 @defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class})
2652 A C expression that places additional restrictions on the register class
2653 to use when it is necessary to copy value @var{x} into a register in class
2654 @var{class}. The value is a register class; perhaps @var{class}, or perhaps
2655 another, smaller class. On many machines, the following definition is
2659 #define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS
2662 Sometimes returning a more restrictive class makes better code. For
2663 example, on the 68000, when @var{x} is an integer constant that is in range
2664 for a @samp{moveq} instruction, the value of this macro is always
2665 @code{DATA_REGS} as long as @var{class} includes the data registers.
2666 Requiring a data register guarantees that a @samp{moveq} will be used.
2668 One case where @code{PREFERRED_RELOAD_CLASS} must not return
2669 @var{class} is if @var{x} is a legitimate constant which cannot be
2670 loaded into some register class. By returning @code{NO_REGS} you can
2671 force @var{x} into a memory location. For example, rs6000 can load
2672 immediate values into general-purpose registers, but does not have an
2673 instruction for loading an immediate value into a floating-point
2674 register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when
2675 @var{x} is a floating-point constant. If the constant cannot be loaded
2676 into any kind of register, code generation will be better if
2677 @code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead
2678 of using @code{TARGET_PREFERRED_RELOAD_CLASS}.
2680 If an insn has pseudos in it after register allocation, reload will go
2681 through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS}
2682 to find the best one. Returning @code{NO_REGS}, in this case, makes
2683 reload add a @code{!} in front of the constraint: the x86 back-end uses
2684 this feature to discourage usage of 387 registers when math is done in
2685 the SSE registers (and vice versa).
2688 @deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass})
2689 Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of
2692 The default version of this hook always returns value of @code{rclass}
2695 You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage
2696 reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}.
2699 @defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class})
2700 A C expression that places additional restrictions on the register class
2701 to use when it is necessary to be able to hold a value of mode
2702 @var{mode} in a reload register for which class @var{class} would
2705 Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when
2706 there are certain modes that simply cannot go in certain reload classes.
2708 The value is a register class; perhaps @var{class}, or perhaps another,
2711 Don't define this macro unless the target machine has limitations which
2712 require the macro to do something nontrivial.
2715 @deftypefn {Target Hook} reg_class_t TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, reg_class_t @var{reload_class}, machine_mode @var{reload_mode}, secondary_reload_info *@var{sri})
2716 Many machines have some registers that cannot be copied directly to or
2717 from memory or even from other types of registers. An example is the
2718 @samp{MQ} register, which on most machines, can only be copied to or
2719 from general registers, but not memory. Below, we shall be using the
2720 term 'intermediate register' when a move operation cannot be performed
2721 directly, but has to be done by copying the source into the intermediate
2722 register first, and then copying the intermediate register to the
2723 destination. An intermediate register always has the same mode as
2724 source and destination. Since it holds the actual value being copied,
2725 reload might apply optimizations to re-use an intermediate register
2726 and eliding the copy from the source when it can determine that the
2727 intermediate register still holds the required value.
2729 Another kind of secondary reload is required on some machines which
2730 allow copying all registers to and from memory, but require a scratch
2731 register for stores to some memory locations (e.g., those with symbolic
2732 address on the RT, and those with certain symbolic address on the SPARC
2733 when compiling PIC)@. Scratch registers need not have the same mode
2734 as the value being copied, and usually hold a different value than
2735 that being copied. Special patterns in the md file are needed to
2736 describe how the copy is performed with the help of the scratch register;
2737 these patterns also describe the number, register class(es) and mode(s)
2738 of the scratch register(s).
2740 In some cases, both an intermediate and a scratch register are required.
2742 For input reloads, this target hook is called with nonzero @var{in_p},
2743 and @var{x} is an rtx that needs to be copied to a register of class
2744 @var{reload_class} in @var{reload_mode}. For output reloads, this target
2745 hook is called with zero @var{in_p}, and a register of class @var{reload_class}
2746 needs to be copied to rtx @var{x} in @var{reload_mode}.
2748 If copying a register of @var{reload_class} from/to @var{x} requires
2749 an intermediate register, the hook @code{secondary_reload} should
2750 return the register class required for this intermediate register.
2751 If no intermediate register is required, it should return NO_REGS.
2752 If more than one intermediate register is required, describe the one
2753 that is closest in the copy chain to the reload register.
2755 If scratch registers are needed, you also have to describe how to
2756 perform the copy from/to the reload register to/from this
2757 closest intermediate register. Or if no intermediate register is
2758 required, but still a scratch register is needed, describe the
2759 copy from/to the reload register to/from the reload operand @var{x}.
2761 You do this by setting @code{sri->icode} to the instruction code of a pattern
2762 in the md file which performs the move. Operands 0 and 1 are the output
2763 and input of this copy, respectively. Operands from operand 2 onward are
2764 for scratch operands. These scratch operands must have a mode, and a
2765 single-register-class
2766 @c [later: or memory]
2769 When an intermediate register is used, the @code{secondary_reload}
2770 hook will be called again to determine how to copy the intermediate
2771 register to/from the reload operand @var{x}, so your hook must also
2772 have code to handle the register class of the intermediate operand.
2774 @c [For later: maybe we'll allow multi-alternative reload patterns -
2775 @c the port maintainer could name a mov<mode> pattern that has clobbers -
2776 @c and match the constraints of input and output to determine the required
2777 @c alternative. A restriction would be that constraints used to match
2778 @c against reloads registers would have to be written as register class
2779 @c constraints, or we need a new target macro / hook that tells us if an
2780 @c arbitrary constraint can match an unknown register of a given class.
2781 @c Such a macro / hook would also be useful in other places.]
2784 @var{x} might be a pseudo-register or a @code{subreg} of a
2785 pseudo-register, which could either be in a hard register or in memory.
2786 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2787 in memory and the hard register number if it is in a register.
2789 Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are
2790 currently not supported. For the time being, you will have to continue
2791 to use @code{TARGET_SECONDARY_MEMORY_NEEDED} for that purpose.
2793 @code{copy_cost} also uses this target hook to find out how values are
2794 copied. If you want it to include some extra cost for the need to allocate
2795 (a) scratch register(s), set @code{sri->extra_cost} to the additional cost.
2796 Or if two dependent moves are supposed to have a lower cost than the sum
2797 of the individual moves due to expected fortuitous scheduling and/or special
2798 forwarding logic, you can set @code{sri->extra_cost} to a negative amount.
2801 @defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2802 @defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2803 @defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x})
2804 These macros are obsolete, new ports should use the target hook
2805 @code{TARGET_SECONDARY_RELOAD} instead.
2807 These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD}
2808 target hook. Older ports still define these macros to indicate to the
2809 reload phase that it may
2810 need to allocate at least one register for a reload in addition to the
2811 register to contain the data. Specifically, if copying @var{x} to a
2812 register @var{class} in @var{mode} requires an intermediate register,
2813 you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the
2814 largest register class all of whose registers can be used as
2815 intermediate registers or scratch registers.
2817 If copying a register @var{class} in @var{mode} to @var{x} requires an
2818 intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS}
2819 was supposed to be defined to return the largest register
2820 class required. If the
2821 requirements for input and output reloads were the same, the macro
2822 @code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both
2825 The values returned by these macros are often @code{GENERAL_REGS}.
2826 Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x}
2827 can be directly copied to or from a register of @var{class} in
2828 @var{mode} without requiring a scratch register. Do not define this
2829 macro if it would always return @code{NO_REGS}.
2831 If a scratch register is required (either with or without an
2832 intermediate register), you were supposed to define patterns for
2833 @samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required
2834 (@pxref{Standard Names}. These patterns, which were normally
2835 implemented with a @code{define_expand}, should be similar to the
2836 @samp{mov@var{m}} patterns, except that operand 2 is the scratch
2839 These patterns need constraints for the reload register and scratch
2841 contain a single register class. If the original reload register (whose
2842 class is @var{class}) can meet the constraint given in the pattern, the
2843 value returned by these macros is used for the class of the scratch
2844 register. Otherwise, two additional reload registers are required.
2845 Their classes are obtained from the constraints in the insn pattern.
2847 @var{x} might be a pseudo-register or a @code{subreg} of a
2848 pseudo-register, which could either be in a hard register or in memory.
2849 Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is
2850 in memory and the hard register number if it is in a register.
2852 These macros should not be used in the case where a particular class of
2853 registers can only be copied to memory and not to another class of
2854 registers. In that case, secondary reload registers are not needed and
2855 would not be helpful. Instead, a stack location must be used to perform
2856 the copy and the @code{mov@var{m}} pattern should use memory as an
2857 intermediate storage. This case often occurs between floating-point and
2861 @deftypefn {Target Hook} bool TARGET_SECONDARY_MEMORY_NEEDED (machine_mode @var{mode}, reg_class_t @var{class1}, reg_class_t @var{class2})
2862 Certain machines have the property that some registers cannot be copied
2863 to some other registers without using memory. Define this hook on
2864 those machines to return true if objects of mode @var{m} in registers
2865 of @var{class1} can only be copied to registers of class @var{class2} by
2866 storing a register of @var{class1} into memory and loading that memory
2867 location into a register of @var{class2}. The default definition returns
2868 false for all inputs.
2871 @defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode})
2872 Normally when @code{TARGET_SECONDARY_MEMORY_NEEDED} is defined, the compiler
2873 allocates a stack slot for a memory location needed for register copies.
2874 If this macro is defined, the compiler instead uses the memory location
2875 defined by this macro.
2877 Do not define this macro if you do not define
2878 @code{TARGET_SECONDARY_MEMORY_NEEDED}.
2881 @deftypefn {Target Hook} machine_mode TARGET_SECONDARY_MEMORY_NEEDED_MODE (machine_mode @var{mode})
2882 If @code{TARGET_SECONDARY_MEMORY_NEEDED} tells the compiler to use memory
2883 when moving between two particular registers of mode @var{mode},
2884 this hook specifies the mode that the memory should have.
2886 The default depends on @code{TARGET_LRA_P}. Without LRA, the default
2887 is to use a word-sized mode for integral modes that are smaller than a
2888 a word. This is right thing to do on most machines because it ensures
2889 that all bits of the register are copied and prevents accesses to the
2890 registers in a narrower mode, which some machines prohibit for
2891 floating-point registers.
2893 However, this default behavior is not correct on some machines, such as
2894 the DEC Alpha, that store short integers in floating-point registers
2895 differently than in integer registers. On those machines, the default
2896 widening will not work correctly and you must define this hook to
2897 suppress that widening in some cases. See the file @file{alpha.cc} for
2900 With LRA, the default is to use @var{mode} unmodified.
2903 @deftypefn {Target Hook} void TARGET_SELECT_EARLY_REMAT_MODES (sbitmap @var{modes})
2904 On some targets, certain modes cannot be held in registers around a
2905 standard ABI call and are relatively expensive to spill to the stack.
2906 The early rematerialization pass can help in such cases by aggressively
2907 recomputing values after calls, so that they don't need to be spilled.
2909 This hook returns the set of such modes by setting the associated bits
2910 in @var{modes}. The default implementation selects no modes, which has
2911 the effect of disabling the early rematerialization pass.
2914 @deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass})
2915 A target hook which returns @code{true} if pseudos that have been assigned
2916 to registers of class @var{rclass} would likely be spilled because
2917 registers of @var{rclass} are needed for spill registers.
2919 The default version of this target hook returns @code{true} if @var{rclass}
2920 has exactly one register and @code{false} otherwise. On most machines, this
2921 default should be used. For generally register-starved machines, such as
2922 i386, or machines with right register constraints, such as SH, this hook
2923 can be used to avoid excessive spilling.
2925 This hook is also used by some of the global intra-procedural code
2926 transformations to throtle code motion, to avoid increasing register
2930 @deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, machine_mode @var{mode})
2931 A target hook returns the maximum number of consecutive registers
2932 of class @var{rclass} needed to hold a value of mode @var{mode}.
2934 This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}.
2935 In fact, the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass},
2936 @var{mode})} target hook should be the maximum value of
2937 @code{TARGET_HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno}
2938 values in the class @var{rclass}.
2940 This target hook helps control the handling of multiple-word values
2943 The default version of this target hook returns the size of @var{mode}
2947 @defmac CLASS_MAX_NREGS (@var{class}, @var{mode})
2948 A C expression for the maximum number of consecutive registers
2949 of class @var{class} needed to hold a value of mode @var{mode}.
2951 This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}. In fact,
2952 the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})}
2953 should be the maximum value of @code{TARGET_HARD_REGNO_NREGS (@var{regno},
2954 @var{mode})} for all @var{regno} values in the class @var{class}.
2956 This macro helps control the handling of multiple-word values
2960 @deftypefn {Target Hook} bool TARGET_CAN_CHANGE_MODE_CLASS (machine_mode @var{from}, machine_mode @var{to}, reg_class_t @var{rclass})
2961 This hook returns true if it is possible to bitcast values held in
2962 registers of class @var{rclass} from mode @var{from} to mode @var{to}
2963 and if doing so preserves the low-order bits that are common to both modes.
2964 The result is only meaningful if @var{rclass} has registers that can hold
2965 both @code{from} and @code{to}. The default implementation returns true.
2967 As an example of when such bitcasting is invalid, loading 32-bit integer or
2968 floating-point objects into floating-point registers on Alpha extends them
2969 to 64 bits. Therefore loading a 64-bit object and then storing it as a
2970 32-bit object does not store the low-order 32 bits, as would be the case
2971 for a normal register. Therefore, @file{alpha.h} defines
2972 @code{TARGET_CAN_CHANGE_MODE_CLASS} to return:
2975 (GET_MODE_SIZE (from) == GET_MODE_SIZE (to)
2976 || !reg_classes_intersect_p (FLOAT_REGS, rclass))
2979 Even if storing from a register in mode @var{to} would be valid,
2980 if both @var{from} and @code{raw_reg_mode} for @var{rclass} are wider
2981 than @code{word_mode}, then we must prevent @var{to} narrowing the
2982 mode. This happens when the middle-end assumes that it can load
2983 or store pieces of an @var{N}-word pseudo, and that the pseudo will
2984 eventually be allocated to @var{N} @code{word_mode} hard registers.
2985 Failure to prevent this kind of mode change will result in the
2986 entire @code{raw_reg_mode} being modified instead of the partial
2987 value that the middle-end intended.
2990 @deftypefn {Target Hook} reg_class_t TARGET_IRA_CHANGE_PSEUDO_ALLOCNO_CLASS (int, @var{reg_class_t}, @var{reg_class_t})
2991 A target hook which can change allocno class for given pseudo from
2992 allocno and best class calculated by IRA.
2994 The default version of this target hook always returns given class.
2997 @deftypefn {Target Hook} bool TARGET_LRA_P (void)
2998 A target hook which returns true if we use LRA instead of reload pass.
3000 The default version of this target hook returns true. New ports
3001 should use LRA, and existing ports are encouraged to convert.
3004 @deftypefn {Target Hook} int TARGET_REGISTER_PRIORITY (int)
3005 A target hook which returns the register priority number to which the
3006 register @var{hard_regno} belongs to. The bigger the number, the
3007 more preferable the hard register usage (when all other conditions are
3008 the same). This hook can be used to prefer some hard register over
3009 others in LRA. For example, some x86-64 register usage needs
3010 additional prefix which makes instructions longer. The hook can
3011 return lower priority number for such registers make them less favorable
3012 and as result making the generated code smaller.
3014 The default version of this target hook returns always zero.
3017 @deftypefn {Target Hook} bool TARGET_REGISTER_USAGE_LEVELING_P (void)
3018 A target hook which returns true if we need register usage leveling.
3019 That means if a few hard registers are equally good for the
3020 assignment, we choose the least used hard register. The register
3021 usage leveling may be profitable for some targets. Don't use the
3022 usage leveling for targets with conditional execution or targets
3023 with big register files as it hurts if-conversion and cross-jumping
3026 The default version of this target hook returns always false.
3029 @deftypefn {Target Hook} bool TARGET_DIFFERENT_ADDR_DISPLACEMENT_P (void)
3030 A target hook which returns true if an address with the same structure
3031 can have different maximal legitimate displacement. For example, the
3032 displacement can depend on memory mode or on operand combinations in
3035 The default version of this target hook returns always false.
3038 @deftypefn {Target Hook} bool TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P (rtx @var{subst})
3039 A target hook which returns @code{true} if @var{subst} can't
3040 substitute safely pseudos with equivalent memory values during
3041 register allocation.
3042 The default version of this target hook returns @code{false}.
3043 On most machines, this default should be used. For generally
3044 machines with non orthogonal register usage for addressing, such
3045 as SH, this hook can be used to avoid excessive spilling.
3048 @deftypefn {Target Hook} bool TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT (rtx *@var{offset1}, rtx *@var{offset2}, poly_int64 @var{orig_offset}, machine_mode @var{mode})
3049 This hook tries to split address offset @var{orig_offset} into
3050 two parts: one that should be added to the base address to create
3051 a local anchor point, and an additional offset that can be applied
3052 to the anchor to address a value of mode @var{mode}. The idea is that
3053 the local anchor could be shared by other accesses to nearby locations.
3055 The hook returns true if it succeeds, storing the offset of the
3056 anchor from the base in @var{offset1} and the offset of the final address
3057 from the anchor in @var{offset2}. The default implementation returns false.
3060 @deftypefn {Target Hook} reg_class_t TARGET_SPILL_CLASS (reg_class_t, @var{machine_mode})
3061 This hook defines a class of registers which could be used for spilling
3062 pseudos of the given mode and class, or @code{NO_REGS} if only memory
3063 should be used. Not defining this hook is equivalent to returning
3064 @code{NO_REGS} for all inputs.
3067 @deftypefn {Target Hook} bool TARGET_ADDITIONAL_ALLOCNO_CLASS_P (reg_class_t)
3068 This hook should return @code{true} if given class of registers should
3069 be an allocno class in any way. Usually RA uses only one register
3070 class from all classes containing the same register set. In some
3071 complicated cases, you need to have two or more such classes as
3072 allocno ones for RA correct work. Not defining this hook is
3073 equivalent to returning @code{false} for all inputs.
3076 @deftypefn {Target Hook} scalar_int_mode TARGET_CSTORE_MODE (enum insn_code @var{icode})
3077 This hook defines the machine mode to use for the boolean result of
3078 conditional store patterns. The ICODE argument is the instruction code
3079 for the cstore being performed. Not definiting this hook is the same
3080 as accepting the mode encoded into operand 0 of the cstore expander
3084 @deftypefn {Target Hook} int TARGET_COMPUTE_PRESSURE_CLASSES (enum reg_class *@var{pressure_classes})
3085 A target hook which lets a backend compute the set of pressure classes to
3086 be used by those optimization passes which take register pressure into
3087 account, as opposed to letting IRA compute them. It returns the number of
3088 register classes stored in the array @var{pressure_classes}.
3091 @node Stack and Calling
3092 @section Stack Layout and Calling Conventions
3093 @cindex calling conventions
3095 @c prevent bad page break with this line
3096 This describes the stack layout and calling conventions.
3100 * Exception Handling::
3105 * Register Arguments::
3107 * Aggregate Return::
3112 * Shrink-wrapping separate components::
3113 * Stack Smashing Protection::
3114 * Miscellaneous Register Hooks::
3118 @subsection Basic Stack Layout
3119 @cindex stack frame layout
3120 @cindex frame layout
3122 @c prevent bad page break with this line
3123 Here is the basic stack layout.
3125 @defmac STACK_GROWS_DOWNWARD
3126 Define this macro to be true if pushing a word onto the stack moves the stack
3127 pointer to a smaller address, and false otherwise.
3130 @defmac STACK_PUSH_CODE
3131 This macro defines the operation used when something is pushed
3132 on the stack. In RTL, a push operation will be
3133 @code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})}
3135 The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC},
3136 and @code{POST_INC}. Which of these is correct depends on
3137 the stack direction and on whether the stack pointer points
3138 to the last item on the stack or whether it points to the
3139 space for the next item on the stack.
3141 The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is
3142 true, which is almost always right, and @code{PRE_INC} otherwise,
3143 which is often wrong.
3146 @defmac FRAME_GROWS_DOWNWARD
3147 Define this macro to nonzero value if the addresses of local variable slots
3148 are at negative offsets from the frame pointer.
3151 @defmac ARGS_GROW_DOWNWARD
3152 Define this macro if successive arguments to a function occupy decreasing
3153 addresses on the stack.
3156 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_STARTING_FRAME_OFFSET (void)
3157 This hook returns the offset from the frame pointer to the first local
3158 variable slot to be allocated. If @code{FRAME_GROWS_DOWNWARD}, it is the
3159 offset to @emph{end} of the first slot allocated, otherwise it is the
3160 offset to @emph{beginning} of the first slot allocated. The default
3161 implementation returns 0.
3164 @defmac STACK_ALIGNMENT_NEEDED
3165 Define to zero to disable final alignment of the stack during reload.
3166 The nonzero default for this macro is suitable for most ports.
3168 On ports where @code{TARGET_STARTING_FRAME_OFFSET} is nonzero or where there
3169 is a register save block following the local block that doesn't require
3170 alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable
3171 stack alignment and do it in the backend.
3174 @defmac STACK_POINTER_OFFSET
3175 Offset from the stack pointer register to the first location at which
3176 outgoing arguments are placed. If not specified, the default value of
3177 zero is used. This is the proper value for most machines.
3179 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3180 the first location at which outgoing arguments are placed.
3183 @defmac FIRST_PARM_OFFSET (@var{fundecl})
3184 Offset from the argument pointer register to the first argument's
3185 address. On some machines it may depend on the data type of the
3188 If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above
3189 the first argument's address.
3192 @defmac STACK_DYNAMIC_OFFSET (@var{fundecl})
3193 Offset from the stack pointer register to an item dynamically allocated
3194 on the stack, e.g., by @code{alloca}.
3196 The default value for this macro is @code{STACK_POINTER_OFFSET} plus the
3197 length of the outgoing arguments. The default is correct for most
3198 machines. See @file{function.cc} for details.
3201 @defmac INITIAL_FRAME_ADDRESS_RTX
3202 A C expression whose value is RTL representing the address of the initial
3203 stack frame. This address is passed to @code{RETURN_ADDR_RTX} and
3204 @code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable
3205 default value will be used. Define this macro in order to make frame pointer
3206 elimination work in the presence of @code{__builtin_frame_address (count)} and
3207 @code{__builtin_return_address (count)} for @code{count} not equal to zero.
3210 @defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr})
3211 A C expression whose value is RTL representing the address in a stack
3212 frame where the pointer to the caller's frame is stored. Assume that
3213 @var{frameaddr} is an RTL expression for the address of the stack frame
3216 If you don't define this macro, the default is to return the value
3217 of @var{frameaddr}---that is, the stack frame address is also the
3218 address of the stack word that points to the previous frame.
3221 @defmac SETUP_FRAME_ADDRESSES
3222 A C expression that produces the machine-specific code to
3223 setup the stack so that arbitrary frames can be accessed. For example,
3224 on the SPARC, we must flush all of the register windows to the stack
3225 before we can access arbitrary stack frames. You will seldom need to
3226 define this macro. The default is to do nothing.
3229 @deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void)
3230 This target hook should return an rtx that is used to store
3231 the address of the current frame into the built in @code{setjmp} buffer.
3232 The default value, @code{virtual_stack_vars_rtx}, is correct for most
3233 machines. One reason you may need to define this target hook is if
3234 @code{hard_frame_pointer_rtx} is the appropriate value on your machine.
3237 @defmac FRAME_ADDR_RTX (@var{frameaddr})
3238 A C expression whose value is RTL representing the value of the frame
3239 address for the current frame. @var{frameaddr} is the frame pointer
3240 of the current frame. This is used for __builtin_frame_address.
3241 You need only define this macro if the frame address is not the same
3242 as the frame pointer. Most machines do not need to define it.
3245 @defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr})
3246 A C expression whose value is RTL representing the value of the return
3247 address for the frame @var{count} steps up from the current frame, after
3248 the prologue. @var{frameaddr} is the frame pointer of the @var{count}
3249 frame, or the frame pointer of the @var{count} @minus{} 1 frame if
3250 @code{RETURN_ADDR_IN_PREVIOUS_FRAME} is nonzero.
3252 The value of the expression must always be the correct address when
3253 @var{count} is zero, but may be @code{NULL_RTX} if there is no way to
3254 determine the return address of other frames.
3257 @defmac RETURN_ADDR_IN_PREVIOUS_FRAME
3258 Define this macro to nonzero value if the return address of a particular
3259 stack frame is accessed from the frame pointer of the previous stack
3260 frame. The zero default for this macro is suitable for most ports.
3263 @defmac INCOMING_RETURN_ADDR_RTX
3264 A C expression whose value is RTL representing the location of the
3265 incoming return address at the beginning of any function, before the
3266 prologue. This RTL is either a @code{REG}, indicating that the return
3267 value is saved in @samp{REG}, or a @code{MEM} representing a location in
3270 You only need to define this macro if you want to support call frame
3271 debugging information like that provided by DWARF 2.
3273 If this RTL is a @code{REG}, you should also define
3274 @code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}.
3277 @defmac DWARF_ALT_FRAME_RETURN_COLUMN
3278 A C expression whose value is an integer giving a DWARF 2 column
3279 number that may be used as an alternative return column. The column
3280 must not correspond to any gcc hard register (that is, it must not
3281 be in the range of @code{DWARF_FRAME_REGNUM}).
3283 This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a
3284 general register, but an alternative column needs to be used for signal
3285 frames. Some targets have also used different frame return columns
3289 @defmac DWARF_ZERO_REG
3290 A C expression whose value is an integer giving a DWARF 2 register
3291 number that is considered to always have the value zero. This should
3292 only be defined if the target has an architected zero register, and
3293 someone decided it was a good idea to use that register number to
3294 terminate the stack backtrace. New ports should avoid this.
3297 @deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index})
3298 This target hook allows the backend to emit frame-related insns that
3299 contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging
3300 info engine will invoke it on insns of the form
3302 (set (reg) (unspec [@dots{}] UNSPEC_INDEX))
3306 (set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)).
3308 to let the backend emit the call frame instructions. @var{label} is
3309 the CFI label attached to the insn, @var{pattern} is the pattern of
3310 the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}.
3313 @deftypefn {Target Hook} {unsigned int} TARGET_DWARF_POLY_INDETERMINATE_VALUE (unsigned int @var{i}, unsigned int *@var{factor}, int *@var{offset})
3314 Express the value of @code{poly_int} indeterminate @var{i} as a DWARF
3315 expression, with @var{i} counting from 1. Return the number of a DWARF
3316 register @var{R} and set @samp{*@var{factor}} and @samp{*@var{offset}} such
3317 that the value of the indeterminate is:
3319 value_of(@var{R}) / @var{factor} - @var{offset}
3322 A target only needs to define this hook if it sets
3323 @samp{NUM_POLY_INT_COEFFS} to a value greater than 1.
3326 @defmac INCOMING_FRAME_SP_OFFSET
3327 A C expression whose value is an integer giving the offset, in bytes,
3328 from the value of the stack pointer register to the top of the stack
3329 frame at the beginning of any function, before the prologue. The top of
3330 the frame is defined to be the value of the stack pointer in the
3331 previous frame, just before the call instruction.
3333 You only need to define this macro if you want to support call frame
3334 debugging information like that provided by DWARF 2.
3337 @defmac DEFAULT_INCOMING_FRAME_SP_OFFSET
3338 Like @code{INCOMING_FRAME_SP_OFFSET}, but must be the same for all
3339 functions of the same ABI, and when using GAS @code{.cfi_*} directives
3340 must also agree with the default CFI GAS emits. Define this macro
3341 only if @code{INCOMING_FRAME_SP_OFFSET} can have different values
3342 between different functions of the same ABI or when
3343 @code{INCOMING_FRAME_SP_OFFSET} does not agree with GAS default CFI.
3346 @defmac ARG_POINTER_CFA_OFFSET (@var{fundecl})
3347 A C expression whose value is an integer giving the offset, in bytes,
3348 from the argument pointer to the canonical frame address (cfa). The
3349 final value should coincide with that calculated by
3350 @code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable
3351 during virtual register instantiation.
3353 The default value for this macro is
3354 @code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size},
3355 which is correct for most machines; in general, the arguments are found
3356 immediately before the stack frame. Note that this is not the case on
3357 some targets that save registers into the caller's frame, such as SPARC
3358 and rs6000, and so such targets need to define this macro.
3360 You only need to define this macro if the default is incorrect, and you
3361 want to support call frame debugging information like that provided by
3365 @defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl})
3366 If defined, a C expression whose value is an integer giving the offset
3367 in bytes from the frame pointer to the canonical frame address (cfa).
3368 The final value should coincide with that calculated by
3369 @code{INCOMING_FRAME_SP_OFFSET}.
3371 Normally the CFA is calculated as an offset from the argument pointer,
3372 via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is
3373 variable due to the ABI, this may not be possible. If this macro is
3374 defined, it implies that the virtual register instantiation should be
3375 based on the frame pointer instead of the argument pointer. Only one
3376 of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET}
3380 @defmac CFA_FRAME_BASE_OFFSET (@var{fundecl})
3381 If defined, a C expression whose value is an integer giving the offset
3382 in bytes from the canonical frame address (cfa) to the frame base used
3383 in DWARF 2 debug information. The default is zero. A different value
3384 may reduce the size of debug information on some ports.
3387 @node Exception Handling
3388 @subsection Exception Handling Support
3389 @cindex exception handling
3391 @defmac EH_RETURN_DATA_REGNO (@var{N})
3392 A C expression whose value is the @var{N}th register number used for
3393 data by exception handlers, or @code{INVALID_REGNUM} if fewer than
3394 @var{N} registers are usable.
3396 The exception handling library routines communicate with the exception
3397 handlers via a set of agreed upon registers. Ideally these registers
3398 should be call-clobbered; it is possible to use call-saved registers,
3399 but may negatively impact code size. The target must support at least
3400 2 data registers, but should define 4 if there are enough free registers.
3402 You must define this macro if you want to support call frame exception
3403 handling like that provided by DWARF 2.
3406 @defmac EH_RETURN_STACKADJ_RTX
3407 A C expression whose value is RTL representing a location in which
3408 to store a stack adjustment to be applied before function return.
3409 This is used to unwind the stack to an exception handler's call frame.
3410 It will be assigned zero on code paths that return normally.
3412 Typically this is a call-clobbered hard register that is otherwise
3413 untouched by the epilogue, but could also be a stack slot.
3415 Do not define this macro if the stack pointer is saved and restored
3416 by the regular prolog and epilog code in the call frame itself; in
3417 this case, the exception handling library routines will update the
3418 stack location to be restored in place. Otherwise, you must define
3419 this macro if you want to support call frame exception handling like
3420 that provided by DWARF 2.
3423 @defmac EH_RETURN_HANDLER_RTX
3424 A C expression whose value is RTL representing a location in which
3425 to store the address of an exception handler to which we should
3426 return. It will not be assigned on code paths that return normally.
3428 Typically this is the location in the call frame at which the normal
3429 return address is stored. For targets that return by popping an
3430 address off the stack, this might be a memory address just below
3431 the @emph{target} call frame rather than inside the current call
3432 frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already
3433 been assigned, so it may be used to calculate the location of the
3436 Some targets have more complex requirements than storing to an
3437 address calculable during initial code generation. In that case
3438 the @code{eh_return} instruction pattern should be used instead.
3440 If you want to support call frame exception handling, you must
3441 define either this macro or the @code{eh_return} instruction pattern.
3444 @defmac RETURN_ADDR_OFFSET
3445 If defined, an integer-valued C expression for which rtl will be generated
3446 to add it to the exception handler address before it is searched in the
3447 exception handling tables, and to subtract it again from the address before
3448 using it to return to the exception handler.
3451 @defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global})
3452 This macro chooses the encoding of pointers embedded in the exception
3453 handling sections. If at all possible, this should be defined such
3454 that the exception handling section will not require dynamic relocations,
3455 and so may be read-only.
3457 @var{code} is 0 for data, 1 for code labels, 2 for function pointers.
3458 @var{global} is true if the symbol may be affected by dynamic relocations.
3459 The macro should return a combination of the @code{DW_EH_PE_*} defines
3460 as found in @file{dwarf2.h}.
3462 If this macro is not defined, pointers will not be encoded but
3463 represented directly.
3466 @defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done})
3467 This macro allows the target to emit whatever special magic is required
3468 to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}.
3469 Generic code takes care of pc-relative and indirect encodings; this must
3470 be defined if the target uses text-relative or data-relative encodings.
3472 This is a C statement that branches to @var{done} if the format was
3473 handled. @var{encoding} is the format chosen, @var{size} is the number
3474 of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF}
3478 @defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs})
3479 This macro allows the target to add CPU and operating system specific
3480 code to the call-frame unwinder for use when there is no unwind data
3481 available. The most common reason to implement this macro is to unwind
3482 through signal frames.
3484 This macro is called from @code{uw_frame_state_for} in
3485 @file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and
3486 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3487 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra}
3488 for the address of the code being executed and @code{context->cfa} for
3489 the stack pointer value. If the frame can be decoded, the register
3490 save addresses should be updated in @var{fs} and the macro should
3491 evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded,
3492 the macro should evaluate to @code{_URC_END_OF_STACK}.
3494 For proper signal handling in Java this macro is accompanied by
3495 @code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers.
3498 @defmac MD_HANDLE_UNWABI (@var{context}, @var{fs})
3499 This macro allows the target to add operating system specific code to the
3500 call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive,
3501 usually used for signal or interrupt frames.
3503 This macro is called from @code{uw_update_context} in libgcc's
3504 @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context};
3505 @var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi}
3506 for the abi and context in the @code{.unwabi} directive. If the
3507 @code{.unwabi} directive can be handled, the register save addresses should
3508 be updated in @var{fs}.
3511 @defmac TARGET_USES_WEAK_UNWIND_INFO
3512 A C expression that evaluates to true if the target requires unwind
3513 info to be given comdat linkage. Define it to be @code{1} if comdat
3514 linkage is necessary. The default is @code{0}.
3517 @node Stack Checking
3518 @subsection Specifying How Stack Checking is Done
3520 GCC will check that stack references are within the boundaries of the
3521 stack, if the option @option{-fstack-check} is specified, in one of
3526 If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC
3527 will assume that you have arranged for full stack checking to be done
3528 at appropriate places in the configuration files. GCC will not do
3529 other special processing.
3532 If @code{STACK_CHECK_BUILTIN} is zero and the value of the
3533 @code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume
3534 that you have arranged for static stack checking (checking of the
3535 static stack frame of functions) to be done at appropriate places
3536 in the configuration files. GCC will only emit code to do dynamic
3537 stack checking (checking on dynamic stack allocations) using the third
3541 If neither of the above are true, GCC will generate code to periodically
3542 ``probe'' the stack pointer using the values of the macros defined below.
3545 If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined,
3546 GCC will change its allocation strategy for large objects if the option
3547 @option{-fstack-check} is specified: they will always be allocated
3548 dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes.
3550 @defmac STACK_CHECK_BUILTIN
3551 A nonzero value if stack checking is done by the configuration files in a
3552 machine-dependent manner. You should define this macro if stack checking
3553 is required by the ABI of your machine or if you would like to do stack
3554 checking in some more efficient way than the generic approach. The default
3555 value of this macro is zero.
3558 @defmac STACK_CHECK_STATIC_BUILTIN
3559 A nonzero value if static stack checking is done by the configuration files
3560 in a machine-dependent manner. You should define this macro if you would
3561 like to do static stack checking in some more efficient way than the generic
3562 approach. The default value of this macro is zero.
3565 @defmac STACK_CHECK_PROBE_INTERVAL_EXP
3566 An integer specifying the interval at which GCC must generate stack probe
3567 instructions, defined as 2 raised to this integer. You will normally
3568 define this macro so that the interval be no larger than the size of
3569 the ``guard pages'' at the end of a stack area. The default value
3570 of 12 (4096-byte interval) is suitable for most systems.
3573 @defmac STACK_CHECK_MOVING_SP
3574 An integer which is nonzero if GCC should move the stack pointer page by page
3575 when doing probes. This can be necessary on systems where the stack pointer
3576 contains the bottom address of the memory area accessible to the executing
3577 thread at any point in time. In this situation an alternate signal stack
3578 is required in order to be able to recover from a stack overflow. The
3579 default value of this macro is zero.
3582 @defmac STACK_CHECK_PROTECT
3583 The number of bytes of stack needed to recover from a stack overflow, for
3584 languages where such a recovery is supported. The default value of 4KB/8KB
3585 with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and
3586 8KB/12KB with other exception handling mechanisms should be adequate for most
3587 architectures and operating systems.
3590 The following macros are relevant only if neither STACK_CHECK_BUILTIN
3591 nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether
3592 in the opposite case.
3594 @defmac STACK_CHECK_MAX_FRAME_SIZE
3595 The maximum size of a stack frame, in bytes. GCC will generate probe
3596 instructions in non-leaf functions to ensure at least this many bytes of
3597 stack are available. If a stack frame is larger than this size, stack
3598 checking will not be reliable and GCC will issue a warning. The
3599 default is chosen so that GCC only generates one instruction on most
3600 systems. You should normally not change the default value of this macro.
3603 @defmac STACK_CHECK_FIXED_FRAME_SIZE
3604 GCC uses this value to generate the above warning message. It
3605 represents the amount of fixed frame used by a function, not including
3606 space for any callee-saved registers, temporaries and user variables.
3607 You need only specify an upper bound for this amount and will normally
3608 use the default of four words.
3611 @defmac STACK_CHECK_MAX_VAR_SIZE
3612 The maximum size, in bytes, of an object that GCC will place in the
3613 fixed area of the stack frame when the user specifies
3614 @option{-fstack-check}.
3615 GCC computed the default from the values of the above macros and you will
3616 normally not need to override that default.
3619 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_STACK_CLASH_PROTECTION_ALLOCA_PROBE_RANGE (void)
3620 Some targets have an ABI defined interval for which no probing needs to be done.
3621 When a probe does need to be done this same interval is used as the probe distance
3622 up when doing stack clash protection for alloca.
3623 On such targets this value can be set to override the default probing up interval.
3624 Define this variable to return nonzero if such a probe range is required or zero otherwise.
3625 Defining this hook also requires your functions which make use of alloca to have at least 8 byes
3626 of outgoing arguments. If this is not the case the stack will be corrupted.
3627 You need not define this macro if it would always have the value zero.
3631 @node Frame Registers
3632 @subsection Registers That Address the Stack Frame
3634 @c prevent bad page break with this line
3635 This discusses registers that address the stack frame.
3637 @defmac STACK_POINTER_REGNUM
3638 The register number of the stack pointer register, which must also be a
3639 fixed register according to @code{FIXED_REGISTERS}. On most machines,
3640 the hardware determines which register this is.
3643 @defmac FRAME_POINTER_REGNUM
3644 The register number of the frame pointer register, which is used to
3645 access automatic variables in the stack frame. On some machines, the
3646 hardware determines which register this is. On other machines, you can
3647 choose any register you wish for this purpose.
3650 @defmac HARD_FRAME_POINTER_REGNUM
3651 On some machines the offset between the frame pointer and starting
3652 offset of the automatic variables is not known until after register
3653 allocation has been done (for example, because the saved registers are
3654 between these two locations). On those machines, define
3655 @code{FRAME_POINTER_REGNUM} the number of a special, fixed register to
3656 be used internally until the offset is known, and define
3657 @code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number
3658 used for the frame pointer.
3660 You should define this macro only in the very rare circumstances when it
3661 is not possible to calculate the offset between the frame pointer and
3662 the automatic variables until after register allocation has been
3663 completed. When this macro is defined, you must also indicate in your
3664 definition of @code{ELIMINABLE_REGS} how to eliminate
3665 @code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM}
3666 or @code{STACK_POINTER_REGNUM}.
3668 Do not define this macro if it would be the same as
3669 @code{FRAME_POINTER_REGNUM}.
3672 @defmac ARG_POINTER_REGNUM
3673 The register number of the arg pointer register, which is used to access
3674 the function's argument list. On some machines, this is the same as the
3675 frame pointer register. On some machines, the hardware determines which
3676 register this is. On other machines, you can choose any register you
3677 wish for this purpose. If this is not the same register as the frame
3678 pointer register, then you must mark it as a fixed register according to
3679 @code{FIXED_REGISTERS}, or arrange to be able to eliminate it
3680 (@pxref{Elimination}).
3683 @defmac HARD_FRAME_POINTER_IS_FRAME_POINTER
3684 Define this to a preprocessor constant that is nonzero if
3685 @code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be
3686 the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM
3687 == FRAME_POINTER_REGNUM)}; you only need to define this macro if that
3688 definition is not suitable for use in preprocessor conditionals.
3691 @defmac HARD_FRAME_POINTER_IS_ARG_POINTER
3692 Define this to a preprocessor constant that is nonzero if
3693 @code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the
3694 same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM ==
3695 ARG_POINTER_REGNUM)}; you only need to define this macro if that
3696 definition is not suitable for use in preprocessor conditionals.
3699 @defmac RETURN_ADDRESS_POINTER_REGNUM
3700 The register number of the return address pointer register, which is used to
3701 access the current function's return address from the stack. On some
3702 machines, the return address is not at a fixed offset from the frame
3703 pointer or stack pointer or argument pointer. This register can be defined
3704 to point to the return address on the stack, and then be converted by
3705 @code{ELIMINABLE_REGS} into either the frame pointer or stack pointer.
3707 Do not define this macro unless there is no other way to get the return
3708 address from the stack.
3711 @defmac STATIC_CHAIN_REGNUM
3712 @defmacx STATIC_CHAIN_INCOMING_REGNUM
3713 Register numbers used for passing a function's static chain pointer. If
3714 register windows are used, the register number as seen by the called
3715 function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register
3716 number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If
3717 these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need
3720 The static chain register need not be a fixed register.
3722 If the static chain is passed in memory, these macros should not be
3723 defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used.
3726 @deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl_or_type}, bool @var{incoming_p})
3727 This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for
3728 targets that may use different static chain locations for different
3729 nested functions. This may be required if the target has function
3730 attributes that affect the calling conventions of the function and
3731 those calling conventions use different static chain locations.
3733 The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al.
3735 If the static chain is passed in memory, this hook should be used to
3736 provide rtx giving @code{mem} expressions that denote where they are stored.
3737 Often the @code{mem} expression as seen by the caller will be at an offset
3738 from the stack pointer and the @code{mem} expression as seen by the callee
3739 will be at an offset from the frame pointer.
3740 @findex stack_pointer_rtx
3741 @findex frame_pointer_rtx
3742 @findex arg_pointer_rtx
3743 The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and
3744 @code{arg_pointer_rtx} will have been initialized and should be used
3745 to refer to those items.
3748 @defmac DWARF_FRAME_REGISTERS
3749 This macro specifies the maximum number of hard registers that can be
3750 saved in a call frame. This is used to size data structures used in
3751 DWARF2 exception handling.
3753 Prior to GCC 3.0, this macro was needed in order to establish a stable
3754 exception handling ABI in the face of adding new hard registers for ISA
3755 extensions. In GCC 3.0 and later, the EH ABI is insulated from changes
3756 in the number of hard registers. Nevertheless, this macro can still be
3757 used to reduce the runtime memory requirements of the exception handling
3758 routines, which can be substantial if the ISA contains a lot of
3759 registers that are not call-saved.
3761 If this macro is not defined, it defaults to
3762 @code{FIRST_PSEUDO_REGISTER}.
3765 @defmac PRE_GCC3_DWARF_FRAME_REGISTERS
3767 This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided
3768 for backward compatibility in pre GCC 3.0 compiled code.
3770 If this macro is not defined, it defaults to
3771 @code{DWARF_FRAME_REGISTERS}.
3774 @defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno})
3776 Define this macro if the target's representation for dwarf registers
3777 is different than the internal representation for unwind column.
3778 Given a dwarf register, this macro should return the internal unwind
3779 column number to use instead.
3782 @defmac DWARF_FRAME_REGNUM (@var{regno})
3784 Define this macro if the target's representation for dwarf registers
3785 used in .eh_frame or .debug_frame is different from that used in other
3786 debug info sections. Given a GCC hard register number, this macro
3787 should return the .eh_frame register number. The default is
3788 @code{DEBUGGER_REGISTER_NUMBER (@var{regno})}.
3792 @defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh})
3794 Define this macro to map register numbers held in the call frame info
3795 that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that
3796 should be output in .debug_frame (@code{@var{for_eh}} is zero) and
3797 .eh_frame (@code{@var{for_eh}} is nonzero). The default is to
3798 return @code{@var{regno}}.
3802 @defmac REG_VALUE_IN_UNWIND_CONTEXT
3804 Define this macro if the target stores register values as
3805 @code{_Unwind_Word} type in unwind context. It should be defined if
3806 target register size is larger than the size of @code{void *}. The
3807 default is to store register values as @code{void *} type.
3811 @defmac ASSUME_EXTENDED_UNWIND_CONTEXT
3813 Define this macro to be 1 if the target always uses extended unwind
3814 context with version, args_size and by_value fields. If it is undefined,
3815 it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is
3816 defined and 0 otherwise.
3820 @defmac DWARF_LAZY_REGISTER_VALUE (@var{regno}, @var{value})
3821 Define this macro if the target has pseudo DWARF registers whose
3822 values need to be computed lazily on demand by the unwinder (such as when
3823 referenced in a CFA expression). The macro returns true if @var{regno}
3824 is such a register and stores its value in @samp{*@var{value}} if so.
3828 @subsection Eliminating Frame Pointer and Arg Pointer
3830 @c prevent bad page break with this line
3831 This is about eliminating the frame pointer and arg pointer.
3833 @deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void)
3834 This target hook should return @code{true} if a function must have and use
3835 a frame pointer. This target hook is called in the reload pass. If its return
3836 value is @code{true} the function will have a frame pointer.
3838 This target hook can in principle examine the current function and decide
3839 according to the facts, but on most machines the constant @code{false} or the
3840 constant @code{true} suffices. Use @code{false} when the machine allows code
3841 to be generated with no frame pointer, and doing so saves some time or space.
3842 Use @code{true} when there is no possible advantage to avoiding a frame
3845 In certain cases, the compiler does not know how to produce valid code
3846 without a frame pointer. The compiler recognizes those cases and
3847 automatically gives the function a frame pointer regardless of what
3848 @code{targetm.frame_pointer_required} returns. You don't need to worry about
3851 In a function that does not require a frame pointer, the frame pointer
3852 register can be allocated for ordinary usage, unless you mark it as a
3853 fixed register. See @code{FIXED_REGISTERS} for more information.
3855 Default return value is @code{false}.
3858 @defmac ELIMINABLE_REGS
3859 This macro specifies a table of register pairs used to eliminate
3860 unneeded registers that point into the stack frame.
3862 The definition of this macro is a list of structure initializations, each
3863 of which specifies an original and replacement register.
3865 On some machines, the position of the argument pointer is not known until
3866 the compilation is completed. In such a case, a separate hard register
3867 must be used for the argument pointer. This register can be eliminated by
3868 replacing it with either the frame pointer or the argument pointer,
3869 depending on whether or not the frame pointer has been eliminated.
3871 In this case, you might specify:
3873 #define ELIMINABLE_REGS \
3874 @{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \
3875 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \
3876 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@}
3879 Note that the elimination of the argument pointer with the stack pointer is
3880 specified first since that is the preferred elimination.
3883 @deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg})
3884 This target hook should return @code{true} if the compiler is allowed to
3885 try to replace register number @var{from_reg} with register number
3886 @var{to_reg}. This target hook will usually be @code{true}, since most of the
3887 cases preventing register elimination are things that the compiler already
3890 Default return value is @code{true}.
3893 @defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var})
3894 This macro returns the initial difference between the specified pair
3895 of registers. The value would be computed from information
3896 such as the result of @code{get_frame_size ()} and the tables of
3897 registers @code{df_regs_ever_live_p} and @code{call_used_regs}.
3900 @deftypefn {Target Hook} void TARGET_COMPUTE_FRAME_LAYOUT (void)
3901 This target hook is called once each time the frame layout needs to be
3902 recalculated. The calculations can be cached by the target and can then
3903 be used by @code{INITIAL_ELIMINATION_OFFSET} instead of re-computing the
3904 layout on every invocation of that hook. This is particularly useful
3905 for targets that have an expensive frame layout function. Implementing
3906 this callback is optional.
3909 @node Stack Arguments
3910 @subsection Passing Function Arguments on the Stack
3911 @cindex arguments on stack
3912 @cindex stack arguments
3914 The macros in this section control how arguments are passed
3915 on the stack. See the following section for other macros that
3916 control passing certain arguments in registers.
3918 @deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype})
3919 This target hook returns @code{true} if an argument declared in a
3920 prototype as an integral type smaller than @code{int} should actually be
3921 passed as an @code{int}. In addition to avoiding errors in certain
3922 cases of mismatch, it also makes for better code on certain machines.
3923 The default is to not promote prototypes.
3926 @deftypefn {Target Hook} bool TARGET_PUSH_ARGUMENT (unsigned int @var{npush})
3927 This target hook returns @code{true} if push instructions will be
3928 used to pass outgoing arguments. When the push instruction usage is
3929 optional, @var{npush} is nonzero to indicate the number of bytes to
3930 push. Otherwise, @var{npush} is zero. If the target machine does not
3931 have a push instruction or push instruction should be avoided,
3932 @code{false} should be returned. That directs GCC to use an alternate
3933 strategy: to allocate the entire argument block and then store the
3934 arguments into it. If this target hook may return @code{true},
3935 @code{PUSH_ROUNDING} must be defined.
3938 @defmac PUSH_ARGS_REVERSED
3939 A C expression. If nonzero, function arguments will be evaluated from
3940 last to first, rather than from first to last. If this macro is not
3941 defined, it defaults to @code{PUSH_ARGS} on targets where the stack
3942 and args grow in opposite directions, and 0 otherwise.
3945 @defmac PUSH_ROUNDING (@var{npushed})
3946 A C expression that is the number of bytes actually pushed onto the
3947 stack when an instruction attempts to push @var{npushed} bytes.
3949 On some machines, the definition
3952 #define PUSH_ROUNDING(BYTES) (BYTES)
3956 will suffice. But on other machines, instructions that appear
3957 to push one byte actually push two bytes in an attempt to maintain
3958 alignment. Then the definition should be
3961 #define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1)
3964 If the value of this macro has a type, it should be an unsigned type.
3967 @findex outgoing_args_size
3968 @findex crtl->outgoing_args_size
3969 @defmac ACCUMULATE_OUTGOING_ARGS
3970 A C expression. If nonzero, the maximum amount of space required for outgoing arguments
3971 will be computed and placed into
3972 @code{crtl->outgoing_args_size}. No space will be pushed
3973 onto the stack for each call; instead, the function prologue should
3974 increase the stack frame size by this amount.
3976 Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS}
3980 @defmac REG_PARM_STACK_SPACE (@var{fndecl})
3981 Define this macro if functions should assume that stack space has been
3982 allocated for arguments even when their values are passed in
3985 The value of this macro is the size, in bytes, of the area reserved for
3986 arguments passed in registers for the function represented by @var{fndecl},
3987 which can be zero if GCC is calling a library function.
3988 The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself
3991 This space can be allocated by the caller, or be a part of the
3992 machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says
3995 @c above is overfull. not sure what to do. --mew 5feb93 did
3996 @c something, not sure if it looks good. --mew 10feb93
3998 @defmac INCOMING_REG_PARM_STACK_SPACE (@var{fndecl})
3999 Like @code{REG_PARM_STACK_SPACE}, but for incoming register arguments.
4000 Define this macro if space guaranteed when compiling a function body
4001 is different to space required when making a call, a situation that
4002 can arise with K&R style function definitions.
4005 @defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype})
4006 Define this to a nonzero value if it is the responsibility of the
4007 caller to allocate the area reserved for arguments passed in registers
4008 when calling a function of @var{fntype}. @var{fntype} may be NULL
4009 if the function called is a library function.
4011 If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls
4012 whether the space for these arguments counts in the value of
4013 @code{crtl->outgoing_args_size}.
4016 @defmac STACK_PARMS_IN_REG_PARM_AREA
4017 Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the
4018 stack parameters don't skip the area specified by it.
4019 @c i changed this, makes more sens and it should have taken care of the
4020 @c overfull.. not as specific, tho. --mew 5feb93
4022 Normally, when a parameter is not passed in registers, it is placed on the
4023 stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro
4024 suppresses this behavior and causes the parameter to be passed on the
4025 stack in its natural location.
4028 @deftypefn {Target Hook} poly_int64 TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, poly_int64 @var{size})
4029 This target hook returns the number of bytes of its own arguments that
4030 a function pops on returning, or 0 if the function pops no arguments
4031 and the caller must therefore pop them all after the function returns.
4033 @var{fundecl} is a C variable whose value is a tree node that describes
4034 the function in question. Normally it is a node of type
4035 @code{FUNCTION_DECL} that describes the declaration of the function.
4036 From this you can obtain the @code{DECL_ATTRIBUTES} of the function.
4038 @var{funtype} is a C variable whose value is a tree node that
4039 describes the function in question. Normally it is a node of type
4040 @code{FUNCTION_TYPE} that describes the data type of the function.
4041 From this it is possible to obtain the data types of the value and
4042 arguments (if known).
4044 When a call to a library function is being considered, @var{fundecl}
4045 will contain an identifier node for the library function. Thus, if
4046 you need to distinguish among various library functions, you can do so
4047 by their names. Note that ``library function'' in this context means
4048 a function used to perform arithmetic, whose name is known specially
4049 in the compiler and was not mentioned in the C code being compiled.
4051 @var{size} is the number of bytes of arguments passed on the
4052 stack. If a variable number of bytes is passed, it is zero, and
4053 argument popping will always be the responsibility of the calling function.
4055 On the VAX, all functions always pop their arguments, so the definition
4056 of this macro is @var{size}. On the 68000, using the standard
4057 calling convention, no functions pop their arguments, so the value of
4058 the macro is always 0 in this case. But an alternative calling
4059 convention is available in which functions that take a fixed number of
4060 arguments pop them but other functions (such as @code{printf}) pop
4061 nothing (the caller pops all). When this convention is in use,
4062 @var{funtype} is examined to determine whether a function takes a fixed
4063 number of arguments.
4066 @defmac CALL_POPS_ARGS (@var{cum})
4067 A C expression that should indicate the number of bytes a call sequence
4068 pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS}
4069 when compiling a function call.
4071 @var{cum} is the variable in which all arguments to the called function
4072 have been accumulated.
4074 On certain architectures, such as the SH5, a call trampoline is used
4075 that pops certain registers off the stack, depending on the arguments
4076 that have been passed to the function. Since this is a property of the
4077 call site, not of the called function, @code{RETURN_POPS_ARGS} is not
4081 @node Register Arguments
4082 @subsection Passing Arguments in Registers
4083 @cindex arguments in registers
4084 @cindex registers arguments
4086 This section describes the macros which let you control how various
4087 types of arguments are passed in registers or how they are arranged in
4090 @deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, const function_arg_info @var{&arg})
4091 Return an RTX indicating whether function argument @var{arg} is passed
4092 in a register and if so, which register. Argument @var{ca} summarizes all
4093 the previous arguments.
4095 The return value is usually either a @code{reg} RTX for the hard
4096 register in which to pass the argument, or zero to pass the argument
4099 The value of the expression can also be a @code{parallel} RTX@. This is
4100 used when an argument is passed in multiple locations. The mode of the
4101 @code{parallel} should be the mode of the entire argument. The
4102 @code{parallel} holds any number of @code{expr_list} pairs; each one
4103 describes where part of the argument is passed. In each
4104 @code{expr_list} the first operand must be a @code{reg} RTX for the hard
4105 register in which to pass this part of the argument, and the mode of the
4106 register RTX indicates how large this part of the argument is. The
4107 second operand of the @code{expr_list} is a @code{const_int} which gives
4108 the offset in bytes into the entire argument of where this part starts.
4109 As a special exception the first @code{expr_list} in the @code{parallel}
4110 RTX may have a first operand of zero. This indicates that the entire
4111 argument is also stored on the stack.
4113 The last time this hook is called, it is called with @code{MODE ==
4114 VOIDmode}, and its result is passed to the @code{call} or @code{call_value}
4115 pattern as operands 2 and 3 respectively.
4117 @cindex @file{stdarg.h} and register arguments
4118 The usual way to make the ISO library @file{stdarg.h} work on a
4119 machine where some arguments are usually passed in registers, is to
4120 cause nameless arguments to be passed on the stack instead. This is
4121 done by making @code{TARGET_FUNCTION_ARG} return 0 whenever
4122 @var{named} is @code{false}.
4124 @cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG}
4125 @cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG}
4126 You may use the hook @code{targetm.calls.must_pass_in_stack}
4127 in the definition of this macro to determine if this argument is of a
4128 type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE}
4129 is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an
4130 argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is
4131 defined, the argument will be computed in the stack and then loaded into
4135 @deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (const function_arg_info @var{&arg})
4136 This target hook should return @code{true} if we should not pass @var{arg}
4137 solely in registers. The file @file{expr.h} defines a
4138 definition that is usually appropriate, refer to @file{expr.h} for additional
4142 @deftypefn {Target Hook} rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t @var{ca}, const function_arg_info @var{&arg})
4143 Define this hook if the caller and callee on the target have different
4144 views of where arguments are passed. Also define this hook if there are
4145 functions that are never directly called, but are invoked by the hardware
4146 and which have nonstandard calling conventions.
4148 In this case @code{TARGET_FUNCTION_ARG} computes the register in
4149 which the caller passes the value, and
4150 @code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar
4151 fashion to tell the function being called where the arguments will
4154 @code{TARGET_FUNCTION_INCOMING_ARG} can also return arbitrary address
4155 computation using hard register, which can be forced into a register,
4156 so that it can be used to pass special arguments.
4158 If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined,
4159 @code{TARGET_FUNCTION_ARG} serves both purposes.
4162 @deftypefn {Target Hook} bool TARGET_USE_PSEUDO_PIC_REG (void)
4163 This hook should return 1 in case pseudo register should be created
4164 for pic_offset_table_rtx during function expand.
4167 @deftypefn {Target Hook} void TARGET_INIT_PIC_REG (void)
4168 Perform a target dependent initialization of pic_offset_table_rtx.
4169 This hook is called at the start of register allocation.
4172 @deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, const function_arg_info @var{&arg})
4173 This target hook returns the number of bytes at the beginning of an
4174 argument that must be put in registers. The value must be zero for
4175 arguments that are passed entirely in registers or that are entirely
4176 pushed on the stack.
4178 On some machines, certain arguments must be passed partially in
4179 registers and partially in memory. On these machines, typically the
4180 first few words of arguments are passed in registers, and the rest
4181 on the stack. If a multi-word argument (a @code{double} or a
4182 structure) crosses that boundary, its first few words must be passed
4183 in registers and the rest must be pushed. This macro tells the
4184 compiler when this occurs, and how many bytes should go in registers.
4186 @code{TARGET_FUNCTION_ARG} for these arguments should return the first
4187 register to be used by the caller for this argument; likewise
4188 @code{TARGET_FUNCTION_INCOMING_ARG}, for the called function.
4191 @deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (cumulative_args_t @var{cum}, const function_arg_info @var{&arg})
4192 This target hook should return @code{true} if argument @var{arg} at the
4193 position indicated by @var{cum} should be passed by reference. This
4194 predicate is queried after target independent reasons for being
4195 passed by reference, such as @code{TREE_ADDRESSABLE (@var{arg}.type)}.
4197 If the hook returns true, a copy of that argument is made in memory and a
4198 pointer to the argument is passed instead of the argument itself.
4199 The pointer is passed in whatever way is appropriate for passing a pointer
4203 @deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, const function_arg_info @var{&arg})
4204 The function argument described by the parameters to this hook is
4205 known to be passed by reference. The hook should return true if the
4206 function argument should be copied by the callee instead of copied
4209 For any argument for which the hook returns true, if it can be
4210 determined that the argument is not modified, then a copy need
4213 The default version of this hook always returns false.
4216 @defmac CUMULATIVE_ARGS
4217 A C type for declaring a variable that is used as the first argument
4218 of @code{TARGET_FUNCTION_ARG} and other related values. For some
4219 target machines, the type @code{int} suffices and can hold the number
4220 of bytes of argument so far.
4222 There is no need to record in @code{CUMULATIVE_ARGS} anything about the
4223 arguments that have been passed on the stack. The compiler has other
4224 variables to keep track of that. For target machines on which all
4225 arguments are passed on the stack, there is no need to store anything in
4226 @code{CUMULATIVE_ARGS}; however, the data structure must exist and
4227 should not be empty, so use @code{int}.
4230 @defmac OVERRIDE_ABI_FORMAT (@var{fndecl})
4231 If defined, this macro is called before generating any code for a
4232 function, but after the @var{cfun} descriptor for the function has been
4233 created. The back end may use this macro to update @var{cfun} to
4234 reflect an ABI other than that which would normally be used by default.
4235 If the compiler is generating code for a compiler-generated function,
4236 @var{fndecl} may be @code{NULL}.
4239 @defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args})
4240 A C statement (sans semicolon) for initializing the variable
4241 @var{cum} for the state at the beginning of the argument list. The
4242 variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype}
4243 is the tree node for the data type of the function which will receive
4244 the args, or 0 if the args are to a compiler support library function.
4245 For direct calls that are not libcalls, @var{fndecl} contain the
4246 declaration node of the function. @var{fndecl} is also set when
4247 @code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function
4248 being compiled. @var{n_named_args} is set to the number of named
4249 arguments, including a structure return address if it is passed as a
4250 parameter, when making a call. When processing incoming arguments,
4251 @var{n_named_args} is set to @minus{}1.
4253 When processing a call to a compiler support library function,
4254 @var{libname} identifies which one. It is a @code{symbol_ref} rtx which
4255 contains the name of the function, as a string. @var{libname} is 0 when
4256 an ordinary C function call is being processed. Thus, each time this
4257 macro is called, either @var{libname} or @var{fntype} is nonzero, but
4258 never both of them at once.
4261 @defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname})
4262 Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls,
4263 it gets a @code{MODE} argument instead of @var{fntype}, that would be
4264 @code{NULL}. @var{indirect} would always be zero, too. If this macro
4265 is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname,
4266 0)} is used instead.
4269 @defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname})
4270 Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of
4271 finding the arguments for the function being compiled. If this macro is
4272 undefined, @code{INIT_CUMULATIVE_ARGS} is used instead.
4274 The value passed for @var{libname} is always 0, since library routines
4275 with special calling conventions are never compiled with GCC@. The
4276 argument @var{libname} exists for symmetry with
4277 @code{INIT_CUMULATIVE_ARGS}.
4278 @c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe.
4279 @c --mew 5feb93 i switched the order of the sentences. --mew 10feb93
4282 @deftypefn {Target Hook} void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t @var{ca}, const function_arg_info @var{&arg})
4283 This hook updates the summarizer variable pointed to by @var{ca} to
4284 advance past argument @var{arg} in the argument list. Once this is done,
4285 the variable @var{cum} is suitable for analyzing the @emph{following}
4286 argument with @code{TARGET_FUNCTION_ARG}, etc.
4288 This hook need not do anything if the argument in question was passed
4289 on the stack. The compiler knows how to track the amount of stack space
4290 used for arguments without any special help.
4293 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_FUNCTION_ARG_OFFSET (machine_mode @var{mode}, const_tree @var{type})
4294 This hook returns the number of bytes to add to the offset of an
4295 argument of type @var{type} and mode @var{mode} when passed in memory.
4296 This is needed for the SPU, which passes @code{char} and @code{short}
4297 arguments in the preferred slot that is in the middle of the quad word
4298 instead of starting at the top. The default implementation returns 0.
4301 @deftypefn {Target Hook} pad_direction TARGET_FUNCTION_ARG_PADDING (machine_mode @var{mode}, const_tree @var{type})
4302 This hook determines whether, and in which direction, to pad out
4303 an argument of mode @var{mode} and type @var{type}. It returns
4304 @code{PAD_UPWARD} to insert padding above the argument, @code{PAD_DOWNWARD}
4305 to insert padding below the argument, or @code{PAD_NONE} to inhibit padding.
4307 The @emph{amount} of padding is not controlled by this hook, but by
4308 @code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is always just enough
4309 to reach the next multiple of that boundary.
4311 This hook has a default definition that is right for most systems.
4312 For little-endian machines, the default is to pad upward. For
4313 big-endian machines, the default is to pad downward for an argument of
4314 constant size shorter than an @code{int}, and upward otherwise.
4317 @defmac PAD_VARARGS_DOWN
4318 If defined, a C expression which determines whether the default
4319 implementation of va_arg will attempt to pad down before reading the
4320 next argument, if that argument is smaller than its aligned space as
4321 controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such
4322 arguments are padded down if @code{BYTES_BIG_ENDIAN} is true.
4325 @defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first})
4326 Specify padding for the last element of a block move between registers and
4327 memory. @var{first} is nonzero if this is the only element. Defining this
4328 macro allows better control of register function parameters on big-endian
4329 machines, without using @code{PARALLEL} rtl. In particular,
4330 @code{MUST_PASS_IN_STACK} need not test padding and mode of types in
4331 registers, as there is no longer a "wrong" part of a register; For example,
4332 a three byte aggregate may be passed in the high part of a register if so
4336 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (machine_mode @var{mode}, const_tree @var{type})
4337 This hook returns the alignment boundary, in bits, of an argument
4338 with the specified mode and type. The default hook returns
4339 @code{PARM_BOUNDARY} for all arguments.
4342 @deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (machine_mode @var{mode}, const_tree @var{type})
4343 Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY},
4344 which is the default value for this hook. You can define this hook to
4345 return a different value if an argument size must be rounded to a larger
4349 @defmac FUNCTION_ARG_REGNO_P (@var{regno})
4350 A C expression that is nonzero if @var{regno} is the number of a hard
4351 register in which function arguments are sometimes passed. This does
4352 @emph{not} include implicit arguments such as the static chain and
4353 the structure-value address. On many machines, no registers can be
4354 used for this purpose since all function arguments are pushed on the
4358 @deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type})
4359 This hook should return true if parameter of type @var{type} are passed
4360 as two scalar parameters. By default, GCC will attempt to pack complex
4361 arguments into the target's word size. Some ABIs require complex arguments
4362 to be split and treated as their individual components. For example, on
4363 AIX64, complex floats should be passed in a pair of floating point
4364 registers, even though a complex float would fit in one 64-bit floating
4367 The default value of this hook is @code{NULL}, which is treated as always
4371 @deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void)
4372 This hook returns a type node for @code{va_list} for the target.
4373 The default version of the hook returns @code{void*}.
4376 @deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree})
4377 This target hook is used in function @code{c_common_nodes_and_builtins}
4378 to iterate through the target specific builtin types for va_list. The
4379 variable @var{idx} is used as iterator. @var{pname} has to be a pointer
4380 to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed
4382 The arguments @var{pname} and @var{ptree} are used to store the result of
4383 this macro and are set to the name of the va_list builtin type and its
4385 If the return value of this macro is zero, then there is no more element.
4386 Otherwise the @var{IDX} should be increased for the next call of this
4387 macro to iterate through all types.
4390 @deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl})
4391 This hook returns the va_list type of the calling convention specified by
4393 The default version of this hook returns @code{va_list_type_node}.
4396 @deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type})
4397 This hook returns the va_list type of the calling convention specified by the
4398 type of @var{type}. If @var{type} is not a valid va_list type, it returns
4402 @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})
4403 This hook performs target-specific gimplification of
4404 @code{VA_ARG_EXPR}. The first two parameters correspond to the
4405 arguments to @code{va_arg}; the latter two are as in
4406 @code{gimplify.cc:gimplify_expr}.
4409 @deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (scalar_int_mode @var{mode})
4410 Define this to return nonzero if the port can handle pointers
4411 with machine mode @var{mode}. The default version of this
4412 hook returns true for both @code{ptr_mode} and @code{Pmode}.
4415 @deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (ao_ref *@var{ref})
4416 Define this to return nonzero if the memory reference @var{ref}
4417 may alias with the system C library errno location. The default
4418 version of this hook assumes the system C library errno location
4419 is either a declaration of type int or accessed by dereferencing
4423 @deftypefn {Target Hook} machine_mode TARGET_TRANSLATE_MODE_ATTRIBUTE (machine_mode @var{mode})
4424 Define this hook if during mode attribute processing, the port should
4425 translate machine_mode @var{mode} to another mode. For example, rs6000's
4426 @code{KFmode}, when it is the same as @code{TFmode}.
4428 The default version of the hook returns that mode that was passed in.
4431 @deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (scalar_mode @var{mode})
4432 Define this to return nonzero if the port is prepared to handle
4433 insns involving scalar mode @var{mode}. For a scalar mode to be
4434 considered supported, all the basic arithmetic and comparisons
4437 The default version of this hook returns true for any mode
4438 required to handle the basic C types (as defined by the port).
4439 Included here are the double-word arithmetic supported by the
4440 code in @file{optabs.cc}.
4443 @deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (machine_mode @var{mode})
4444 Define this to return nonzero if the port is prepared to handle
4445 insns involving vector mode @var{mode}. At the very least, it
4446 must have move patterns for this mode.
4449 @deftypefn {Target Hook} bool TARGET_COMPATIBLE_VECTOR_TYPES_P (const_tree @var{type1}, const_tree @var{type2})
4450 Return true if there is no target-specific reason for treating
4451 vector types @var{type1} and @var{type2} as distinct types. The caller
4452 has already checked for target-independent reasons, meaning that the
4453 types are known to have the same mode, to have the same number of elements,
4454 and to have what the caller considers to be compatible element types.
4456 The main reason for defining this hook is to reject pairs of types
4457 that are handled differently by the target's calling convention.
4458 For example, when a new @var{N}-bit vector architecture is added
4459 to a target, the target may want to handle normal @var{N}-bit
4460 @code{VECTOR_TYPE} arguments and return values in the same way as
4461 before, to maintain backwards compatibility. However, it may also
4462 provide new, architecture-specific @code{VECTOR_TYPE}s that are passed
4463 and returned in a more efficient way. It is then important to maintain
4464 a distinction between the ``normal'' @code{VECTOR_TYPE}s and the new
4465 architecture-specific ones.
4467 The default implementation returns true, which is correct for most targets.
4470 @deftypefn {Target Hook} opt_machine_mode TARGET_ARRAY_MODE (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems})
4471 Return the mode that GCC should use for an array that has
4472 @var{nelems} elements, with each element having mode @var{mode}.
4473 Return no mode if the target has no special requirements. In the
4474 latter case, GCC looks for an integer mode of the appropriate size
4475 if available and uses BLKmode otherwise. Usually the search for the
4476 integer mode is limited to @code{MAX_FIXED_MODE_SIZE}, but the
4477 @code{TARGET_ARRAY_MODE_SUPPORTED_P} hook allows a larger mode to be
4478 used in specific cases.
4480 The main use of this hook is to specify that an array of vectors should
4481 also have a vector mode. The default implementation returns no mode.
4484 @deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems})
4485 Return true if GCC should try to use a scalar mode to store an array
4486 of @var{nelems} elements, given that each element has mode @var{mode}.
4487 Returning true here overrides the usual @code{MAX_FIXED_MODE} limit
4488 and allows GCC to use any defined integer mode.
4490 One use of this hook is to support vector load and store operations
4491 that operate on several homogeneous vectors. For example, ARM NEON
4492 has operations like:
4495 int8x8x3_t vld3_s8 (const int8_t *)
4498 where the return type is defined as:
4501 typedef struct int8x8x3_t
4507 If this hook allows @code{val} to have a scalar mode, then
4508 @code{int8x8x3_t} can have the same mode. GCC can then store
4509 @code{int8x8x3_t}s in registers rather than forcing them onto the stack.
4512 @deftypefn {Target Hook} bool TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P (scalar_float_mode @var{mode})
4513 Define this to return nonzero if libgcc provides support for the
4514 floating-point mode @var{mode}, which is known to pass
4515 @code{TARGET_SCALAR_MODE_SUPPORTED_P}. The default version of this
4516 hook returns true for all of @code{SFmode}, @code{DFmode},
4517 @code{XFmode} and @code{TFmode}, if such modes exist.
4520 @deftypefn {Target Hook} opt_scalar_float_mode TARGET_FLOATN_MODE (int @var{n}, bool @var{extended})
4521 Define this to return the machine mode to use for the type
4522 @code{_Float@var{n}}, if @var{extended} is false, or the type
4523 @code{_Float@var{n}x}, if @var{extended} is true. If such a type is not
4524 supported, return @code{opt_scalar_float_mode ()}. The default version of
4525 this hook returns @code{SFmode} for @code{_Float32}, @code{DFmode} for
4526 @code{_Float64} and @code{_Float32x} and @code{TFmode} for
4527 @code{_Float128}, if those modes exist and satisfy the requirements for
4528 those types and pass @code{TARGET_SCALAR_MODE_SUPPORTED_P} and
4529 @code{TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P}; for @code{_Float64x}, it
4530 returns the first of @code{XFmode} and @code{TFmode} that exists and
4531 satisfies the same requirements; for other types, it returns
4532 @code{opt_scalar_float_mode ()}. The hook is only called for values
4533 of @var{n} and @var{extended} that are valid according to
4534 ISO/IEC TS 18661-3:2015; that is, @var{n} is one of 32, 64, 128, or,
4535 if @var{extended} is false, 16 or greater than 128 and a multiple of 32.
4538 @deftypefn {Target Hook} bool TARGET_FLOATN_BUILTIN_P (int @var{func})
4539 Define this to return true if the @code{_Float@var{n}} and
4540 @code{_Float@var{n}x} built-in functions should implicitly enable the
4541 built-in function without the @code{__builtin_} prefix in addition to the
4542 normal built-in function with the @code{__builtin_} prefix. The default is
4543 to only enable built-in functions without the @code{__builtin_} prefix for
4544 the GNU C langauge. In strict ANSI/ISO mode, the built-in function without
4545 the @code{__builtin_} prefix is not enabled. The argument @code{FUNC} is the
4546 @code{enum built_in_function} id of the function to be enabled.
4549 @deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (machine_mode @var{mode})
4550 Define this to return nonzero for machine modes for which the port has
4551 small register classes. If this target hook returns nonzero for a given
4552 @var{mode}, the compiler will try to minimize the lifetime of registers
4553 in @var{mode}. The hook may be called with @code{VOIDmode} as argument.
4554 In this case, the hook is expected to return nonzero if it returns nonzero
4557 On some machines, it is risky to let hard registers live across arbitrary
4558 insns. Typically, these machines have instructions that require values
4559 to be in specific registers (like an accumulator), and reload will fail
4560 if the required hard register is used for another purpose across such an
4563 Passes before reload do not know which hard registers will be used
4564 in an instruction, but the machine modes of the registers set or used in
4565 the instruction are already known. And for some machines, register
4566 classes are small for, say, integer registers but not for floating point
4567 registers. For example, the AMD x86-64 architecture requires specific
4568 registers for the legacy x86 integer instructions, but there are many
4569 SSE registers for floating point operations. On such targets, a good
4570 strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P}
4571 machine modes but zero for the SSE register classes.
4573 The default version of this hook returns false for any mode. It is always
4574 safe to redefine this hook to return with a nonzero value. But if you
4575 unnecessarily define it, you will reduce the amount of optimizations
4576 that can be performed in some cases. If you do not define this hook
4577 to return a nonzero value when it is required, the compiler will run out
4578 of spill registers and print a fatal error message.
4582 @subsection How Scalar Function Values Are Returned
4583 @cindex return values in registers
4584 @cindex values, returned by functions
4585 @cindex scalars, returned as values
4587 This section discusses the macros that control returning scalars as
4588 values---values that can fit in registers.
4590 @deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing})
4592 Define this to return an RTX representing the place where a function
4593 returns or receives a value of data type @var{ret_type}, a tree node
4594 representing a data type. @var{fn_decl_or_type} is a tree node
4595 representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a
4596 function being called. If @var{outgoing} is false, the hook should
4597 compute the register in which the caller will see the return value.
4598 Otherwise, the hook should return an RTX representing the place where
4599 a function returns a value.
4601 On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant.
4602 (Actually, on most machines, scalar values are returned in the same
4603 place regardless of mode.) The value of the expression is usually a
4604 @code{reg} RTX for the hard register where the return value is stored.
4605 The value can also be a @code{parallel} RTX, if the return value is in
4606 multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the
4607 @code{parallel} form. Note that the callee will populate every
4608 location specified in the @code{parallel}, but if the first element of
4609 the @code{parallel} contains the whole return value, callers will use
4610 that element as the canonical location and ignore the others. The m68k
4611 port uses this type of @code{parallel} to return pointers in both
4612 @samp{%a0} (the canonical location) and @samp{%d0}.
4614 If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply
4615 the same promotion rules specified in @code{PROMOTE_MODE} if
4616 @var{valtype} is a scalar type.
4618 If the precise function being called is known, @var{func} is a tree
4619 node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null
4620 pointer. This makes it possible to use a different value-returning
4621 convention for specific functions when all their calls are
4624 Some target machines have ``register windows'' so that the register in
4625 which a function returns its value is not the same as the one in which
4626 the caller sees the value. For such machines, you should return
4627 different RTX depending on @var{outgoing}.
4629 @code{TARGET_FUNCTION_VALUE} is not used for return values with
4630 aggregate data types, because these are returned in another way. See
4631 @code{TARGET_STRUCT_VALUE_RTX} and related macros, below.
4634 @defmac FUNCTION_VALUE (@var{valtype}, @var{func})
4635 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for
4636 a new target instead.
4639 @defmac LIBCALL_VALUE (@var{mode})
4640 A C expression to create an RTX representing the place where a library
4641 function returns a value of mode @var{mode}.
4643 Note that ``library function'' in this context means a compiler
4644 support routine, used to perform arithmetic, whose name is known
4645 specially by the compiler and was not mentioned in the C code being
4649 @deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (machine_mode @var{mode}, const_rtx @var{fun})
4650 Define this hook if the back-end needs to know the name of the libcall
4651 function in order to determine where the result should be returned.
4653 The mode of the result is given by @var{mode} and the name of the called
4654 library function is given by @var{fun}. The hook should return an RTX
4655 representing the place where the library function result will be returned.
4657 If this hook is not defined, then LIBCALL_VALUE will be used.
4660 @defmac FUNCTION_VALUE_REGNO_P (@var{regno})
4661 A C expression that is nonzero if @var{regno} is the number of a hard
4662 register in which the values of called function may come back.
4664 A register whose use for returning values is limited to serving as the
4665 second of a pair (for a value of type @code{double}, say) need not be
4666 recognized by this macro. So for most machines, this definition
4670 #define FUNCTION_VALUE_REGNO_P(N) ((N) == 0)
4673 If the machine has register windows, so that the caller and the called
4674 function use different registers for the return value, this macro
4675 should recognize only the caller's register numbers.
4677 This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P}
4678 for a new target instead.
4681 @deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno})
4682 A target hook that return @code{true} if @var{regno} is the number of a hard
4683 register in which the values of called function may come back.
4685 A register whose use for returning values is limited to serving as the
4686 second of a pair (for a value of type @code{double}, say) need not be
4687 recognized by this target hook.
4689 If the machine has register windows, so that the caller and the called
4690 function use different registers for the return value, this target hook
4691 should recognize only the caller's register numbers.
4693 If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used.
4696 @defmac APPLY_RESULT_SIZE
4697 Define this macro if @samp{untyped_call} and @samp{untyped_return}
4698 need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for
4699 saving and restoring an arbitrary return value.
4702 @deftypevr {Target Hook} bool TARGET_OMIT_STRUCT_RETURN_REG
4703 Normally, when a function returns a structure by memory, the address
4704 is passed as an invisible pointer argument, but the compiler also
4705 arranges to return the address from the function like it would a normal
4706 pointer return value. Define this to true if that behavior is
4707 undesirable on your target.
4710 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type})
4711 This hook should return true if values of type @var{type} are returned
4712 at the most significant end of a register (in other words, if they are
4713 padded at the least significant end). You can assume that @var{type}
4714 is returned in a register; the caller is required to check this.
4716 Note that the register provided by @code{TARGET_FUNCTION_VALUE} must
4717 be able to hold the complete return value. For example, if a 1-, 2-
4718 or 3-byte structure is returned at the most significant end of a
4719 4-byte register, @code{TARGET_FUNCTION_VALUE} should provide an
4723 @node Aggregate Return
4724 @subsection How Large Values Are Returned
4725 @cindex aggregates as return values
4726 @cindex large return values
4727 @cindex returning aggregate values
4728 @cindex structure value address
4730 When a function value's mode is @code{BLKmode} (and in some other
4731 cases), the value is not returned according to
4732 @code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the
4733 caller passes the address of a block of memory in which the value
4734 should be stored. This address is called the @dfn{structure value
4737 This section describes how to control returning structure values in
4740 @deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype})
4741 This target hook should return a nonzero value to say to return the
4742 function value in memory, just as large structures are always returned.
4743 Here @var{type} will be the data type of the value, and @var{fntype}
4744 will be the type of the function doing the returning, or @code{NULL} for
4747 Note that values of mode @code{BLKmode} must be explicitly handled
4748 by this function. Also, the option @option{-fpcc-struct-return}
4749 takes effect regardless of this macro. On most systems, it is
4750 possible to leave the hook undefined; this causes a default
4751 definition to be used, whose value is the constant 1 for @code{BLKmode}
4752 values, and 0 otherwise.
4754 Do not use this hook to indicate that structures and unions should always
4755 be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN}
4759 @defmac DEFAULT_PCC_STRUCT_RETURN
4760 Define this macro to be 1 if all structure and union return values must be
4761 in memory. Since this results in slower code, this should be defined
4762 only if needed for compatibility with other compilers or with an ABI@.
4763 If you define this macro to be 0, then the conventions used for structure
4764 and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY}
4767 If not defined, this defaults to the value 1.
4770 @deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming})
4771 This target hook should return the location of the structure value
4772 address (normally a @code{mem} or @code{reg}), or 0 if the address is
4773 passed as an ``invisible'' first argument. Note that @var{fndecl} may
4774 be @code{NULL}, for libcalls. You do not need to define this target
4775 hook if the address is always passed as an ``invisible'' first
4778 On some architectures the place where the structure value address
4779 is found by the called function is not the same place that the
4780 caller put it. This can be due to register windows, or it could
4781 be because the function prologue moves it to a different place.
4782 @var{incoming} is @code{1} or @code{2} when the location is needed in
4783 the context of the called function, and @code{0} in the context of
4786 If @var{incoming} is nonzero and the address is to be found on the
4787 stack, return a @code{mem} which refers to the frame pointer. If
4788 @var{incoming} is @code{2}, the result is being used to fetch the
4789 structure value address at the beginning of a function. If you need
4790 to emit adjusting code, you should do it at this point.
4793 @defmac PCC_STATIC_STRUCT_RETURN
4794 Define this macro if the usual system convention on the target machine
4795 for returning structures and unions is for the called function to return
4796 the address of a static variable containing the value.
4798 Do not define this if the usual system convention is for the caller to
4799 pass an address to the subroutine.
4801 This macro has effect in @option{-fpcc-struct-return} mode, but it does
4802 nothing when you use @option{-freg-struct-return} mode.
4805 @deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_RESULT_MODE (int @var{regno})
4806 This target hook returns the mode to be used when accessing raw return
4807 registers in @code{__builtin_return}. Define this macro if the value
4808 in @var{reg_raw_mode} is not correct.
4811 @deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_ARG_MODE (int @var{regno})
4812 This target hook returns the mode to be used when accessing raw argument
4813 registers in @code{__builtin_apply_args}. Define this macro if the value
4814 in @var{reg_raw_mode} is not correct.
4817 @deftypefn {Target Hook} bool TARGET_EMPTY_RECORD_P (const_tree @var{type})
4818 This target hook returns true if the type is an empty record. The default
4819 is to return @code{false}.
4822 @deftypefn {Target Hook} void TARGET_WARN_PARAMETER_PASSING_ABI (cumulative_args_t @var{ca}, tree @var{type})
4823 This target hook warns about the change in empty class parameter passing
4828 @subsection Caller-Saves Register Allocation
4830 If you enable it, GCC can save registers around function calls. This
4831 makes it possible to use call-clobbered registers to hold variables that
4832 must live across calls.
4834 @defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs})
4835 A C expression specifying which mode is required for saving @var{nregs}
4836 of a pseudo-register in call-clobbered hard register @var{regno}. If
4837 @var{regno} is unsuitable for caller save, @code{VOIDmode} should be
4838 returned. For most machines this macro need not be defined since GCC
4839 will select the smallest suitable mode.
4842 @node Function Entry
4843 @subsection Function Entry and Exit
4844 @cindex function entry and exit
4848 This section describes the macros that output function entry
4849 (@dfn{prologue}) and exit (@dfn{epilogue}) code.
4851 @deftypefn {Target Hook} void TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY (FILE *@var{file}, unsigned HOST_WIDE_INT @var{patch_area_size}, bool @var{record_p})
4852 Generate a patchable area at the function start, consisting of
4853 @var{patch_area_size} NOP instructions. If the target supports named
4854 sections and if @var{record_p} is true, insert a pointer to the current
4855 location in the table of patchable functions. The default implementation
4856 of the hook places the table of pointers in the special section named
4857 @code{__patchable_function_entries}.
4860 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file})
4861 If defined, a function that outputs the assembler code for entry to a
4862 function. The prologue is responsible for setting up the stack frame,
4863 initializing the frame pointer register, saving registers that must be
4864 saved, and allocating @var{size} additional bytes of storage for the
4865 local variables. @var{file} is a stdio stream to which the assembler
4866 code should be output.
4868 The label for the beginning of the function need not be output by this
4869 macro. That has already been done when the macro is run.
4871 @findex regs_ever_live
4872 To determine which registers to save, the macro can refer to the array
4873 @code{regs_ever_live}: element @var{r} is nonzero if hard register
4874 @var{r} is used anywhere within the function. This implies the function
4875 prologue should save register @var{r}, provided it is not one of the
4876 call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use
4877 @code{regs_ever_live}.)
4879 On machines that have ``register windows'', the function entry code does
4880 not save on the stack the registers that are in the windows, even if
4881 they are supposed to be preserved by function calls; instead it takes
4882 appropriate steps to ``push'' the register stack, if any non-call-used
4883 registers are used in the function.
4885 @findex frame_pointer_needed
4886 On machines where functions may or may not have frame-pointers, the
4887 function entry code must vary accordingly; it must set up the frame
4888 pointer if one is wanted, and not otherwise. To determine whether a
4889 frame pointer is in wanted, the macro can refer to the variable
4890 @code{frame_pointer_needed}. The variable's value will be 1 at run
4891 time in a function that needs a frame pointer. @xref{Elimination}.
4893 The function entry code is responsible for allocating any stack space
4894 required for the function. This stack space consists of the regions
4895 listed below. In most cases, these regions are allocated in the
4896 order listed, with the last listed region closest to the top of the
4897 stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and
4898 the highest address if it is not defined). You can use a different order
4899 for a machine if doing so is more convenient or required for
4900 compatibility reasons. Except in cases where required by standard
4901 or by a debugger, there is no reason why the stack layout used by GCC
4902 need agree with that used by other compilers for a machine.
4905 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file})
4906 If defined, a function that outputs assembler code at the end of a
4907 prologue. This should be used when the function prologue is being
4908 emitted as RTL, and you have some extra assembler that needs to be
4909 emitted. @xref{prologue instruction pattern}.
4912 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file})
4913 If defined, a function that outputs assembler code at the start of an
4914 epilogue. This should be used when the function epilogue is being
4915 emitted as RTL, and you have some extra assembler that needs to be
4916 emitted. @xref{epilogue instruction pattern}.
4919 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file})
4920 If defined, a function that outputs the assembler code for exit from a
4921 function. The epilogue is responsible for restoring the saved
4922 registers and stack pointer to their values when the function was
4923 called, and returning control to the caller. This macro takes the
4924 same argument as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the
4925 registers to restore are determined from @code{regs_ever_live} and
4926 @code{CALL_USED_REGISTERS} in the same way.
4928 On some machines, there is a single instruction that does all the work
4929 of returning from the function. On these machines, give that
4930 instruction the name @samp{return} and do not define the macro
4931 @code{TARGET_ASM_FUNCTION_EPILOGUE} at all.
4933 Do not define a pattern named @samp{return} if you want the
4934 @code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target
4935 switches to control whether return instructions or epilogues are used,
4936 define a @samp{return} pattern with a validity condition that tests the
4937 target switches appropriately. If the @samp{return} pattern's validity
4938 condition is false, epilogues will be used.
4940 On machines where functions may or may not have frame-pointers, the
4941 function exit code must vary accordingly. Sometimes the code for these
4942 two cases is completely different. To determine whether a frame pointer
4943 is wanted, the macro can refer to the variable
4944 @code{frame_pointer_needed}. The variable's value will be 1 when compiling
4945 a function that needs a frame pointer.
4947 Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and
4948 @code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially.
4949 The C variable @code{current_function_is_leaf} is nonzero for such a
4950 function. @xref{Leaf Functions}.
4952 On some machines, some functions pop their arguments on exit while
4953 others leave that for the caller to do. For example, the 68020 when
4954 given @option{-mrtd} pops arguments in functions that take a fixed
4955 number of arguments.
4958 @findex crtl->args.pops_args
4959 Your definition of the macro @code{RETURN_POPS_ARGS} decides which
4960 functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE}
4961 needs to know what was decided. The number of bytes of the current
4962 function's arguments that this function should pop is available in
4963 @code{crtl->args.pops_args}. @xref{Scalar Return}.
4968 @findex pretend_args_size
4969 @findex crtl->args.pretend_args_size
4970 A region of @code{crtl->args.pretend_args_size} bytes of
4971 uninitialized space just underneath the first argument arriving on the
4972 stack. (This may not be at the very start of the allocated stack region
4973 if the calling sequence has pushed anything else since pushing the stack
4974 arguments. But usually, on such machines, nothing else has been pushed
4975 yet, because the function prologue itself does all the pushing.) This
4976 region is used on machines where an argument may be passed partly in
4977 registers and partly in memory, and, in some cases to support the
4978 features in @code{<stdarg.h>}.
4981 An area of memory used to save certain registers used by the function.
4982 The size of this area, which may also include space for such things as
4983 the return address and pointers to previous stack frames, is
4984 machine-specific and usually depends on which registers have been used
4985 in the function. Machines with register windows often do not require
4989 A region of at least @var{size} bytes, possibly rounded up to an allocation
4990 boundary, to contain the local variables of the function. On some machines,
4991 this region and the save area may occur in the opposite order, with the
4992 save area closer to the top of the stack.
4995 @cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames
4996 Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of
4997 @code{crtl->outgoing_args_size} bytes to be used for outgoing
4998 argument lists of the function. @xref{Stack Arguments}.
5001 @defmac EXIT_IGNORE_STACK
5002 Define this macro as a C expression that is nonzero if the return
5003 instruction or the function epilogue ignores the value of the stack
5004 pointer; in other words, if it is safe to delete an instruction to
5005 adjust the stack pointer before a return from the function. The
5008 Note that this macro's value is relevant only for functions for which
5009 frame pointers are maintained. It is never safe to delete a final
5010 stack adjustment in a function that has no frame pointer, and the
5011 compiler knows this regardless of @code{EXIT_IGNORE_STACK}.
5014 @defmac EPILOGUE_USES (@var{regno})
5015 Define this macro as a C expression that is nonzero for registers that are
5016 used by the epilogue or the @samp{return} pattern. The stack and frame
5017 pointer registers are already assumed to be used as needed.
5020 @defmac EH_USES (@var{regno})
5021 Define this macro as a C expression that is nonzero for registers that are
5022 used by the exception handling mechanism, and so should be considered live
5023 on entry to an exception edge.
5026 @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})
5027 A function that outputs the assembler code for a thunk
5028 function, used to implement C++ virtual function calls with multiple
5029 inheritance. The thunk acts as a wrapper around a virtual function,
5030 adjusting the implicit object parameter before handing control off to
5033 First, emit code to add the integer @var{delta} to the location that
5034 contains the incoming first argument. Assume that this argument
5035 contains a pointer, and is the one used to pass the @code{this} pointer
5036 in C++. This is the incoming argument @emph{before} the function prologue,
5037 e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of
5038 all other incoming arguments.
5040 Then, if @var{vcall_offset} is nonzero, an additional adjustment should be
5041 made after adding @code{delta}. In particular, if @var{p} is the
5042 adjusted pointer, the following adjustment should be made:
5045 p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)]
5048 After the additions, emit code to jump to @var{function}, which is a
5049 @code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does
5050 not touch the return address. Hence returning from @var{FUNCTION} will
5051 return to whoever called the current @samp{thunk}.
5053 The effect must be as if @var{function} had been called directly with
5054 the adjusted first argument. This macro is responsible for emitting all
5055 of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE}
5056 and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked.
5058 The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function}
5059 have already been extracted from it.) It might possibly be useful on
5060 some targets, but probably not.
5062 If you do not define this macro, the target-independent code in the C++
5063 front end will generate a less efficient heavyweight thunk that calls
5064 @var{function} instead of jumping to it. The generic approach does
5065 not support varargs.
5068 @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})
5069 A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able
5070 to output the assembler code for the thunk function specified by the
5071 arguments it is passed, and false otherwise. In the latter case, the
5072 generic approach will be used by the C++ front end, with the limitations
5077 @subsection Generating Code for Profiling
5078 @cindex profiling, code generation
5080 These macros will help you generate code for profiling.
5082 @defmac FUNCTION_PROFILER (@var{file}, @var{labelno})
5083 A C statement or compound statement to output to @var{file} some
5084 assembler code to call the profiling subroutine @code{mcount}.
5087 The details of how @code{mcount} expects to be called are determined by
5088 your operating system environment, not by GCC@. To figure them out,
5089 compile a small program for profiling using the system's installed C
5090 compiler and look at the assembler code that results.
5092 Older implementations of @code{mcount} expect the address of a counter
5093 variable to be loaded into some register. The name of this variable is
5094 @samp{LP} followed by the number @var{labelno}, so you would generate
5095 the name using @samp{LP%d} in a @code{fprintf}.
5098 @defmac PROFILE_HOOK
5099 A C statement or compound statement to output to @var{file} some assembly
5100 code to call the profiling subroutine @code{mcount} even the target does
5101 not support profiling.
5104 @defmac NO_PROFILE_COUNTERS
5105 Define this macro to be an expression with a nonzero value if the
5106 @code{mcount} subroutine on your system does not need a counter variable
5107 allocated for each function. This is true for almost all modern
5108 implementations. If you define this macro, you must not use the
5109 @var{labelno} argument to @code{FUNCTION_PROFILER}.
5112 @defmac PROFILE_BEFORE_PROLOGUE
5113 Define this macro if the code for function profiling should come before
5114 the function prologue. Normally, the profiling code comes after.
5117 @deftypefn {Target Hook} bool TARGET_KEEP_LEAF_WHEN_PROFILED (void)
5118 This target hook returns true if the target wants the leaf flag for
5119 the current function to stay true even if it calls mcount. This might
5120 make sense for targets using the leaf flag only to determine whether a
5121 stack frame needs to be generated or not and for which the call to
5122 mcount is generated before the function prologue.
5126 @subsection Permitting tail calls
5129 @deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp})
5130 True if it is OK to do sibling call optimization for the specified
5131 call expression @var{exp}. @var{decl} will be the called function,
5132 or @code{NULL} if this is an indirect call.
5134 It is not uncommon for limitations of calling conventions to prevent
5135 tail calls to functions outside the current unit of translation, or
5136 during PIC compilation. The hook is used to enforce these restrictions,
5137 as the @code{sibcall} md pattern cannot fail, or fall over to a
5138 ``normal'' call. The criteria for successful sibling call optimization
5139 may vary greatly between different architectures.
5142 @deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs})
5143 Add any hard registers to @var{regs} that are live on entry to the
5144 function. This hook only needs to be defined to provide registers that
5145 cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved
5146 registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM,
5147 TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES,
5148 FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM.
5151 @deftypefn {Target Hook} void TARGET_SET_UP_BY_PROLOGUE (struct hard_reg_set_container *@var{})
5152 This hook should add additional registers that are computed by the prologue
5153 to the hard regset for shrink-wrapping optimization purposes.
5156 @deftypefn {Target Hook} bool TARGET_WARN_FUNC_RETURN (tree)
5157 True if a function's return statements should be checked for matching
5158 the function's return type. This includes checking for falling off the end
5159 of a non-void function. Return false if no such check should be made.
5162 @node Shrink-wrapping separate components
5163 @subsection Shrink-wrapping separate components
5164 @cindex shrink-wrapping separate components
5166 The prologue may perform a variety of target dependent tasks such as
5167 saving callee-saved registers, saving the return address, aligning the
5168 stack, creating a stack frame, initializing the PIC register, setting
5169 up the static chain, etc.
5171 On some targets some of these tasks may be independent of others and
5172 thus may be shrink-wrapped separately. These independent tasks are
5173 referred to as components and are handled generically by the target
5174 independent parts of GCC.
5176 Using the following hooks those prologue or epilogue components can be
5177 shrink-wrapped separately, so that the initialization (and possibly
5178 teardown) those components do is not done as frequently on execution
5179 paths where this would unnecessary.
5181 What exactly those components are is up to the target code; the generic
5182 code treats them abstractly, as a bit in an @code{sbitmap}. These
5183 @code{sbitmap}s are allocated by the @code{shrink_wrap.get_separate_components}
5184 and @code{shrink_wrap.components_for_bb} hooks, and deallocated by the
5187 @deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS (void)
5188 This hook should return an @code{sbitmap} with the bits set for those
5189 components that can be separately shrink-wrapped in the current function.
5190 Return @code{NULL} if the current function should not get any separate
5192 Don't define this hook if it would always return @code{NULL}.
5193 If it is defined, the other hooks in this group have to be defined as well.
5196 @deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB (basic_block)
5197 This hook should return an @code{sbitmap} with the bits set for those
5198 components where either the prologue component has to be executed before
5199 the @code{basic_block}, or the epilogue component after it, or both.
5202 @deftypefn {Target Hook} void TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS (sbitmap @var{components}, edge @var{e}, sbitmap @var{edge_components}, bool @var{is_prologue})
5203 This hook should clear the bits in the @var{components} bitmap for those
5204 components in @var{edge_components} that the target cannot handle on edge
5205 @var{e}, where @var{is_prologue} says if this is for a prologue or an
5209 @deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS (sbitmap)
5210 Emit prologue insns for the components indicated by the parameter.
5213 @deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS (sbitmap)
5214 Emit epilogue insns for the components indicated by the parameter.
5217 @deftypefn {Target Hook} void TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS (sbitmap)
5218 Mark the components in the parameter as handled, so that the
5219 @code{prologue} and @code{epilogue} named patterns know to ignore those
5220 components. The target code should not hang on to the @code{sbitmap}, it
5221 will be deleted after this call.
5224 @node Stack Smashing Protection
5225 @subsection Stack smashing protection
5226 @cindex stack smashing protection
5228 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void)
5229 This hook returns a @code{DECL} node for the external variable to use
5230 for the stack protection guard. This variable is initialized by the
5231 runtime to some random value and is used to initialize the guard value
5232 that is placed at the top of the local stack frame. The type of this
5233 variable must be @code{ptr_type_node}.
5235 The default version of this hook creates a variable called
5236 @samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}.
5239 @deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void)
5240 This hook returns a @code{CALL_EXPR} that alerts the runtime that the
5241 stack protect guard variable has been modified. This expression should
5242 involve a call to a @code{noreturn} function.
5244 The default version of this hook invokes a function called
5245 @samp{__stack_chk_fail}, taking no arguments. This function is
5246 normally defined in @file{libgcc2.c}.
5249 @deftypefn {Target Hook} bool TARGET_STACK_PROTECT_RUNTIME_ENABLED_P (void)
5250 Returns true if the target wants GCC's default stack protect runtime support,
5251 otherwise return false. The default implementation always returns true.
5254 @deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts})
5255 Whether this target supports splitting the stack when the options
5256 described in @var{opts} have been passed. This is called
5257 after options have been parsed, so the target may reject splitting
5258 the stack in some configurations. The default version of this hook
5259 returns false. If @var{report} is true, this function may issue a warning
5260 or error; if @var{report} is false, it must simply return a value
5263 @deftypefn {Common Target Hook} {vec<const char *>} TARGET_GET_VALID_OPTION_VALUES (int @var{option_code}, const char *@var{prefix})
5264 The hook is used for options that have a non-trivial list of
5265 possible option values. OPTION_CODE is option code of opt_code
5266 enum type. PREFIX is used for bash completion and allows an implementation
5267 to return more specific completion based on the prefix. All string values
5268 should be allocated from heap memory and consumers should release them.
5269 The result will be pruned to cases with PREFIX if not NULL.
5272 @node Miscellaneous Register Hooks
5273 @subsection Miscellaneous register hooks
5274 @cindex miscellaneous register hooks
5276 @deftypevr {Target Hook} bool TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS
5277 Set to true if each call that binds to a local definition explicitly
5278 clobbers or sets all non-fixed registers modified by performing the call.
5279 That is, by the call pattern itself, or by code that might be inserted by the
5280 linker (e.g.@: stubs, veneers, branch islands), but not including those
5281 modifiable by the callee. The affected registers may be mentioned explicitly
5282 in the call pattern, or included as clobbers in CALL_INSN_FUNCTION_USAGE.
5283 The default version of this hook is set to false. The purpose of this hook
5284 is to enable the fipa-ra optimization.
5288 @section Implementing the Varargs Macros
5289 @cindex varargs implementation
5291 GCC comes with an implementation of @code{<varargs.h>} and
5292 @code{<stdarg.h>} that work without change on machines that pass arguments
5293 on the stack. Other machines require their own implementations of
5294 varargs, and the two machine independent header files must have
5295 conditionals to include it.
5297 ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in
5298 the calling convention for @code{va_start}. The traditional
5299 implementation takes just one argument, which is the variable in which
5300 to store the argument pointer. The ISO implementation of
5301 @code{va_start} takes an additional second argument. The user is
5302 supposed to write the last named argument of the function here.
5304 However, @code{va_start} should not use this argument. The way to find
5305 the end of the named arguments is with the built-in functions described
5308 @defmac __builtin_saveregs ()
5309 Use this built-in function to save the argument registers in memory so
5310 that the varargs mechanism can access them. Both ISO and traditional
5311 versions of @code{va_start} must use @code{__builtin_saveregs}, unless
5312 you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead.
5314 On some machines, @code{__builtin_saveregs} is open-coded under the
5315 control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On
5316 other machines, it calls a routine written in assembler language,
5317 found in @file{libgcc2.c}.
5319 Code generated for the call to @code{__builtin_saveregs} appears at the
5320 beginning of the function, as opposed to where the call to
5321 @code{__builtin_saveregs} is written, regardless of what the code is.
5322 This is because the registers must be saved before the function starts
5323 to use them for its own purposes.
5324 @c i rewrote the first sentence above to fix an overfull hbox. --mew
5328 @defmac __builtin_next_arg (@var{lastarg})
5329 This builtin returns the address of the first anonymous stack
5330 argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it
5331 returns the address of the location above the first anonymous stack
5332 argument. Use it in @code{va_start} to initialize the pointer for
5333 fetching arguments from the stack. Also use it in @code{va_start} to
5334 verify that the second parameter @var{lastarg} is the last named argument
5335 of the current function.
5338 @defmac __builtin_classify_type (@var{object})
5339 Since each machine has its own conventions for which data types are
5340 passed in which kind of register, your implementation of @code{va_arg}
5341 has to embody these conventions. The easiest way to categorize the
5342 specified data type is to use @code{__builtin_classify_type} together
5343 with @code{sizeof} and @code{__alignof__}.
5345 @code{__builtin_classify_type} ignores the value of @var{object},
5346 considering only its data type. It returns an integer describing what
5347 kind of type that is---integer, floating, pointer, structure, and so on.
5349 The file @file{typeclass.h} defines an enumeration that you can use to
5350 interpret the values of @code{__builtin_classify_type}.
5353 These machine description macros help implement varargs:
5355 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void)
5356 If defined, this hook produces the machine-specific code for a call to
5357 @code{__builtin_saveregs}. This code will be moved to the very
5358 beginning of the function, before any parameter access are made. The
5359 return value of this function should be an RTX that contains the value
5360 to use as the return of @code{__builtin_saveregs}.
5363 @deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t @var{args_so_far}, const function_arg_info @var{&arg}, int *@var{pretend_args_size}, int @var{second_time})
5364 This target hook offers an alternative to using
5365 @code{__builtin_saveregs} and defining the hook
5366 @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous
5367 register arguments into the stack so that all the arguments appear to
5368 have been passed consecutively on the stack. Once this is done, you can
5369 use the standard implementation of varargs that works for machines that
5370 pass all their arguments on the stack.
5372 The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data
5373 structure, containing the values that are obtained after processing the
5374 named arguments. The argument @var{arg} describes the last of these named
5377 The target hook should do two things: first, push onto the stack all the
5378 argument registers @emph{not} used for the named arguments, and second,
5379 store the size of the data thus pushed into the @code{int}-valued
5380 variable pointed to by @var{pretend_args_size}. The value that you
5381 store here will serve as additional offset for setting up the stack
5384 Because you must generate code to push the anonymous arguments at
5385 compile time without knowing their data types,
5386 @code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that
5387 have just a single category of argument register and use it uniformly
5390 If the argument @var{second_time} is nonzero, it means that the
5391 arguments of the function are being analyzed for the second time. This
5392 happens for an inline function, which is not actually compiled until the
5393 end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should
5394 not generate any instructions in this case.
5397 @deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca})
5398 Define this hook to return @code{true} if the location where a function
5399 argument is passed depends on whether or not it is a named argument.
5401 This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG}
5402 is set for varargs and stdarg functions. If this hook returns
5403 @code{true}, the @var{named} argument is always true for named
5404 arguments, and false for unnamed arguments. If it returns @code{false},
5405 but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true},
5406 then all arguments are treated as named. Otherwise, all named arguments
5407 except the last are treated as named.
5409 You need not define this hook if it always returns @code{false}.
5412 @deftypefn {Target Hook} void TARGET_CALL_ARGS (rtx, @var{tree})
5413 While generating RTL for a function call, this target hook is invoked once
5414 for each argument passed to the function, either a register returned by
5415 @code{TARGET_FUNCTION_ARG} or a memory location. It is called just
5416 before the point where argument registers are stored. The type of the
5417 function to be called is also passed as the second argument; it is
5418 @code{NULL_TREE} for libcalls. The @code{TARGET_END_CALL_ARGS} hook is
5419 invoked just after the code to copy the return reg has been emitted.
5420 This functionality can be used to perform special setup of call argument
5421 registers if a target needs it.
5422 For functions without arguments, the hook is called once with @code{pc_rtx}
5423 passed instead of an argument register.
5424 Most ports do not need to implement anything for this hook.
5427 @deftypefn {Target Hook} void TARGET_END_CALL_ARGS (void)
5428 This target hook is invoked while generating RTL for a function call,
5429 just after the point where the return reg is copied into a pseudo. It
5430 signals that all the call argument and return registers for the just
5431 emitted call are now no longer in use.
5432 Most ports do not need to implement anything for this hook.
5435 @deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca})
5436 If you need to conditionally change ABIs so that one works with
5437 @code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither
5438 @code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was
5439 defined, then define this hook to return @code{true} if
5440 @code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise.
5441 Otherwise, you should not define this hook.
5445 @section Support for Nested Functions
5446 @cindex support for nested functions
5447 @cindex trampolines for nested functions
5448 @cindex descriptors for nested functions
5449 @cindex nested functions, support for
5451 Taking the address of a nested function requires special compiler
5452 handling to ensure that the static chain register is loaded when
5453 the function is invoked via an indirect call.
5455 GCC has traditionally supported nested functions by creating an
5456 executable @dfn{trampoline} at run time when the address of a nested
5457 function is taken. This is a small piece of code which normally
5458 resides on the stack, in the stack frame of the containing function.
5459 The trampoline loads the static chain register and then jumps to the
5460 real address of the nested function.
5462 The use of trampolines requires an executable stack, which is a
5463 security risk. To avoid this problem, GCC also supports another
5464 strategy: using descriptors for nested functions. Under this model,
5465 taking the address of a nested function results in a pointer to a
5466 non-executable function descriptor object. Initializing the static chain
5467 from the descriptor is handled at indirect call sites.
5469 On some targets, including HPPA and IA-64, function descriptors may be
5470 mandated by the ABI or be otherwise handled in a target-specific way
5471 by the back end in its code generation strategy for indirect calls.
5472 GCC also provides its own generic descriptor implementation to support the
5473 @option{-fno-trampolines} option. In this case runtime detection of
5474 function descriptors at indirect call sites relies on descriptor
5475 pointers being tagged with a bit that is never set in bare function
5476 addresses. Since GCC's generic function descriptors are
5477 not ABI-compliant, this option is typically used only on a
5478 per-language basis (notably by Ada) or when it can otherwise be
5479 applied to the whole program.
5481 For languages other than Ada, the @code{-ftrampolines} and
5482 @code{-fno-trampolines} options currently have no effect, and
5483 trampolines are always generated on platforms that need them
5484 for nested functions.
5486 Define the following hook if your backend either implements ABI-specified
5487 descriptor support, or can use GCC's generic descriptor implementation
5488 for nested functions.
5490 @deftypevr {Target Hook} int TARGET_CUSTOM_FUNCTION_DESCRIPTORS
5491 If the target can use GCC's generic descriptor mechanism for nested
5492 functions, define this hook to a power of 2 representing an unused bit
5493 in function pointers which can be used to differentiate descriptors at
5494 run time. This value gives the number of bytes by which descriptor
5495 pointers are misaligned compared to function pointers. For example, on
5496 targets that require functions to be aligned to a 4-byte boundary, a
5497 value of either 1 or 2 is appropriate unless the architecture already
5498 reserves the bit for another purpose, such as on ARM.
5500 Define this hook to 0 if the target implements ABI support for
5501 function descriptors in its standard calling sequence, like for example
5504 Using descriptors for nested functions
5505 eliminates the need for trampolines that reside on the stack and require
5506 it to be made executable.
5509 The following macros tell GCC how to generate code to allocate and
5510 initialize an executable trampoline. You can also use this interface
5511 if your back end needs to create ABI-specified non-executable descriptors; in
5512 this case the "trampoline" created is the descriptor containing data only.
5514 The instructions in an executable trampoline must do two things: load
5515 a constant address into the static chain register, and jump to the real
5516 address of the nested function. On CISC machines such as the m68k,
5517 this requires two instructions, a move immediate and a jump. Then the
5518 two addresses exist in the trampoline as word-long immediate operands.
5519 On RISC machines, it is often necessary to load each address into a
5520 register in two parts. Then pieces of each address form separate
5523 The code generated to initialize the trampoline must store the variable
5524 parts---the static chain value and the function address---into the
5525 immediate operands of the instructions. On a CISC machine, this is
5526 simply a matter of copying each address to a memory reference at the
5527 proper offset from the start of the trampoline. On a RISC machine, it
5528 may be necessary to take out pieces of the address and store them
5531 @deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f})
5532 This hook is called by @code{assemble_trampoline_template} to output,
5533 on the stream @var{f}, assembler code for a block of data that contains
5534 the constant parts of a trampoline. This code should not include a
5535 label---the label is taken care of automatically.
5537 If you do not define this hook, it means no template is needed
5538 for the target. Do not define this hook on systems where the block move
5539 code to copy the trampoline into place would be larger than the code
5540 to generate it on the spot.
5543 @defmac TRAMPOLINE_SECTION
5544 Return the section into which the trampoline template is to be placed
5545 (@pxref{Sections}). The default value is @code{readonly_data_section}.
5548 @defmac TRAMPOLINE_SIZE
5549 A C expression for the size in bytes of the trampoline, as an integer.
5552 @defmac TRAMPOLINE_ALIGNMENT
5553 Alignment required for trampolines, in bits.
5555 If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT}
5556 is used for aligning trampolines.
5559 @deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain})
5560 This hook is called to initialize a trampoline.
5561 @var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl}
5562 is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an
5563 RTX for the static chain value that should be passed to the function
5566 If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the
5567 first thing this hook should do is emit a block move into @var{m_tramp}
5568 from the memory block returned by @code{assemble_trampoline_template}.
5569 Note that the block move need only cover the constant parts of the
5570 trampoline. If the target isolates the variable parts of the trampoline
5571 to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied.
5573 If the target requires any other actions, such as flushing caches
5574 (possibly calling function maybe_emit_call_builtin___clear_cache) or
5575 enabling stack execution, these actions should be performed after
5576 initializing the trampoline proper.
5579 @deftypefn {Target Hook} void TARGET_EMIT_CALL_BUILTIN___CLEAR_CACHE (rtx @var{begin}, rtx @var{end})
5580 On targets that do not define a @code{clear_cache} insn expander,
5581 but that define the @code{CLEAR_CACHE_INSN} macro,
5582 maybe_emit_call_builtin___clear_cache relies on this target hook
5583 to clear an address range in the instruction cache.
5585 The default implementation calls the @code{__clear_cache} builtin,
5586 taking the assembler name from the builtin declaration. Overriding
5587 definitions may call alternate functions, with alternate calling
5588 conventions, or emit alternate RTX to perform the job.
5591 @deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr})
5592 This hook should perform any machine-specific adjustment in
5593 the address of the trampoline. Its argument contains the address of the
5594 memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case
5595 the address to be used for a function call should be different from the
5596 address at which the template was stored, the different address should
5597 be returned; otherwise @var{addr} should be returned unchanged.
5598 If this hook is not defined, @var{addr} will be used for function calls.
5601 Implementing trampolines is difficult on many machines because they have
5602 separate instruction and data caches. Writing into a stack location
5603 fails to clear the memory in the instruction cache, so when the program
5604 jumps to that location, it executes the old contents.
5606 Here are two possible solutions. One is to clear the relevant parts of
5607 the instruction cache whenever a trampoline is set up. The other is to
5608 make all trampolines identical, by having them jump to a standard
5609 subroutine. The former technique makes trampoline execution faster; the
5610 latter makes initialization faster.
5612 To clear the instruction cache when a trampoline is initialized, define
5613 the following macro.
5615 @defmac CLEAR_INSN_CACHE (@var{beg}, @var{end})
5616 If defined, expands to a C expression clearing the @emph{instruction
5617 cache} in the specified interval. The definition of this macro would
5618 typically be a series of @code{asm} statements. Both @var{beg} and
5619 @var{end} are pointer expressions.
5622 To use a standard subroutine, define the following macro. In addition,
5623 you must make sure that the instructions in a trampoline fill an entire
5624 cache line with identical instructions, or else ensure that the
5625 beginning of the trampoline code is always aligned at the same point in
5626 its cache line. Look in @file{m68k.h} as a guide.
5628 @defmac TRANSFER_FROM_TRAMPOLINE
5629 Define this macro if trampolines need a special subroutine to do their
5630 work. The macro should expand to a series of @code{asm} statements
5631 which will be compiled with GCC@. They go in a library function named
5632 @code{__transfer_from_trampoline}.
5634 If you need to avoid executing the ordinary prologue code of a compiled
5635 C function when you jump to the subroutine, you can do so by placing a
5636 special label of your own in the assembler code. Use one @code{asm}
5637 statement to generate an assembler label, and another to make the label
5638 global. Then trampolines can use that label to jump directly to your
5639 special assembler code.
5643 @section Implicit Calls to Library Routines
5644 @cindex library subroutine names
5645 @cindex @file{libgcc.a}
5647 @c prevent bad page break with this line
5648 Here is an explanation of implicit calls to library routines.
5650 @defmac DECLARE_LIBRARY_RENAMES
5651 This macro, if defined, should expand to a piece of C code that will get
5652 expanded when compiling functions for libgcc.a. It can be used to
5653 provide alternate names for GCC's internal library functions if there
5654 are ABI-mandated names that the compiler should provide.
5657 @findex set_optab_libfunc
5658 @findex init_one_libfunc
5659 @deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void)
5660 This hook should declare additional library routines or rename
5661 existing ones, using the functions @code{set_optab_libfunc} and
5662 @code{init_one_libfunc} defined in @file{optabs.cc}.
5663 @code{init_optabs} calls this macro after initializing all the normal
5666 The default is to do nothing. Most ports don't need to define this hook.
5669 @deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX
5670 If false (the default), internal library routines start with two
5671 underscores. If set to true, these routines start with @code{__gnu_}
5672 instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This
5673 currently only affects functions defined in @file{libgcc2.c}. If this
5674 is set to true, the @file{tm.h} file must also
5675 @code{#define LIBGCC2_GNU_PREFIX}.
5678 @defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison})
5679 This macro should return @code{true} if the library routine that
5680 implements the floating point comparison operator @var{comparison} in
5681 mode @var{mode} will return a boolean, and @var{false} if it will
5684 GCC's own floating point libraries return tristates from the
5685 comparison operators, so the default returns false always. Most ports
5686 don't need to define this macro.
5689 @defmac TARGET_LIB_INT_CMP_BIASED
5690 This macro should evaluate to @code{true} if the integer comparison
5691 functions (like @code{__cmpdi2}) return 0 to indicate that the first
5692 operand is smaller than the second, 1 to indicate that they are equal,
5693 and 2 to indicate that the first operand is greater than the second.
5694 If this macro evaluates to @code{false} the comparison functions return
5695 @minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines
5696 in @file{libgcc.a}, you do not need to define this macro.
5699 @defmac TARGET_HAS_NO_HW_DIVIDE
5700 This macro should be defined if the target has no hardware divide
5701 instructions. If this macro is defined, GCC will use an algorithm which
5702 make use of simple logical and arithmetic operations for 64-bit
5703 division. If the macro is not defined, GCC will use an algorithm which
5704 make use of a 64-bit by 32-bit divide primitive.
5707 @cindex @code{EDOM}, implicit usage
5710 The value of @code{EDOM} on the target machine, as a C integer constant
5711 expression. If you don't define this macro, GCC does not attempt to
5712 deposit the value of @code{EDOM} into @code{errno} directly. Look in
5713 @file{/usr/include/errno.h} to find the value of @code{EDOM} on your
5716 If you do not define @code{TARGET_EDOM}, then compiled code reports
5717 domain errors by calling the library function and letting it report the
5718 error. If mathematical functions on your system use @code{matherr} when
5719 there is an error, then you should leave @code{TARGET_EDOM} undefined so
5720 that @code{matherr} is used normally.
5723 @cindex @code{errno}, implicit usage
5724 @defmac GEN_ERRNO_RTX
5725 Define this macro as a C expression to create an rtl expression that
5726 refers to the global ``variable'' @code{errno}. (On certain systems,
5727 @code{errno} may not actually be a variable.) If you don't define this
5728 macro, a reasonable default is used.
5731 @deftypefn {Target Hook} bool TARGET_LIBC_HAS_FUNCTION (enum function_class @var{fn_class}, tree @var{type})
5732 This hook determines whether a function from a class of functions
5733 @var{fn_class} is present in the target C library. If @var{type} is NULL,
5734 the caller asks for support for all standard (float, double, long double)
5735 types. If @var{type} is non-NULL, the caller asks for support for a
5739 @deftypefn {Target Hook} bool TARGET_LIBC_HAS_FAST_FUNCTION (int @var{fcode})
5740 This hook determines whether a function from a class of functions
5741 @code{(enum function_class)}@var{fcode} has a fast implementation.
5744 @defmac NEXT_OBJC_RUNTIME
5745 Set this macro to 1 to use the "NeXT" Objective-C message sending conventions
5746 by default. This calling convention involves passing the object, the selector
5747 and the method arguments all at once to the method-lookup library function.
5748 This is the usual setting when targeting Darwin/Mac OS X systems, which have
5749 the NeXT runtime installed.
5751 If the macro is set to 0, the "GNU" Objective-C message sending convention
5752 will be used by default. This convention passes just the object and the
5753 selector to the method-lookup function, which returns a pointer to the method.
5755 In either case, it remains possible to select code-generation for the alternate
5756 scheme, by means of compiler command line switches.
5759 @node Addressing Modes
5760 @section Addressing Modes
5761 @cindex addressing modes
5763 @c prevent bad page break with this line
5764 This is about addressing modes.
5766 @defmac HAVE_PRE_INCREMENT
5767 @defmacx HAVE_PRE_DECREMENT
5768 @defmacx HAVE_POST_INCREMENT
5769 @defmacx HAVE_POST_DECREMENT
5770 A C expression that is nonzero if the machine supports pre-increment,
5771 pre-decrement, post-increment, or post-decrement addressing respectively.
5774 @defmac HAVE_PRE_MODIFY_DISP
5775 @defmacx HAVE_POST_MODIFY_DISP
5776 A C expression that is nonzero if the machine supports pre- or
5777 post-address side-effect generation involving constants other than
5778 the size of the memory operand.
5781 @defmac HAVE_PRE_MODIFY_REG
5782 @defmacx HAVE_POST_MODIFY_REG
5783 A C expression that is nonzero if the machine supports pre- or
5784 post-address side-effect generation involving a register displacement.
5787 @defmac CONSTANT_ADDRESS_P (@var{x})
5788 A C expression that is 1 if the RTX @var{x} is a constant which
5789 is a valid address. On most machines the default definition of
5790 @code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)}
5791 is acceptable, but a few machines are more restrictive as to which
5792 constant addresses are supported.
5795 @defmac CONSTANT_P (@var{x})
5796 @code{CONSTANT_P}, which is defined by target-independent code,
5797 accepts integer-values expressions whose values are not explicitly
5798 known, such as @code{symbol_ref}, @code{label_ref}, and @code{high}
5799 expressions and @code{const} arithmetic expressions, in addition to
5800 @code{const_int} and @code{const_double} expressions.
5803 @defmac MAX_REGS_PER_ADDRESS
5804 A number, the maximum number of registers that can appear in a valid
5805 memory address. Note that it is up to you to specify a value equal to
5806 the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever
5810 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{x}, bool @var{strict})
5811 A function that returns whether @var{x} (an RTX) is a legitimate memory
5812 address on the target machine for a memory operand of mode @var{mode}.
5814 Legitimate addresses are defined in two variants: a strict variant and a
5815 non-strict one. The @var{strict} parameter chooses which variant is
5816 desired by the caller.
5818 The strict variant is used in the reload pass. It must be defined so
5819 that any pseudo-register that has not been allocated a hard register is
5820 considered a memory reference. This is because in contexts where some
5821 kind of register is required, a pseudo-register with no hard register
5822 must be rejected. For non-hard registers, the strict variant should look
5823 up the @code{reg_renumber} array; it should then proceed using the hard
5824 register number in the array, or treat the pseudo as a memory reference
5825 if the array holds @code{-1}.
5827 The non-strict variant is used in other passes. It must be defined to
5828 accept all pseudo-registers in every context where some kind of
5829 register is required.
5831 Normally, constant addresses which are the sum of a @code{symbol_ref}
5832 and an integer are stored inside a @code{const} RTX to mark them as
5833 constant. Therefore, there is no need to recognize such sums
5834 specifically as legitimate addresses. Normally you would simply
5835 recognize any @code{const} as legitimate.
5837 Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant
5838 sums that are not marked with @code{const}. It assumes that a naked
5839 @code{plus} indicates indexing. If so, then you @emph{must} reject such
5840 naked constant sums as illegitimate addresses, so that none of them will
5841 be given to @code{PRINT_OPERAND_ADDRESS}.
5843 @cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation
5844 On some machines, whether a symbolic address is legitimate depends on
5845 the section that the address refers to. On these machines, define the
5846 target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information
5847 into the @code{symbol_ref}, and then check for it here. When you see a
5848 @code{const}, you will have to look inside it to find the
5849 @code{symbol_ref} in order to determine the section. @xref{Assembler
5852 @cindex @code{GO_IF_LEGITIMATE_ADDRESS}
5853 Some ports are still using a deprecated legacy substitute for
5854 this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro
5858 #define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label})
5862 and should @code{goto @var{label}} if the address @var{x} is a valid
5863 address on the target machine for a memory operand of mode @var{mode}.
5865 @findex REG_OK_STRICT
5866 Compiler source files that want to use the strict variant of this
5867 macro define the macro @code{REG_OK_STRICT}. You should use an
5868 @code{#ifdef REG_OK_STRICT} conditional to define the strict variant in
5869 that case and the non-strict variant otherwise.
5871 Using the hook is usually simpler because it limits the number of
5872 files that are recompiled when changes are made.
5875 @defmac TARGET_MEM_CONSTRAINT
5876 A single character to be used instead of the default @code{'m'}
5877 character for general memory addresses. This defines the constraint
5878 letter which matches the memory addresses accepted by
5879 @code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to
5880 support new address formats in your back end without changing the
5881 semantics of the @code{'m'} constraint. This is necessary in order to
5882 preserve functionality of inline assembly constructs using the
5883 @code{'m'} constraint.
5886 @defmac FIND_BASE_TERM (@var{x})
5887 A C expression to determine the base term of address @var{x},
5888 or to provide a simplified version of @var{x} from which @file{alias.cc}
5889 can easily find the base term. This macro is used in only two places:
5890 @code{find_base_value} and @code{find_base_term} in @file{alias.cc}.
5892 It is always safe for this macro to not be defined. It exists so
5893 that alias analysis can understand machine-dependent addresses.
5895 The typical use of this macro is to handle addresses containing
5896 a label_ref or symbol_ref within an UNSPEC@.
5899 @deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode})
5900 This hook is given an invalid memory address @var{x} for an
5901 operand of mode @var{mode} and should try to return a valid memory
5904 @findex break_out_memory_refs
5905 @var{x} will always be the result of a call to @code{break_out_memory_refs},
5906 and @var{oldx} will be the operand that was given to that function to produce
5909 The code of the hook should not alter the substructure of
5910 @var{x}. If it transforms @var{x} into a more legitimate form, it
5911 should return the new @var{x}.
5913 It is not necessary for this hook to come up with a legitimate address,
5914 with the exception of native TLS addresses (@pxref{Emulated TLS}).
5915 The compiler has standard ways of doing so in all cases. In fact, if
5916 the target supports only emulated TLS, it
5917 is safe to omit this hook or make it return @var{x} if it cannot find
5918 a valid way to legitimize the address. But often a machine-dependent
5919 strategy can generate better code.
5922 @defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win})
5923 A C compound statement that attempts to replace @var{x}, which is an address
5924 that needs reloading, with a valid memory address for an operand of mode
5925 @var{mode}. @var{win} will be a C statement label elsewhere in the code.
5926 It is not necessary to define this macro, but it might be useful for
5927 performance reasons.
5929 For example, on the i386, it is sometimes possible to use a single
5930 reload register instead of two by reloading a sum of two pseudo
5931 registers into a register. On the other hand, for number of RISC
5932 processors offsets are limited so that often an intermediate address
5933 needs to be generated in order to address a stack slot. By defining
5934 @code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses
5935 generated for adjacent some stack slots can be made identical, and thus
5938 @emph{Note}: This macro should be used with caution. It is necessary
5939 to know something of how reload works in order to effectively use this,
5940 and it is quite easy to produce macros that build in too much knowledge
5941 of reload internals.
5943 @emph{Note}: This macro must be able to reload an address created by a
5944 previous invocation of this macro. If it fails to handle such addresses
5945 then the compiler may generate incorrect code or abort.
5948 The macro definition should use @code{push_reload} to indicate parts that
5949 need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually
5950 suitable to be passed unaltered to @code{push_reload}.
5952 The code generated by this macro must not alter the substructure of
5953 @var{x}. If it transforms @var{x} into a more legitimate form, it
5954 should assign @var{x} (which will always be a C variable) a new value.
5955 This also applies to parts that you change indirectly by calling
5958 @findex strict_memory_address_p
5959 The macro definition may use @code{strict_memory_address_p} to test if
5960 the address has become legitimate.
5963 If you want to change only a part of @var{x}, one standard way of doing
5964 this is to use @code{copy_rtx}. Note, however, that it unshares only a
5965 single level of rtl. Thus, if the part to be changed is not at the
5966 top level, you'll need to replace first the top level.
5967 It is not necessary for this macro to come up with a legitimate
5968 address; but often a machine-dependent strategy can generate better code.
5971 @deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr}, addr_space_t @var{addrspace})
5972 This hook returns @code{true} if memory address @var{addr} in address
5973 space @var{addrspace} can have
5974 different meanings depending on the machine mode of the memory
5975 reference it is used for or if the address is valid for some modes
5978 Autoincrement and autodecrement addresses typically have mode-dependent
5979 effects because the amount of the increment or decrement is the size
5980 of the operand being addressed. Some machines have other mode-dependent
5981 addresses. Many RISC machines have no mode-dependent addresses.
5983 You may assume that @var{addr} is a valid address for the machine.
5985 The default version of this hook returns @code{false}.
5988 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (machine_mode @var{mode}, rtx @var{x})
5989 This hook returns true if @var{x} is a legitimate constant for a
5990 @var{mode}-mode immediate operand on the target machine. You can assume that
5991 @var{x} satisfies @code{CONSTANT_P}, so you need not check this.
5993 The default definition returns true.
5996 @deftypefn {Target Hook} bool TARGET_PRECOMPUTE_TLS_P (machine_mode @var{mode}, rtx @var{x})
5997 This hook returns true if @var{x} is a TLS operand on the target
5998 machine that should be pre-computed when used as the argument in a call.
5999 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
6002 The default definition returns false.
6005 @deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x})
6006 This hook is used to undo the possibly obfuscating effects of the
6007 @code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target
6008 macros. Some backend implementations of these macros wrap symbol
6009 references inside an @code{UNSPEC} rtx to represent PIC or similar
6010 addressing modes. This target hook allows GCC's optimizers to understand
6011 the semantics of these opaque @code{UNSPEC}s by converting them back
6012 into their original form.
6015 @deftypefn {Target Hook} bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx @var{x})
6016 This hook should return true if @var{x} should not be emitted into
6020 @deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (machine_mode @var{mode}, rtx @var{x})
6021 This hook should return true if @var{x} is of a form that cannot (or
6022 should not) be spilled to the constant pool. @var{mode} is the mode
6025 The default version of this hook returns false.
6027 The primary reason to define this hook is to prevent reload from
6028 deciding that a non-legitimate constant would be better reloaded
6029 from the constant pool instead of spilling and reloading a register
6030 holding the constant. This restriction is often true of addresses
6031 of TLS symbols for various targets.
6034 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (machine_mode @var{mode}, const_rtx @var{x})
6035 This hook should return true if pool entries for constant @var{x} can
6036 be placed in an @code{object_block} structure. @var{mode} is the mode
6039 The default version returns false for all constants.
6042 @deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_DECL_P (const_tree @var{decl})
6043 This hook should return true if pool entries for @var{decl} should
6044 be placed in an @code{object_block} structure.
6046 The default version returns true for all decls.
6049 @deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (tree @var{fndecl})
6050 This hook should return the DECL of a function that implements the
6051 reciprocal of the machine-specific builtin function @var{fndecl}, or
6052 @code{NULL_TREE} if such a function is not available.
6055 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void)
6056 This hook should return the DECL of a function @var{f} that given an
6057 address @var{addr} as an argument returns a mask @var{m} that can be
6058 used to extract from two vectors the relevant data that resides in
6059 @var{addr} in case @var{addr} is not properly aligned.
6061 The autovectorizer, when vectorizing a load operation from an address
6062 @var{addr} that may be unaligned, will generate two vector loads from
6063 the two aligned addresses around @var{addr}. It then generates a
6064 @code{REALIGN_LOAD} operation to extract the relevant data from the
6065 two loaded vectors. The first two arguments to @code{REALIGN_LOAD},
6066 @var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and
6067 the third argument, @var{OFF}, defines how the data will be extracted
6068 from these two vectors: if @var{OFF} is 0, then the returned vector is
6069 @var{v2}; otherwise, the returned vector is composed from the last
6070 @var{VS}-@var{OFF} elements of @var{v1} concatenated to the first
6071 @var{OFF} elements of @var{v2}.
6073 If this hook is defined, the autovectorizer will generate a call
6074 to @var{f} (using the DECL tree that this hook returns) and will
6075 use the return value of @var{f} as the argument @var{OFF} to
6076 @code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f}
6077 should comply with the semantics expected by @code{REALIGN_LOAD}
6079 If this hook is not defined, then @var{addr} will be used as
6080 the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low
6081 log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered.
6084 @deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign})
6085 Returns cost of different scalar or vector statements for vectorization cost model.
6086 For vector memory operations the cost may depend on type (@var{vectype}) and
6087 misalignment value (@var{misalign}).
6090 @deftypefn {Target Hook} poly_uint64 TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT (const_tree @var{type})
6091 This hook returns the preferred alignment in bits for accesses to
6092 vectors of type @var{type} in vectorized code. This might be less than
6093 or greater than the ABI-defined value returned by
6094 @code{TARGET_VECTOR_ALIGNMENT}. It can be equal to the alignment of
6095 a single element, in which case the vectorizer will not try to optimize
6098 The default hook returns @code{TYPE_ALIGN (@var{type})}, which is
6099 correct for most targets.
6102 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed})
6103 Return true if vector alignment is reachable (by peeling N iterations)
6104 for the given scalar type @var{type}. @var{is_packed} is false if the scalar
6105 access using @var{type} is known to be naturally aligned.
6108 @deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST (machine_mode @var{mode}, machine_mode @var{op_mode}, rtx @var{output}, rtx @var{in0}, rtx @var{in1}, const vec_perm_indices @var{&sel})
6109 This hook is used to test whether the target can permute up to two
6110 vectors of mode @var{op_mode} using the permutation vector @code{sel},
6111 producing a vector of mode @var{mode}. The hook is also used to emit such
6114 When the hook is being used to test whether the target supports a permutation,
6115 @var{in0}, @var{in1}, and @var{out} are all null. When the hook is being used
6116 to emit a permutation, @var{in0} and @var{in1} are the source vectors of mode
6117 @var{op_mode} and @var{out} is the destination vector of mode @var{mode}.
6118 @var{in1} is the same as @var{in0} if @var{sel} describes a permutation on one
6119 vector instead of two.
6121 Return true if the operation is possible, emitting instructions for it
6122 if rtxes are provided.
6124 @cindex @code{vec_perm@var{m}} instruction pattern
6125 If the hook returns false for a mode with multibyte elements, GCC will
6126 try the equivalent byte operation. If that also fails, it will try forcing
6127 the selector into a register and using the @var{vec_perm@var{mode}}
6128 instruction pattern. There is no need for the hook to handle these two
6129 implementation approaches itself.
6132 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (unsigned @var{code}, tree @var{vec_type_out}, tree @var{vec_type_in})
6133 This hook should return the decl of a function that implements the
6134 vectorized variant of the function with the @code{combined_fn} code
6135 @var{code} or @code{NULL_TREE} if such a function is not available.
6136 The return type of the vectorized function shall be of vector type
6137 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
6140 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in})
6141 This hook should return the decl of a function that implements the
6142 vectorized variant of target built-in function @code{fndecl}. The
6143 return type of the vectorized function shall be of vector type
6144 @var{vec_type_out} and the argument types should be @var{vec_type_in}.
6147 @deftypefn {Target Hook} bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT (machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed})
6148 This hook should return true if the target supports misaligned vector
6149 store/load of a specific factor denoted in the @var{misalignment}
6150 parameter. The vector store/load should be of machine mode @var{mode} and
6151 the elements in the vectors should be of type @var{type}. @var{is_packed}
6152 parameter is true if the memory access is defined in a packed struct.
6155 @deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_PREFERRED_SIMD_MODE (scalar_mode @var{mode})
6156 This hook should return the preferred mode for vectorizing scalar
6157 mode @var{mode}. The default is
6158 equal to @code{word_mode}, because the vectorizer can do some
6159 transformations even in absence of specialized @acronym{SIMD} hardware.
6162 @deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_SPLIT_REDUCTION (machine_mode)
6163 This hook should return the preferred mode to split the final reduction
6164 step on @var{mode} to. The reduction is then carried out reducing upper
6165 against lower halves of vectors recursively until the specified mode is
6166 reached. The default is @var{mode} which means no splitting.
6169 @deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES (vector_modes *@var{modes}, bool @var{all})
6170 If using the mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}
6171 is not the only approach worth considering, this hook should add one mode to
6172 @var{modes} for each useful alternative approach. These modes are then
6173 passed to @code{TARGET_VECTORIZE_RELATED_MODE} to obtain the vector mode
6174 for a given element mode.
6176 The modes returned in @var{modes} should use the smallest element mode
6177 possible for the vectorization approach that they represent, preferring
6178 integer modes over floating-poing modes in the event of a tie. The first
6179 mode should be the @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} for its
6182 If @var{all} is true, add suitable vector modes even when they are generally
6183 not expected to be worthwhile.
6185 The hook returns a bitmask of flags that control how the modes in
6186 @var{modes} are used. The flags are:
6188 @item VECT_COMPARE_COSTS
6189 Tells the loop vectorizer to try all the provided modes and pick the one
6190 with the lowest cost. By default the vectorizer will choose the first
6194 The hook does not need to do anything if the vector returned by
6195 @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} is the only one relevant
6196 for autovectorization. The default implementation adds no modes and
6200 @deftypefn {Target Hook} opt_machine_mode TARGET_VECTORIZE_RELATED_MODE (machine_mode @var{vector_mode}, scalar_mode @var{element_mode}, poly_uint64 @var{nunits})
6201 If a piece of code is using vector mode @var{vector_mode} and also wants
6202 to operate on elements of mode @var{element_mode}, return the vector mode
6203 it should use for those elements. If @var{nunits} is nonzero, ensure that
6204 the mode has exactly @var{nunits} elements, otherwise pick whichever vector
6205 size pairs the most naturally with @var{vector_mode}. Return an empty
6206 @code{opt_machine_mode} if there is no supported vector mode with the
6207 required properties.
6209 There is no prescribed way of handling the case in which @var{nunits}
6210 is zero. One common choice is to pick a vector mode with the same size
6211 as @var{vector_mode}; this is the natural choice if the target has a
6212 fixed vector size. Another option is to choose a vector mode with the
6213 same number of elements as @var{vector_mode}; this is the natural choice
6214 if the target has a fixed number of elements. Alternatively, the hook
6215 might choose a middle ground, such as trying to keep the number of
6216 elements as similar as possible while applying maximum and minimum
6219 The default implementation uses @code{mode_for_vector} to find the
6220 requested mode, returning a mode with the same size as @var{vector_mode}
6221 when @var{nunits} is zero. This is the correct behavior for most targets.
6224 @deftypefn {Target Hook} opt_machine_mode TARGET_VECTORIZE_GET_MASK_MODE (machine_mode @var{mode})
6225 Return the mode to use for a vector mask that holds one boolean
6226 result for each element of vector mode @var{mode}. The returned mask mode
6227 can be a vector of integers (class @code{MODE_VECTOR_INT}), a vector of
6228 booleans (class @code{MODE_VECTOR_BOOL}) or a scalar integer (class
6229 @code{MODE_INT}). Return an empty @code{opt_machine_mode} if no such
6232 The default implementation returns a @code{MODE_VECTOR_INT} with the
6233 same size and number of elements as @var{mode}, if such a mode exists.
6236 @deftypefn {Target Hook} bool TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE (unsigned @var{ifn})
6237 This hook returns true if masked internal function @var{ifn} (really of
6238 type @code{internal_fn}) should be considered expensive when the mask is
6239 all zeros. GCC can then try to branch around the instruction instead.
6242 @deftypefn {Target Hook} {class vector_costs *} TARGET_VECTORIZE_CREATE_COSTS (vec_info *@var{vinfo}, bool @var{costing_for_scalar})
6243 This hook should initialize target-specific data structures in preparation
6244 for modeling the costs of vectorizing a loop or basic block. The default
6245 allocates three unsigned integers for accumulating costs for the prologue,
6246 body, and epilogue of the loop or basic block. If @var{loop_info} is
6247 non-NULL, it identifies the loop being vectorized; otherwise a single block
6248 is being vectorized. If @var{costing_for_scalar} is true, it indicates the
6249 current cost model is for the scalar version of a loop or block; otherwise
6250 it is for the vector version.
6253 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale})
6254 Target builtin that implements vector gather operation. @var{mem_vectype}
6255 is the vector type of the load and @var{index_type} is scalar type of
6256 the index, scaled by @var{scale}.
6257 The default is @code{NULL_TREE} which means to not vectorize gather
6261 @deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_SCATTER (const_tree @var{vectype}, const_tree @var{index_type}, int @var{scale})
6262 Target builtin that implements vector scatter operation. @var{vectype}
6263 is the vector type of the store and @var{index_type} is scalar type of
6264 the index, scaled by @var{scale}.
6265 The default is @code{NULL_TREE} which means to not vectorize scatter
6269 @deftypefn {Target Hook} int TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN (struct cgraph_node *@var{}, struct cgraph_simd_clone *@var{}, @var{tree}, @var{int})
6270 This hook should set @var{vecsize_mangle}, @var{vecsize_int}, @var{vecsize_float}
6271 fields in @var{simd_clone} structure pointed by @var{clone_info} argument and also
6272 @var{simdlen} field if it was previously 0.
6273 @var{vecsize_mangle} is a marker for the backend only. @var{vecsize_int} and
6274 @var{vecsize_float} should be left zero on targets where the number of lanes is
6275 not determined by the bitsize (in which case @var{simdlen} is always used).
6276 The hook should return 0 if SIMD clones shouldn't be emitted,
6277 or number of @var{vecsize_mangle} variants that should be emitted.
6280 @deftypefn {Target Hook} void TARGET_SIMD_CLONE_ADJUST (struct cgraph_node *@var{})
6281 This hook should add implicit @code{attribute(target("..."))} attribute
6282 to SIMD clone @var{node} if needed.
6285 @deftypefn {Target Hook} int TARGET_SIMD_CLONE_USABLE (struct cgraph_node *@var{})
6286 This hook should return -1 if SIMD clone @var{node} shouldn't be used
6287 in vectorized loops in current function, or non-negative number if it is
6288 usable. In that case, the smaller the number is, the more desirable it is
6292 @deftypefn {Target Hook} int TARGET_SIMT_VF (void)
6293 Return number of threads in SIMT thread group on the target.
6296 @deftypefn {Target Hook} int TARGET_OMP_DEVICE_KIND_ARCH_ISA (enum omp_device_kind_arch_isa @var{trait}, const char *@var{name})
6297 Return 1 if @var{trait} @var{name} is present in the OpenMP context's
6298 device trait set, return 0 if not present in any OpenMP context in the
6299 whole translation unit, or -1 if not present in the current OpenMP context
6300 but might be present in another OpenMP context in the same TU.
6303 @deftypefn {Target Hook} bool TARGET_GOACC_VALIDATE_DIMS (tree @var{decl}, int *@var{dims}, int @var{fn_level}, unsigned @var{used})
6304 This hook should check the launch dimensions provided for an OpenACC
6305 compute region, or routine. Defaulted values are represented as -1
6306 and non-constant values as 0. The @var{fn_level} is negative for the
6307 function corresponding to the compute region. For a routine it is the
6308 outermost level at which partitioned execution may be spawned. The hook
6309 should verify non-default values. If DECL is NULL, global defaults
6310 are being validated and unspecified defaults should be filled in.
6311 Diagnostics should be issued as appropriate. Return
6312 true, if changes have been made. You must override this hook to
6313 provide dimensions larger than 1.
6316 @deftypefn {Target Hook} int TARGET_GOACC_DIM_LIMIT (int @var{axis})
6317 This hook should return the maximum size of a particular dimension,
6318 or zero if unbounded.
6321 @deftypefn {Target Hook} bool TARGET_GOACC_FORK_JOIN (gcall *@var{call}, const int *@var{dims}, bool @var{is_fork})
6322 This hook can be used to convert IFN_GOACC_FORK and IFN_GOACC_JOIN
6323 function calls to target-specific gimple, or indicate whether they
6324 should be retained. It is executed during the oacc_device_lower pass.
6325 It should return true, if the call should be retained. It should
6326 return false, if it is to be deleted (either because target-specific
6327 gimple has been inserted before it, or there is no need for it).
6328 The default hook returns false, if there are no RTL expanders for them.
6331 @deftypefn {Target Hook} void TARGET_GOACC_REDUCTION (gcall *@var{call})
6332 This hook is used by the oacc_transform pass to expand calls to the
6333 @var{GOACC_REDUCTION} internal function, into a sequence of gimple
6334 instructions. @var{call} is gimple statement containing the call to
6335 the function. This hook removes statement @var{call} after the
6336 expanded sequence has been inserted. This hook is also responsible
6337 for allocating any storage for reductions when necessary.
6340 @deftypefn {Target Hook} tree TARGET_PREFERRED_ELSE_VALUE (unsigned @var{ifn}, tree @var{type}, unsigned @var{nops}, tree *@var{ops})
6341 This hook returns the target's preferred final argument for a call
6342 to conditional internal function @var{ifn} (really of type
6343 @code{internal_fn}). @var{type} specifies the return type of the
6344 function and @var{ops} are the operands to the conditional operation,
6345 of which there are @var{nops}.
6347 For example, if @var{ifn} is @code{IFN_COND_ADD}, the hook returns
6348 a value of type @var{type} that should be used when @samp{@var{ops}[0]}
6349 and @samp{@var{ops}[1]} are conditionally added together.
6351 This hook is only relevant if the target supports conditional patterns
6352 like @code{cond_add@var{m}}. The default implementation returns a zero
6353 constant of type @var{type}.
6356 @deftypefn {Target Hook} tree TARGET_GOACC_ADJUST_PRIVATE_DECL (location_t @var{loc}, tree @var{var}, int @var{level})
6357 This hook, if defined, is used by accelerator target back-ends to adjust
6358 OpenACC variable declarations that should be made private to the given
6359 parallelism level (i.e. @code{GOMP_DIM_GANG}, @code{GOMP_DIM_WORKER} or
6360 @code{GOMP_DIM_VECTOR}). A typical use for this hook is to force variable
6361 declarations at the @code{gang} level to reside in GPU shared memory.
6362 @var{loc} may be used for diagnostic purposes.
6364 You may also use the @code{TARGET_GOACC_EXPAND_VAR_DECL} hook if the
6365 adjusted variable declaration needs to be expanded to RTL in a non-standard
6369 @deftypefn {Target Hook} rtx TARGET_GOACC_EXPAND_VAR_DECL (tree @var{var})
6370 This hook, if defined, is used by accelerator target back-ends to expand
6371 specially handled kinds of @code{VAR_DECL} expressions. A particular use is
6372 to place variables with specific attributes inside special accelarator
6373 memories. A return value of @code{NULL} indicates that the target does not
6374 handle this @code{VAR_DECL}, and normal RTL expanding is resumed.
6376 Only define this hook if your accelerator target needs to expand certain
6377 @code{VAR_DECL} nodes in a way that differs from the default. You can also adjust
6378 private variables at OpenACC device-lowering time using the
6379 @code{TARGET_GOACC_ADJUST_PRIVATE_DECL} target hook.
6382 @deftypefn {Target Hook} tree TARGET_GOACC_CREATE_WORKER_BROADCAST_RECORD (tree @var{rec}, bool @var{sender}, const char *@var{name}, unsigned HOST_WIDE_INT @var{offset})
6383 Create a record used to propagate local-variable state from an active
6384 worker to other workers. A possible implementation might adjust the type
6385 of REC to place the new variable in shared GPU memory.
6387 Presence of this target hook indicates that middle end neutering/broadcasting
6391 @deftypefn {Target Hook} void TARGET_GOACC_SHARED_MEM_LAYOUT (unsigned HOST_WIDE_INT *@var{}, unsigned HOST_WIDE_INT *@var{}, @var{int[]}, unsigned @var{HOST_WIDE_INT[]}, unsigned @var{HOST_WIDE_INT[]})
6392 Lay out a fixed shared-memory region on the target. The LO and HI
6393 arguments should be set to a range of addresses that can be used for worker
6394 broadcasting. The dimensions, reduction size and gang-private size
6395 arguments are for the current offload region.
6398 @node Anchored Addresses
6399 @section Anchored Addresses
6400 @cindex anchored addresses
6401 @cindex @option{-fsection-anchors}
6403 GCC usually addresses every static object as a separate entity.
6404 For example, if we have:
6408 int foo (void) @{ return a + b + c; @}
6411 the code for @code{foo} will usually calculate three separate symbolic
6412 addresses: those of @code{a}, @code{b} and @code{c}. On some targets,
6413 it would be better to calculate just one symbolic address and access
6414 the three variables relative to it. The equivalent pseudocode would
6420 register int *xr = &x;
6421 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
6425 (which isn't valid C). We refer to shared addresses like @code{x} as
6426 ``section anchors''. Their use is controlled by @option{-fsection-anchors}.
6428 The hooks below describe the target properties that GCC needs to know
6429 in order to make effective use of section anchors. It won't use
6430 section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET}
6431 or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value.
6433 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET
6434 The minimum offset that should be applied to a section anchor.
6435 On most targets, it should be the smallest offset that can be
6436 applied to a base register while still giving a legitimate address
6437 for every mode. The default value is 0.
6440 @deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET
6441 Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive)
6442 offset that should be applied to section anchors. The default
6446 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x})
6447 Write the assembly code to define section anchor @var{x}, which is a
6448 @code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true.
6449 The hook is called with the assembly output position set to the beginning
6450 of @code{SYMBOL_REF_BLOCK (@var{x})}.
6452 If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses
6453 it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}.
6454 If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition
6455 is @code{NULL}, which disables the use of section anchors altogether.
6458 @deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x})
6459 Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF}
6460 @var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and
6461 @samp{!SYMBOL_REF_ANCHOR_P (@var{x})}.
6463 The default version is correct for most targets, but you might need to
6464 intercept this hook to handle things like target-specific attributes
6465 or target-specific sections.
6468 @node Condition Code
6469 @section Condition Code Status
6470 @cindex condition code status
6472 Condition codes in GCC are represented as registers,
6473 which provides better schedulability for
6474 architectures that do have a condition code register, but on which
6475 most instructions do not affect it. The latter category includes
6478 Implicit clobbering would pose a strong restriction on the placement of
6479 the definition and use of the condition code. In the past the definition
6480 and use were always adjacent. However, recent changes to support trapping
6481 arithmetic may result in the definition and user being in different blocks.
6482 Thus, there may be a @code{NOTE_INSN_BASIC_BLOCK} between them. Additionally,
6483 the definition may be the source of exception handling edges.
6485 These restrictions can prevent important
6486 optimizations on some machines. For example, on the IBM RS/6000, there
6487 is a delay for taken branches unless the condition code register is set
6488 three instructions earlier than the conditional branch. The instruction
6489 scheduler cannot perform this optimization if it is not permitted to
6490 separate the definition and use of the condition code register.
6492 If there is a specific
6493 condition code register in the machine, use a hard register. If the
6494 condition code or comparison result can be placed in any general register,
6495 or if there are multiple condition registers, use a pseudo register.
6496 Registers used to store the condition code value will usually have a mode
6497 that is in class @code{MODE_CC}.
6499 Alternatively, you can use @code{BImode} if the comparison operator is
6500 specified already in the compare instruction. In this case, you are not
6501 interested in most macros in this section.
6504 * MODE_CC Condition Codes:: Modern representation of condition codes.
6507 @node MODE_CC Condition Codes
6508 @subsection Representation of condition codes using registers
6512 @defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y})
6513 On many machines, the condition code may be produced by other instructions
6514 than compares, for example the branch can use directly the condition
6515 code set by a subtract instruction. However, on some machines
6516 when the condition code is set this way some bits (such as the overflow
6517 bit) are not set in the same way as a test instruction, so that a different
6518 branch instruction must be used for some conditional branches. When
6519 this happens, use the machine mode of the condition code register to
6520 record different formats of the condition code register. Modes can
6521 also be used to record which compare instruction (e.g.@: a signed or an
6522 unsigned comparison) produced the condition codes.
6524 If other modes than @code{CCmode} are required, add them to
6525 @file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose
6526 a mode given an operand of a compare. This is needed because the modes
6527 have to be chosen not only during RTL generation but also, for example,
6528 by instruction combination. The result of @code{SELECT_CC_MODE} should
6529 be consistent with the mode used in the patterns; for example to support
6530 the case of the add on the SPARC discussed above, we have the pattern
6536 (plus:SI (match_operand:SI 0 "register_operand" "%r")
6537 (match_operand:SI 1 "arith_operand" "rI"))
6544 together with a @code{SELECT_CC_MODE} that returns @code{CCNZmode}
6545 for comparisons whose argument is a @code{plus}:
6548 #define SELECT_CC_MODE(OP,X,Y) \
6549 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \
6550 ? ((OP == LT || OP == LE || OP == GT || OP == GE) \
6551 ? CCFPEmode : CCFPmode) \
6552 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \
6553 || GET_CODE (X) == NEG || GET_CODE (x) == ASHIFT) \
6554 ? CCNZmode : CCmode))
6557 Another reason to use modes is to retain information on which operands
6558 were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in
6561 You should define this macro if and only if you define extra CC modes
6562 in @file{@var{machine}-modes.def}.
6565 @deftypefn {Target Hook} void TARGET_CANONICALIZE_COMPARISON (int *@var{code}, rtx *@var{op0}, rtx *@var{op1}, bool @var{op0_preserve_value})
6566 On some machines not all possible comparisons are defined, but you can
6567 convert an invalid comparison into a valid one. For example, the Alpha
6568 does not have a @code{GT} comparison, but you can use an @code{LT}
6569 comparison instead and swap the order of the operands.
6571 On such machines, implement this hook to do any required conversions.
6572 @var{code} is the initial comparison code and @var{op0} and @var{op1}
6573 are the left and right operands of the comparison, respectively. If
6574 @var{op0_preserve_value} is @code{true} the implementation is not
6575 allowed to change the value of @var{op0} since the value might be used
6576 in RTXs which aren't comparisons. E.g. the implementation is not
6577 allowed to swap operands in that case.
6579 GCC will not assume that the comparison resulting from this macro is
6580 valid but will see if the resulting insn matches a pattern in the
6583 You need not to implement this hook if it would never change the
6584 comparison code or operands.
6587 @defmac REVERSIBLE_CC_MODE (@var{mode})
6588 A C expression whose value is one if it is always safe to reverse a
6589 comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE}
6590 can ever return @var{mode} for a floating-point inequality comparison,
6591 then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero.
6593 You need not define this macro if it would always returns zero or if the
6594 floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}.
6595 For example, here is the definition used on the SPARC, where floating-point
6596 inequality comparisons are given either @code{CCFPEmode} or @code{CCFPmode}:
6599 #define REVERSIBLE_CC_MODE(MODE) \
6600 ((MODE) != CCFPEmode && (MODE) != CCFPmode)
6604 @defmac REVERSE_CONDITION (@var{code}, @var{mode})
6605 A C expression whose value is reversed condition code of the @var{code} for
6606 comparison done in CC_MODE @var{mode}. The macro is used only in case
6607 @code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case
6608 machine has some non-standard way how to reverse certain conditionals. For
6609 instance in case all floating point conditions are non-trapping, compiler may
6610 freely convert unordered compares to ordered ones. Then definition may look
6614 #define REVERSE_CONDITION(CODE, MODE) \
6615 ((MODE) != CCFPmode ? reverse_condition (CODE) \
6616 : reverse_condition_maybe_unordered (CODE))
6620 @deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2})
6621 On targets which use a hard
6622 register rather than a pseudo-register to hold condition codes, the
6623 regular CSE passes are often not able to identify cases in which the
6624 hard register is set to a common value. Use this hook to enable a
6625 small pass which optimizes such cases. This hook should return true
6626 to enable this pass, and it should set the integers to which its
6627 arguments point to the hard register numbers used for condition codes.
6628 When there is only one such register, as is true on most systems, the
6629 integer pointed to by @var{p2} should be set to
6630 @code{INVALID_REGNUM}.
6632 The default version of this hook returns false.
6635 @deftypefn {Target Hook} machine_mode TARGET_CC_MODES_COMPATIBLE (machine_mode @var{m1}, machine_mode @var{m2})
6636 On targets which use multiple condition code modes in class
6637 @code{MODE_CC}, it is sometimes the case that a comparison can be
6638 validly done in more than one mode. On such a system, define this
6639 target hook to take two mode arguments and to return a mode in which
6640 both comparisons may be validly done. If there is no such mode,
6641 return @code{VOIDmode}.
6643 The default version of this hook checks whether the modes are the
6644 same. If they are, it returns that mode. If they are different, it
6645 returns @code{VOIDmode}.
6648 @deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM
6649 If the target has a dedicated flags register, and it needs to use the
6650 post-reload comparison elimination pass, or the delay slot filler pass,
6651 then this value should be set appropriately.
6655 @section Describing Relative Costs of Operations
6656 @cindex costs of instructions
6657 @cindex relative costs
6658 @cindex speed of instructions
6660 These macros let you describe the relative speed of various operations
6661 on the target machine.
6663 @defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to})
6664 A C expression for the cost of moving data of mode @var{mode} from a
6665 register in class @var{from} to one in class @var{to}. The classes are
6666 expressed using the enumeration values such as @code{GENERAL_REGS}. A
6667 value of 2 is the default; other values are interpreted relative to
6670 It is not required that the cost always equal 2 when @var{from} is the
6671 same as @var{to}; on some machines it is expensive to move between
6672 registers if they are not general registers.
6674 If reload sees an insn consisting of a single @code{set} between two
6675 hard registers, and if @code{REGISTER_MOVE_COST} applied to their
6676 classes returns a value of 2, reload does not check to ensure that the
6677 constraints of the insn are met. Setting a cost of other than 2 will
6678 allow reload to verify that the constraints are met. You should do this
6679 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6681 These macros are obsolete, new ports should use the target hook
6682 @code{TARGET_REGISTER_MOVE_COST} instead.
6685 @deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to})
6686 This target hook should return the cost of moving data of mode @var{mode}
6687 from a register in class @var{from} to one in class @var{to}. The classes
6688 are expressed using the enumeration values such as @code{GENERAL_REGS}.
6689 A value of 2 is the default; other values are interpreted relative to
6692 It is not required that the cost always equal 2 when @var{from} is the
6693 same as @var{to}; on some machines it is expensive to move between
6694 registers if they are not general registers.
6696 If reload sees an insn consisting of a single @code{set} between two
6697 hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their
6698 classes returns a value of 2, reload does not check to ensure that the
6699 constraints of the insn are met. Setting a cost of other than 2 will
6700 allow reload to verify that the constraints are met. You should do this
6701 if the @samp{mov@var{m}} pattern's constraints do not allow such copying.
6703 The default version of this function returns 2.
6706 @defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in})
6707 A C expression for the cost of moving data of mode @var{mode} between a
6708 register of class @var{class} and memory; @var{in} is zero if the value
6709 is to be written to memory, nonzero if it is to be read in. This cost
6710 is relative to those in @code{REGISTER_MOVE_COST}. If moving between
6711 registers and memory is more expensive than between two registers, you
6712 should define this macro to express the relative cost.
6714 If you do not define this macro, GCC uses a default cost of 4 plus
6715 the cost of copying via a secondary reload register, if one is
6716 needed. If your machine requires a secondary reload register to copy
6717 between memory and a register of @var{class} but the reload mechanism is
6718 more complex than copying via an intermediate, define this macro to
6719 reflect the actual cost of the move.
6721 GCC defines the function @code{memory_move_secondary_cost} if
6722 secondary reloads are needed. It computes the costs due to copying via
6723 a secondary register. If your machine copies from memory using a
6724 secondary register in the conventional way but the default base value of
6725 4 is not correct for your machine, define this macro to add some other
6726 value to the result of that function. The arguments to that function
6727 are the same as to this macro.
6729 These macros are obsolete, new ports should use the target hook
6730 @code{TARGET_MEMORY_MOVE_COST} instead.
6733 @deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in})
6734 This target hook should return the cost of moving data of mode @var{mode}
6735 between a register of class @var{rclass} and memory; @var{in} is @code{false}
6736 if the value is to be written to memory, @code{true} if it is to be read in.
6737 This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}.
6738 If moving between registers and memory is more expensive than between two
6739 registers, you should add this target hook to express the relative cost.
6741 If you do not add this target hook, GCC uses a default cost of 4 plus
6742 the cost of copying via a secondary reload register, if one is
6743 needed. If your machine requires a secondary reload register to copy
6744 between memory and a register of @var{rclass} but the reload mechanism is
6745 more complex than copying via an intermediate, use this target hook to
6746 reflect the actual cost of the move.
6748 GCC defines the function @code{memory_move_secondary_cost} if
6749 secondary reloads are needed. It computes the costs due to copying via
6750 a secondary register. If your machine copies from memory using a
6751 secondary register in the conventional way but the default base value of
6752 4 is not correct for your machine, use this target hook to add some other
6753 value to the result of that function. The arguments to that function
6754 are the same as to this target hook.
6757 @defmac BRANCH_COST (@var{speed_p}, @var{predictable_p})
6758 A C expression for the cost of a branch instruction. A value of 1 is
6759 the default; other values are interpreted relative to that. Parameter
6760 @var{speed_p} is true when the branch in question should be optimized
6761 for speed. When it is false, @code{BRANCH_COST} should return a value
6762 optimal for code size rather than performance. @var{predictable_p} is
6763 true for well-predicted branches. On many architectures the
6764 @code{BRANCH_COST} can be reduced then.
6767 Here are additional macros which do not specify precise relative costs,
6768 but only that certain actions are more expensive than GCC would
6771 @defmac SLOW_BYTE_ACCESS
6772 Define this macro as a C expression which is nonzero if accessing less
6773 than a word of memory (i.e.@: a @code{char} or a @code{short}) is no
6774 faster than accessing a word of memory, i.e., if such access
6775 require more than one instruction or if there is no difference in cost
6776 between byte and (aligned) word loads.
6778 When this macro is not defined, the compiler will access a field by
6779 finding the smallest containing object; when it is defined, a fullword
6780 load will be used if alignment permits. Unless bytes accesses are
6781 faster than word accesses, using word accesses is preferable since it
6782 may eliminate subsequent memory access if subsequent accesses occur to
6783 other fields in the same word of the structure, but to different bytes.
6786 @deftypefn {Target Hook} bool TARGET_SLOW_UNALIGNED_ACCESS (machine_mode @var{mode}, unsigned int @var{align})
6787 This hook returns true if memory accesses described by the
6788 @var{mode} and @var{alignment} parameters have a cost many times greater
6789 than aligned accesses, for example if they are emulated in a trap handler.
6790 This hook is invoked only for unaligned accesses, i.e.@: when
6791 @code{@var{alignment} < GET_MODE_ALIGNMENT (@var{mode})}.
6793 When this hook returns true, the compiler will act as if
6794 @code{STRICT_ALIGNMENT} were true when generating code for block
6795 moves. This can cause significantly more instructions to be produced.
6796 Therefore, do not make this hook return true if unaligned accesses only
6797 add a cycle or two to the time for a memory access.
6799 The hook must return true whenever @code{STRICT_ALIGNMENT} is true.
6800 The default implementation returns @code{STRICT_ALIGNMENT}.
6803 @defmac MOVE_RATIO (@var{speed})
6804 The threshold of number of scalar memory-to-memory move insns, @emph{below}
6805 which a sequence of insns should be generated instead of a
6806 string move insn or a library call. Increasing the value will always
6807 make code faster, but eventually incurs high cost in increased code size.
6809 Note that on machines where the corresponding move insn is a
6810 @code{define_expand} that emits a sequence of insns, this macro counts
6811 the number of such sequences.
6813 The parameter @var{speed} is true if the code is currently being
6814 optimized for speed rather than size.
6816 If you don't define this, a reasonable default is used.
6819 @deftypefn {Target Hook} bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned HOST_WIDE_INT @var{size}, unsigned int @var{alignment}, enum by_pieces_operation @var{op}, bool @var{speed_p})
6820 GCC will attempt several strategies when asked to copy between
6821 two areas of memory, or to set, clear or store to memory, for example
6822 when copying a @code{struct}. The @code{by_pieces} infrastructure
6823 implements such memory operations as a sequence of load, store or move
6824 insns. Alternate strategies are to expand the
6825 @code{cpymem} or @code{setmem} optabs, to emit a library call, or to emit
6826 unit-by-unit, loop-based operations.
6828 This target hook should return true if, for a memory operation with a
6829 given @var{size} and @var{alignment}, using the @code{by_pieces}
6830 infrastructure is expected to result in better code generation.
6831 Both @var{size} and @var{alignment} are measured in terms of storage
6834 The parameter @var{op} is one of: @code{CLEAR_BY_PIECES},
6835 @code{MOVE_BY_PIECES}, @code{SET_BY_PIECES}, @code{STORE_BY_PIECES} or
6836 @code{COMPARE_BY_PIECES}. These describe the type of memory operation
6837 under consideration.
6839 The parameter @var{speed_p} is true if the code is currently being
6840 optimized for speed rather than size.
6842 Returning true for higher values of @var{size} can improve code generation
6843 for speed if the target does not provide an implementation of the
6844 @code{cpymem} or @code{setmem} standard names, if the @code{cpymem} or
6845 @code{setmem} implementation would be more expensive than a sequence of
6846 insns, or if the overhead of a library call would dominate that of
6847 the body of the memory operation.
6849 Returning true for higher values of @code{size} may also cause an increase
6850 in code size, for example where the number of insns emitted to perform a
6851 move would be greater than that of a library call.
6854 @deftypefn {Target Hook} bool TARGET_OVERLAP_OP_BY_PIECES_P (void)
6855 This target hook should return true if when the @code{by_pieces}
6856 infrastructure is used, an offset adjusted unaligned memory operation
6857 in the smallest integer mode for the last piece operation of a memory
6858 region can be generated to avoid doing more than one smaller operations.
6861 @deftypefn {Target Hook} int TARGET_COMPARE_BY_PIECES_BRANCH_RATIO (machine_mode @var{mode})
6862 When expanding a block comparison in MODE, gcc can try to reduce the
6863 number of branches at the expense of more memory operations. This hook
6864 allows the target to override the default choice. It should return the
6865 factor by which branches should be reduced over the plain expansion with
6866 one comparison per @var{mode}-sized piece. A port can also prevent a
6867 particular mode from being used for block comparisons by returning a
6868 negative number from this hook.
6871 @defmac MOVE_MAX_PIECES
6872 A C expression used by @code{move_by_pieces} to determine the largest unit
6873 a load or store used to copy memory is. Defaults to @code{MOVE_MAX}.
6876 @defmac STORE_MAX_PIECES
6877 A C expression used by @code{store_by_pieces} to determine the largest unit
6878 a store used to memory is. Defaults to @code{MOVE_MAX_PIECES}, or two times
6879 the size of @code{HOST_WIDE_INT}, whichever is smaller.
6882 @defmac COMPARE_MAX_PIECES
6883 A C expression used by @code{compare_by_pieces} to determine the largest unit
6884 a load or store used to compare memory is. Defaults to
6885 @code{MOVE_MAX_PIECES}.
6888 @defmac CLEAR_RATIO (@var{speed})
6889 The threshold of number of scalar move insns, @emph{below} which a sequence
6890 of insns should be generated to clear memory instead of a string clear insn
6891 or a library call. Increasing the value will always make code faster, but
6892 eventually incurs high cost in increased code size.
6894 The parameter @var{speed} is true if the code is currently being
6895 optimized for speed rather than size.
6897 If you don't define this, a reasonable default is used.
6900 @defmac SET_RATIO (@var{speed})
6901 The threshold of number of scalar move insns, @emph{below} which a sequence
6902 of insns should be generated to set memory to a constant value, instead of
6903 a block set insn or a library call.
6904 Increasing the value will always make code faster, but
6905 eventually incurs high cost in increased code size.
6907 The parameter @var{speed} is true if the code is currently being
6908 optimized for speed rather than size.
6910 If you don't define this, it defaults to the value of @code{MOVE_RATIO}.
6913 @defmac USE_LOAD_POST_INCREMENT (@var{mode})
6914 A C expression used to determine whether a load postincrement is a good
6915 thing to use for a given mode. Defaults to the value of
6916 @code{HAVE_POST_INCREMENT}.
6919 @defmac USE_LOAD_POST_DECREMENT (@var{mode})
6920 A C expression used to determine whether a load postdecrement is a good
6921 thing to use for a given mode. Defaults to the value of
6922 @code{HAVE_POST_DECREMENT}.
6925 @defmac USE_LOAD_PRE_INCREMENT (@var{mode})
6926 A C expression used to determine whether a load preincrement is a good
6927 thing to use for a given mode. Defaults to the value of
6928 @code{HAVE_PRE_INCREMENT}.
6931 @defmac USE_LOAD_PRE_DECREMENT (@var{mode})
6932 A C expression used to determine whether a load predecrement is a good
6933 thing to use for a given mode. Defaults to the value of
6934 @code{HAVE_PRE_DECREMENT}.
6937 @defmac USE_STORE_POST_INCREMENT (@var{mode})
6938 A C expression used to determine whether a store postincrement is a good
6939 thing to use for a given mode. Defaults to the value of
6940 @code{HAVE_POST_INCREMENT}.
6943 @defmac USE_STORE_POST_DECREMENT (@var{mode})
6944 A C expression used to determine whether a store postdecrement is a good
6945 thing to use for a given mode. Defaults to the value of
6946 @code{HAVE_POST_DECREMENT}.
6949 @defmac USE_STORE_PRE_INCREMENT (@var{mode})
6950 This macro is used to determine whether a store preincrement is a good
6951 thing to use for a given mode. Defaults to the value of
6952 @code{HAVE_PRE_INCREMENT}.
6955 @defmac USE_STORE_PRE_DECREMENT (@var{mode})
6956 This macro is used to determine whether a store predecrement is a good
6957 thing to use for a given mode. Defaults to the value of
6958 @code{HAVE_PRE_DECREMENT}.
6961 @defmac NO_FUNCTION_CSE
6962 Define this macro to be true if it is as good or better to call a constant
6963 function address than to call an address kept in a register.
6966 @defmac LOGICAL_OP_NON_SHORT_CIRCUIT
6967 Define this macro if a non-short-circuit operation produced by
6968 @samp{fold_range_test ()} is optimal. This macro defaults to true if
6969 @code{BRANCH_COST} is greater than or equal to the value 2.
6972 @deftypefn {Target Hook} bool TARGET_OPTAB_SUPPORTED_P (int @var{op}, machine_mode @var{mode1}, machine_mode @var{mode2}, optimization_type @var{opt_type})
6973 Return true if the optimizers should use optab @var{op} with
6974 modes @var{mode1} and @var{mode2} for optimization type @var{opt_type}.
6975 The optab is known to have an associated @file{.md} instruction
6976 whose C condition is true. @var{mode2} is only meaningful for conversion
6977 optabs; for direct optabs it is a copy of @var{mode1}.
6979 For example, when called with @var{op} equal to @code{rint_optab} and
6980 @var{mode1} equal to @code{DFmode}, the hook should say whether the
6981 optimizers should use optab @code{rintdf2}.
6983 The default hook returns true for all inputs.
6986 @deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, machine_mode @var{mode}, int @var{outer_code}, int @var{opno}, int *@var{total}, bool @var{speed})
6987 This target hook describes the relative costs of RTL expressions.
6989 The cost may depend on the precise form of the expression, which is
6990 available for examination in @var{x}, and the fact that @var{x} appears
6991 as operand @var{opno} of an expression with rtx code @var{outer_code}.
6992 That is, the hook can assume that there is some rtx @var{y} such
6993 that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that
6994 either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or
6995 (b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}.
6997 @var{mode} is @var{x}'s machine mode, or for cases like @code{const_int} that
6998 do not have a mode, the mode in which @var{x} is used.
7000 In implementing this hook, you can use the construct
7001 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
7004 On entry to the hook, @code{*@var{total}} contains a default estimate
7005 for the cost of the expression. The hook should modify this value as
7006 necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)}
7007 for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus
7008 operations, and @code{COSTS_N_INSNS (1)} for all other operations.
7010 When optimizing for code size, i.e.@: when @code{speed} is
7011 false, this target hook should be used to estimate the relative
7012 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
7014 The hook returns true when all subexpressions of @var{x} have been
7015 processed, and false when @code{rtx_cost} should recurse.
7018 @deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, machine_mode @var{mode}, addr_space_t @var{as}, bool @var{speed})
7019 This hook computes the cost of an addressing mode that contains
7020 @var{address}. If not defined, the cost is computed from
7021 the @var{address} expression and the @code{TARGET_RTX_COST} hook.
7023 For most CISC machines, the default cost is a good approximation of the
7024 true cost of the addressing mode. However, on RISC machines, all
7025 instructions normally have the same length and execution time. Hence
7026 all addresses will have equal costs.
7028 In cases where more than one form of an address is known, the form with
7029 the lowest cost will be used. If multiple forms have the same, lowest,
7030 cost, the one that is the most complex will be used.
7032 For example, suppose an address that is equal to the sum of a register
7033 and a constant is used twice in the same basic block. When this macro
7034 is not defined, the address will be computed in a register and memory
7035 references will be indirect through that register. On machines where
7036 the cost of the addressing mode containing the sum is no higher than
7037 that of a simple indirect reference, this will produce an additional
7038 instruction and possibly require an additional register. Proper
7039 specification of this macro eliminates this overhead for such machines.
7041 This hook is never called with an invalid address.
7043 On machines where an address involving more than one register is as
7044 cheap as an address computation involving only one register, defining
7045 @code{TARGET_ADDRESS_COST} to reflect this can cause two registers to
7046 be live over a region of code where only one would have been if
7047 @code{TARGET_ADDRESS_COST} were not defined in that manner. This effect
7048 should be considered in the definition of this macro. Equivalent costs
7049 should probably only be given to addresses with different numbers of
7050 registers on machines with lots of registers.
7053 @deftypefn {Target Hook} int TARGET_INSN_COST (rtx_insn *@var{insn}, bool @var{speed})
7054 This target hook describes the relative costs of RTL instructions.
7056 In implementing this hook, you can use the construct
7057 @code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast
7060 When optimizing for code size, i.e.@: when @code{speed} is
7061 false, this target hook should be used to estimate the relative
7062 size cost of an expression, again relative to @code{COSTS_N_INSNS}.
7065 @deftypefn {Target Hook} {unsigned int} TARGET_MAX_NOCE_IFCVT_SEQ_COST (edge @var{e})
7066 This hook returns a value in the same units as @code{TARGET_RTX_COSTS},
7067 giving the maximum acceptable cost for a sequence generated by the RTL
7068 if-conversion pass when conditional execution is not available.
7069 The RTL if-conversion pass attempts to convert conditional operations
7070 that would require a branch to a series of unconditional operations and
7071 @code{mov@var{mode}cc} insns. This hook returns the maximum cost of the
7072 unconditional instructions and the @code{mov@var{mode}cc} insns.
7073 RTL if-conversion is cancelled if the cost of the converted sequence
7074 is greater than the value returned by this hook.
7076 @code{e} is the edge between the basic block containing the conditional
7077 branch to the basic block which would be executed if the condition
7080 The default implementation of this hook uses the
7081 @code{max-rtl-if-conversion-[un]predictable} parameters if they are set,
7082 and uses a multiple of @code{BRANCH_COST} otherwise.
7085 @deftypefn {Target Hook} bool TARGET_NOCE_CONVERSION_PROFITABLE_P (rtx_insn *@var{seq}, struct noce_if_info *@var{if_info})
7086 This hook returns true if the instruction sequence @code{seq} is a good
7087 candidate as a replacement for the if-convertible sequence described in
7091 @deftypefn {Target Hook} bool TARGET_NEW_ADDRESS_PROFITABLE_P (rtx @var{memref}, rtx_insn * @var{insn}, rtx @var{new_addr})
7092 Return @code{true} if it is profitable to replace the address in
7093 @var{memref} with @var{new_addr}. This allows targets to prevent the
7094 scheduler from undoing address optimizations. The instruction containing the
7095 memref is @var{insn}. The default implementation returns @code{true}.
7098 @deftypefn {Target Hook} bool TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P (void)
7099 This predicate controls the use of the eager delay slot filler to disallow
7100 speculatively executed instructions being placed in delay slots. Targets
7101 such as certain MIPS architectures possess both branches with and without
7102 delay slots. As the eager delay slot filler can decrease performance,
7103 disabling it is beneficial when ordinary branches are available. Use of
7104 delay slot branches filled using the basic filler is often still desirable
7105 as the delay slot can hide a pipeline bubble.
7108 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_ESTIMATED_POLY_VALUE (poly_int64 @var{val}, poly_value_estimate_kind @var{kind})
7109 Return an estimate of the runtime value of @var{val}, for use in
7110 things like cost calculations or profiling frequencies. @var{kind} is used
7111 to ask for the minimum, maximum, and likely estimates of the value through
7112 the @code{POLY_VALUE_MIN}, @code{POLY_VALUE_MAX} and
7113 @code{POLY_VALUE_LIKELY} values. The default
7114 implementation returns the lowest possible value of @var{val}.
7118 @section Adjusting the Instruction Scheduler
7120 The instruction scheduler may need a fair amount of machine-specific
7121 adjustment in order to produce good code. GCC provides several target
7122 hooks for this purpose. It is usually enough to define just a few of
7123 them: try the first ones in this list first.
7125 @deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void)
7126 This hook returns the maximum number of instructions that can ever
7127 issue at the same time on the target machine. The default is one.
7128 Although the insn scheduler can define itself the possibility of issue
7129 an insn on the same cycle, the value can serve as an additional
7130 constraint to issue insns on the same simulated processor cycle (see
7131 hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}).
7132 This value must be constant over the entire compilation. If you need
7133 it to vary depending on what the instructions are, you must use
7134 @samp{TARGET_SCHED_VARIABLE_ISSUE}.
7137 @deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx_insn *@var{insn}, int @var{more})
7138 This hook is executed by the scheduler after it has scheduled an insn
7139 from the ready list. It should return the number of insns which can
7140 still be issued in the current cycle. The default is
7141 @samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and
7142 @code{USE}, which normally are not counted against the issue rate.
7143 You should define this hook if some insns take more machine resources
7144 than others, so that fewer insns can follow them in the same cycle.
7145 @var{file} is either a null pointer, or a stdio stream to write any
7146 debug output to. @var{verbose} is the verbose level provided by
7147 @option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that
7151 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx_insn *@var{insn}, int @var{dep_type1}, rtx_insn *@var{dep_insn}, int @var{cost}, unsigned int @var{dw})
7152 This function corrects the value of @var{cost} based on the
7153 relationship between @var{insn} and @var{dep_insn} through a
7154 dependence of type dep_type, and strength @var{dw}. It should return the new
7155 value. The default is to make no adjustment to @var{cost}. This can be
7156 used for example to specify to the scheduler using the traditional pipeline
7157 description that an output- or anti-dependence does not incur the same cost
7158 as a data-dependence. If the scheduler using the automaton based pipeline
7159 description, the cost of anti-dependence is zero and the cost of
7160 output-dependence is maximum of one and the difference of latency
7161 times of the first and the second insns. If these values are not
7162 acceptable, you could use the hook to modify them too. See also
7163 @pxref{Processor pipeline description}.
7166 @deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx_insn *@var{insn}, int @var{priority})
7167 This hook adjusts the integer scheduling priority @var{priority} of
7168 @var{insn}. It should return the new priority. Increase the priority to
7169 execute @var{insn} earlier, reduce the priority to execute @var{insn}
7170 later. Do not define this hook if you do not need to adjust the
7171 scheduling priorities of insns.
7174 @deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock})
7175 This hook is executed by the scheduler after it has scheduled the ready
7176 list, to allow the machine description to reorder it (for example to
7177 combine two small instructions together on @samp{VLIW} machines).
7178 @var{file} is either a null pointer, or a stdio stream to write any
7179 debug output to. @var{verbose} is the verbose level provided by
7180 @option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready
7181 list of instructions that are ready to be scheduled. @var{n_readyp} is
7182 a pointer to the number of elements in the ready list. The scheduler
7183 reads the ready list in reverse order, starting with
7184 @var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock}
7185 is the timer tick of the scheduler. You may modify the ready list and
7186 the number of ready insns. The return value is the number of insns that
7187 can issue this cycle; normally this is just @code{issue_rate}. See also
7188 @samp{TARGET_SCHED_REORDER2}.
7191 @deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock})
7192 Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That
7193 function is called whenever the scheduler starts a new cycle. This one
7194 is called once per iteration over a cycle, immediately after
7195 @samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and
7196 return the number of insns to be scheduled in the same cycle. Defining
7197 this hook can be useful if there are frequent situations where
7198 scheduling one insn causes other insns to become ready in the same
7199 cycle. These other insns can then be taken into account properly.
7202 @deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_P (void)
7203 This hook is used to check whether target platform supports macro fusion.
7206 @deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_PAIR_P (rtx_insn *@var{prev}, rtx_insn *@var{curr})
7207 This hook is used to check whether two insns should be macro fused for
7208 a target microarchitecture. If this hook returns true for the given insn pair
7209 (@var{prev} and @var{curr}), the scheduler will put them into a sched
7210 group, and they will not be scheduled apart. The two insns will be either
7211 two SET insns or a compare and a conditional jump and this hook should
7212 validate any dependencies needed to fuse the two insns together.
7215 @deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx_insn *@var{head}, rtx_insn *@var{tail})
7216 This hook is called after evaluation forward dependencies of insns in
7217 chain given by two parameter values (@var{head} and @var{tail}
7218 correspondingly) but before insns scheduling of the insn chain. For
7219 example, it can be used for better insn classification if it requires
7220 analysis of dependencies. This hook can use backward and forward
7221 dependencies of the insn scheduler because they are already
7225 @deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready})
7226 This hook is executed by the scheduler at the beginning of each block of
7227 instructions that are to be scheduled. @var{file} is either a null
7228 pointer, or a stdio stream to write any debug output to. @var{verbose}
7229 is the verbose level provided by @option{-fsched-verbose-@var{n}}.
7230 @var{max_ready} is the maximum number of insns in the current scheduling
7231 region that can be live at the same time. This can be used to allocate
7232 scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}.
7235 @deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose})
7236 This hook is executed by the scheduler at the end of each block of
7237 instructions that are to be scheduled. It can be used to perform
7238 cleanup of any actions done by the other scheduling hooks. @var{file}
7239 is either a null pointer, or a stdio stream to write any debug output
7240 to. @var{verbose} is the verbose level provided by
7241 @option{-fsched-verbose-@var{n}}.
7244 @deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid})
7245 This hook is executed by the scheduler after function level initializations.
7246 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
7247 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
7248 @var{old_max_uid} is the maximum insn uid when scheduling begins.
7251 @deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose})
7252 This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}.
7253 @var{file} is either a null pointer, or a stdio stream to write any debug output to.
7254 @var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}.
7257 @deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void)
7258 The hook returns an RTL insn. The automaton state used in the
7259 pipeline hazard recognizer is changed as if the insn were scheduled
7260 when the new simulated processor cycle starts. Usage of the hook may
7261 simplify the automaton pipeline description for some @acronym{VLIW}
7262 processors. If the hook is defined, it is used only for the automaton
7263 based pipeline description. The default is not to change the state
7264 when the new simulated processor cycle starts.
7267 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void)
7268 The hook can be used to initialize data used by the previous hook.
7271 @deftypefn {Target Hook} {rtx_insn *} TARGET_SCHED_DFA_POST_CYCLE_INSN (void)
7272 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
7273 to changed the state as if the insn were scheduled when the new
7274 simulated processor cycle finishes.
7277 @deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void)
7278 The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but
7279 used to initialize data used by the previous hook.
7282 @deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void)
7283 The hook to notify target that the current simulated cycle is about to finish.
7284 The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used
7285 to change the state in more complicated situations - e.g., when advancing
7286 state on a single insn is not enough.
7289 @deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void)
7290 The hook to notify target that new simulated cycle has just started.
7291 The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used
7292 to change the state in more complicated situations - e.g., when advancing
7293 state on a single insn is not enough.
7296 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void)
7297 This hook controls better choosing an insn from the ready insn queue
7298 for the @acronym{DFA}-based insn scheduler. Usually the scheduler
7299 chooses the first insn from the queue. If the hook returns a positive
7300 value, an additional scheduler code tries all permutations of
7301 @samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()}
7302 subsequent ready insns to choose an insn whose issue will result in
7303 maximal number of issued insns on the same cycle. For the
7304 @acronym{VLIW} processor, the code could actually solve the problem of
7305 packing simple insns into the @acronym{VLIW} insn. Of course, if the
7306 rules of @acronym{VLIW} packing are described in the automaton.
7308 This code also could be used for superscalar @acronym{RISC}
7309 processors. Let us consider a superscalar @acronym{RISC} processor
7310 with 3 pipelines. Some insns can be executed in pipelines @var{A} or
7311 @var{B}, some insns can be executed only in pipelines @var{B} or
7312 @var{C}, and one insn can be executed in pipeline @var{B}. The
7313 processor may issue the 1st insn into @var{A} and the 2nd one into
7314 @var{B}. In this case, the 3rd insn will wait for freeing @var{B}
7315 until the next cycle. If the scheduler issues the 3rd insn the first,
7316 the processor could issue all 3 insns per cycle.
7318 Actually this code demonstrates advantages of the automaton based
7319 pipeline hazard recognizer. We try quickly and easy many insn
7320 schedules to choose the best one.
7322 The default is no multipass scheduling.
7325 @deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx_insn *@var{insn}, int @var{ready_index})
7327 This hook controls what insns from the ready insn queue will be
7328 considered for the multipass insn scheduling. If the hook returns
7329 zero for @var{insn}, the insn will be considered in multipass scheduling.
7330 Positive return values will remove @var{insn} from consideration on
7331 the current round of multipass scheduling.
7332 Negative return values will remove @var{insn} from consideration for given
7334 Backends should be careful about returning non-zero for highest priority
7335 instruction at position 0 in the ready list. @var{ready_index} is passed
7336 to allow backends make correct judgements.
7338 The default is that any ready insns can be chosen to be issued.
7341 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, bool @var{first_cycle_insn_p})
7342 This hook prepares the target backend for a new round of multipass
7346 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, rtx_insn *@var{insn}, const void *@var{prev_data})
7347 This hook is called when multipass scheduling evaluates instruction INSN.
7350 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, signed char *@var{ready_try}, int @var{n_ready})
7351 This is called when multipass scheduling backtracks from evaluation of
7355 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data})
7356 This hook notifies the target about the result of the concluded current
7357 round of multipass scheduling.
7360 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data})
7361 This hook initializes target-specific data used in multipass scheduling.
7364 @deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data})
7365 This hook finalizes target-specific data used in multipass scheduling.
7368 @deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *@var{dump}, int @var{verbose}, rtx_insn *@var{insn}, int @var{last_clock}, int @var{clock}, int *@var{sort_p})
7369 This hook is called by the insn scheduler before issuing @var{insn}
7370 on cycle @var{clock}. If the hook returns nonzero,
7371 @var{insn} is not issued on this processor cycle. Instead,
7372 the processor cycle is advanced. If *@var{sort_p}
7373 is zero, the insn ready queue is not sorted on the new cycle
7374 start as usually. @var{dump} and @var{verbose} specify the file and
7375 verbosity level to use for debugging output.
7376 @var{last_clock} and @var{clock} are, respectively, the
7377 processor cycle on which the previous insn has been issued,
7378 and the current processor cycle.
7381 @deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance})
7382 This hook is used to define which dependences are considered costly by
7383 the target, so costly that it is not advisable to schedule the insns that
7384 are involved in the dependence too close to one another. The parameters
7385 to this hook are as follows: The first parameter @var{_dep} is the dependence
7386 being evaluated. The second parameter @var{cost} is the cost of the
7387 dependence as estimated by the scheduler, and the third
7388 parameter @var{distance} is the distance in cycles between the two insns.
7389 The hook returns @code{true} if considering the distance between the two
7390 insns the dependence between them is considered costly by the target,
7391 and @code{false} otherwise.
7393 Defining this hook can be useful in multiple-issue out-of-order machines,
7394 where (a) it's practically hopeless to predict the actual data/resource
7395 delays, however: (b) there's a better chance to predict the actual grouping
7396 that will be formed, and (c) correctly emulating the grouping can be very
7397 important. In such targets one may want to allow issuing dependent insns
7398 closer to one another---i.e., closer than the dependence distance; however,
7399 not in cases of ``costly dependences'', which this hooks allows to define.
7402 @deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void)
7403 This hook is called by the insn scheduler after emitting a new instruction to
7404 the instruction stream. The hook notifies a target backend to extend its
7405 per instruction data structures.
7408 @deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void)
7409 Return a pointer to a store large enough to hold target scheduling context.
7412 @deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p})
7413 Initialize store pointed to by @var{tc} to hold target scheduling context.
7414 It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the
7415 beginning of the block. Otherwise, copy the current context into @var{tc}.
7418 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc})
7419 Copy target scheduling context pointed to by @var{tc} to the current context.
7422 @deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc})
7423 Deallocate internal data in target scheduling context pointed to by @var{tc}.
7426 @deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc})
7427 Deallocate a store for target scheduling context pointed to by @var{tc}.
7430 @deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx_insn *@var{insn}, unsigned int @var{dep_status}, rtx *@var{new_pat})
7431 This hook is called by the insn scheduler when @var{insn} has only
7432 speculative dependencies and therefore can be scheduled speculatively.
7433 The hook is used to check if the pattern of @var{insn} has a speculative
7434 version and, in case of successful check, to generate that speculative
7435 pattern. The hook should return 1, if the instruction has a speculative form,
7436 or @minus{}1, if it doesn't. @var{request} describes the type of requested
7437 speculation. If the return value equals 1 then @var{new_pat} is assigned
7438 the generated speculative pattern.
7441 @deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (unsigned int @var{dep_status})
7442 This hook is called by the insn scheduler during generation of recovery code
7443 for @var{insn}. It should return @code{true}, if the corresponding check
7444 instruction should branch to recovery code, or @code{false} otherwise.
7447 @deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx_insn *@var{insn}, rtx_insn *@var{label}, unsigned int @var{ds})
7448 This hook is called by the insn scheduler to generate a pattern for recovery
7449 check instruction. If @var{mutate_p} is zero, then @var{insn} is a
7450 speculative instruction for which the check should be generated.
7451 @var{label} is either a label of a basic block, where recovery code should
7452 be emitted, or a null pointer, when requested check doesn't branch to
7453 recovery code (a simple check). If @var{mutate_p} is nonzero, then
7454 a pattern for a branchy check corresponding to a simple check denoted by
7455 @var{insn} should be generated. In this case @var{label} can't be null.
7458 @deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info})
7459 This hook is used by the insn scheduler to find out what features should be
7461 The structure *@var{spec_info} should be filled in by the target.
7462 The structure describes speculation types that can be used in the scheduler.
7465 @deftypefn {Target Hook} bool TARGET_SCHED_CAN_SPECULATE_INSN (rtx_insn *@var{insn})
7466 Some instructions should never be speculated by the schedulers, usually
7467 because the instruction is too expensive to get this wrong. Often such
7468 instructions have long latency, and often they are not fully modeled in the
7469 pipeline descriptions. This hook should return @code{false} if @var{insn}
7470 should not be speculated.
7473 @deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g})
7474 This hook is called by the swing modulo scheduler to calculate a
7475 resource-based lower bound which is based on the resources available in
7476 the machine and the resources required by each instruction. The target
7477 backend can use @var{g} to calculate such bound. A very simple lower
7478 bound will be used in case this hook is not implemented: the total number
7479 of instructions divided by the issue rate.
7482 @deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx_insn *@var{insn}, int @var{x})
7483 This hook is called by Haifa Scheduler. It returns true if dispatch scheduling
7484 is supported in hardware and the condition specified in the parameter is true.
7487 @deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx_insn *@var{insn}, int @var{x})
7488 This hook is called by Haifa Scheduler. It performs the operation specified
7489 in its second parameter.
7492 @deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE
7493 True if the processor has an exposed pipeline, which means that not just
7494 the order of instructions is important for correctness when scheduling, but
7495 also the latencies of operations.
7498 @deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, machine_mode @var{mode})
7499 This hook is called by tree reassociator to determine a level of
7500 parallelism required in output calculations chain.
7503 @deftypefn {Target Hook} void TARGET_SCHED_FUSION_PRIORITY (rtx_insn *@var{insn}, int @var{max_pri}, int *@var{fusion_pri}, int *@var{pri})
7504 This hook is called by scheduling fusion pass. It calculates fusion
7505 priorities for each instruction passed in by parameter. The priorities
7506 are returned via pointer parameters.
7508 @var{insn} is the instruction whose priorities need to be calculated.
7509 @var{max_pri} is the maximum priority can be returned in any cases.
7510 @var{fusion_pri} is the pointer parameter through which @var{insn}'s
7511 fusion priority should be calculated and returned.
7512 @var{pri} is the pointer parameter through which @var{insn}'s priority
7513 should be calculated and returned.
7515 Same @var{fusion_pri} should be returned for instructions which should
7516 be scheduled together. Different @var{pri} should be returned for
7517 instructions with same @var{fusion_pri}. @var{fusion_pri} is the major
7518 sort key, @var{pri} is the minor sort key. All instructions will be
7519 scheduled according to the two priorities. All priorities calculated
7520 should be between 0 (exclusive) and @var{max_pri} (inclusive). To avoid
7521 false dependencies, @var{fusion_pri} of instructions which need to be
7522 scheduled together should be smaller than @var{fusion_pri} of irrelevant
7525 Given below example:
7538 On targets like ARM/AArch64, the two pairs of consecutive loads should be
7539 merged. Since peephole2 pass can't help in this case unless consecutive
7540 loads are actually next to each other in instruction flow. That's where
7541 this scheduling fusion pass works. This hook calculates priority for each
7542 instruction based on its fustion type, like:
7545 ldr r10, [r1, 4] ; fusion_pri=99, pri=96
7546 add r4, r4, r10 ; fusion_pri=100, pri=100
7547 ldr r15, [r2, 8] ; fusion_pri=98, pri=92
7548 sub r5, r5, r15 ; fusion_pri=100, pri=100
7549 ldr r11, [r1, 0] ; fusion_pri=99, pri=100
7550 add r4, r4, r11 ; fusion_pri=100, pri=100
7551 ldr r16, [r2, 12] ; fusion_pri=98, pri=88
7552 sub r5, r5, r16 ; fusion_pri=100, pri=100
7555 Scheduling fusion pass then sorts all ready to issue instructions according
7556 to the priorities. As a result, instructions of same fusion type will be
7557 pushed together in instruction flow, like:
7570 Now peephole2 pass can simply merge the two pairs of loads.
7572 Since scheduling fusion pass relies on peephole2 to do real fusion
7573 work, it is only enabled by default when peephole2 is in effect.
7575 This is firstly introduced on ARM/AArch64 targets, please refer to
7576 the hook implementation for how different fusion types are supported.
7579 @deftypefn {Target Hook} void TARGET_EXPAND_DIVMOD_LIBFUNC (rtx @var{libfunc}, machine_mode @var{mode}, rtx @var{op0}, rtx @var{op1}, rtx *@var{quot}, rtx *@var{rem})
7580 Define this hook for enabling divmod transform if the port does not have
7581 hardware divmod insn but defines target-specific divmod libfuncs.
7585 @section Dividing the Output into Sections (Texts, Data, @dots{})
7586 @c the above section title is WAY too long. maybe cut the part between
7587 @c the (...)? --mew 10feb93
7589 An object file is divided into sections containing different types of
7590 data. In the most common case, there are three sections: the @dfn{text
7591 section}, which holds instructions and read-only data; the @dfn{data
7592 section}, which holds initialized writable data; and the @dfn{bss
7593 section}, which holds uninitialized data. Some systems have other kinds
7596 @file{varasm.cc} provides several well-known sections, such as
7597 @code{text_section}, @code{data_section} and @code{bss_section}.
7598 The normal way of controlling a @code{@var{foo}_section} variable
7599 is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro,
7600 as described below. The macros are only read once, when @file{varasm.cc}
7601 initializes itself, so their values must be run-time constants.
7602 They may however depend on command-line flags.
7604 @emph{Note:} Some run-time files, such @file{crtstuff.c}, also make
7605 use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them
7606 to be string literals.
7608 Some assemblers require a different string to be written every time a
7609 section is selected. If your assembler falls into this category, you
7610 should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use
7611 @code{get_unnamed_section} to set up the sections.
7613 You must always create a @code{text_section}, either by defining
7614 @code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section}
7615 in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of
7616 @code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not
7617 create a distinct @code{readonly_data_section}, the default is to
7618 reuse @code{text_section}.
7620 All the other @file{varasm.cc} sections are optional, and are null
7621 if the target does not provide them.
7623 @defmac TEXT_SECTION_ASM_OP
7624 A C expression whose value is a string, including spacing, containing the
7625 assembler operation that should precede instructions and read-only data.
7626 Normally @code{"\t.text"} is right.
7629 @defmac HOT_TEXT_SECTION_NAME
7630 If defined, a C string constant for the name of the section containing most
7631 frequently executed functions of the program. If not defined, GCC will provide
7632 a default definition if the target supports named sections.
7635 @defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME
7636 If defined, a C string constant for the name of the section containing unlikely
7637 executed functions in the program.
7640 @defmac DATA_SECTION_ASM_OP
7641 A C expression whose value is a string, including spacing, containing the
7642 assembler operation to identify the following data as writable initialized
7643 data. Normally @code{"\t.data"} is right.
7646 @defmac SDATA_SECTION_ASM_OP
7647 If defined, a C expression whose value is a string, including spacing,
7648 containing the assembler operation to identify the following data as
7649 initialized, writable small data.
7652 @defmac READONLY_DATA_SECTION_ASM_OP
7653 A C expression whose value is a string, including spacing, containing the
7654 assembler operation to identify the following data as read-only initialized
7658 @defmac BSS_SECTION_ASM_OP
7659 If defined, a C expression whose value is a string, including spacing,
7660 containing the assembler operation to identify the following data as
7661 uninitialized global data. If not defined, and
7662 @code{ASM_OUTPUT_ALIGNED_BSS} not defined,
7663 uninitialized global data will be output in the data section if
7664 @option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be
7668 @defmac SBSS_SECTION_ASM_OP
7669 If defined, a C expression whose value is a string, including spacing,
7670 containing the assembler operation to identify the following data as
7671 uninitialized, writable small data.
7674 @defmac TLS_COMMON_ASM_OP
7675 If defined, a C expression whose value is a string containing the
7676 assembler operation to identify the following data as thread-local
7677 common data. The default is @code{".tls_common"}.
7680 @defmac TLS_SECTION_ASM_FLAG
7681 If defined, a C expression whose value is a character constant
7682 containing the flag used to mark a section as a TLS section. The
7683 default is @code{'T'}.
7686 @defmac INIT_SECTION_ASM_OP
7687 If defined, a C expression whose value is a string, including spacing,
7688 containing the assembler operation to identify the following data as
7689 initialization code. If not defined, GCC will assume such a section does
7690 not exist. This section has no corresponding @code{init_section}
7691 variable; it is used entirely in runtime code.
7694 @defmac FINI_SECTION_ASM_OP
7695 If defined, a C expression whose value is a string, including spacing,
7696 containing the assembler operation to identify the following data as
7697 finalization code. If not defined, GCC will assume such a section does
7698 not exist. This section has no corresponding @code{fini_section}
7699 variable; it is used entirely in runtime code.
7702 @defmac INIT_ARRAY_SECTION_ASM_OP
7703 If defined, a C expression whose value is a string, including spacing,
7704 containing the assembler operation to identify the following data as
7705 part of the @code{.init_array} (or equivalent) section. If not
7706 defined, GCC will assume such a section does not exist. Do not define
7707 both this macro and @code{INIT_SECTION_ASM_OP}.
7710 @defmac FINI_ARRAY_SECTION_ASM_OP
7711 If defined, a C expression whose value is a string, including spacing,
7712 containing the assembler operation to identify the following data as
7713 part of the @code{.fini_array} (or equivalent) section. If not
7714 defined, GCC will assume such a section does not exist. Do not define
7715 both this macro and @code{FINI_SECTION_ASM_OP}.
7718 @defmac MACH_DEP_SECTION_ASM_FLAG
7719 If defined, a C expression whose value is a character constant
7720 containing the flag used to mark a machine-dependent section. This
7721 corresponds to the @code{SECTION_MACH_DEP} section flag.
7724 @defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function})
7725 If defined, an ASM statement that switches to a different section
7726 via @var{section_op}, calls @var{function}, and switches back to
7727 the text section. This is used in @file{crtstuff.c} if
7728 @code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls
7729 to initialization and finalization functions from the init and fini
7730 sections. By default, this macro uses a simple function call. Some
7731 ports need hand-crafted assembly code to avoid dependencies on
7732 registers initialized in the function prologue or to ensure that
7733 constant pools don't end up too far way in the text section.
7736 @defmac TARGET_LIBGCC_SDATA_SECTION
7737 If defined, a string which names the section into which small
7738 variables defined in crtstuff and libgcc should go. This is useful
7739 when the target has options for optimizing access to small data, and
7740 you want the crtstuff and libgcc routines to be conservative in what
7741 they expect of your application yet liberal in what your application
7742 expects. For example, for targets with a @code{.sdata} section (like
7743 MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't
7744 require small data support from your application, but use this macro
7745 to put small data into @code{.sdata} so that your application can
7746 access these variables whether it uses small data or not.
7749 @defmac FORCE_CODE_SECTION_ALIGN
7750 If defined, an ASM statement that aligns a code section to some
7751 arbitrary boundary. This is used to force all fragments of the
7752 @code{.init} and @code{.fini} sections to have to same alignment
7753 and thus prevent the linker from having to add any padding.
7756 @defmac JUMP_TABLES_IN_TEXT_SECTION
7757 Define this macro to be an expression with a nonzero value if jump
7758 tables (for @code{tablejump} insns) should be output in the text
7759 section, along with the assembler instructions. Otherwise, the
7760 readonly data section is used.
7762 This macro is irrelevant if there is no separate readonly data section.
7765 @deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void)
7766 Define this hook if you need to do something special to set up the
7767 @file{varasm.cc} sections, or if your target has some special sections
7768 of its own that you need to create.
7770 GCC calls this hook after processing the command line, but before writing
7771 any assembly code, and before calling any of the section-returning hooks
7775 @deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void)
7776 Return a mask describing how relocations should be treated when
7777 selecting sections. Bit 1 should be set if global relocations
7778 should be placed in a read-write section; bit 0 should be set if
7779 local relocations should be placed in a read-write section.
7781 The default version of this function returns 3 when @option{-fpic}
7782 is in effect, and 0 otherwise. The hook is typically redefined
7783 when the target cannot support (some kinds of) dynamic relocations
7784 in read-only sections even in executables.
7787 @deftypefn {Target Hook} bool TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC (void)
7788 Return true to generate ADDR_DIF_VEC table
7789 or false to generate ADDR_VEC table for jumps in case of -fPIC.
7791 The default version of this function returns true if flag_pic
7792 equals true and false otherwise
7795 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align})
7796 Return the section into which @var{exp} should be placed. You can
7797 assume that @var{exp} is either a @code{VAR_DECL} node or a constant of
7798 some sort. @var{reloc} indicates whether the initial value of @var{exp}
7799 requires link-time relocations. Bit 0 is set when variable contains
7800 local relocations only, while bit 1 is set for global relocations.
7801 @var{align} is the constant alignment in bits.
7803 The default version of this function takes care of putting read-only
7804 variables in @code{readonly_data_section}.
7806 See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}.
7809 @defmac USE_SELECT_SECTION_FOR_FUNCTIONS
7810 Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called
7811 for @code{FUNCTION_DECL}s as well as for variables and constants.
7813 In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the
7814 function has been determined to be likely to be called, and nonzero if
7815 it is unlikely to be called.
7818 @deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc})
7819 Build up a unique section name, expressed as a @code{STRING_CST} node,
7820 and assign it to @samp{DECL_SECTION_NAME (@var{decl})}.
7821 As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether
7822 the initial value of @var{exp} requires link-time relocations.
7824 The default version of this function appends the symbol name to the
7825 ELF section name that would normally be used for the symbol. For
7826 example, the function @code{foo} would be placed in @code{.text.foo}.
7827 Whatever the actual target object format, this is often good enough.
7830 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl}, bool @var{relocatable})
7831 Return the readonly data or reloc readonly data section associated with
7832 @samp{DECL_SECTION_NAME (@var{decl})}. @var{relocatable} selects the latter
7834 The default version of this function selects @code{.gnu.linkonce.r.name} if
7835 the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name}
7836 or @code{.data.rel.ro.name} if function is in @code{.text.name}, and
7837 the normal readonly-data or reloc readonly data section otherwise.
7840 @deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX
7841 Usually, the compiler uses the prefix @code{".rodata"} to construct
7842 section names for mergeable constant data. Define this macro to override
7843 the string if a different section name should be used.
7846 @deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void)
7847 Return the section that should be used for transactional memory clone
7851 @deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align})
7852 Return the section into which a constant @var{x}, of mode @var{mode},
7853 should be placed. You can assume that @var{x} is some kind of
7854 constant in RTL@. The argument @var{mode} is redundant except in the
7855 case of a @code{const_int} rtx. @var{align} is the constant alignment
7858 The default version of this function takes care of putting symbolic
7859 constants in @code{flag_pic} mode in @code{data_section} and everything
7860 else in @code{readonly_data_section}.
7863 @deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id})
7864 Define this hook if you need to postprocess the assembler name generated
7865 by target-independent code. The @var{id} provided to this hook will be
7866 the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C,
7867 or the mangled name of the @var{decl} in C++). The return value of the
7868 hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on
7869 your target system. The default implementation of this hook just
7870 returns the @var{id} provided.
7873 @deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p})
7874 Define this hook if references to a symbol or a constant must be
7875 treated differently depending on something about the variable or
7876 function named by the symbol (such as what section it is in).
7878 The hook is executed immediately after rtl has been created for
7879 @var{decl}, which may be a variable or function declaration or
7880 an entry in the constant pool. In either case, @var{rtl} is the
7881 rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})}
7882 in this hook; that field may not have been initialized yet.
7884 In the case of a constant, it is safe to assume that the rtl is
7885 a @code{mem} whose address is a @code{symbol_ref}. Most decls
7886 will also have this form, but that is not guaranteed. Global
7887 register variables, for instance, will have a @code{reg} for their
7888 rtl. (Normally the right thing to do with such unusual rtl is
7891 The @var{new_decl_p} argument will be true if this is the first time
7892 that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will
7893 be false for subsequent invocations, which will happen for duplicate
7894 declarations. Whether or not anything must be done for the duplicate
7895 declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}.
7896 @var{new_decl_p} is always true when the hook is called for a constant.
7898 @cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO}
7899 The usual thing for this hook to do is to record flags in the
7900 @code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}.
7901 Historically, the name string was modified if it was necessary to
7902 encode more than one bit of information, but this practice is now
7903 discouraged; use @code{SYMBOL_REF_FLAGS}.
7905 The default definition of this hook, @code{default_encode_section_info}
7906 in @file{varasm.cc}, sets a number of commonly-useful bits in
7907 @code{SYMBOL_REF_FLAGS}. Check whether the default does what you need
7908 before overriding it.
7911 @deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name})
7912 Decode @var{name} and return the real name part, sans
7913 the characters that @code{TARGET_ENCODE_SECTION_INFO}
7917 @deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp})
7918 Returns true if @var{exp} should be placed into a ``small data'' section.
7919 The default version of this hook always returns false.
7922 @deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION
7923 Contains the value true if the target places read-only
7924 ``small data'' into a separate section. The default value is false.
7927 @deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void)
7928 It returns true if target wants profile code emitted before prologue.
7930 The default version of this hook use the target macro
7931 @code{PROFILE_BEFORE_PROLOGUE}.
7934 @deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp})
7935 Returns true if @var{exp} names an object for which name resolution
7936 rules must resolve to the current ``module'' (dynamic shared library
7937 or executable image).
7939 The default version of this hook implements the name resolution rules
7940 for ELF, which has a looser model of global name binding than other
7941 currently supported object file formats.
7944 @deftypevr {Target Hook} bool TARGET_HAVE_TLS
7945 Contains the value true if the target supports thread-local storage.
7946 The default value is false.
7951 @section Position Independent Code
7952 @cindex position independent code
7955 This section describes macros that help implement generation of position
7956 independent code. Simply defining these macros is not enough to
7957 generate valid PIC; you must also add support to the hook
7958 @code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro
7959 @code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You
7960 must modify the definition of @samp{movsi} to do something appropriate
7961 when the source operand contains a symbolic address. You may also
7962 need to alter the handling of switch statements so that they use
7964 @c i rearranged the order of the macros above to try to force one of
7965 @c them to the next line, to eliminate an overfull hbox. --mew 10feb93
7967 @defmac PIC_OFFSET_TABLE_REGNUM
7968 The register number of the register used to address a table of static
7969 data addresses in memory. In some cases this register is defined by a
7970 processor's ``application binary interface'' (ABI)@. When this macro
7971 is defined, RTL is generated for this register once, as with the stack
7972 pointer and frame pointer registers. If this macro is not defined, it
7973 is up to the machine-dependent files to allocate such a register (if
7974 necessary). Note that this register must be fixed when in use (e.g.@:
7975 when @code{flag_pic} is true).
7978 @defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED
7979 A C expression that is nonzero if the register defined by
7980 @code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined,
7981 the default is zero. Do not define
7982 this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined.
7985 @defmac LEGITIMATE_PIC_OPERAND_P (@var{x})
7986 A C expression that is nonzero if @var{x} is a legitimate immediate
7987 operand on the target machine when generating position independent code.
7988 You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not
7989 check this. You can also assume @var{flag_pic} is true, so you need not
7990 check it either. You need not define this macro if all constants
7991 (including @code{SYMBOL_REF}) can be immediate operands when generating
7992 position independent code.
7995 @node Assembler Format
7996 @section Defining the Output Assembler Language
7998 This section describes macros whose principal purpose is to describe how
7999 to write instructions in assembler language---rather than what the
8003 * File Framework:: Structural information for the assembler file.
8004 * Data Output:: Output of constants (numbers, strings, addresses).
8005 * Uninitialized Data:: Output of uninitialized variables.
8006 * Label Output:: Output and generation of labels.
8007 * Initialization:: General principles of initialization
8008 and termination routines.
8009 * Macros for Initialization::
8010 Specific macros that control the handling of
8011 initialization and termination routines.
8012 * Instruction Output:: Output of actual instructions.
8013 * Dispatch Tables:: Output of jump tables.
8014 * Exception Region Output:: Output of exception region code.
8015 * Alignment Output:: Pseudo ops for alignment and skipping data.
8018 @node File Framework
8019 @subsection The Overall Framework of an Assembler File
8020 @cindex assembler format
8021 @cindex output of assembler code
8023 @c prevent bad page break with this line
8024 This describes the overall framework of an assembly file.
8026 @findex default_file_start
8027 @deftypefn {Target Hook} void TARGET_ASM_FILE_START (void)
8028 Output to @code{asm_out_file} any text which the assembler expects to
8029 find at the beginning of a file. The default behavior is controlled
8030 by two flags, documented below. Unless your target's assembler is
8031 quite unusual, if you override the default, you should call
8032 @code{default_file_start} at some point in your target hook. This
8033 lets other target files rely on these variables.
8036 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF
8037 If this flag is true, the text of the macro @code{ASM_APP_OFF} will be
8038 printed as the very first line in the assembly file, unless
8039 @option{-fverbose-asm} is in effect. (If that macro has been defined
8040 to the empty string, this variable has no effect.) With the normal
8041 definition of @code{ASM_APP_OFF}, the effect is to notify the GNU
8042 assembler that it need not bother stripping comments or extra
8043 whitespace from its input. This allows it to work a bit faster.
8045 The default is false. You should not set it to true unless you have
8046 verified that your port does not generate any extra whitespace or
8047 comments that will cause GAS to issue errors in NO_APP mode.
8050 @deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE
8051 If this flag is true, @code{output_file_directive} will be called
8052 for the primary source file, immediately after printing
8053 @code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect
8054 this to be done. The default is false.
8057 @deftypefn {Target Hook} void TARGET_ASM_FILE_END (void)
8058 Output to @code{asm_out_file} any text which the assembler expects
8059 to find at the end of a file. The default is to output nothing.
8062 @deftypefun void file_end_indicate_exec_stack ()
8063 Some systems use a common convention, the @samp{.note.GNU-stack}
8064 special section, to indicate whether or not an object file relies on
8065 the stack being executable. If your system uses this convention, you
8066 should define @code{TARGET_ASM_FILE_END} to this function. If you
8067 need to do other things in that hook, have your hook function call
8071 @deftypefn {Target Hook} void TARGET_ASM_LTO_START (void)
8072 Output to @code{asm_out_file} any text which the assembler expects
8073 to find at the start of an LTO section. The default is to output
8077 @deftypefn {Target Hook} void TARGET_ASM_LTO_END (void)
8078 Output to @code{asm_out_file} any text which the assembler expects
8079 to find at the end of an LTO section. The default is to output
8083 @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void)
8084 Output to @code{asm_out_file} any text which is needed before emitting
8085 unwind info and debug info at the end of a file. Some targets emit
8086 here PIC setup thunks that cannot be emitted at the end of file,
8087 because they couldn't have unwind info then. The default is to output
8091 @defmac ASM_COMMENT_START
8092 A C string constant describing how to begin a comment in the target
8093 assembler language. The compiler assumes that the comment will end at
8094 the end of the line.
8098 A C string constant for text to be output before each @code{asm}
8099 statement or group of consecutive ones. Normally this is
8100 @code{"#APP"}, which is a comment that has no effect on most
8101 assemblers but tells the GNU assembler that it must check the lines
8102 that follow for all valid assembler constructs.
8106 A C string constant for text to be output after each @code{asm}
8107 statement or group of consecutive ones. Normally this is
8108 @code{"#NO_APP"}, which tells the GNU assembler to resume making the
8109 time-saving assumptions that are valid for ordinary compiler output.
8112 @defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name})
8113 A C statement to output COFF information or DWARF debugging information
8114 which indicates that filename @var{name} is the current source file to
8115 the stdio stream @var{stream}.
8117 This macro need not be defined if the standard form of output
8118 for the file format in use is appropriate.
8121 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name})
8122 Output DWARF debugging information which indicates that filename
8123 @var{name} is the current source file to the stdio stream @var{file}.
8125 This target hook need not be defined if the standard form of output
8126 for the file format in use is appropriate.
8129 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_IDENT (const char *@var{name})
8130 Output a string based on @var{name}, suitable for the @samp{#ident}
8131 directive, or the equivalent directive or pragma in non-C-family languages.
8132 If this hook is not defined, nothing is output for the @samp{#ident}
8136 @defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string})
8137 A C statement to output the string @var{string} to the stdio stream
8138 @var{stream}. If you do not call the function @code{output_quoted_string}
8139 in your config files, GCC will only call it to output filenames to
8140 the assembler source. So you can use it to canonicalize the format
8141 of the filename using this macro.
8144 @deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl})
8145 Output assembly directives to switch to section @var{name}. The section
8146 should have attributes as specified by @var{flags}, which is a bit mask
8147 of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl}
8148 is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which
8149 this section is associated.
8152 @deftypefn {Target Hook} bool TARGET_ASM_ELF_FLAGS_NUMERIC (unsigned int @var{flags}, unsigned int *@var{num})
8153 This hook can be used to encode ELF section flags for which no letter
8154 code has been defined in the assembler. It is called by
8155 @code{default_asm_named_section} whenever the section flags need to be
8156 emitted in the assembler output. If the hook returns true, then the
8157 numerical value for ELF section flags should be calculated from
8158 @var{flags} and saved in @var{*num}; the value is printed out instead of the
8159 normal sequence of letter codes. If the hook is not defined, or if it
8160 returns false, then @var{num} is ignored and the traditional letter sequence
8164 @deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit})
8165 Return preferred text (sub)section for function @var{decl}.
8166 Main purpose of this function is to separate cold, normal and hot
8167 functions. @var{startup} is true when function is known to be used only
8168 at startup (from static constructors or it is @code{main()}).
8169 @var{exit} is true when function is known to be used only at exit
8170 (from static destructors).
8171 Return NULL if function should go to default text section.
8174 @deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold})
8175 Used by the target to emit any assembler directives or additional
8176 labels needed when a function is partitioned between different
8177 sections. Output should be written to @var{file}. The function
8178 decl is available as @var{decl} and the new section is `cold' if
8179 @var{new_is_cold} is @code{true}.
8182 @deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS
8183 This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}.
8184 It must not be modified by command-line option processing.
8187 @anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}
8188 @deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS
8189 This flag is true if we can create zeroed data by switching to a BSS
8190 section and then using @code{ASM_OUTPUT_SKIP} to allocate the space.
8191 This is true on most ELF targets.
8194 @deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc})
8195 Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION}
8196 based on a variable or function decl, a section name, and whether or not the
8197 declaration's initializer may contain runtime relocations. @var{decl} may be
8198 null, in which case read-write data should be assumed.
8200 The default version of this function handles choosing code vs data,
8201 read-only vs read-write data, and @code{flag_pic}. You should only
8202 need to override this if your target has special flags that might be
8203 set via @code{__attribute__}.
8206 @deftypefn {Target Hook} void TARGET_ASM_RECORD_GCC_SWITCHES (const char *@var{})
8207 Provides the target with the ability to record the gcc command line
8208 switches provided as argument.
8210 By default this hook is set to NULL, but an example implementation is
8211 provided for ELF based targets. Called @var{elf_record_gcc_switches},
8212 it records the switches as ASCII text inside a new, string mergeable
8213 section in the assembler output file. The name of the new section is
8214 provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target
8218 @deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION
8219 This is the name of the section that will be created by the example
8220 ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target
8226 @subsection Output of Data
8229 @deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP
8230 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP
8231 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PSI_OP
8232 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP
8233 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PDI_OP
8234 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP
8235 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PTI_OP
8236 @deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP
8237 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP
8238 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PSI_OP
8239 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP
8240 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PDI_OP
8241 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP
8242 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PTI_OP
8243 @deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP
8244 These hooks specify assembly directives for creating certain kinds
8245 of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a
8246 byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an
8247 aligned two-byte object, and so on. Any of the hooks may be
8248 @code{NULL}, indicating that no suitable directive is available.
8250 The compiler will print these strings at the start of a new line,
8251 followed immediately by the object's initial value. In most cases,
8252 the string should contain a tab, a pseudo-op, and then another tab.
8255 @deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p})
8256 The @code{assemble_integer} function uses this hook to output an
8257 integer object. @var{x} is the object's value, @var{size} is its size
8258 in bytes and @var{aligned_p} indicates whether it is aligned. The
8259 function should return @code{true} if it was able to output the
8260 object. If it returns false, @code{assemble_integer} will try to
8261 split the object into smaller parts.
8263 The default implementation of this hook will use the
8264 @code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false}
8265 when the relevant string is @code{NULL}.
8268 @deftypefn {Target Hook} void TARGET_ASM_DECL_END (void)
8269 Define this hook if the target assembler requires a special marker to
8270 terminate an initialized variable declaration.
8273 @deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x})
8274 A target hook to recognize @var{rtx} patterns that @code{output_addr_const}
8275 can't deal with, and output assembly code to @var{file} corresponding to
8276 the pattern @var{x}. This may be used to allow machine-dependent
8277 @code{UNSPEC}s to appear within constants.
8279 If target hook fails to recognize a pattern, it must return @code{false},
8280 so that a standard error message is printed. If it prints an error message
8281 itself, by calling, for example, @code{output_operand_lossage}, it may just
8285 @defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len})
8286 A C statement to output to the stdio stream @var{stream} an assembler
8287 instruction to assemble a string constant containing the @var{len}
8288 bytes at @var{ptr}. @var{ptr} will be a C expression of type
8289 @code{char *} and @var{len} a C expression of type @code{int}.
8291 If the assembler has a @code{.ascii} pseudo-op as found in the
8292 Berkeley Unix assembler, do not define the macro
8293 @code{ASM_OUTPUT_ASCII}.
8296 @defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n})
8297 A C statement to output word @var{n} of a function descriptor for
8298 @var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS}
8299 is defined, and is otherwise unused.
8302 @defmac CONSTANT_POOL_BEFORE_FUNCTION
8303 You may define this macro as a C expression. You should define the
8304 expression to have a nonzero value if GCC should output the constant
8305 pool for a function before the code for the function, or a zero value if
8306 GCC should output the constant pool after the function. If you do
8307 not define this macro, the usual case, GCC will output the constant
8308 pool before the function.
8311 @defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size})
8312 A C statement to output assembler commands to define the start of the
8313 constant pool for a function. @var{funname} is a string giving
8314 the name of the function. Should the return type of the function
8315 be required, it can be obtained via @var{fundecl}. @var{size}
8316 is the size, in bytes, of the constant pool that will be written
8317 immediately after this call.
8319 If no constant-pool prefix is required, the usual case, this macro need
8323 @defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto})
8324 A C statement (with or without semicolon) to output a constant in the
8325 constant pool, if it needs special treatment. (This macro need not do
8326 anything for RTL expressions that can be output normally.)
8328 The argument @var{file} is the standard I/O stream to output the
8329 assembler code on. @var{x} is the RTL expression for the constant to
8330 output, and @var{mode} is the machine mode (in case @var{x} is a
8331 @samp{const_int}). @var{align} is the required alignment for the value
8332 @var{x}; you should output an assembler directive to force this much
8335 The argument @var{labelno} is a number to use in an internal label for
8336 the address of this pool entry. The definition of this macro is
8337 responsible for outputting the label definition at the proper place.
8338 Here is how to do this:
8341 @code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno});
8344 When you output a pool entry specially, you should end with a
8345 @code{goto} to the label @var{jumpto}. This will prevent the same pool
8346 entry from being output a second time in the usual manner.
8348 You need not define this macro if it would do nothing.
8351 @defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size})
8352 A C statement to output assembler commands to at the end of the constant
8353 pool for a function. @var{funname} is a string giving the name of the
8354 function. Should the return type of the function be required, you can
8355 obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the
8356 constant pool that GCC wrote immediately before this call.
8358 If no constant-pool epilogue is required, the usual case, you need not
8362 @defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR})
8363 Define this macro as a C expression which is nonzero if @var{C} is
8364 used as a logical line separator by the assembler. @var{STR} points
8365 to the position in the string where @var{C} was found; this can be used if
8366 a line separator uses multiple characters.
8368 If you do not define this macro, the default is that only
8369 the character @samp{;} is treated as a logical line separator.
8372 @deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN
8373 @deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN
8374 These target hooks are C string constants, describing the syntax in the
8375 assembler for grouping arithmetic expressions. If not overridden, they
8376 default to normal parentheses, which is correct for most assemblers.
8379 These macros are provided by @file{real.h} for writing the definitions
8380 of @code{ASM_OUTPUT_DOUBLE} and the like:
8382 @defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l})
8383 @defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l})
8384 @defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l})
8385 @defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l})
8386 @defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l})
8387 @defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l})
8388 These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the
8389 target's floating point representation, and store its bit pattern in
8390 the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and
8391 @code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a
8392 simple @code{long int}. For the others, it should be an array of
8393 @code{long int}. The number of elements in this array is determined
8394 by the size of the desired target floating point data type: 32 bits of
8395 it go in each @code{long int} array element. Each array element holds
8396 32 bits of the result, even if @code{long int} is wider than 32 bits
8397 on the host machine.
8399 The array element values are designed so that you can print them out
8400 using @code{fprintf} in the order they should appear in the target
8404 @node Uninitialized Data
8405 @subsection Output of Uninitialized Variables
8407 Each of the macros in this section is used to do the whole job of
8408 outputting a single uninitialized variable.
8410 @defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded})
8411 A C statement (sans semicolon) to output to the stdio stream
8412 @var{stream} the assembler definition of a common-label named
8413 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
8414 is the size rounded up to whatever alignment the caller wants. It is
8415 possible that @var{size} may be zero, for instance if a struct with no
8416 other member than a zero-length array is defined. In this case, the
8417 backend must output a symbol definition that allocates at least one
8418 byte, both so that the address of the resulting object does not compare
8419 equal to any other, and because some object formats cannot even express
8420 the concept of a zero-sized common symbol, as that is how they represent
8421 an ordinary undefined external.
8423 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
8424 output the name itself; before and after that, output the additional
8425 assembler syntax for defining the name, and a newline.
8427 This macro controls how the assembler definitions of uninitialized
8428 common global variables are output.
8431 @defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment})
8432 Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a
8433 separate, explicit argument. If you define this macro, it is used in
8434 place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in
8435 handling the required alignment of the variable. The alignment is specified
8436 as the number of bits.
8439 @defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
8440 Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the
8441 variable to be output, if there is one, or @code{NULL_TREE} if there
8442 is no corresponding variable. If you define this macro, GCC will use it
8443 in place of both @code{ASM_OUTPUT_COMMON} and
8444 @code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see
8445 the variable's decl in order to chose what to output.
8448 @defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
8449 A C statement (sans semicolon) to output to the stdio stream
8450 @var{stream} the assembler definition of uninitialized global @var{decl} named
8451 @var{name} whose size is @var{size} bytes. The variable @var{alignment}
8452 is the alignment specified as the number of bits.
8454 Try to use function @code{asm_output_aligned_bss} defined in file
8455 @file{varasm.cc} when defining this macro. If unable, use the expression
8456 @code{assemble_name (@var{stream}, @var{name})} to output the name itself;
8457 before and after that, output the additional assembler syntax for defining
8458 the name, and a newline.
8460 There are two ways of handling global BSS@. One is to define this macro.
8461 The other is to have @code{TARGET_ASM_SELECT_SECTION} return a
8462 switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}).
8463 You do not need to do both.
8465 Some languages do not have @code{common} data, and require a
8466 non-common form of global BSS in order to handle uninitialized globals
8467 efficiently. C++ is one example of this. However, if the target does
8468 not support global BSS, the front end may choose to make globals
8469 common in order to save space in the object file.
8472 @defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded})
8473 A C statement (sans semicolon) to output to the stdio stream
8474 @var{stream} the assembler definition of a local-common-label named
8475 @var{name} whose size is @var{size} bytes. The variable @var{rounded}
8476 is the size rounded up to whatever alignment the caller wants.
8478 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
8479 output the name itself; before and after that, output the additional
8480 assembler syntax for defining the name, and a newline.
8482 This macro controls how the assembler definitions of uninitialized
8483 static variables are output.
8486 @defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment})
8487 Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a
8488 separate, explicit argument. If you define this macro, it is used in
8489 place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in
8490 handling the required alignment of the variable. The alignment is specified
8491 as the number of bits.
8494 @defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment})
8495 Like @code{ASM_OUTPUT_ALIGNED_LOCAL} except that @var{decl} of the
8496 variable to be output, if there is one, or @code{NULL_TREE} if there
8497 is no corresponding variable. If you define this macro, GCC will use it
8498 in place of both @code{ASM_OUTPUT_LOCAL} and
8499 @code{ASM_OUTPUT_ALIGNED_LOCAL}. Define this macro when you need to see
8500 the variable's decl in order to chose what to output.
8504 @subsection Output and Generation of Labels
8506 @c prevent bad page break with this line
8507 This is about outputting labels.
8509 @findex assemble_name
8510 @defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name})
8511 A C statement (sans semicolon) to output to the stdio stream
8512 @var{stream} the assembler definition of a label named @var{name}.
8513 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
8514 output the name itself; before and after that, output the additional
8515 assembler syntax for defining the name, and a newline. A default
8516 definition of this macro is provided which is correct for most systems.
8519 @defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl})
8520 A C statement (sans semicolon) to output to the stdio stream
8521 @var{stream} the assembler definition of a label named @var{name} of
8523 Use the expression @code{assemble_name (@var{stream}, @var{name})} to
8524 output the name itself; before and after that, output the additional
8525 assembler syntax for defining the name, and a newline. A default
8526 definition of this macro is provided which is correct for most systems.
8528 If this macro is not defined, then the function name is defined in the
8529 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
8532 @findex assemble_name_raw
8533 @defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name})
8534 Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known
8535 to refer to a compiler-generated label. The default definition uses
8536 @code{assemble_name_raw}, which is like @code{assemble_name} except
8537 that it is more efficient.
8541 A C string containing the appropriate assembler directive to specify the
8542 size of a symbol, without any arguments. On systems that use ELF, the
8543 default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other
8544 systems, the default is not to define this macro.
8546 Define this macro only if it is correct to use the default definitions
8547 of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE}
8548 for your system. If you need your own custom definitions of those
8549 macros, or if you do not need explicit symbol sizes at all, do not
8553 @defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size})
8554 A C statement (sans semicolon) to output to the stdio stream
8555 @var{stream} a directive telling the assembler that the size of the
8556 symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}.
8557 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
8561 @defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name})
8562 A C statement (sans semicolon) to output to the stdio stream
8563 @var{stream} a directive telling the assembler to calculate the size of
8564 the symbol @var{name} by subtracting its address from the current
8567 If you define @code{SIZE_ASM_OP}, a default definition of this macro is
8568 provided. The default assumes that the assembler recognizes a special
8569 @samp{.} symbol as referring to the current address, and can calculate
8570 the difference between this and another symbol. If your assembler does
8571 not recognize @samp{.} or cannot do calculations with it, you will need
8572 to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique.
8575 @defmac NO_DOLLAR_IN_LABEL
8576 Define this macro if the assembler does not accept the character
8577 @samp{$} in label names. By default constructors and destructors in
8578 G++ have @samp{$} in the identifiers. If this macro is defined,
8579 @samp{.} is used instead.
8582 @defmac NO_DOT_IN_LABEL
8583 Define this macro if the assembler does not accept the character
8584 @samp{.} in label names. By default constructors and destructors in G++
8585 have names that use @samp{.}. If this macro is defined, these names
8586 are rewritten to avoid @samp{.}.
8590 A C string containing the appropriate assembler directive to specify the
8591 type of a symbol, without any arguments. On systems that use ELF, the
8592 default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other
8593 systems, the default is not to define this macro.
8595 Define this macro only if it is correct to use the default definition of
8596 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
8597 custom definition of this macro, or if you do not need explicit symbol
8598 types at all, do not define this macro.
8601 @defmac TYPE_OPERAND_FMT
8602 A C string which specifies (using @code{printf} syntax) the format of
8603 the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the
8604 default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems,
8605 the default is not to define this macro.
8607 Define this macro only if it is correct to use the default definition of
8608 @code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own
8609 custom definition of this macro, or if you do not need explicit symbol
8610 types at all, do not define this macro.
8613 @defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type})
8614 A C statement (sans semicolon) to output to the stdio stream
8615 @var{stream} a directive telling the assembler that the type of the
8616 symbol @var{name} is @var{type}. @var{type} is a C string; currently,
8617 that string is always either @samp{"function"} or @samp{"object"}, but
8618 you should not count on this.
8620 If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default
8621 definition of this macro is provided.
8624 @defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
8625 A C statement (sans semicolon) to output to the stdio stream
8626 @var{stream} any text necessary for declaring the name @var{name} of a
8627 function which is being defined. This macro is responsible for
8628 outputting the label definition (perhaps using
8629 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
8630 @code{FUNCTION_DECL} tree node representing the function.
8632 If this macro is not defined, then the function name is defined in the
8633 usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}).
8635 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
8639 @defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
8640 A C statement (sans semicolon) to output to the stdio stream
8641 @var{stream} any text necessary for declaring the size of a function
8642 which is being defined. The argument @var{name} is the name of the
8643 function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node
8644 representing the function.
8646 If this macro is not defined, then the function size is not defined.
8648 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
8652 @defmac ASM_DECLARE_COLD_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl})
8653 A C statement (sans semicolon) to output to the stdio stream
8654 @var{stream} any text necessary for declaring the name @var{name} of a
8655 cold function partition which is being defined. This macro is responsible
8656 for outputting the label definition (perhaps using
8657 @code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the
8658 @code{FUNCTION_DECL} tree node representing the function.
8660 If this macro is not defined, then the cold partition name is defined in the
8661 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
8663 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition
8667 @defmac ASM_DECLARE_COLD_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl})
8668 A C statement (sans semicolon) to output to the stdio stream
8669 @var{stream} any text necessary for declaring the size of a cold function
8670 partition which is being defined. The argument @var{name} is the name of the
8671 cold partition of the function. The argument @var{decl} is the
8672 @code{FUNCTION_DECL} tree node representing the function.
8674 If this macro is not defined, then the partition size is not defined.
8676 You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition
8680 @defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl})
8681 A C statement (sans semicolon) to output to the stdio stream
8682 @var{stream} any text necessary for declaring the name @var{name} of an
8683 initialized variable which is being defined. This macro must output the
8684 label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument
8685 @var{decl} is the @code{VAR_DECL} tree node representing the variable.
8687 If this macro is not defined, then the variable name is defined in the
8688 usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}).
8690 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or
8691 @code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro.
8694 @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})
8695 A target hook to output to the stdio stream @var{file} any text necessary
8696 for declaring the name @var{name} of a constant which is being defined. This
8697 target hook is responsible for outputting the label definition (perhaps using
8698 @code{assemble_label}). The argument @var{exp} is the value of the constant,
8699 and @var{size} is the size of the constant in bytes. The @var{name}
8700 will be an internal label.
8702 The default version of this target hook, define the @var{name} in the
8703 usual manner as a label (by means of @code{assemble_label}).
8705 You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook.
8708 @defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name})
8709 A C statement (sans semicolon) to output to the stdio stream
8710 @var{stream} any text necessary for claiming a register @var{regno}
8711 for a global variable @var{decl} with name @var{name}.
8713 If you don't define this macro, that is equivalent to defining it to do
8717 @defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend})
8718 A C statement (sans semicolon) to finish up declaring a variable name
8719 once the compiler has processed its initializer fully and thus has had a
8720 chance to determine the size of an array when controlled by an
8721 initializer. This is used on systems where it's necessary to declare
8722 something about the size of the object.
8724 If you don't define this macro, that is equivalent to defining it to do
8727 You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or
8728 @code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro.
8731 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name})
8732 This target hook is a function to output to the stdio stream
8733 @var{stream} some commands that will make the label @var{name} global;
8734 that is, available for reference from other files.
8736 The default implementation relies on a proper definition of
8737 @code{GLOBAL_ASM_OP}.
8740 @deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl})
8741 This target hook is a function to output to the stdio stream
8742 @var{stream} some commands that will make the name associated with @var{decl}
8743 global; that is, available for reference from other files.
8745 The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook.
8748 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_UNDEFINED_DECL (FILE *@var{stream}, const char *@var{name}, const_tree @var{decl})
8749 This target hook is a function to output to the stdio stream
8750 @var{stream} some commands that will declare the name associated with
8751 @var{decl} which is not defined in the current translation unit. Most
8752 assemblers do not require anything to be output in this case.
8755 @defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name})
8756 A C statement (sans semicolon) to output to the stdio stream
8757 @var{stream} some commands that will make the label @var{name} weak;
8758 that is, available for reference from other files but only used if
8759 no other definition is available. Use the expression
8760 @code{assemble_name (@var{stream}, @var{name})} to output the name
8761 itself; before and after that, output the additional assembler syntax
8762 for making that name weak, and a newline.
8764 If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not
8765 support weak symbols and you should not define the @code{SUPPORTS_WEAK}
8769 @defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value})
8770 Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and
8771 @code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function
8772 or variable decl. If @var{value} is not @code{NULL}, this C statement
8773 should output to the stdio stream @var{stream} assembler code which
8774 defines (equates) the weak symbol @var{name} to have the value
8775 @var{value}. If @var{value} is @code{NULL}, it should output commands
8776 to make @var{name} weak.
8779 @defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value})
8780 Outputs a directive that enables @var{name} to be used to refer to
8781 symbol @var{value} with weak-symbol semantics. @code{decl} is the
8782 declaration of @code{name}.
8785 @defmac SUPPORTS_WEAK
8786 A preprocessor constant expression which evaluates to true if the target
8787 supports weak symbols.
8789 If you don't define this macro, @file{defaults.h} provides a default
8790 definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL}
8791 is defined, the default definition is @samp{1}; otherwise, it is @samp{0}.
8794 @defmac TARGET_SUPPORTS_WEAK
8795 A C expression which evaluates to true if the target supports weak symbols.
8797 If you don't define this macro, @file{defaults.h} provides a default
8798 definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define
8799 this macro if you want to control weak symbol support with a compiler
8800 flag such as @option{-melf}.
8803 @defmac MAKE_DECL_ONE_ONLY (@var{decl})
8804 A C statement (sans semicolon) to mark @var{decl} to be emitted as a
8805 public symbol such that extra copies in multiple translation units will
8806 be discarded by the linker. Define this macro if your object file
8807 format provides support for this concept, such as the @samp{COMDAT}
8808 section flags in the Microsoft Windows PE/COFF format, and this support
8809 requires changes to @var{decl}, such as putting it in a separate section.
8812 @defmac SUPPORTS_ONE_ONLY
8813 A C expression which evaluates to true if the target supports one-only
8816 If you don't define this macro, @file{varasm.cc} provides a default
8817 definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default
8818 definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if
8819 you want to control one-only symbol support with a compiler flag, or if
8820 setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to
8821 be emitted as one-only.
8824 @deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility})
8825 This target hook is a function to output to @var{asm_out_file} some
8826 commands that will make the symbol(s) associated with @var{decl} have
8827 hidden, protected or internal visibility as specified by @var{visibility}.
8830 @defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC
8831 A C expression that evaluates to true if the target's linker expects
8832 that weak symbols do not appear in a static archive's table of contents.
8833 The default is @code{0}.
8835 Leaving weak symbols out of an archive's table of contents means that,
8836 if a symbol will only have a definition in one translation unit and
8837 will have undefined references from other translation units, that
8838 symbol should not be weak. Defining this macro to be nonzero will
8839 thus have the effect that certain symbols that would normally be weak
8840 (explicit template instantiations, and vtables for polymorphic classes
8841 with noninline key methods) will instead be nonweak.
8843 The C++ ABI requires this macro to be zero. Define this macro for
8844 targets where full C++ ABI compliance is impossible and where linker
8845 restrictions require weak symbols to be left out of a static archive's
8849 @defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name})
8850 A C statement (sans semicolon) to output to the stdio stream
8851 @var{stream} any text necessary for declaring the name of an external
8852 symbol named @var{name} which is referenced in this compilation but
8853 not defined. The value of @var{decl} is the tree node for the
8856 This macro need not be defined if it does not need to output anything.
8857 The GNU assembler and most Unix assemblers don't require anything.
8860 @deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref})
8861 This target hook is a function to output to @var{asm_out_file} an assembler
8862 pseudo-op to declare a library function name external. The name of the
8863 library function is given by @var{symref}, which is a @code{symbol_ref}.
8866 @deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol})
8867 This target hook is a function to output to @var{asm_out_file} an assembler
8868 directive to annotate @var{symbol} as used. The Darwin target uses the
8869 .no_dead_code_strip directive.
8872 @defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name})
8873 A C statement (sans semicolon) to output to the stdio stream
8874 @var{stream} a reference in assembler syntax to a label named
8875 @var{name}. This should add @samp{_} to the front of the name, if that
8876 is customary on your operating system, as it is in most Berkeley Unix
8877 systems. This macro is used in @code{assemble_name}.
8880 @deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name})
8881 Given a symbol @var{name}, perform same mangling as @code{varasm.cc}'s
8882 @code{assemble_name}, but in memory rather than to a file stream, returning
8883 result as an @code{IDENTIFIER_NODE}. Required for correct LTO symtabs. The
8884 default implementation calls the @code{TARGET_STRIP_NAME_ENCODING} hook and
8885 then prepends the @code{USER_LABEL_PREFIX}, if any.
8888 @defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym})
8889 A C statement (sans semicolon) to output a reference to
8890 @code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name}
8891 will be used to output the name of the symbol. This macro may be used
8892 to modify the way a symbol is referenced depending on information
8893 encoded by @code{TARGET_ENCODE_SECTION_INFO}.
8896 @defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf})
8897 A C statement (sans semicolon) to output a reference to @var{buf}, the
8898 result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined,
8899 @code{assemble_name} will be used to output the name of the symbol.
8900 This macro is not used by @code{output_asm_label}, or the @code{%l}
8901 specifier that calls it; the intention is that this macro should be set
8902 when it is necessary to output a label differently when its address is
8906 @deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno})
8907 A function to output to the stdio stream @var{stream} a label whose
8908 name is made from the string @var{prefix} and the number @var{labelno}.
8910 It is absolutely essential that these labels be distinct from the labels
8911 used for user-level functions and variables. Otherwise, certain programs
8912 will have name conflicts with internal labels.
8914 It is desirable to exclude internal labels from the symbol table of the
8915 object file. Most assemblers have a naming convention for labels that
8916 should be excluded; on many systems, the letter @samp{L} at the
8917 beginning of a label has this effect. You should find out what
8918 convention your system uses, and follow it.
8920 The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}.
8923 @defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num})
8924 A C statement to output to the stdio stream @var{stream} a debug info
8925 label whose name is made from the string @var{prefix} and the number
8926 @var{num}. This is useful for VLIW targets, where debug info labels
8927 may need to be treated differently than branch target labels. On some
8928 systems, branch target labels must be at the beginning of instruction
8929 bundles, but debug info labels can occur in the middle of instruction
8932 If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be
8936 @defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num})
8937 A C statement to store into the string @var{string} a label whose name
8938 is made from the string @var{prefix} and the number @var{num}.
8940 This string, when output subsequently by @code{assemble_name}, should
8941 produce the output that @code{(*targetm.asm_out.internal_label)} would produce
8942 with the same @var{prefix} and @var{num}.
8944 If the string begins with @samp{*}, then @code{assemble_name} will
8945 output the rest of the string unchanged. It is often convenient for
8946 @code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the
8947 string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets
8948 to output the string, and may change it. (Of course,
8949 @code{ASM_OUTPUT_LABELREF} is also part of your machine description, so
8950 you should know what it does on your machine.)
8953 @defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number})
8954 A C expression to assign to @var{outvar} (which is a variable of type
8955 @code{char *}) a newly allocated string made from the string
8956 @var{name} and the number @var{number}, with some suitable punctuation
8957 added. Use @code{alloca} to get space for the string.
8959 The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to
8960 produce an assembler label for an internal static variable whose name is
8961 @var{name}. Therefore, the string must be such as to result in valid
8962 assembler code. The argument @var{number} is different each time this
8963 macro is executed; it prevents conflicts between similarly-named
8964 internal static variables in different scopes.
8966 Ideally this string should not be a valid C identifier, to prevent any
8967 conflict with the user's own symbols. Most assemblers allow periods
8968 or percent signs in assembler symbols; putting at least one of these
8969 between the name and the number will suffice.
8971 If this macro is not defined, a default definition will be provided
8972 which is correct for most systems.
8975 @defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value})
8976 A C statement to output to the stdio stream @var{stream} assembler code
8977 which defines (equates) the symbol @var{name} to have the value @var{value}.
8980 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8981 correct for most systems.
8984 @defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value})
8985 A C statement to output to the stdio stream @var{stream} assembler code
8986 which defines (equates) the symbol whose tree node is @var{decl_of_name}
8987 to have the value of the tree node @var{decl_of_value}. This macro will
8988 be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if
8989 the tree nodes are available.
8992 If @code{SET_ASM_OP} is defined, a default definition is provided which is
8993 correct for most systems.
8996 @defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value})
8997 A C statement that evaluates to true if the assembler code which defines
8998 (equates) the symbol whose tree node is @var{decl_of_name} to have the value
8999 of the tree node @var{decl_of_value} should be emitted near the end of the
9000 current compilation unit. The default is to not defer output of defines.
9001 This macro affects defines output by @samp{ASM_OUTPUT_DEF} and
9002 @samp{ASM_OUTPUT_DEF_FROM_DECLS}.
9005 @defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value})
9006 A C statement to output to the stdio stream @var{stream} assembler code
9007 which defines (equates) the weak symbol @var{name} to have the value
9008 @var{value}. If @var{value} is @code{NULL}, it defines @var{name} as
9009 an undefined weak symbol.
9011 Define this macro if the target only supports weak aliases; define
9012 @code{ASM_OUTPUT_DEF} instead if possible.
9015 @defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name})
9016 Define this macro to override the default assembler names used for
9017 Objective-C methods.
9019 The default name is a unique method number followed by the name of the
9020 class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of
9021 the category is also included in the assembler name (e.g.@:
9024 These names are safe on most systems, but make debugging difficult since
9025 the method's selector is not present in the name. Therefore, particular
9026 systems define other ways of computing names.
9028 @var{buf} is an expression of type @code{char *} which gives you a
9029 buffer in which to store the name; its length is as long as
9030 @var{class_name}, @var{cat_name} and @var{sel_name} put together, plus
9031 50 characters extra.
9033 The argument @var{is_inst} specifies whether the method is an instance
9034 method or a class method; @var{class_name} is the name of the class;
9035 @var{cat_name} is the name of the category (or @code{NULL} if the method is not
9036 in a category); and @var{sel_name} is the name of the selector.
9038 On systems where the assembler can handle quoted names, you can use this
9039 macro to provide more human-readable names.
9042 @node Initialization
9043 @subsection How Initialization Functions Are Handled
9044 @cindex initialization routines
9045 @cindex termination routines
9046 @cindex constructors, output of
9047 @cindex destructors, output of
9049 The compiled code for certain languages includes @dfn{constructors}
9050 (also called @dfn{initialization routines})---functions to initialize
9051 data in the program when the program is started. These functions need
9052 to be called before the program is ``started''---that is to say, before
9053 @code{main} is called.
9055 Compiling some languages generates @dfn{destructors} (also called
9056 @dfn{termination routines}) that should be called when the program
9059 To make the initialization and termination functions work, the compiler
9060 must output something in the assembler code to cause those functions to
9061 be called at the appropriate time. When you port the compiler to a new
9062 system, you need to specify how to do this.
9064 There are two major ways that GCC currently supports the execution of
9065 initialization and termination functions. Each way has two variants.
9066 Much of the structure is common to all four variations.
9068 @findex __CTOR_LIST__
9069 @findex __DTOR_LIST__
9070 The linker must build two lists of these functions---a list of
9071 initialization functions, called @code{__CTOR_LIST__}, and a list of
9072 termination functions, called @code{__DTOR_LIST__}.
9074 Each list always begins with an ignored function pointer (which may hold
9075 0, @minus{}1, or a count of the function pointers after it, depending on
9076 the environment). This is followed by a series of zero or more function
9077 pointers to constructors (or destructors), followed by a function
9078 pointer containing zero.
9080 Depending on the operating system and its executable file format, either
9081 @file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup
9082 time and exit time. Constructors are called in reverse order of the
9083 list; destructors in forward order.
9085 The best way to handle static constructors works only for object file
9086 formats which provide arbitrarily-named sections. A section is set
9087 aside for a list of constructors, and another for a list of destructors.
9088 Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each
9089 object file that defines an initialization function also puts a word in
9090 the constructor section to point to that function. The linker
9091 accumulates all these words into one contiguous @samp{.ctors} section.
9092 Termination functions are handled similarly.
9094 This method will be chosen as the default by @file{target-def.h} if
9095 @code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not
9096 support arbitrary sections, but does support special designated
9097 constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP}
9098 and @code{DTORS_SECTION_ASM_OP} to achieve the same effect.
9100 When arbitrary sections are available, there are two variants, depending
9101 upon how the code in @file{crtstuff.c} is called. On systems that
9102 support a @dfn{.init} section which is executed at program startup,
9103 parts of @file{crtstuff.c} are compiled into that section. The
9104 program is linked by the @command{gcc} driver like this:
9107 ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o
9110 The prologue of a function (@code{__init}) appears in the @code{.init}
9111 section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise
9112 for the function @code{__fini} in the @dfn{.fini} section. Normally these
9113 files are provided by the operating system or by the GNU C library, but
9114 are provided by GCC for a few targets.
9116 The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets)
9117 compiled from @file{crtstuff.c}. They contain, among other things, code
9118 fragments within the @code{.init} and @code{.fini} sections that branch
9119 to routines in the @code{.text} section. The linker will pull all parts
9120 of a section together, which results in a complete @code{__init} function
9121 that invokes the routines we need at startup.
9123 To use this variant, you must define the @code{INIT_SECTION_ASM_OP}
9126 If no init section is available, when GCC compiles any function called
9127 @code{main} (or more accurately, any function designated as a program
9128 entry point by the language front end calling @code{expand_main_function}),
9129 it inserts a procedure call to @code{__main} as the first executable code
9130 after the function prologue. The @code{__main} function is defined
9131 in @file{libgcc2.c} and runs the global constructors.
9133 In file formats that don't support arbitrary sections, there are again
9134 two variants. In the simplest variant, the GNU linker (GNU @code{ld})
9135 and an `a.out' format must be used. In this case,
9136 @code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs}
9137 entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__},
9138 and with the address of the void function containing the initialization
9139 code as its value. The GNU linker recognizes this as a request to add
9140 the value to a @dfn{set}; the values are accumulated, and are eventually
9141 placed in the executable as a vector in the format described above, with
9142 a leading (ignored) count and a trailing zero element.
9143 @code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init
9144 section is available, the absence of @code{INIT_SECTION_ASM_OP} causes
9145 the compilation of @code{main} to call @code{__main} as above, starting
9146 the initialization process.
9148 The last variant uses neither arbitrary sections nor the GNU linker.
9149 This is preferable when you want to do dynamic linking and when using
9150 file formats which the GNU linker does not support, such as `ECOFF'@. In
9151 this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and
9152 termination functions are recognized simply by their names. This requires
9153 an extra program in the linkage step, called @command{collect2}. This program
9154 pretends to be the linker, for use with GCC; it does its job by running
9155 the ordinary linker, but also arranges to include the vectors of
9156 initialization and termination functions. These functions are called
9157 via @code{__main} as described above. In order to use this method,
9158 @code{use_collect2} must be defined in the target in @file{config.gcc}.
9161 The following section describes the specific macros that control and
9162 customize the handling of initialization and termination functions.
9165 @node Macros for Initialization
9166 @subsection Macros Controlling Initialization Routines
9168 Here are the macros that control how the compiler handles initialization
9169 and termination functions:
9171 @defmac INIT_SECTION_ASM_OP
9172 If defined, a C string constant, including spacing, for the assembler
9173 operation to identify the following data as initialization code. If not
9174 defined, GCC will assume such a section does not exist. When you are
9175 using special sections for initialization and termination functions, this
9176 macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to
9177 run the initialization functions.
9180 @defmac HAS_INIT_SECTION
9181 If defined, @code{main} will not call @code{__main} as described above.
9182 This macro should be defined for systems that control start-up code
9183 on a symbol-by-symbol basis, such as OSF/1, and should not
9184 be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}.
9187 @defmac LD_INIT_SWITCH
9188 If defined, a C string constant for a switch that tells the linker that
9189 the following symbol is an initialization routine.
9192 @defmac LD_FINI_SWITCH
9193 If defined, a C string constant for a switch that tells the linker that
9194 the following symbol is a finalization routine.
9197 @defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func})
9198 If defined, a C statement that will write a function that can be
9199 automatically called when a shared library is loaded. The function
9200 should call @var{func}, which takes no arguments. If not defined, and
9201 the object format requires an explicit initialization function, then a
9202 function called @code{_GLOBAL__DI} will be generated.
9204 This function and the following one are used by collect2 when linking a
9205 shared library that needs constructors or destructors, or has DWARF2
9206 exception tables embedded in the code.
9209 @defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func})
9210 If defined, a C statement that will write a function that can be
9211 automatically called when a shared library is unloaded. The function
9212 should call @var{func}, which takes no arguments. If not defined, and
9213 the object format requires an explicit finalization function, then a
9214 function called @code{_GLOBAL__DD} will be generated.
9217 @defmac INVOKE__main
9218 If defined, @code{main} will call @code{__main} despite the presence of
9219 @code{INIT_SECTION_ASM_OP}. This macro should be defined for systems
9220 where the init section is not actually run automatically, but is still
9221 useful for collecting the lists of constructors and destructors.
9224 @defmac SUPPORTS_INIT_PRIORITY
9225 If nonzero, the C++ @code{init_priority} attribute is supported and the
9226 compiler should emit instructions to control the order of initialization
9227 of objects. If zero, the compiler will issue an error message upon
9228 encountering an @code{init_priority} attribute.
9231 @deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS
9232 This value is true if the target supports some ``native'' method of
9233 collecting constructors and destructors to be run at startup and exit.
9234 It is false if we must use @command{collect2}.
9237 @deftypevr {Target Hook} bool TARGET_DTORS_FROM_CXA_ATEXIT
9238 This value is true if the target wants destructors to be queued to be
9239 run from __cxa_atexit. If this is the case then, for each priority level,
9240 a new constructor will be entered that registers the destructors for that
9241 level with __cxa_atexit (and there will be no destructors emitted).
9242 It is false the method implied by @code{have_ctors_dtors} is used.
9245 @deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority})
9246 If defined, a function that outputs assembler code to arrange to call
9247 the function referenced by @var{symbol} at initialization time.
9249 Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking
9250 no arguments and with no return value. If the target supports initialization
9251 priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY};
9252 otherwise it must be @code{DEFAULT_INIT_PRIORITY}.
9254 If this macro is not defined by the target, a suitable default will
9255 be chosen if (1) the target supports arbitrary section names, (2) the
9256 target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2}
9260 @deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority})
9261 This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination
9262 functions rather than initialization functions.
9265 If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine
9266 generated for the generated object file will have static linkage.
9268 If your system uses @command{collect2} as the means of processing
9269 constructors, then that program normally uses @command{nm} to scan
9270 an object file for constructor functions to be called.
9272 On certain kinds of systems, you can define this macro to make
9273 @command{collect2} work faster (and, in some cases, make it work at all):
9275 @defmac OBJECT_FORMAT_COFF
9276 Define this macro if the system uses COFF (Common Object File Format)
9277 object files, so that @command{collect2} can assume this format and scan
9278 object files directly for dynamic constructor/destructor functions.
9280 This macro is effective only in a native compiler; @command{collect2} as
9281 part of a cross compiler always uses @command{nm} for the target machine.
9284 @defmac REAL_NM_FILE_NAME
9285 Define this macro as a C string constant containing the file name to use
9286 to execute @command{nm}. The default is to search the path normally for
9291 @command{collect2} calls @command{nm} to scan object files for static
9292 constructors and destructors and LTO info. By default, @option{-n} is
9293 passed. Define @code{NM_FLAGS} to a C string constant if other options
9294 are needed to get the same output format as GNU @command{nm -n}
9298 If your system supports shared libraries and has a program to list the
9299 dynamic dependencies of a given library or executable, you can define
9300 these macros to enable support for running initialization and
9301 termination functions in shared libraries:
9304 Define this macro to a C string constant containing the name of the program
9305 which lists dynamic dependencies, like @command{ldd} under SunOS 4.
9308 @defmac PARSE_LDD_OUTPUT (@var{ptr})
9309 Define this macro to be C code that extracts filenames from the output
9310 of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable
9311 of type @code{char *} that points to the beginning of a line of output
9312 from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the
9313 code must advance @var{ptr} to the beginning of the filename on that
9314 line. Otherwise, it must set @var{ptr} to @code{NULL}.
9317 @defmac SHLIB_SUFFIX
9318 Define this macro to a C string constant containing the default shared
9319 library extension of the target (e.g., @samp{".so"}). @command{collect2}
9320 strips version information after this suffix when generating global
9321 constructor and destructor names. This define is only needed on targets
9322 that use @command{collect2} to process constructors and destructors.
9325 @node Instruction Output
9326 @subsection Output of Assembler Instructions
9328 @c prevent bad page break with this line
9329 This describes assembler instruction output.
9331 @defmac REGISTER_NAMES
9332 A C initializer containing the assembler's names for the machine
9333 registers, each one as a C string constant. This is what translates
9334 register numbers in the compiler into assembler language.
9337 @defmac ADDITIONAL_REGISTER_NAMES
9338 If defined, a C initializer for an array of structures containing a name
9339 and a register number. This macro defines additional names for hard
9340 registers, thus allowing the @code{asm} option in declarations to refer
9341 to registers using alternate names.
9344 @defmac OVERLAPPING_REGISTER_NAMES
9345 If defined, a C initializer for an array of structures containing a
9346 name, a register number and a count of the number of consecutive
9347 machine registers the name overlaps. This macro defines additional
9348 names for hard registers, thus allowing the @code{asm} option in
9349 declarations to refer to registers using alternate names. Unlike
9350 @code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the
9351 register name implies multiple underlying registers.
9353 This macro should be used when it is important that a clobber in an
9354 @code{asm} statement clobbers all the underlying values implied by the
9355 register name. For example, on ARM, clobbering the double-precision
9356 VFP register ``d0'' implies clobbering both single-precision registers
9360 @defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr})
9361 Define this macro if you are using an unusual assembler that
9362 requires different names for the machine instructions.
9364 The definition is a C statement or statements which output an
9365 assembler instruction opcode to the stdio stream @var{stream}. The
9366 macro-operand @var{ptr} is a variable of type @code{char *} which
9367 points to the opcode name in its ``internal'' form---the form that is
9368 written in the machine description. The definition should output the
9369 opcode name to @var{stream}, performing any translation you desire, and
9370 increment the variable @var{ptr} to point at the end of the opcode
9371 so that it will not be output twice.
9373 In fact, your macro definition may process less than the entire opcode
9374 name, or more than the opcode name; but if you want to process text
9375 that includes @samp{%}-sequences to substitute operands, you must take
9376 care of the substitution yourself. Just be sure to increment
9377 @var{ptr} over whatever text should not be output normally.
9379 @findex recog_data.operand
9380 If you need to look at the operand values, they can be found as the
9381 elements of @code{recog_data.operand}.
9383 If the macro definition does nothing, the instruction is output
9387 @defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands})
9388 If defined, a C statement to be executed just prior to the output of
9389 assembler code for @var{insn}, to modify the extracted operands so
9390 they will be output differently.
9392 Here the argument @var{opvec} is the vector containing the operands
9393 extracted from @var{insn}, and @var{noperands} is the number of
9394 elements of the vector which contain meaningful data for this insn.
9395 The contents of this vector are what will be used to convert the insn
9396 template into assembler code, so you can change the assembler output
9397 by changing the contents of the vector.
9399 This macro is useful when various assembler syntaxes share a single
9400 file of instruction patterns; by defining this macro differently, you
9401 can cause a large class of instructions to be output differently (such
9402 as with rearranged operands). Naturally, variations in assembler
9403 syntax affecting individual insn patterns ought to be handled by
9404 writing conditional output routines in those patterns.
9406 If this macro is not defined, it is equivalent to a null statement.
9409 @deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx_insn *@var{insn}, rtx *@var{opvec}, int @var{noperands})
9410 If defined, this target hook is a function which is executed just after the
9411 output of assembler code for @var{insn}, to change the mode of the assembler
9414 Here the argument @var{opvec} is the vector containing the operands
9415 extracted from @var{insn}, and @var{noperands} is the number of
9416 elements of the vector which contain meaningful data for this insn.
9417 The contents of this vector are what was used to convert the insn
9418 template into assembler code, so you can change the assembler mode
9419 by checking the contents of the vector.
9422 @defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code})
9423 A C compound statement to output to stdio stream @var{stream} the
9424 assembler syntax for an instruction operand @var{x}. @var{x} is an
9427 @var{code} is a value that can be used to specify one of several ways
9428 of printing the operand. It is used when identical operands must be
9429 printed differently depending on the context. @var{code} comes from
9430 the @samp{%} specification that was used to request printing of the
9431 operand. If the specification was just @samp{%@var{digit}} then
9432 @var{code} is 0; if the specification was @samp{%@var{ltr}
9433 @var{digit}} then @var{code} is the ASCII code for @var{ltr}.
9436 If @var{x} is a register, this macro should print the register's name.
9437 The names can be found in an array @code{reg_names} whose type is
9438 @code{char *[]}. @code{reg_names} is initialized from
9439 @code{REGISTER_NAMES}.
9441 When the machine description has a specification @samp{%@var{punct}}
9442 (a @samp{%} followed by a punctuation character), this macro is called
9443 with a null pointer for @var{x} and the punctuation character for
9447 @defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code})
9448 A C expression which evaluates to true if @var{code} is a valid
9449 punctuation character for use in the @code{PRINT_OPERAND} macro. If
9450 @code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no
9451 punctuation characters (except for the standard one, @samp{%}) are used
9455 @defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x})
9456 A C compound statement to output to stdio stream @var{stream} the
9457 assembler syntax for an instruction operand that is a memory reference
9458 whose address is @var{x}. @var{x} is an RTL expression.
9460 @cindex @code{TARGET_ENCODE_SECTION_INFO} usage
9461 On some machines, the syntax for a symbolic address depends on the
9462 section that the address refers to. On these machines, define the hook
9463 @code{TARGET_ENCODE_SECTION_INFO} to store the information into the
9464 @code{symbol_ref}, and then check for it here. @xref{Assembler
9468 @findex dbr_sequence_length
9469 @defmac DBR_OUTPUT_SEQEND (@var{file})
9470 A C statement, to be executed after all slot-filler instructions have
9471 been output. If necessary, call @code{dbr_sequence_length} to
9472 determine the number of slots filled in a sequence (zero if not
9473 currently outputting a sequence), to decide how many no-ops to output,
9476 Don't define this macro if it has nothing to do, but it is helpful in
9477 reading assembly output if the extent of the delay sequence is made
9478 explicit (e.g.@: with white space).
9481 @findex final_sequence
9482 Note that output routines for instructions with delay slots must be
9483 prepared to deal with not being output as part of a sequence
9484 (i.e.@: when the scheduling pass is not run, or when no slot fillers could be
9485 found.) The variable @code{final_sequence} is null when not
9486 processing a sequence, otherwise it contains the @code{sequence} rtx
9490 @defmac REGISTER_PREFIX
9491 @defmacx LOCAL_LABEL_PREFIX
9492 @defmacx USER_LABEL_PREFIX
9493 @defmacx IMMEDIATE_PREFIX
9494 If defined, C string expressions to be used for the @samp{%R}, @samp{%L},
9495 @samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see
9496 @file{final.cc}). These are useful when a single @file{md} file must
9497 support multiple assembler formats. In that case, the various @file{tm.h}
9498 files can define these macros differently.
9501 @defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format})
9502 If defined this macro should expand to a series of @code{case}
9503 statements which will be parsed inside the @code{switch} statement of
9504 the @code{asm_fprintf} function. This allows targets to define extra
9505 printf formats which may useful when generating their assembler
9506 statements. Note that uppercase letters are reserved for future
9507 generic extensions to asm_fprintf, and so are not available to target
9508 specific code. The output file is given by the parameter @var{file}.
9509 The varargs input pointer is @var{argptr} and the rest of the format
9510 string, starting the character after the one that is being switched
9511 upon, is pointed to by @var{format}.
9514 @defmac ASSEMBLER_DIALECT
9515 If your target supports multiple dialects of assembler language (such as
9516 different opcodes), define this macro as a C expression that gives the
9517 numeric index of the assembler language dialect to use, with zero as the
9520 If this macro is defined, you may use constructs of the form
9522 @samp{@{option0|option1|option2@dots{}@}}
9525 in the output templates of patterns (@pxref{Output Template}) or in the
9526 first argument of @code{asm_fprintf}. This construct outputs
9527 @samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of
9528 @code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters
9529 within these strings retain their usual meaning. If there are fewer
9530 alternatives within the braces than the value of
9531 @code{ASSEMBLER_DIALECT}, the construct outputs nothing. If it's needed
9532 to print curly braces or @samp{|} character in assembler output directly,
9533 @samp{%@{}, @samp{%@}} and @samp{%|} can be used.
9535 If you do not define this macro, the characters @samp{@{}, @samp{|} and
9536 @samp{@}} do not have any special meaning when used in templates or
9537 operands to @code{asm_fprintf}.
9539 Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX},
9540 @code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express
9541 the variations in assembler language syntax with that mechanism. Define
9542 @code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax
9543 if the syntax variant are larger and involve such things as different
9544 opcodes or operand order.
9547 @defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno})
9548 A C expression to output to @var{stream} some assembler code
9549 which will push hard register number @var{regno} onto the stack.
9550 The code need not be optimal, since this macro is used only when
9554 @defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno})
9555 A C expression to output to @var{stream} some assembler code
9556 which will pop hard register number @var{regno} off of the stack.
9557 The code need not be optimal, since this macro is used only when
9561 @node Dispatch Tables
9562 @subsection Output of Dispatch Tables
9564 @c prevent bad page break with this line
9565 This concerns dispatch tables.
9567 @cindex dispatch table
9568 @defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel})
9569 A C statement to output to the stdio stream @var{stream} an assembler
9570 pseudo-instruction to generate a difference between two labels.
9571 @var{value} and @var{rel} are the numbers of two internal labels. The
9572 definitions of these labels are output using
9573 @code{(*targetm.asm_out.internal_label)}, and they must be printed in the same
9574 way here. For example,
9577 fprintf (@var{stream}, "\t.word L%d-L%d\n",
9578 @var{value}, @var{rel})
9581 You must provide this macro on machines where the addresses in a
9582 dispatch table are relative to the table's own address. If defined, GCC
9583 will also use this macro on all machines when producing PIC@.
9584 @var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the
9585 mode and flags can be read.
9588 @defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value})
9589 This macro should be provided on machines where the addresses
9590 in a dispatch table are absolute.
9592 The definition should be a C statement to output to the stdio stream
9593 @var{stream} an assembler pseudo-instruction to generate a reference to
9594 a label. @var{value} is the number of an internal label whose
9595 definition is output using @code{(*targetm.asm_out.internal_label)}.
9599 fprintf (@var{stream}, "\t.word L%d\n", @var{value})
9603 @defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table})
9604 Define this if the label before a jump-table needs to be output
9605 specially. The first three arguments are the same as for
9606 @code{(*targetm.asm_out.internal_label)}; the fourth argument is the
9607 jump-table which follows (a @code{jump_table_data} containing an
9608 @code{addr_vec} or @code{addr_diff_vec}).
9610 This feature is used on system V to output a @code{swbeg} statement
9613 If this macro is not defined, these labels are output with
9614 @code{(*targetm.asm_out.internal_label)}.
9617 @defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table})
9618 Define this if something special must be output at the end of a
9619 jump-table. The definition should be a C statement to be executed
9620 after the assembler code for the table is written. It should write
9621 the appropriate code to stdio stream @var{stream}. The argument
9622 @var{table} is the jump-table insn, and @var{num} is the label-number
9623 of the preceding label.
9625 If this macro is not defined, nothing special is output at the end of
9629 @deftypefn {Target Hook} void TARGET_ASM_POST_CFI_STARTPROC (FILE *@var{}, @var{tree})
9630 This target hook is used to emit assembly strings required by the target
9631 after the .cfi_startproc directive. The first argument is the file stream to
9632 write the strings to and the second argument is the function's declaration. The
9633 expected use is to add more .cfi_* directives.
9635 The default is to not output any assembly strings.
9638 @deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty})
9639 This target hook emits a label at the beginning of each FDE@. It
9640 should be defined on targets where FDEs need special labels, and it
9641 should write the appropriate label, for the FDE associated with the
9642 function declaration @var{decl}, to the stdio stream @var{stream}.
9643 The third argument, @var{for_eh}, is a boolean: true if this is for an
9644 exception table. The fourth argument, @var{empty}, is a boolean:
9645 true if this is a placeholder label for an omitted FDE@.
9647 The default is that FDEs are not given nonlocal labels.
9650 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream})
9651 This target hook emits a label at the beginning of the exception table.
9652 It should be defined on targets where it is desirable for the table
9653 to be broken up according to function.
9655 The default is that no label is emitted.
9658 @deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality})
9659 If the target implements @code{TARGET_ASM_UNWIND_EMIT}, this hook may be
9660 used to emit a directive to install a personality hook into the unwind
9661 info. This hook should not be used if dwarf2 unwind info is used.
9664 @deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx_insn *@var{insn})
9665 This target hook emits assembly directives required to unwind the
9666 given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO}
9667 returns @code{UI_TARGET}.
9670 @deftypefn {Target Hook} rtx TARGET_ASM_MAKE_EH_SYMBOL_INDIRECT (rtx @var{origsymbol}, bool @var{pubvis})
9671 If necessary, modify personality and LSDA references to handle indirection.
9672 The original symbol is in @code{origsymbol} and if @code{pubvis} is true
9673 the symbol is visible outside the TU.
9676 @deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN
9677 True if the @code{TARGET_ASM_UNWIND_EMIT} hook should be called before
9678 the assembly for @var{insn} has been emitted, false if the hook should
9679 be called afterward.
9682 @deftypefn {Target Hook} bool TARGET_ASM_SHOULD_RESTORE_CFA_STATE (void)
9683 For DWARF-based unwind frames, two CFI instructions provide for save and
9684 restore of register state. GCC maintains the current frame address (CFA)
9685 separately from the register bank but the unwinder in libgcc preserves this
9686 state along with the registers (and this is expected by the code that writes
9687 the unwind frames). This hook allows the target to specify that the CFA data
9688 is not saved/restored along with the registers by the target unwinder so that
9689 suitable additional instructions should be emitted to restore it.
9692 @node Exception Region Output
9693 @subsection Assembler Commands for Exception Regions
9695 @c prevent bad page break with this line
9697 This describes commands marking the start and the end of an exception
9700 @defmac EH_FRAME_SECTION_NAME
9701 If defined, a C string constant for the name of the section containing
9702 exception handling frame unwind information. If not defined, GCC will
9703 provide a default definition if the target supports named sections.
9704 @file{crtstuff.c} uses this macro to switch to the appropriate section.
9706 You should define this symbol if your target supports DWARF 2 frame
9707 unwind information and the default definition does not work.
9710 @defmac EH_FRAME_THROUGH_COLLECT2
9711 If defined, DWARF 2 frame unwind information will identified by
9712 specially named labels. The collect2 process will locate these
9713 labels and generate code to register the frames.
9715 This might be necessary, for instance, if the system linker will not
9716 place the eh_frames in-between the sentinals from @file{crtstuff.c},
9717 or if the system linker does garbage collection and sections cannot
9718 be marked as not to be collected.
9721 @defmac EH_TABLES_CAN_BE_READ_ONLY
9722 Define this macro to 1 if your target is such that no frame unwind
9723 information encoding used with non-PIC code will ever require a
9724 runtime relocation, but the linker may not support merging read-only
9725 and read-write sections into a single read-write section.
9728 @defmac MASK_RETURN_ADDR
9729 An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so
9730 that it does not contain any extraneous set bits in it.
9733 @defmac DWARF2_UNWIND_INFO
9734 Define this macro to 0 if your target supports DWARF 2 frame unwind
9735 information, but it does not yet work with exception handling.
9736 Otherwise, if your target supports this information (if it defines
9737 @code{INCOMING_RETURN_ADDR_RTX} and @code{OBJECT_FORMAT_ELF}),
9738 GCC will provide a default definition of 1.
9741 @deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts})
9742 This hook defines the mechanism that will be used for exception handling
9743 by the target. If the target has ABI specified unwind tables, the hook
9744 should return @code{UI_TARGET}. If the target is to use the
9745 @code{setjmp}/@code{longjmp}-based exception handling scheme, the hook
9746 should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind
9747 information, the hook should return @code{UI_DWARF2}.
9749 A target may, if exceptions are disabled, choose to return @code{UI_NONE}.
9750 This may end up simplifying other parts of target-specific code. The
9751 default implementation of this hook never returns @code{UI_NONE}.
9753 Note that the value returned by this hook should be constant. It should
9754 not depend on anything except the command-line switches described by
9755 @var{opts}. In particular, the
9756 setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor
9757 macros and builtin functions related to exception handling are set up
9758 depending on this setting.
9760 The default implementation of the hook first honors the
9761 @option{--enable-sjlj-exceptions} configure option, then
9762 @code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If
9763 @code{DWARF2_UNWIND_INFO} depends on command-line options, the target
9764 must define this hook so that @var{opts} is used correctly.
9767 @deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT
9768 This variable should be set to @code{true} if the target ABI requires unwinding
9769 tables even when exceptions are not used. It must not be modified by
9770 command-line option processing.
9773 @defmac DONT_USE_BUILTIN_SETJMP
9774 Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme
9775 should use the @code{setjmp}/@code{longjmp} functions from the C library
9776 instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery.
9779 @defmac JMP_BUF_SIZE
9780 This macro has no effect unless @code{DONT_USE_BUILTIN_SETJMP} is also
9781 defined. Define this macro if the default size of @code{jmp_buf} buffer
9782 for the @code{setjmp}/@code{longjmp}-based exception handling mechanism
9783 is not large enough, or if it is much too large.
9784 The default size is @code{FIRST_PSEUDO_REGISTER * sizeof(void *)}.
9787 @defmac DWARF_CIE_DATA_ALIGNMENT
9788 This macro need only be defined if the target might save registers in the
9789 function prologue at an offset to the stack pointer that is not aligned to
9790 @code{UNITS_PER_WORD}. The definition should be the negative minimum
9791 alignment if @code{STACK_GROWS_DOWNWARD} is true, and the positive
9792 minimum alignment otherwise. @xref{DWARF}. Only applicable if
9793 the target supports DWARF 2 frame unwind information.
9796 @deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO
9797 Contains the value true if the target should add a zero word onto the
9798 end of a Dwarf-2 frame info section when used for exception handling.
9799 Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and
9803 @deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg})
9804 Given a register, this hook should return a parallel of registers to
9805 represent where to find the register pieces. Define this hook if the
9806 register and its mode are represented in Dwarf in non-contiguous
9807 locations, or if the register should be represented in more than one
9808 register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}.
9809 If not defined, the default is to return @code{NULL_RTX}.
9812 @deftypefn {Target Hook} machine_mode TARGET_DWARF_FRAME_REG_MODE (int @var{regno})
9813 Given a register, this hook should return the mode which the
9814 corresponding Dwarf frame register should have. This is normally
9815 used to return a smaller mode than the raw mode to prevent call
9816 clobbered parts of a register altering the frame register size
9819 @deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address})
9820 If some registers are represented in Dwarf-2 unwind information in
9821 multiple pieces, define this hook to fill in information about the
9822 sizes of those pieces in the table used by the unwinder at runtime.
9823 It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after
9824 filling in a single size corresponding to each hard register;
9825 @var{address} is the address of the table.
9828 @deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym})
9829 This hook is used to output a reference from a frame unwinding table to
9830 the type_info object identified by @var{sym}. It should return @code{true}
9831 if the reference was output. Returning @code{false} will cause the
9832 reference to be output using the normal Dwarf2 routines.
9835 @deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER
9836 This flag should be set to @code{true} on targets that use an ARM EABI
9837 based unwinding library, and @code{false} on other targets. This effects
9838 the format of unwinding tables, and how the unwinder in entered after
9839 running a cleanup. The default is @code{false}.
9842 @node Alignment Output
9843 @subsection Assembler Commands for Alignment
9845 @c prevent bad page break with this line
9846 This describes commands for alignment.
9848 @defmac JUMP_ALIGN (@var{label})
9849 The alignment (log base 2) to put in front of @var{label}, which is
9850 a common destination of jumps and has no fallthru incoming edge.
9852 This macro need not be defined if you don't want any special alignment
9853 to be done at such a time. Most machine descriptions do not currently
9856 Unless it's necessary to inspect the @var{label} parameter, it is better
9857 to set the variable @var{align_jumps} in the target's
9858 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9859 selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation.
9862 @defmac LABEL_ALIGN_AFTER_BARRIER (@var{label})
9863 The alignment (log base 2) to put in front of @var{label}, which follows
9866 This macro need not be defined if you don't want any special alignment
9867 to be done at such a time. Most machine descriptions do not currently
9871 @defmac LOOP_ALIGN (@var{label})
9872 The alignment (log base 2) to put in front of @var{label} that heads
9873 a frequently executed basic block (usually the header of a loop).
9875 This macro need not be defined if you don't want any special alignment
9876 to be done at such a time. Most machine descriptions do not currently
9879 Unless it's necessary to inspect the @var{label} parameter, it is better
9880 to set the variable @code{align_loops} in the target's
9881 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9882 selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation.
9885 @defmac LABEL_ALIGN (@var{label})
9886 The alignment (log base 2) to put in front of @var{label}.
9887 If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment,
9888 the maximum of the specified values is used.
9890 Unless it's necessary to inspect the @var{label} parameter, it is better
9891 to set the variable @code{align_labels} in the target's
9892 @code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's
9893 selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation.
9896 @defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes})
9897 A C statement to output to the stdio stream @var{stream} an assembler
9898 instruction to advance the location counter by @var{nbytes} bytes.
9899 Those bytes should be zero when loaded. @var{nbytes} will be a C
9900 expression of type @code{unsigned HOST_WIDE_INT}.
9903 @defmac ASM_NO_SKIP_IN_TEXT
9904 Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the
9905 text section because it fails to put zeros in the bytes that are skipped.
9906 This is true on many Unix systems, where the pseudo--op to skip bytes
9907 produces no-op instructions rather than zeros when used in the text
9911 @defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power})
9912 A C statement to output to the stdio stream @var{stream} an assembler
9913 command to advance the location counter to a multiple of 2 to the
9914 @var{power} bytes. @var{power} will be a C expression of type @code{int}.
9917 @defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power})
9918 Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used
9919 for padding, if necessary.
9922 @defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip})
9923 A C statement to output to the stdio stream @var{stream} an assembler
9924 command to advance the location counter to a multiple of 2 to the
9925 @var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to
9926 satisfy the alignment request. @var{power} and @var{max_skip} will be
9927 a C expression of type @code{int}.
9931 @node Debugging Info
9932 @section Controlling Debugging Information Format
9934 @c prevent bad page break with this line
9935 This describes how to specify debugging information.
9938 * All Debuggers:: Macros that affect all debugging formats uniformly.
9939 * DWARF:: Macros for DWARF format.
9940 * VMS Debug:: Macros for VMS debug format.
9941 * CTF Debug:: Macros for CTF debug format.
9942 * BTF Debug:: Macros for BTF debug format.
9946 @subsection Macros Affecting All Debugging Formats
9948 @c prevent bad page break with this line
9949 These macros affect all debugging formats.
9951 @defmac DEBUGGER_REGISTER_NUMBER (@var{regno})
9952 A C expression that returns the debugger register number for the compiler
9953 register number @var{regno}. In the default macro provided, the value
9954 of this expression will be @var{regno} itself. But sometimes there are
9955 some registers that the compiler knows about and debugger does not, or vice
9956 versa. In such cases, some register may need to have one number in the
9957 compiler and another for debugger@.
9959 If two registers have consecutive numbers inside GCC, and they can be
9960 used as a pair to hold a multiword value, then they @emph{must} have
9961 consecutive numbers after renumbering with @code{DEBUGGER_REGISTER_NUMBER}.
9962 Otherwise, debuggers will be unable to access such a pair, because they
9963 expect register pairs to be consecutive in their own numbering scheme.
9965 If you find yourself defining @code{DEBUGGER_REGISTER_NUMBER} in way that
9966 does not preserve register pairs, then what you must do instead is
9967 redefine the actual register numbering scheme.
9970 @defmac DEBUGGER_AUTO_OFFSET (@var{x})
9971 A C expression that returns the integer offset value for an automatic
9972 variable having address @var{x} (an RTL expression). The default
9973 computation assumes that @var{x} is based on the frame-pointer and
9974 gives the offset from the frame-pointer. This is required for targets
9975 that produce debugging output for debugger and allow the frame-pointer to be
9976 eliminated when the @option{-g} option is used.
9979 @defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x})
9980 A C expression that returns the integer offset value for an argument
9981 having address @var{x} (an RTL expression). The nominal offset is
9985 @defmac PREFERRED_DEBUGGING_TYPE
9986 A C expression that returns the type of debugging output GCC should
9987 produce when the user specifies just @option{-g}. Define
9988 this if you have arranged for GCC to support more than one format of
9989 debugging output. Currently, the allowable values are
9990 @code{DWARF2_DEBUG}, @code{VMS_DEBUG},
9991 and @code{VMS_AND_DWARF2_DEBUG}.
9993 When the user specifies @option{-ggdb}, GCC normally also uses the
9994 value of this macro to select the debugging output format, but with two
9995 exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the
9996 value @code{DWARF2_DEBUG}.
9998 The value of this macro only affects the default debugging output; the
9999 user can always get a specific type of output by using @option{-gdwarf-2},
10003 @defmac DEFAULT_GDB_EXTENSIONS
10004 Define this macro to control whether GCC should by default generate
10005 GDB's extended version of debugging information. If you don't define the
10006 macro, the default is 1: always generate the extended information
10007 if there is any occasion to.
10012 @subsection Macros for DWARF Output
10014 @c prevent bad page break with this line
10015 Here are macros for DWARF output.
10017 @defmac DWARF2_DEBUGGING_INFO
10018 Define this macro if GCC should produce dwarf version 2 format
10019 debugging output in response to the @option{-g} option.
10021 To support optional call frame debugging information, you must also
10022 define @code{INCOMING_RETURN_ADDR_RTX} and either set
10023 @code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the
10024 prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save}
10025 as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't.
10028 @deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function})
10029 Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to
10030 be emitted for each function. Instead of an integer return the enum
10031 value for the @code{DW_CC_} tag.
10034 @defmac DWARF2_FRAME_INFO
10035 Define this macro to a nonzero value if GCC should always output
10036 Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO}
10037 (@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and
10038 exceptions are enabled, GCC will output this information not matter
10039 how you define @code{DWARF2_FRAME_INFO}.
10042 @deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void)
10043 This hook defines the mechanism that will be used for describing frame
10044 unwind information to the debugger. Normally the hook will return
10045 @code{UI_DWARF2} if DWARF 2 debug information is enabled, and
10046 return @code{UI_NONE} otherwise.
10048 A target may return @code{UI_DWARF2} even when DWARF 2 debug information
10049 is disabled in order to always output DWARF 2 frame information.
10051 A target may return @code{UI_TARGET} if it has ABI specified unwind tables.
10052 This will suppress generation of the normal debug frame unwind information.
10055 @defmac DWARF2_ASM_LINE_DEBUG_INFO
10056 Define this macro to be a nonzero value if the assembler can generate Dwarf 2
10057 line debug info sections. This will result in much more compact line number
10058 tables, and hence is desirable if it works.
10061 @defmac DWARF2_ASM_VIEW_DEBUG_INFO
10062 Define this macro to be a nonzero value if the assembler supports view
10063 assignment and verification in @code{.loc}. If it does not, but the
10064 user enables location views, the compiler may have to fallback to
10065 internal line number tables.
10068 @deftypefn {Target Hook} int TARGET_RESET_LOCATION_VIEW (rtx_insn *@var{})
10069 This hook, if defined, enables -ginternal-reset-location-views, and
10070 uses its result to override cases in which the estimated min insn
10071 length might be nonzero even when a PC advance (i.e., a view reset)
10072 cannot be taken for granted.
10074 If the hook is defined, it must return a positive value to indicate
10075 the insn definitely advances the PC, and so the view number can be
10076 safely assumed to be reset; a negative value to mean the insn
10077 definitely does not advance the PC, and os the view number must not
10078 be reset; or zero to decide based on the estimated insn length.
10080 If insn length is to be regarded as reliable, set the hook to
10081 @code{hook_int_rtx_insn_0}.
10084 @deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS
10085 True if the @code{.debug_pubtypes} and @code{.debug_pubnames} sections
10086 should be emitted. These sections are not used on most platforms, and
10087 in particular GDB does not use them.
10090 @deftypevr {Target Hook} bool TARGET_DELAY_SCHED2
10091 True if sched2 is not to be run at its normal place.
10092 This usually means it will be run as part of machine-specific reorg.
10095 @deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK
10096 True if vartrack is not to be run at its normal place.
10097 This usually means it will be run as part of machine-specific reorg.
10100 @deftypevr {Target Hook} bool TARGET_NO_REGISTER_ALLOCATION
10101 True if register allocation and the passes
10102 following it should not be run. Usually true only for virtual assembler
10106 @defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
10107 A C statement to issue assembly directives that create a difference
10108 @var{lab1} minus @var{lab2}, using an integer of the given @var{size}.
10111 @defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2})
10112 A C statement to issue assembly directives that create a difference
10113 between the two given labels in system defined units, e.g.@: instruction
10114 slots on IA64 VMS, using an integer of the given size.
10117 @defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{offset}, @var{section})
10118 A C statement to issue assembly directives that create a
10119 section-relative reference to the given @var{label} plus @var{offset}, using
10120 an integer of the given @var{size}. The label is known to be defined in the
10121 given @var{section}.
10124 @defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label})
10125 A C statement to issue assembly directives that create a self-relative
10126 reference to the given @var{label}, using an integer of the given @var{size}.
10129 @defmac ASM_OUTPUT_DWARF_DATAREL (@var{stream}, @var{size}, @var{label})
10130 A C statement to issue assembly directives that create a reference to the
10131 given @var{label} relative to the dbase, using an integer of the given @var{size}.
10134 @defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label})
10135 A C statement to issue assembly directives that create a reference to
10136 the DWARF table identifier @var{label} from the current section. This
10137 is used on some systems to avoid garbage collecting a DWARF table which
10138 is referenced by a function.
10141 @deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x})
10142 If defined, this target hook is a function which outputs a DTP-relative
10143 reference to the given TLS symbol of the specified size.
10148 @subsection Macros for VMS Debug Format
10150 @c prevent bad page break with this line
10151 Here are macros for VMS debug format.
10153 @defmac VMS_DEBUGGING_INFO
10154 Define this macro if GCC should produce debugging output for VMS
10155 in response to the @option{-g} option. The default behavior for VMS
10156 is to generate minimal debug info for a traceback in the absence of
10157 @option{-g} unless explicitly overridden with @option{-g0}. This
10158 behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and
10159 @code{TARGET_OPTION_OVERRIDE}.
10164 @subsection Macros for CTF Debug Format
10166 @c prevent bad page break with this line
10167 Here are macros for CTF debug format.
10169 @defmac CTF_DEBUGGING_INFO
10170 Define this macro if GCC should produce debugging output in CTF debug
10171 format in response to the @option{-gctf} option.
10176 @subsection Macros for BTF Debug Format
10178 @c prevent bad page break with this line
10179 Here are macros for BTF debug format.
10181 @defmac BTF_DEBUGGING_INFO
10182 Define this macro if GCC should produce debugging output in BTF debug
10183 format in response to the @option{-gbtf} option.
10186 @node Floating Point
10187 @section Cross Compilation and Floating Point
10188 @cindex cross compilation and floating point
10189 @cindex floating point and cross compilation
10191 While all modern machines use twos-complement representation for integers,
10192 there are a variety of representations for floating point numbers. This
10193 means that in a cross-compiler the representation of floating point numbers
10194 in the compiled program may be different from that used in the machine
10195 doing the compilation.
10197 Because different representation systems may offer different amounts of
10198 range and precision, all floating point constants must be represented in
10199 the target machine's format. Therefore, the cross compiler cannot
10200 safely use the host machine's floating point arithmetic; it must emulate
10201 the target's arithmetic. To ensure consistency, GCC always uses
10202 emulation to work with floating point values, even when the host and
10203 target floating point formats are identical.
10205 The following macros are provided by @file{real.h} for the compiler to
10206 use. All parts of the compiler which generate or optimize
10207 floating-point calculations must use these macros. They may evaluate
10208 their operands more than once, so operands must not have side effects.
10210 @defmac REAL_VALUE_TYPE
10211 The C data type to be used to hold a floating point value in the target
10212 machine's format. Typically this is a @code{struct} containing an
10213 array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque
10217 @deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x})
10218 Truncates @var{x} to a signed integer, rounding toward zero.
10221 @deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x})
10222 Truncates @var{x} to an unsigned integer, rounding toward zero. If
10223 @var{x} is negative, returns zero.
10226 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, machine_mode @var{mode})
10227 Converts @var{string} into a floating point number in the target machine's
10228 representation for mode @var{mode}. This routine can handle both
10229 decimal and hexadecimal floating point constants, using the syntax
10230 defined by the C language for both.
10233 @deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x})
10234 Returns 1 if @var{x} is negative (including negative zero), 0 otherwise.
10237 @deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x})
10238 Determines whether @var{x} represents infinity (positive or negative).
10241 @deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x})
10242 Determines whether @var{x} represents a ``NaN'' (not-a-number).
10245 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x})
10246 Returns the negative of the floating point value @var{x}.
10249 @deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x})
10250 Returns the absolute value of @var{x}.
10253 @node Mode Switching
10254 @section Mode Switching Instructions
10255 @cindex mode switching
10256 The following macros control mode switching optimizations:
10258 @defmac OPTIMIZE_MODE_SWITCHING (@var{entity})
10259 Define this macro if the port needs extra instructions inserted for mode
10260 switching in an optimizing compilation.
10262 For an example, the SH4 can perform both single and double precision
10263 floating point operations, but to perform a single precision operation,
10264 the FPSCR PR bit has to be cleared, while for a double precision
10265 operation, this bit has to be set. Changing the PR bit requires a general
10266 purpose register as a scratch register, hence these FPSCR sets have to
10267 be inserted before reload, i.e.@: you cannot put this into instruction emitting
10268 or @code{TARGET_MACHINE_DEPENDENT_REORG}.
10270 You can have multiple entities that are mode-switched, and select at run time
10271 which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should
10272 return nonzero for any @var{entity} that needs mode-switching.
10273 If you define this macro, you also have to define
10274 @code{NUM_MODES_FOR_MODE_SWITCHING}, @code{TARGET_MODE_NEEDED},
10275 @code{TARGET_MODE_PRIORITY} and @code{TARGET_MODE_EMIT}.
10276 @code{TARGET_MODE_AFTER}, @code{TARGET_MODE_ENTRY}, and @code{TARGET_MODE_EXIT}
10280 @defmac NUM_MODES_FOR_MODE_SWITCHING
10281 If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as
10282 initializer for an array of integers. Each initializer element
10283 N refers to an entity that needs mode switching, and specifies the number
10284 of different modes that might need to be set for this entity.
10285 The position of the initializer in the initializer---starting counting at
10286 zero---determines the integer that is used to refer to the mode-switched
10287 entity in question.
10288 In macros that take mode arguments / yield a mode result, modes are
10289 represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode
10290 switch is needed / supplied.
10293 @deftypefn {Target Hook} void TARGET_MODE_EMIT (int @var{entity}, int @var{mode}, int @var{prev_mode}, HARD_REG_SET @var{regs_live})
10294 Generate one or more insns to set @var{entity} to @var{mode}.
10295 @var{hard_reg_live} is the set of hard registers live at the point where
10296 the insn(s) are to be inserted. @var{prev_moxde} indicates the mode
10297 to switch from. Sets of a lower numbered entity will be emitted before
10298 sets of a higher numbered entity to a mode of the same or lower priority.
10301 @deftypefn {Target Hook} int TARGET_MODE_NEEDED (int @var{entity}, rtx_insn *@var{insn})
10302 @var{entity} is an integer specifying a mode-switched entity.
10303 If @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro
10304 to return an integer value not larger than the corresponding element
10305 in @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity}
10306 must be switched into prior to the execution of @var{insn}.
10309 @deftypefn {Target Hook} int TARGET_MODE_AFTER (int @var{entity}, int @var{mode}, rtx_insn *@var{insn})
10310 @var{entity} is an integer specifying a mode-switched entity.
10311 If this macro is defined, it is evaluated for every @var{insn} during mode
10312 switching. It determines the mode that an insn results
10313 in (if different from the incoming mode).
10316 @deftypefn {Target Hook} int TARGET_MODE_ENTRY (int @var{entity})
10317 If this macro is defined, it is evaluated for every @var{entity} that
10318 needs mode switching. It should evaluate to an integer, which is a mode
10319 that @var{entity} is assumed to be switched to at function entry.
10320 If @code{TARGET_MODE_ENTRY} is defined then @code{TARGET_MODE_EXIT}
10324 @deftypefn {Target Hook} int TARGET_MODE_EXIT (int @var{entity})
10325 If this macro is defined, it is evaluated for every @var{entity} that
10326 needs mode switching. It should evaluate to an integer, which is a mode
10327 that @var{entity} is assumed to be switched to at function exit.
10328 If @code{TARGET_MODE_EXIT} is defined then @code{TARGET_MODE_ENTRY}
10332 @deftypefn {Target Hook} int TARGET_MODE_PRIORITY (int @var{entity}, int @var{n})
10333 This macro specifies the order in which modes for @var{entity}
10334 are processed. 0 is the highest priority,
10335 @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the lowest.
10336 The value of the macro should be an integer designating a mode
10337 for @var{entity}. For any fixed @var{entity}, @code{mode_priority}
10338 (@var{entity}, @var{n}) shall be a bijection in 0 @dots{}
10339 @code{num_modes_for_mode_switching[@var{entity}] - 1}.
10342 @node Target Attributes
10343 @section Defining target-specific uses of @code{__attribute__}
10344 @cindex target attributes
10345 @cindex machine attributes
10346 @cindex attributes, target-specific
10348 Target-specific attributes may be defined for functions, data and types.
10349 These are described using the following target hooks; they also need to
10350 be documented in @file{extend.texi}.
10352 @deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE
10353 If defined, this target hook points to an array of @samp{struct
10354 attribute_spec} (defined in @file{tree-core.h}) specifying the machine
10355 specific attributes for this target and some of the restrictions on the
10356 entities to which these attributes are applied and the arguments they
10360 @deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name})
10361 If defined, this target hook is a function which returns true if the
10362 machine-specific attribute named @var{name} expects an identifier
10363 given as its first argument to be passed on as a plain identifier, not
10364 subjected to name lookup. If this is not defined, the default is
10365 false for all machine-specific attributes.
10368 @deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2})
10369 If defined, this target hook is a function which returns zero if the attributes on
10370 @var{type1} and @var{type2} are incompatible, one if they are compatible,
10371 and two if they are nearly compatible (which causes a warning to be
10372 generated). If this is not defined, machine-specific attributes are
10373 supposed always to be compatible.
10376 @deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type})
10377 If defined, this target hook is a function which assigns default attributes to
10378 the newly defined @var{type}.
10381 @deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2})
10382 Define this target hook if the merging of type attributes needs special
10383 handling. If defined, the result is a list of the combined
10384 @code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed
10385 that @code{comptypes} has already been called and returned 1. This
10386 function may call @code{merge_attributes} to handle machine-independent
10390 @deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl})
10391 Define this target hook if the merging of decl attributes needs special
10392 handling. If defined, the result is a list of the combined
10393 @code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}.
10394 @var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of
10395 when this is needed are when one attribute overrides another, or when an
10396 attribute is nullified by a subsequent definition. This function may
10397 call @code{merge_attributes} to handle machine-independent merging.
10399 @findex TARGET_DLLIMPORT_DECL_ATTRIBUTES
10400 If the only target-specific handling you require is @samp{dllimport}
10401 for Microsoft Windows targets, you should define the macro
10402 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler
10403 will then define a function called
10404 @code{merge_dllimport_decl_attributes} which can then be defined as
10405 the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also
10406 add @code{handle_dll_attribute} in the attribute table for your port
10407 to perform initial processing of the @samp{dllimport} and
10408 @samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and
10409 @file{i386/i386.cc}, for example.
10412 @deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl})
10413 @var{decl} is a variable or function with @code{__attribute__((dllimport))}
10414 specified. Use this hook if the target needs to add extra validation
10415 checks to @code{handle_dll_attribute}.
10418 @defmac TARGET_DECLSPEC
10419 Define this macro to a nonzero value if you want to treat
10420 @code{__declspec(X)} as equivalent to @code{__attribute((X))}. By
10421 default, this behavior is enabled only for targets that define
10422 @code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation
10423 of @code{__declspec} is via a built-in macro, but you should not rely
10424 on this implementation detail.
10427 @deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr})
10428 Define this target hook if you want to be able to add attributes to a decl
10429 when it is being created. This is normally useful for back ends which
10430 wish to implement a pragma by using the attributes which correspond to
10431 the pragma's effect. The @var{node} argument is the decl which is being
10432 created. The @var{attr_ptr} argument is a pointer to the attribute list
10433 for this decl. The list itself should not be modified, since it may be
10434 shared with other decls, but attributes may be chained on the head of
10435 the list and @code{*@var{attr_ptr}} modified to point to the new
10436 attributes, or a copy of the list may be made if further changes are
10440 @deftypefn {Target Hook} tree TARGET_HANDLE_GENERIC_ATTRIBUTE (tree *@var{node}, tree @var{name}, tree @var{args}, int @var{flags}, bool *@var{no_add_attrs})
10441 Define this target hook if you want to be able to perform additional
10442 target-specific processing of an attribute which is handled generically
10443 by a front end. The arguments are the same as those which are passed to
10444 attribute handlers. So far this only affects the @var{noinit} and
10445 @var{section} attribute.
10448 @deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl})
10450 This target hook returns @code{true} if it is OK to inline @var{fndecl}
10451 into the current function, despite its having target-specific
10452 attributes, @code{false} otherwise. By default, if a function has a
10453 target specific attribute attached to it, it will not be inlined.
10456 @deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags})
10457 This hook is called to parse @code{attribute(target("..."))}, which
10458 allows setting target-specific options on individual functions.
10459 These function-specific options may differ
10460 from the options specified on the command line. The hook should return
10461 @code{true} if the options are valid.
10463 The hook should set the @code{DECL_FUNCTION_SPECIFIC_TARGET} field in
10464 the function declaration to hold a pointer to a target-specific
10465 @code{struct cl_target_option} structure.
10468 @deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr}, struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set})
10469 This hook is called to save any additional target-specific information
10470 in the @code{struct cl_target_option} structure for function-specific
10471 options from the @code{struct gcc_options} structure.
10472 @xref{Option file format}.
10475 @deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set}, struct cl_target_option *@var{ptr})
10476 This hook is called to restore any additional target-specific
10477 information in the @code{struct cl_target_option} structure for
10478 function-specific options to the @code{struct gcc_options} structure.
10481 @deftypefn {Target Hook} void TARGET_OPTION_POST_STREAM_IN (struct cl_target_option *@var{ptr})
10482 This hook is called to update target-specific information in the
10483 @code{struct cl_target_option} structure after it is streamed in from
10487 @deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr})
10488 This hook is called to print any additional target-specific
10489 information in the @code{struct cl_target_option} structure for
10490 function-specific options.
10493 @deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target})
10494 This target hook parses the options for @code{#pragma GCC target}, which
10495 sets the target-specific options for functions that occur later in the
10496 input stream. The options accepted should be the same as those handled by the
10497 @code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook.
10500 @deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void)
10501 Sometimes certain combinations of command options do not make sense on
10502 a particular target machine. You can override the hook
10503 @code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called
10504 once just after all the command options have been parsed.
10506 Don't use this hook to turn on various extra optimizations for
10507 @option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for.
10509 If you need to do something whenever the optimization level is
10510 changed via the optimize attribute or pragma, see
10511 @code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}
10514 @deftypefn {Target Hook} bool TARGET_OPTION_FUNCTION_VERSIONS (tree @var{decl1}, tree @var{decl2})
10515 This target hook returns @code{true} if @var{DECL1} and @var{DECL2} are
10516 versions of the same function. @var{DECL1} and @var{DECL2} are function
10517 versions if and only if they have the same function signature and
10518 different target specific attributes, that is, they are compiled for
10519 different target machines.
10522 @deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee})
10523 This target hook returns @code{false} if the @var{caller} function
10524 cannot inline @var{callee}, based on target specific information. By
10525 default, inlining is not allowed if the callee function has function
10526 specific target options and the caller does not use the same options.
10529 @deftypefn {Target Hook} bool TARGET_UPDATE_IPA_FN_TARGET_INFO (unsigned int& @var{info}, const gimple* @var{stmt})
10530 Allow target to analyze all gimple statements for the given function to
10531 record and update some target specific information for inlining. A typical
10532 example is that a caller with one isa feature disabled is normally not
10533 allowed to inline a callee with that same isa feature enabled even which is
10534 attributed by always_inline, but with the conservative analysis on all
10535 statements of the callee if we are able to guarantee the callee does not
10536 exploit any instructions from the mismatch isa feature, it would be safe to
10537 allow the caller to inline the callee.
10538 @var{info} is one @code{unsigned int} value to record information in which
10539 one set bit indicates one corresponding feature is detected in the analysis,
10540 @var{stmt} is the statement being analyzed. Return true if target still
10541 need to analyze the subsequent statements, otherwise return false to stop
10542 subsequent analysis.
10543 The default version of this hook returns false.
10546 @deftypefn {Target Hook} bool TARGET_NEED_IPA_FN_TARGET_INFO (const_tree @var{decl}, unsigned int& @var{info})
10547 Allow target to check early whether it is necessary to analyze all gimple
10548 statements in the given function to update target specific information for
10549 inlining. See hook @code{update_ipa_fn_target_info} for usage example of
10550 target specific information. This hook is expected to be invoked ahead of
10551 the iterating with hook @code{update_ipa_fn_target_info}.
10552 @var{decl} is the function being analyzed, @var{info} is the same as what
10553 in hook @code{update_ipa_fn_target_info}, target can do one time update
10554 into @var{info} without iterating for some case. Return true if target
10555 decides to analyze all gimple statements to collect information, otherwise
10557 The default version of this hook returns false.
10560 @deftypefn {Target Hook} void TARGET_RELAYOUT_FUNCTION (tree @var{fndecl})
10561 This target hook fixes function @var{fndecl} after attributes are processed.
10562 Default does nothing. On ARM, the default function's alignment is updated
10563 with the attribute target.
10567 @section Emulating TLS
10568 @cindex Emulated TLS
10570 For targets whose psABI does not provide Thread Local Storage via
10571 specific relocations and instruction sequences, an emulation layer is
10572 used. A set of target hooks allows this emulation layer to be
10573 configured for the requirements of a particular target. For instance
10574 the psABI may in fact specify TLS support in terms of an emulation
10577 The emulation layer works by creating a control object for every TLS
10578 object. To access the TLS object, a lookup function is provided
10579 which, when given the address of the control object, will return the
10580 address of the current thread's instance of the TLS object.
10582 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS
10583 Contains the name of the helper function that uses a TLS control
10584 object to locate a TLS instance. The default causes libgcc's
10585 emulated TLS helper function to be used.
10588 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON
10589 Contains the name of the helper function that should be used at
10590 program startup to register TLS objects that are implicitly
10591 initialized to zero. If this is @code{NULL}, all TLS objects will
10592 have explicit initializers. The default causes libgcc's emulated TLS
10593 registration function to be used.
10596 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION
10597 Contains the name of the section in which TLS control variables should
10598 be placed. The default of @code{NULL} allows these to be placed in
10602 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION
10603 Contains the name of the section in which TLS initializers should be
10604 placed. The default of @code{NULL} allows these to be placed in any
10608 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX
10609 Contains the prefix to be prepended to TLS control variable names.
10610 The default of @code{NULL} uses a target-specific prefix.
10613 @deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX
10614 Contains the prefix to be prepended to TLS initializer objects. The
10615 default of @code{NULL} uses a target-specific prefix.
10618 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name})
10619 Specifies a function that generates the FIELD_DECLs for a TLS control
10620 object type. @var{type} is the RECORD_TYPE the fields are for and
10621 @var{name} should be filled with the structure tag, if the default of
10622 @code{__emutls_object} is unsuitable. The default creates a type suitable
10623 for libgcc's emulated TLS function.
10626 @deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr})
10627 Specifies a function that generates the CONSTRUCTOR to initialize a
10628 TLS control object. @var{var} is the TLS control object, @var{decl}
10629 is the TLS object and @var{tmpl_addr} is the address of the
10630 initializer. The default initializes libgcc's emulated TLS control object.
10633 @deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED
10634 Specifies whether the alignment of TLS control variable objects is
10635 fixed and should not be increased as some backends may do to optimize
10636 single objects. The default is false.
10639 @deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS
10640 Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor
10641 may be used to describe emulated TLS control objects.
10644 @node MIPS Coprocessors
10645 @section Defining coprocessor specifics for MIPS targets.
10646 @cindex MIPS coprocessor-definition macros
10648 The MIPS specification allows MIPS implementations to have as many as 4
10649 coprocessors, each with as many as 32 private registers. GCC supports
10650 accessing these registers and transferring values between the registers
10651 and memory using asm-ized variables. For example:
10654 register unsigned int cp0count asm ("c0r1");
10660 (``c0r1'' is the default name of register 1 in coprocessor 0; alternate
10661 names may be added as described below, or the default names may be
10662 overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.)
10664 Coprocessor registers are assumed to be epilogue-used; sets to them will
10665 be preserved even if it does not appear that the register is used again
10666 later in the function.
10668 Another note: according to the MIPS spec, coprocessor 1 (if present) is
10669 the FPU@. One accesses COP1 registers through standard mips
10670 floating-point support; they are not included in this mechanism.
10673 @section Parameters for Precompiled Header Validity Checking
10674 @cindex parameters, precompiled headers
10676 @deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz})
10677 This hook returns a pointer to the data needed by
10678 @code{TARGET_PCH_VALID_P} and sets
10679 @samp{*@var{sz}} to the size of the data in bytes.
10682 @deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz})
10683 This hook checks whether the options used to create a PCH file are
10684 compatible with the current settings. It returns @code{NULL}
10685 if so and a suitable error message if not. Error messages will
10686 be presented to the user and must be localized using @samp{_(@var{msg})}.
10688 @var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY}
10689 when the PCH file was created and @var{sz} is the size of that data in bytes.
10690 It's safe to assume that the data was created by the same version of the
10691 compiler, so no format checking is needed.
10693 The default definition of @code{default_pch_valid_p} should be
10694 suitable for most targets.
10697 @deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags})
10698 If this hook is nonnull, the default implementation of
10699 @code{TARGET_PCH_VALID_P} will use it to check for compatible values
10700 of @code{target_flags}. @var{pch_flags} specifies the value that
10701 @code{target_flags} had when the PCH file was created. The return
10702 value is the same as for @code{TARGET_PCH_VALID_P}.
10705 @deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void)
10706 Called before writing out a PCH file. If the target has some
10707 garbage-collected data that needs to be in a particular state on PCH loads,
10708 it can use this hook to enforce that state. Very few targets need
10709 to do anything here.
10713 @section C++ ABI parameters
10714 @cindex parameters, c++ abi
10716 @deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void)
10717 Define this hook to override the integer type used for guard variables.
10718 These are used to implement one-time construction of static objects. The
10719 default is long_long_integer_type_node.
10722 @deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void)
10723 This hook determines how guard variables are used. It should return
10724 @code{false} (the default) if the first byte should be used. A return value of
10725 @code{true} indicates that only the least significant bit should be used.
10728 @deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type})
10729 This hook returns the size of the cookie to use when allocating an array
10730 whose elements have the indicated @var{type}. Assumes that it is already
10731 known that a cookie is needed. The default is
10732 @code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the
10733 IA64/Generic C++ ABI@.
10736 @deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void)
10737 This hook should return @code{true} if the element size should be stored in
10738 array cookies. The default is to return @code{false}.
10741 @deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export})
10742 If defined by a backend this hook allows the decision made to export
10743 class @var{type} to be overruled. Upon entry @var{import_export}
10744 will contain 1 if the class is going to be exported, @minus{}1 if it is going
10745 to be imported and 0 otherwise. This function should return the
10746 modified value and perform any other actions necessary to support the
10747 backend's targeted operating system.
10750 @deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void)
10751 This hook should return @code{true} if constructors and destructors return
10752 the address of the object created/destroyed. The default is to return
10756 @deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void)
10757 This hook returns true if the key method for a class (i.e., the method
10758 which, if defined in the current translation unit, causes the virtual
10759 table to be emitted) may be an inline function. Under the standard
10760 Itanium C++ ABI the key method may be an inline function so long as
10761 the function is not declared inline in the class definition. Under
10762 some variants of the ABI, an inline function can never be the key
10763 method. The default is to return @code{true}.
10766 @deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl})
10767 @var{decl} is a virtual table, virtual table table, typeinfo object,
10768 or other similar implicit class data object that will be emitted with
10769 external linkage in this translation unit. No ELF visibility has been
10770 explicitly specified. If the target needs to specify a visibility
10771 other than that of the containing class, use this hook to set
10772 @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}.
10775 @deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void)
10776 This hook returns true (the default) if virtual tables and other
10777 similar implicit class data objects are always COMDAT if they have
10778 external linkage. If this hook returns false, then class data for
10779 classes whose virtual table will be emitted in only one translation
10780 unit will not be COMDAT.
10783 @deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void)
10784 This hook returns true (the default) if the RTTI information for
10785 the basic types which is defined in the C++ runtime should always
10786 be COMDAT, false if it should not be COMDAT.
10789 @deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void)
10790 This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI)
10791 should be used to register static destructors when @option{-fuse-cxa-atexit}
10792 is in effect. The default is to return false to use @code{__cxa_atexit}.
10795 @deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void)
10796 This hook returns true if the target @code{atexit} function can be used
10797 in the same manner as @code{__cxa_atexit} to register C++ static
10798 destructors. This requires that @code{atexit}-registered functions in
10799 shared libraries are run in the correct order when the libraries are
10800 unloaded. The default is to return false.
10803 @deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type})
10804 @var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just
10805 been defined. Use this hook to make adjustments to the class (eg, tweak
10806 visibility or perform any other required target modifications).
10809 @deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl})
10810 Return target-specific mangling context of @var{decl} or @code{NULL_TREE}.
10813 @node D Language and ABI
10814 @section D ABI parameters
10815 @cindex parameters, d abi
10817 @deftypefn {D Target Hook} void TARGET_D_CPU_VERSIONS (void)
10818 Declare all environmental version identifiers relating to the target CPU
10819 using the function @code{builtin_version}, which takes a string representing
10820 the name of the version. Version identifiers predefined by this hook apply
10821 to all modules that are being compiled and imported.
10824 @deftypefn {D Target Hook} void TARGET_D_OS_VERSIONS (void)
10825 Similarly to @code{TARGET_D_CPU_VERSIONS}, but is used for versions
10826 relating to the target operating system.
10829 @deftypefn {D Target Hook} void TARGET_D_REGISTER_CPU_TARGET_INFO (void)
10830 Register all target information keys relating to the target CPU using the
10831 function @code{d_add_target_info_handlers}, which takes a
10832 @samp{struct d_target_info_spec} (defined in @file{d/d-target.h}). The keys
10833 added by this hook are made available at compile time by the
10834 @code{__traits(getTargetInfo)} extension, the result is an expression
10835 describing the requested target information.
10838 @deftypefn {D Target Hook} void TARGET_D_REGISTER_OS_TARGET_INFO (void)
10839 Same as @code{TARGET_D_CPU_TARGET_INFO}, but is used for keys relating to
10840 the target operating system.
10843 @deftypevr {D Target Hook} {const char *} TARGET_D_MINFO_SECTION
10844 Contains the name of the section in which module info references should be
10845 placed. This section is expected to be bracketed by two symbols to indicate
10846 the start and end address of the section, so that the runtime library can
10847 collect all modules for each loaded shared library and executable. The
10848 default value of @code{NULL} disables the use of sections altogether.
10851 @deftypevr {D Target Hook} {const char *} TARGET_D_MINFO_START_NAME
10852 If @code{TARGET_D_MINFO_SECTION} is defined, then this must also be defined
10853 as the name of the symbol indicating the start address of the module info
10857 @deftypevr {D Target Hook} {const char *} TARGET_D_MINFO_END_NAME
10858 If @code{TARGET_D_MINFO_SECTION} is defined, then this must also be defined
10859 as the name of the symbol indicating the end address of the module info
10863 @deftypefn {D Target Hook} bool TARGET_D_HAS_STDCALL_CONVENTION (unsigned int *@var{link_system}, unsigned int *@var{link_windows})
10864 Returns @code{true} if the target supports the stdcall calling convention.
10865 The hook should also set @var{link_system} to @code{1} if the @code{stdcall}
10866 attribute should be applied to functions with @code{extern(System)} linkage,
10867 and @var{link_windows} to @code{1} to apply @code{stdcall} to functions with
10868 @code{extern(Windows)} linkage.
10871 @deftypevr {D Target Hook} bool TARGET_D_TEMPLATES_ALWAYS_COMDAT
10872 This flag is true if instantiated functions and variables are always COMDAT
10873 if they have external linkage. If this flag is false, then instantiated
10874 decls will be emitted as weak symbols. The default is @code{false}.
10877 @node Named Address Spaces
10878 @section Adding support for named address spaces
10879 @cindex named address spaces
10881 The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275
10882 standards committee, @cite{Programming Languages - C - Extensions to
10883 support embedded processors}, specifies a syntax for embedded
10884 processors to specify alternate address spaces. You can configure a
10885 GCC port to support section 5.1 of the draft report to add support for
10886 address spaces other than the default address space. These address
10887 spaces are new keywords that are similar to the @code{volatile} and
10888 @code{const} type attributes.
10890 Pointers to named address spaces can have a different size than
10891 pointers to the generic address space.
10893 For example, the SPU port uses the @code{__ea} address space to refer
10894 to memory in the host processor, rather than memory local to the SPU
10895 processor. Access to memory in the @code{__ea} address space involves
10896 issuing DMA operations to move data between the host processor and the
10897 local processor memory address space. Pointers in the @code{__ea}
10898 address space are either 32 bits or 64 bits based on the
10899 @option{-mea32} or @option{-mea64} switches (native SPU pointers are
10902 Internally, address spaces are represented as a small integer in the
10903 range 0 to 15 with address space 0 being reserved for the generic
10906 To register a named address space qualifier keyword with the C front end,
10907 the target may call the @code{c_register_addr_space} routine. For example,
10908 the SPU port uses the following to declare @code{__ea} as the keyword for
10909 named address space #1:
10911 #define ADDR_SPACE_EA 1
10912 c_register_addr_space ("__ea", ADDR_SPACE_EA);
10915 @deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space})
10916 Define this to return the machine mode to use for pointers to
10917 @var{address_space} if the target supports named address spaces.
10918 The default version of this hook returns @code{ptr_mode}.
10921 @deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space})
10922 Define this to return the machine mode to use for addresses in
10923 @var{address_space} if the target supports named address spaces.
10924 The default version of this hook returns @code{Pmode}.
10927 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (scalar_int_mode @var{mode}, addr_space_t @var{as})
10928 Define this to return nonzero if the port can handle pointers
10929 with machine mode @var{mode} to address space @var{as}. This target
10930 hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook,
10931 except that it includes explicit named address space support. The default
10932 version of this hook returns true for the modes returned by either the
10933 @code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE}
10934 target hooks for the given address space.
10937 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{exp}, bool @var{strict}, addr_space_t @var{as})
10938 Define this to return true if @var{exp} is a valid address for mode
10939 @var{mode} in the named address space @var{as}. The @var{strict}
10940 parameter says whether strict addressing is in effect after reload has
10941 finished. This target hook is the same as the
10942 @code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes
10943 explicit named address space support.
10946 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}, addr_space_t @var{as})
10947 Define this to modify an invalid address @var{x} to be a valid address
10948 with mode @var{mode} in the named address space @var{as}. This target
10949 hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook,
10950 except that it includes explicit named address space support.
10953 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset})
10954 Define this to return whether the @var{subset} named address space is
10955 contained within the @var{superset} named address space. Pointers to
10956 a named address space that is a subset of another named address space
10957 will be converted automatically without a cast if used together in
10958 arithmetic operations. Pointers to a superset address space can be
10959 converted to pointers to a subset address space via explicit casts.
10962 @deftypefn {Target Hook} bool TARGET_ADDR_SPACE_ZERO_ADDRESS_VALID (addr_space_t @var{as})
10963 Define this to modify the default handling of address 0 for the
10964 address space. Return true if 0 should be considered a valid address.
10967 @deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type})
10968 Define this to convert the pointer expression represented by the RTL
10969 @var{op} with type @var{from_type} that points to a named address
10970 space to a new pointer expression with type @var{to_type} that points
10971 to a different named address space. When this hook it called, it is
10972 guaranteed that one of the two address spaces is a subset of the other,
10973 as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook.
10976 @deftypefn {Target Hook} int TARGET_ADDR_SPACE_DEBUG (addr_space_t @var{as})
10977 Define this to define how the address space is encoded in dwarf.
10978 The result is the value to be used with @code{DW_AT_address_class}.
10981 @deftypefn {Target Hook} void TARGET_ADDR_SPACE_DIAGNOSE_USAGE (addr_space_t @var{as}, location_t @var{loc})
10982 Define this hook if the availability of an address space depends on
10983 command line options and some diagnostics should be printed when the
10984 address space is used. This hook is called during parsing and allows
10985 to emit a better diagnostic compared to the case where the address space
10986 was not registered with @code{c_register_addr_space}. @var{as} is
10987 the address space as registered with @code{c_register_addr_space}.
10988 @var{loc} is the location of the address space qualifier token.
10989 The default implementation does nothing.
10993 @section Miscellaneous Parameters
10994 @cindex parameters, miscellaneous
10996 @c prevent bad page break with this line
10997 Here are several miscellaneous parameters.
10999 @defmac HAS_LONG_COND_BRANCH
11000 Define this boolean macro to indicate whether or not your architecture
11001 has conditional branches that can span all of memory. It is used in
11002 conjunction with an optimization that partitions hot and cold basic
11003 blocks into separate sections of the executable. If this macro is
11004 set to false, gcc will convert any conditional branches that attempt
11005 to cross between sections into unconditional branches or indirect jumps.
11008 @defmac HAS_LONG_UNCOND_BRANCH
11009 Define this boolean macro to indicate whether or not your architecture
11010 has unconditional branches that can span all of memory. It is used in
11011 conjunction with an optimization that partitions hot and cold basic
11012 blocks into separate sections of the executable. If this macro is
11013 set to false, gcc will convert any unconditional branches that attempt
11014 to cross between sections into indirect jumps.
11017 @defmac CASE_VECTOR_MODE
11018 An alias for a machine mode name. This is the machine mode that
11019 elements of a jump-table should have.
11022 @defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body})
11023 Optional: return the preferred mode for an @code{addr_diff_vec}
11024 when the minimum and maximum offset are known. If you define this,
11025 it enables extra code in branch shortening to deal with @code{addr_diff_vec}.
11026 To make this work, you also have to define @code{INSN_ALIGN} and
11027 make the alignment for @code{addr_diff_vec} explicit.
11028 The @var{body} argument is provided so that the offset_unsigned and scale
11029 flags can be updated.
11032 @defmac CASE_VECTOR_PC_RELATIVE
11033 Define this macro to be a C expression to indicate when jump-tables
11034 should contain relative addresses. You need not define this macro if
11035 jump-tables never contain relative addresses, or jump-tables should
11036 contain relative addresses only when @option{-fPIC} or @option{-fPIC}
11040 @deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void)
11041 This function return the smallest number of different values for which it
11042 is best to use a jump-table instead of a tree of conditional branches.
11043 The default is four for machines with a @code{casesi} instruction and
11044 five otherwise. This is best for most machines.
11047 @defmac WORD_REGISTER_OPERATIONS
11048 Define this macro to 1 if operations between registers with integral mode
11049 smaller than a word are always performed on the entire register. To be
11050 more explicit, if you start with a pair of @code{word_mode} registers with
11051 known values and you do a subword, for example @code{QImode}, addition on
11052 the low part of the registers, then the compiler may consider that the
11053 result has a known value in @code{word_mode} too if the macro is defined
11054 to 1. Most RISC machines have this property and most CISC machines do not.
11057 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_ARITHMETIC_PRECISION (void)
11058 On some RISC architectures with 64-bit registers, the processor also
11059 maintains 32-bit condition codes that make it possible to do real 32-bit
11060 arithmetic, although the operations are performed on the full registers.
11062 On such architectures, defining this hook to 32 tells the compiler to try
11063 using 32-bit arithmetical operations setting the condition codes instead
11064 of doing full 64-bit arithmetic.
11066 More generally, define this hook on RISC architectures if you want the
11067 compiler to try using arithmetical operations setting the condition codes
11068 with a precision lower than the word precision.
11070 You need not define this hook if @code{WORD_REGISTER_OPERATIONS} is not
11074 @defmac LOAD_EXTEND_OP (@var{mem_mode})
11075 Define this macro to be a C expression indicating when insns that read
11076 memory in @var{mem_mode}, an integral mode narrower than a word, set the
11077 bits outside of @var{mem_mode} to be either the sign-extension or the
11078 zero-extension of the data read. Return @code{SIGN_EXTEND} for values
11079 of @var{mem_mode} for which the
11080 insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and
11081 @code{UNKNOWN} for other modes.
11083 This macro is not called with @var{mem_mode} non-integral or with a width
11084 greater than or equal to @code{BITS_PER_WORD}, so you may return any
11085 value in this case. Do not define this macro if it would always return
11086 @code{UNKNOWN}. On machines where this macro is defined, you will normally
11087 define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}.
11089 You may return a non-@code{UNKNOWN} value even if for some hard registers
11090 the sign extension is not performed, if for the @code{REGNO_REG_CLASS}
11091 of these hard registers @code{TARGET_CAN_CHANGE_MODE_CLASS} returns false
11092 when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any
11093 integral mode larger than this but not larger than @code{word_mode}.
11095 You must return @code{UNKNOWN} if for some hard registers that allow this
11096 mode, @code{TARGET_CAN_CHANGE_MODE_CLASS} says that they cannot change to
11097 @code{word_mode}, but that they can change to another integral mode that
11098 is larger then @var{mem_mode} but still smaller than @code{word_mode}.
11101 @defmac SHORT_IMMEDIATES_SIGN_EXTEND
11102 Define this macro to 1 if loading short immediate values into registers sign
11106 @deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (machine_mode @var{mode})
11107 When @option{-ffast-math} is in effect, GCC tries to optimize
11108 divisions by the same divisor, by turning them into multiplications by
11109 the reciprocal. This target hook specifies the minimum number of divisions
11110 that should be there for GCC to perform the optimization for a variable
11111 of mode @var{mode}. The default implementation returns 3 if the machine
11112 has an instruction for the division, and 2 if it does not.
11116 The maximum number of bytes that a single instruction can move quickly
11117 between memory and registers or between two memory locations.
11120 @defmac MAX_MOVE_MAX
11121 The maximum number of bytes that a single instruction can move quickly
11122 between memory and registers or between two memory locations. If this
11123 is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the
11124 constant value that is the largest value that @code{MOVE_MAX} can have
11128 @defmac SHIFT_COUNT_TRUNCATED
11129 A C expression that is nonzero if on this machine the number of bits
11130 actually used for the count of a shift operation is equal to the number
11131 of bits needed to represent the size of the object being shifted. When
11132 this macro is nonzero, the compiler will assume that it is safe to omit
11133 a sign-extend, zero-extend, and certain bitwise `and' instructions that
11134 truncates the count of a shift operation. On machines that have
11135 instructions that act on bit-fields at variable positions, which may
11136 include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED}
11137 also enables deletion of truncations of the values that serve as
11138 arguments to bit-field instructions.
11140 If both types of instructions truncate the count (for shifts) and
11141 position (for bit-field operations), or if no variable-position bit-field
11142 instructions exist, you should define this macro.
11144 However, on some machines, such as the 80386 and the 680x0, truncation
11145 only applies to shift operations and not the (real or pretended)
11146 bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on
11147 such machines. Instead, add patterns to the @file{md} file that include
11148 the implied truncation of the shift instructions.
11150 You need not define this macro if it would always have the value of zero.
11153 @anchor{TARGET_SHIFT_TRUNCATION_MASK}
11154 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (machine_mode @var{mode})
11155 This function describes how the standard shift patterns for @var{mode}
11156 deal with shifts by negative amounts or by more than the width of the mode.
11157 @xref{shift patterns}.
11159 On many machines, the shift patterns will apply a mask @var{m} to the
11160 shift count, meaning that a fixed-width shift of @var{x} by @var{y} is
11161 equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If
11162 this is true for mode @var{mode}, the function should return @var{m},
11163 otherwise it should return 0. A return value of 0 indicates that no
11164 particular behavior is guaranteed.
11166 Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does
11167 @emph{not} apply to general shift rtxes; it applies only to instructions
11168 that are generated by the named shift patterns.
11170 The default implementation of this function returns
11171 @code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED}
11172 and 0 otherwise. This definition is always safe, but if
11173 @code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns
11174 nevertheless truncate the shift count, you may get better code
11178 @deftypefn {Target Hook} bool TARGET_TRULY_NOOP_TRUNCATION (poly_uint64 @var{outprec}, poly_uint64 @var{inprec})
11179 This hook returns true if it is safe to ``convert'' a value of
11180 @var{inprec} bits to one of @var{outprec} bits (where @var{outprec} is
11181 smaller than @var{inprec}) by merely operating on it as if it had only
11182 @var{outprec} bits. The default returns true unconditionally, which
11183 is correct for most machines. When @code{TARGET_TRULY_NOOP_TRUNCATION}
11184 returns false, the machine description should provide a @code{trunc}
11185 optab to specify the RTL that performs the required truncation.
11187 If @code{TARGET_MODES_TIEABLE_P} returns false for a pair of modes,
11188 suboptimal code can result if this hook returns true for the corresponding
11189 mode sizes. Making this hook return false in such cases may improve things.
11192 @deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (scalar_int_mode @var{mode}, scalar_int_mode @var{rep_mode})
11193 The representation of an integral mode can be such that the values
11194 are always extended to a wider integral mode. Return
11195 @code{SIGN_EXTEND} if values of @var{mode} are represented in
11196 sign-extended form to @var{rep_mode}. Return @code{UNKNOWN}
11197 otherwise. (Currently, none of the targets use zero-extended
11198 representation this way so unlike @code{LOAD_EXTEND_OP},
11199 @code{TARGET_MODE_REP_EXTENDED} is expected to return either
11200 @code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends
11201 @var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next
11202 widest integral mode and currently we take advantage of this fact.)
11204 Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN}
11205 value even if the extension is not performed on certain hard registers
11206 as long as for the @code{REGNO_REG_CLASS} of these hard registers
11207 @code{TARGET_CAN_CHANGE_MODE_CLASS} returns false.
11209 Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP}
11210 describe two related properties. If you define
11211 @code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want
11212 to define @code{LOAD_EXTEND_OP (mode)} to return the same type of
11215 In order to enforce the representation of @code{mode},
11216 @code{TARGET_TRULY_NOOP_TRUNCATION} should return false when truncating to
11220 @deftypefn {Target Hook} bool TARGET_SETJMP_PRESERVES_NONVOLATILE_REGS_P (void)
11221 On some targets, it is assumed that the compiler will spill all pseudos
11222 that are live across a call to @code{setjmp}, while other targets treat
11223 @code{setjmp} calls as normal function calls.
11225 This hook returns false if @code{setjmp} calls do not preserve all
11226 non-volatile registers so that gcc that must spill all pseudos that are
11227 live across @code{setjmp} calls. Define this to return true if the
11228 target does not need to spill all pseudos live across @code{setjmp} calls.
11229 The default implementation conservatively assumes all pseudos must be
11230 spilled across @code{setjmp} calls.
11233 @defmac STORE_FLAG_VALUE
11234 A C expression describing the value returned by a comparison operator
11235 with an integral mode and stored by a store-flag instruction
11236 (@samp{cstore@var{mode}4}) when the condition is true. This description must
11237 apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the
11238 comparison operators whose results have a @code{MODE_INT} mode.
11240 A value of 1 or @minus{}1 means that the instruction implementing the
11241 comparison operator returns exactly 1 or @minus{}1 when the comparison is true
11242 and 0 when the comparison is false. Otherwise, the value indicates
11243 which bits of the result are guaranteed to be 1 when the comparison is
11244 true. This value is interpreted in the mode of the comparison
11245 operation, which is given by the mode of the first operand in the
11246 @samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of
11247 @code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by
11250 If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will
11251 generate code that depends only on the specified bits. It can also
11252 replace comparison operators with equivalent operations if they cause
11253 the required bits to be set, even if the remaining bits are undefined.
11254 For example, on a machine whose comparison operators return an
11255 @code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as
11256 @samp{0x80000000}, saying that just the sign bit is relevant, the
11260 (ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0))
11264 can be converted to
11267 (ashift:SI @var{x} (const_int @var{n}))
11271 where @var{n} is the appropriate shift count to move the bit being
11272 tested into the sign bit.
11274 There is no way to describe a machine that always sets the low-order bit
11275 for a true value, but does not guarantee the value of any other bits,
11276 but we do not know of any machine that has such an instruction. If you
11277 are trying to port GCC to such a machine, include an instruction to
11278 perform a logical-and of the result with 1 in the pattern for the
11279 comparison operators and let us know at @email{gcc@@gcc.gnu.org}.
11281 Often, a machine will have multiple instructions that obtain a value
11282 from a comparison (or the condition codes). Here are rules to guide the
11283 choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions
11288 Use the shortest sequence that yields a valid definition for
11289 @code{STORE_FLAG_VALUE}. It is more efficient for the compiler to
11290 ``normalize'' the value (convert it to, e.g., 1 or 0) than for the
11291 comparison operators to do so because there may be opportunities to
11292 combine the normalization with other operations.
11295 For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being
11296 slightly preferred on machines with expensive jumps and 1 preferred on
11300 As a second choice, choose a value of @samp{0x80000001} if instructions
11301 exist that set both the sign and low-order bits but do not define the
11305 Otherwise, use a value of @samp{0x80000000}.
11308 Many machines can produce both the value chosen for
11309 @code{STORE_FLAG_VALUE} and its negation in the same number of
11310 instructions. On those machines, you should also define a pattern for
11311 those cases, e.g., one matching
11314 (set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C})))
11317 Some machines can also perform @code{and} or @code{plus} operations on
11318 condition code values with less instructions than the corresponding
11319 @samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those
11320 machines, define the appropriate patterns. Use the names @code{incscc}
11321 and @code{decscc}, respectively, for the patterns which perform
11322 @code{plus} or @code{minus} operations on condition code values. See
11323 @file{rs6000.md} for some examples. The GNU Superoptimizer can be used to
11324 find such instruction sequences on other machines.
11326 If this macro is not defined, the default value, 1, is used. You need
11327 not define @code{STORE_FLAG_VALUE} if the machine has no store-flag
11328 instructions, or if the value generated by these instructions is 1.
11331 @defmac FLOAT_STORE_FLAG_VALUE (@var{mode})
11332 A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is
11333 returned when comparison operators with floating-point results are true.
11334 Define this macro on machines that have comparison operations that return
11335 floating-point values. If there are no such operations, do not define
11339 @defmac VECTOR_STORE_FLAG_VALUE (@var{mode})
11340 A C expression that gives an rtx representing the nonzero true element
11341 for vector comparisons. The returned rtx should be valid for the inner
11342 mode of @var{mode} which is guaranteed to be a vector mode. Define
11343 this macro on machines that have vector comparison operations that
11344 return a vector result. If there are no such operations, do not define
11345 this macro. Typically, this macro is defined as @code{const1_rtx} or
11346 @code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent
11347 the compiler optimizing such vector comparison operations for the
11351 @defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
11352 @defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value})
11353 A C expression that indicates whether the architecture defines a value
11354 for @code{clz} or @code{ctz} with a zero operand.
11355 A result of @code{0} indicates the value is undefined.
11356 If the value is defined for only the RTL expression, the macro should
11357 evaluate to @code{1}; if the value applies also to the corresponding optab
11358 entry (which is normally the case if it expands directly into
11359 the corresponding RTL), then the macro should evaluate to @code{2}.
11360 In the cases where the value is defined, @var{value} should be set to
11363 If this macro is not defined, the value of @code{clz} or
11364 @code{ctz} at zero is assumed to be undefined.
11366 This macro must be defined if the target's expansion for @code{ffs}
11367 relies on a particular value to get correct results. Otherwise it
11368 is not necessary, though it may be used to optimize some corner cases, and
11369 to provide a default expansion for the @code{ffs} optab.
11371 Note that regardless of this macro the ``definedness'' of @code{clz}
11372 and @code{ctz} at zero do @emph{not} extend to the builtin functions
11373 visible to the user. Thus one may be free to adjust the value at will
11374 to match the target expansion of these operations without fear of
11379 An alias for the machine mode for pointers. On most machines, define
11380 this to be the integer mode corresponding to the width of a hardware
11381 pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines.
11382 On some machines you must define this to be one of the partial integer
11383 modes, such as @code{PSImode}.
11385 The width of @code{Pmode} must be at least as large as the value of
11386 @code{POINTER_SIZE}. If it is not equal, you must define the macro
11387 @code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended
11391 @defmac FUNCTION_MODE
11392 An alias for the machine mode used for memory references to functions
11393 being called, in @code{call} RTL expressions. On most CISC machines,
11394 where an instruction can begin at any byte address, this should be
11395 @code{QImode}. On most RISC machines, where all instructions have fixed
11396 size and alignment, this should be a mode with the same size and alignment
11397 as the machine instruction words - typically @code{SImode} or @code{HImode}.
11400 @defmac STDC_0_IN_SYSTEM_HEADERS
11401 In normal operation, the preprocessor expands @code{__STDC__} to the
11402 constant 1, to signify that GCC conforms to ISO Standard C@. On some
11403 hosts, like Solaris, the system compiler uses a different convention,
11404 where @code{__STDC__} is normally 0, but is 1 if the user specifies
11405 strict conformance to the C Standard.
11407 Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host
11408 convention when processing system header files, but when processing user
11409 files @code{__STDC__} will always expand to 1.
11412 @deftypefn {C Target Hook} {const char *} TARGET_C_PREINCLUDE (void)
11413 Define this hook to return the name of a header file to be included at
11414 the start of all compilations, as if it had been included with
11415 @code{#include <@var{file}>}. If this hook returns @code{NULL}, or is
11416 not defined, or the header is not found, or if the user specifies
11417 @option{-ffreestanding} or @option{-nostdinc}, no header is included.
11419 This hook can be used together with a header provided by the system C
11420 library to implement ISO C requirements for certain macros to be
11421 predefined that describe properties of the whole implementation rather
11422 than just the compiler.
11425 @deftypefn {C Target Hook} bool TARGET_CXX_IMPLICIT_EXTERN_C (const char*@var{})
11426 Define this hook to add target-specific C++ implicit extern C functions.
11427 If this function returns true for the name of a file-scope function, that
11428 function implicitly gets extern "C" linkage rather than whatever language
11429 linkage the declaration would normally have. An example of such function
11430 is WinMain on Win32 targets.
11433 @defmac SYSTEM_IMPLICIT_EXTERN_C
11434 Define this macro if the system header files do not support C++@.
11435 This macro handles system header files by pretending that system
11436 header files are enclosed in @samp{extern "C" @{@dots{}@}}.
11441 @defmac REGISTER_TARGET_PRAGMAS ()
11442 Define this macro if you want to implement any target-specific pragmas.
11443 If defined, it is a C expression which makes a series of calls to
11444 @code{c_register_pragma} or @code{c_register_pragma_with_expansion}
11445 for each pragma. The macro may also do any
11446 setup required for the pragmas.
11448 The primary reason to define this macro is to provide compatibility with
11449 other compilers for the same target. In general, we discourage
11450 definition of target-specific pragmas for GCC@.
11452 If the pragma can be implemented by attributes then you should consider
11453 defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well.
11455 Preprocessor macros that appear on pragma lines are not expanded. All
11456 @samp{#pragma} directives that do not match any registered pragma are
11457 silently ignored, unless the user specifies @option{-Wunknown-pragmas}.
11460 @deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
11461 @deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *))
11463 Each call to @code{c_register_pragma} or
11464 @code{c_register_pragma_with_expansion} establishes one pragma. The
11465 @var{callback} routine will be called when the preprocessor encounters a
11469 #pragma [@var{space}] @var{name} @dots{}
11472 @var{space} is the case-sensitive namespace of the pragma, or
11473 @code{NULL} to put the pragma in the global namespace. The callback
11474 routine receives @var{pfile} as its first argument, which can be passed
11475 on to cpplib's functions if necessary. You can lex tokens after the
11476 @var{name} by calling @code{pragma_lex}. Tokens that are not read by the
11477 callback will be silently ignored. The end of the line is indicated by
11478 a token of type @code{CPP_EOF}. Macro expansion occurs on the
11479 arguments of pragmas registered with
11480 @code{c_register_pragma_with_expansion} but not on the arguments of
11481 pragmas registered with @code{c_register_pragma}.
11483 Note that the use of @code{pragma_lex} is specific to the C and C++
11484 compilers. It will not work in the Java or Fortran compilers, or any
11485 other language compilers for that matter. Thus if @code{pragma_lex} is going
11486 to be called from target-specific code, it must only be done so when
11487 building the C and C++ compilers. This can be done by defining the
11488 variables @code{c_target_objs} and @code{cxx_target_objs} in the
11489 target entry in the @file{config.gcc} file. These variables should name
11490 the target-specific, language-specific object file which contains the
11491 code that uses @code{pragma_lex}. Note it will also be necessary to add a
11492 rule to the makefile fragment pointed to by @code{tmake_file} that shows
11493 how to build this object file.
11496 @defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION
11497 Define this macro if macros should be expanded in the
11498 arguments of @samp{#pragma pack}.
11501 @defmac TARGET_DEFAULT_PACK_STRUCT
11502 If your target requires a structure packing default other than 0 (meaning
11503 the machine default), define this macro to the necessary value (in bytes).
11504 This must be a value that would also be valid to use with
11505 @samp{#pragma pack()} (that is, a small power of two).
11508 @defmac DOLLARS_IN_IDENTIFIERS
11509 Define this macro to control use of the character @samp{$} in
11510 identifier names for the C family of languages. 0 means @samp{$} is
11511 not allowed by default; 1 means it is allowed. 1 is the default;
11512 there is no need to define this macro in that case.
11515 @defmac INSN_SETS_ARE_DELAYED (@var{insn})
11516 Define this macro as a C expression that is nonzero if it is safe for the
11517 delay slot scheduler to place instructions in the delay slot of @var{insn},
11518 even if they appear to use a resource set or clobbered in @var{insn}.
11519 @var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that
11520 every @code{call_insn} has this behavior. On machines where some @code{insn}
11521 or @code{jump_insn} is really a function call and hence has this behavior,
11522 you should define this macro.
11524 You need not define this macro if it would always return zero.
11527 @defmac INSN_REFERENCES_ARE_DELAYED (@var{insn})
11528 Define this macro as a C expression that is nonzero if it is safe for the
11529 delay slot scheduler to place instructions in the delay slot of @var{insn},
11530 even if they appear to set or clobber a resource referenced in @var{insn}.
11531 @var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where
11532 some @code{insn} or @code{jump_insn} is really a function call and its operands
11533 are registers whose use is actually in the subroutine it calls, you should
11534 define this macro. Doing so allows the delay slot scheduler to move
11535 instructions which copy arguments into the argument registers into the delay
11536 slot of @var{insn}.
11538 You need not define this macro if it would always return zero.
11541 @defmac MULTIPLE_SYMBOL_SPACES
11542 Define this macro as a C expression that is nonzero if, in some cases,
11543 global symbols from one translation unit may not be bound to undefined
11544 symbols in another translation unit without user intervention. For
11545 instance, under Microsoft Windows symbols must be explicitly imported
11546 from shared libraries (DLLs).
11548 You need not define this macro if it would always evaluate to zero.
11551 @deftypefn {Target Hook} {rtx_insn *} TARGET_MD_ASM_ADJUST (vec<rtx>& @var{outputs}, vec<rtx>& @var{inputs}, vec<machine_mode>& @var{input_modes}, vec<const char *>& @var{constraints}, vec<rtx>& @var{clobbers}, HARD_REG_SET& @var{clobbered_regs}, location_t @var{loc})
11552 This target hook may add @dfn{clobbers} to @var{clobbers} and
11553 @var{clobbered_regs} for any hard regs the port wishes to automatically
11554 clobber for an asm. The @var{outputs} and @var{inputs} may be inspected
11555 to avoid clobbering a register that is already used by the asm. @var{loc}
11556 is the source location of the asm.
11558 It may modify the @var{outputs}, @var{inputs}, @var{input_modes}, and
11559 @var{constraints} as necessary for other pre-processing. In this case the
11560 return value is a sequence of insns to emit after the asm. Note that
11561 changes to @var{inputs} must be accompanied by the corresponding changes
11562 to @var{input_modes}.
11565 @defmac MATH_LIBRARY
11566 Define this macro as a C string constant for the linker argument to link
11567 in the system math library, minus the initial @samp{"-l"}, or
11568 @samp{""} if the target does not have a
11569 separate math library.
11571 You need only define this macro if the default of @samp{"m"} is wrong.
11574 @defmac LIBRARY_PATH_ENV
11575 Define this macro as a C string constant for the environment variable that
11576 specifies where the linker should look for libraries.
11578 You need only define this macro if the default of @samp{"LIBRARY_PATH"}
11582 @defmac TARGET_POSIX_IO
11583 Define this macro if the target supports the following POSIX@ file
11584 functions, access, mkdir and file locking with fcntl / F_SETLKW@.
11585 Defining @code{TARGET_POSIX_IO} will enable the test coverage code
11586 to use file locking when exiting a program, which avoids race conditions
11587 if the program has forked. It will also create directories at run-time
11588 for cross-profiling.
11591 @defmac MAX_CONDITIONAL_EXECUTE
11593 A C expression for the maximum number of instructions to execute via
11594 conditional execution instructions instead of a branch. A value of
11595 @code{BRANCH_COST}+1 is the default.
11598 @defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr})
11599 Used if the target needs to perform machine-dependent modifications on the
11600 conditionals used for turning basic blocks into conditionally executed code.
11601 @var{ce_info} points to a data structure, @code{struct ce_if_block}, which
11602 contains information about the currently processed blocks. @var{true_expr}
11603 and @var{false_expr} are the tests that are used for converting the
11604 then-block and the else-block, respectively. Set either @var{true_expr} or
11605 @var{false_expr} to a null pointer if the tests cannot be converted.
11608 @defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr})
11609 Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated
11610 if-statements into conditions combined by @code{and} and @code{or} operations.
11611 @var{bb} contains the basic block that contains the test that is currently
11612 being processed and about to be turned into a condition.
11615 @defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn})
11616 A C expression to modify the @var{PATTERN} of an @var{INSN} that is to
11617 be converted to conditional execution format. @var{ce_info} points to
11618 a data structure, @code{struct ce_if_block}, which contains information
11619 about the currently processed blocks.
11622 @defmac IFCVT_MODIFY_FINAL (@var{ce_info})
11623 A C expression to perform any final machine dependent modifications in
11624 converting code to conditional execution. The involved basic blocks
11625 can be found in the @code{struct ce_if_block} structure that is pointed
11626 to by @var{ce_info}.
11629 @defmac IFCVT_MODIFY_CANCEL (@var{ce_info})
11630 A C expression to cancel any machine dependent modifications in
11631 converting code to conditional execution. The involved basic blocks
11632 can be found in the @code{struct ce_if_block} structure that is pointed
11633 to by @var{ce_info}.
11636 @defmac IFCVT_MACHDEP_INIT (@var{ce_info})
11637 A C expression to initialize any machine specific data for if-conversion
11638 of the if-block in the @code{struct ce_if_block} structure that is pointed
11639 to by @var{ce_info}.
11642 @deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void)
11643 If non-null, this hook performs a target-specific pass over the
11644 instruction stream. The compiler will run it at all optimization levels,
11645 just before the point at which it normally does delayed-branch scheduling.
11647 The exact purpose of the hook varies from target to target. Some use
11648 it to do transformations that are necessary for correctness, such as
11649 laying out in-function constant pools or avoiding hardware hazards.
11650 Others use it as an opportunity to do some machine-dependent optimizations.
11652 You need not implement the hook if it has nothing to do. The default
11653 definition is null.
11656 @deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void)
11657 Define this hook if you have any machine-specific built-in functions
11658 that need to be defined. It should be a function that performs the
11661 Machine specific built-in functions can be useful to expand special machine
11662 instructions that would otherwise not normally be generated because
11663 they have no equivalent in the source language (for example, SIMD vector
11664 instructions or prefetch instructions).
11666 To create a built-in function, call the function
11667 @code{lang_hooks.builtin_function}
11668 which is defined by the language front end. You can use any type nodes set
11669 up by @code{build_common_tree_nodes};
11670 only language front ends that use those two functions will call
11671 @samp{TARGET_INIT_BUILTINS}.
11674 @deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p})
11675 Define this hook if you have any machine-specific built-in functions
11676 that need to be defined. It should be a function that returns the
11677 builtin function declaration for the builtin function code @var{code}.
11678 If there is no such builtin and it cannot be initialized at this time
11679 if @var{initialize_p} is true the function should return @code{NULL_TREE}.
11680 If @var{code} is out of range the function should return
11681 @code{error_mark_node}.
11684 @deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, machine_mode @var{mode}, int @var{ignore})
11686 Expand a call to a machine specific built-in function that was set up by
11687 @samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the
11688 function call; the result should go to @var{target} if that is
11689 convenient, and have mode @var{mode} if that is convenient.
11690 @var{subtarget} may be used as the target for computing one of
11691 @var{exp}'s operands. @var{ignore} is nonzero if the value is to be
11692 ignored. This function should return the result of the call to the
11696 @deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist})
11697 Select a replacement for a machine specific built-in function that
11698 was set up by @samp{TARGET_INIT_BUILTINS}. This is done
11699 @emph{before} regular type checking, and so allows the target to
11700 implement a crude form of function overloading. @var{fndecl} is the
11701 declaration of the built-in function. @var{arglist} is the list of
11702 arguments passed to the built-in function. The result is a
11703 complete expression that implements the operation, usually
11704 another @code{CALL_EXPR}.
11705 @var{arglist} really has type @samp{VEC(tree,gc)*}
11708 @deftypefn {Target Hook} bool TARGET_CHECK_BUILTIN_CALL (location_t @var{loc}, vec<location_t> @var{arg_loc}, tree @var{fndecl}, tree @var{orig_fndecl}, unsigned int @var{nargs}, tree *@var{args})
11709 Perform semantic checking on a call to a machine-specific built-in
11710 function after its arguments have been constrained to the function
11711 signature. Return true if the call is valid, otherwise report an error
11714 This hook is called after @code{TARGET_RESOLVE_OVERLOADED_BUILTIN}.
11715 The call was originally to built-in function @var{orig_fndecl},
11716 but after the optional @code{TARGET_RESOLVE_OVERLOADED_BUILTIN}
11717 step is now to built-in function @var{fndecl}. @var{loc} is the
11718 location of the call and @var{args} is an array of function arguments,
11719 of which there are @var{nargs}. @var{arg_loc} specifies the location
11723 @deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore})
11724 Fold a call to a machine specific built-in function that was set up by
11725 @samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the
11726 built-in function. @var{n_args} is the number of arguments passed to
11727 the function; the arguments themselves are pointed to by @var{argp}.
11728 The result is another tree, valid for both GIMPLE and GENERIC,
11729 containing a simplified expression for the call's result. If
11730 @var{ignore} is true the value will be ignored.
11733 @deftypefn {Target Hook} bool TARGET_GIMPLE_FOLD_BUILTIN (gimple_stmt_iterator *@var{gsi})
11734 Fold a call to a machine specific built-in function that was set up
11735 by @samp{TARGET_INIT_BUILTINS}. @var{gsi} points to the gimple
11736 statement holding the function call. Returns true if any change
11737 was made to the GIMPLE stream.
11740 @deftypefn {Target Hook} int TARGET_COMPARE_VERSION_PRIORITY (tree @var{decl1}, tree @var{decl2})
11741 This hook is used to compare the target attributes in two functions to
11742 determine which function's features get higher priority. This is used
11743 during function multi-versioning to figure out the order in which two
11744 versions must be dispatched. A function version with a higher priority
11745 is checked for dispatching earlier. @var{decl1} and @var{decl2} are
11746 the two function decls that will be compared.
11749 @deftypefn {Target Hook} tree TARGET_GET_FUNCTION_VERSIONS_DISPATCHER (void *@var{decl})
11750 This hook is used to get the dispatcher function for a set of function
11751 versions. The dispatcher function is called to invoke the right function
11752 version at run-time. @var{decl} is one version from a set of semantically
11753 identical versions.
11756 @deftypefn {Target Hook} tree TARGET_GENERATE_VERSION_DISPATCHER_BODY (void *@var{arg})
11757 This hook is used to generate the dispatcher logic to invoke the right
11758 function version at run-time for a given set of function versions.
11759 @var{arg} points to the callgraph node of the dispatcher function whose
11760 body must be generated.
11763 @deftypefn {Target Hook} bool TARGET_PREDICT_DOLOOP_P (class loop *@var{loop})
11764 Return true if we can predict it is possible to use a low-overhead loop
11765 for a particular loop. The parameter @var{loop} is a pointer to the loop.
11766 This target hook is required only when the target supports low-overhead
11767 loops, and will help ivopts to make some decisions.
11768 The default version of this hook returns false.
11771 @deftypevr {Target Hook} bool TARGET_HAVE_COUNT_REG_DECR_P
11772 Return true if the target supports hardware count register for decrement
11774 The default value is false.
11777 @deftypevr {Target Hook} int64_t TARGET_DOLOOP_COST_FOR_GENERIC
11778 One IV candidate dedicated for doloop is introduced in IVOPTs, we can
11779 calculate the computation cost of adopting it to any generic IV use by
11780 function get_computation_cost as before. But for targets which have
11781 hardware count register support for decrement and branch, it may have to
11782 move IV value from hardware count register to general purpose register
11783 while doloop IV candidate is used for generic IV uses. It probably takes
11784 expensive penalty. This hook allows target owners to define the cost for
11785 this especially for generic IV uses.
11786 The default value is zero.
11789 @deftypevr {Target Hook} int64_t TARGET_DOLOOP_COST_FOR_ADDRESS
11790 One IV candidate dedicated for doloop is introduced in IVOPTs, we can
11791 calculate the computation cost of adopting it to any address IV use by
11792 function get_computation_cost as before. But for targets which have
11793 hardware count register support for decrement and branch, it may have to
11794 move IV value from hardware count register to general purpose register
11795 while doloop IV candidate is used for address IV uses. It probably takes
11796 expensive penalty. This hook allows target owners to define the cost for
11797 this escpecially for address IV uses.
11798 The default value is zero.
11801 @deftypefn {Target Hook} bool TARGET_CAN_USE_DOLOOP_P (const widest_int @var{&iterations}, const widest_int @var{&iterations_max}, unsigned int @var{loop_depth}, bool @var{entered_at_top})
11802 Return true if it is possible to use low-overhead loops (@code{doloop_end}
11803 and @code{doloop_begin}) for a particular loop. @var{iterations} gives the
11804 exact number of iterations, or 0 if not known. @var{iterations_max} gives
11805 the maximum number of iterations, or 0 if not known. @var{loop_depth} is
11806 the nesting depth of the loop, with 1 for innermost loops, 2 for loops that
11807 contain innermost loops, and so on. @var{entered_at_top} is true if the
11808 loop is only entered from the top.
11810 This hook is only used if @code{doloop_end} is available. The default
11811 implementation returns true. You can use @code{can_use_doloop_if_innermost}
11812 if the loop must be the innermost, and if there are no other restrictions.
11815 @deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const rtx_insn *@var{insn})
11817 Take an instruction in @var{insn} and return NULL if it is valid within a
11818 low-overhead loop, otherwise return a string explaining why doloop
11819 could not be applied.
11821 Many targets use special registers for low-overhead looping. For any
11822 instruction that clobbers these this function should return a string indicating
11823 the reason why the doloop could not be applied.
11824 By default, the RTL loop optimizer does not use a present doloop pattern for
11825 loops containing function calls or branch on table instructions.
11828 @deftypefn {Target Hook} machine_mode TARGET_PREFERRED_DOLOOP_MODE (machine_mode @var{mode})
11829 This hook takes a @var{mode} for a doloop IV, where @code{mode} is the
11830 original mode for the operation. If the target prefers an alternate
11831 @code{mode} for the operation, then this hook should return that mode;
11832 otherwise the original @code{mode} should be returned. For example, on a
11833 64-bit target, @code{DImode} might be preferred over @code{SImode}. Both the
11834 original and the returned modes should be @code{MODE_INT}.
11837 @deftypefn {Target Hook} bool TARGET_LEGITIMATE_COMBINED_INSN (rtx_insn *@var{insn})
11838 Take an instruction in @var{insn} and return @code{false} if the instruction
11839 is not appropriate as a combination of two or more instructions. The
11840 default is to accept all instructions.
11843 @deftypefn {Target Hook} bool TARGET_CAN_FOLLOW_JUMP (const rtx_insn *@var{follower}, const rtx_insn *@var{followee})
11844 FOLLOWER and FOLLOWEE are JUMP_INSN instructions;
11845 return true if FOLLOWER may be modified to follow FOLLOWEE;
11846 false, if it can't.
11847 For example, on some targets, certain kinds of branches can't be made to
11848 follow through a hot/cold partitioning.
11851 @deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code})
11852 This target hook returns @code{true} if @var{x} is considered to be commutative.
11853 Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider
11854 PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code
11855 of the enclosing rtl, if known, otherwise it is UNKNOWN.
11858 @deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg})
11860 When the initial value of a hard register has been copied in a pseudo
11861 register, it is often not necessary to actually allocate another register
11862 to this pseudo register, because the original hard register or a stack slot
11863 it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE}
11864 is called at the start of register allocation once for each hard register
11865 that had its initial value copied by using
11866 @code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}.
11867 Possible values are @code{NULL_RTX}, if you don't want
11868 to do any special allocation, a @code{REG} rtx---that would typically be
11869 the hard register itself, if it is known not to be clobbered---or a
11871 If you are returning a @code{MEM}, this is only a hint for the allocator;
11872 it might decide to use another register anyways.
11873 You may use @code{current_function_is_leaf} or
11874 @code{REG_N_SETS} in the hook to determine if the hard
11875 register in question will not be clobbered.
11876 The default value of this hook is @code{NULL}, which disables any special
11880 @deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags})
11881 This target hook returns nonzero if @var{x}, an @code{unspec} or
11882 @code{unspec_volatile} operation, might cause a trap. Targets can use
11883 this hook to enhance precision of analysis for @code{unspec} and
11884 @code{unspec_volatile} operations. You may call @code{may_trap_p_1}
11885 to analyze inner elements of @var{x} in which case @var{flags} should be
11889 @deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl})
11890 The compiler invokes this hook whenever it changes its current function
11891 context (@code{cfun}). You can define this function if
11892 the back end needs to perform any initialization or reset actions on a
11893 per-function basis. For example, it may be used to implement function
11894 attributes that affect register usage or code generation patterns.
11895 The argument @var{decl} is the declaration for the new function context,
11896 and may be null to indicate that the compiler has left a function context
11897 and is returning to processing at the top level.
11898 The default hook function does nothing.
11900 GCC sets @code{cfun} to a dummy function context during initialization of
11901 some parts of the back end. The hook function is not invoked in this
11902 situation; you need not worry about the hook being invoked recursively,
11903 or when the back end is in a partially-initialized state.
11904 @code{cfun} might be @code{NULL} to indicate processing at top level,
11905 outside of any function scope.
11908 @defmac TARGET_OBJECT_SUFFIX
11909 Define this macro to be a C string representing the suffix for object
11910 files on your target machine. If you do not define this macro, GCC will
11911 use @samp{.o} as the suffix for object files.
11914 @defmac TARGET_EXECUTABLE_SUFFIX
11915 Define this macro to be a C string representing the suffix to be
11916 automatically added to executable files on your target machine. If you
11917 do not define this macro, GCC will use the null string as the suffix for
11921 @defmac COLLECT_EXPORT_LIST
11922 If defined, @code{collect2} will scan the individual object files
11923 specified on its command line and create an export list for the linker.
11924 Define this macro for systems like AIX, where the linker discards
11925 object files that are not referenced from @code{main} and uses export
11929 @deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void)
11930 This target hook returns @code{true} past the point in which new jump
11931 instructions could be created. On machines that require a register for
11932 every jump such as the SHmedia ISA of SH5, this point would typically be
11933 reload, so this target hook should be defined to a function such as:
11937 cannot_modify_jumps_past_reload_p ()
11939 return (reload_completed || reload_in_progress);
11944 @deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void)
11945 This target hook returns true if the target supports conditional execution.
11946 This target hook is required only when the target has several different
11947 modes and they have different conditional execution capability, such as ARM.
11950 @deftypefn {Target Hook} rtx TARGET_GEN_CCMP_FIRST (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, int @var{code}, tree @var{op0}, tree @var{op1})
11951 This function prepares to emit a comparison insn for the first compare in a
11952 sequence of conditional comparisions. It returns an appropriate comparison
11953 with @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}.
11954 The insns to prepare the compare are saved in @var{prep_seq} and the compare
11955 insns are saved in @var{gen_seq}. They will be emitted when all the
11956 compares in the conditional comparision are generated without error.
11957 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}.
11960 @deftypefn {Target Hook} rtx TARGET_GEN_CCMP_NEXT (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, rtx @var{prev}, int @var{cmp_code}, tree @var{op0}, tree @var{op1}, int @var{bit_code})
11961 This function prepares to emit a conditional comparison within a sequence
11962 of conditional comparisons. It returns an appropriate comparison with
11963 @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}.
11964 The insns to prepare the compare are saved in @var{prep_seq} and the compare
11965 insns are saved in @var{gen_seq}. They will be emitted when all the
11966 compares in the conditional comparision are generated without error. The
11967 @var{prev} expression is the result of a prior call to @code{gen_ccmp_first}
11968 or @code{gen_ccmp_next}. It may return @code{NULL} if the combination of
11969 @var{prev} and this comparison is not supported, otherwise the result must
11970 be appropriate for passing to @code{gen_ccmp_next} or @code{cbranch_optab}.
11971 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}.
11972 @var{bit_code} is @code{AND} or @code{IOR}, which is the op on the compares.
11975 @deftypefn {Target Hook} rtx TARGET_GEN_MEMSET_SCRATCH_RTX (machine_mode @var{mode})
11976 This hook should return an rtx for a scratch register in @var{mode} to
11977 be used when expanding memset calls. The backend can use a hard scratch
11978 register to avoid stack realignment when expanding memset. The default
11979 is @code{gen_reg_rtx}.
11982 @deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, class loop *@var{loop})
11983 This target hook returns a new value for the number of times @var{loop}
11984 should be unrolled. The parameter @var{nunroll} is the number of times
11985 the loop is to be unrolled. The parameter @var{loop} is a pointer to
11986 the loop, which is going to be checked for unrolling. This target hook
11987 is required only when the target has special constraints like maximum
11988 number of memory accesses.
11991 @defmac POWI_MAX_MULTS
11992 If defined, this macro is interpreted as a signed integer C expression
11993 that specifies the maximum number of floating point multiplications
11994 that should be emitted when expanding exponentiation by an integer
11995 constant inline. When this value is defined, exponentiation requiring
11996 more than this number of multiplications is implemented by calling the
11997 system library's @code{pow}, @code{powf} or @code{powl} routines.
11998 The default value places no upper bound on the multiplication count.
12001 @deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
12002 This target hook should register any extra include files for the
12003 target. The parameter @var{stdinc} indicates if normal include files
12004 are present. The parameter @var{sysroot} is the system root directory.
12005 The parameter @var{iprefix} is the prefix for the gcc directory.
12008 @deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc})
12009 This target hook should register any extra include files for the
12010 target before any standard headers. The parameter @var{stdinc}
12011 indicates if normal include files are present. The parameter
12012 @var{sysroot} is the system root directory. The parameter
12013 @var{iprefix} is the prefix for the gcc directory.
12016 @deftypefn Macro void TARGET_OPTF (char *@var{path})
12017 This target hook should register special include paths for the target.
12018 The parameter @var{path} is the include to register. On Darwin
12019 systems, this is used for Framework includes, which have semantics
12020 that are different from @option{-I}.
12023 @defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl})
12024 This target macro returns @code{true} if it is safe to use a local alias
12025 for a virtual function @var{fndecl} when constructing thunks,
12026 @code{false} otherwise. By default, the macro returns @code{true} for all
12027 functions, if a target supports aliases (i.e.@: defines
12028 @code{ASM_OUTPUT_DEF}), @code{false} otherwise,
12031 @defmac TARGET_FORMAT_TYPES
12032 If defined, this macro is the name of a global variable containing
12033 target-specific format checking information for the @option{-Wformat}
12034 option. The default is to have no target-specific format checks.
12037 @defmac TARGET_N_FORMAT_TYPES
12038 If defined, this macro is the number of entries in
12039 @code{TARGET_FORMAT_TYPES}.
12042 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES
12043 If defined, this macro is the name of a global variable containing
12044 target-specific format overrides for the @option{-Wformat} option. The
12045 default is to have no target-specific format overrides. If defined,
12046 @code{TARGET_FORMAT_TYPES} must be defined, too.
12049 @defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT
12050 If defined, this macro specifies the number of entries in
12051 @code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}.
12054 @defmac TARGET_OVERRIDES_FORMAT_INIT
12055 If defined, this macro specifies the optional initialization
12056 routine for target specific customizations of the system printf
12057 and scanf formatter settings.
12060 @deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val})
12061 If defined, this macro returns the diagnostic message when it is
12062 illegal to pass argument @var{val} to function @var{funcdecl}
12063 with prototype @var{typelist}.
12066 @deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype})
12067 If defined, this macro returns the diagnostic message when it is
12068 invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL}
12069 if validity should be determined by the front end.
12072 @deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type})
12073 If defined, this macro returns the diagnostic message when it is
12074 invalid to apply operation @var{op} (where unary plus is denoted by
12075 @code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL}
12076 if validity should be determined by the front end.
12079 @deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2})
12080 If defined, this macro returns the diagnostic message when it is
12081 invalid to apply operation @var{op} to operands of types @var{type1}
12082 and @var{type2}, or @code{NULL} if validity should be determined by
12086 @deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type})
12087 If defined, this target hook returns the type to which values of
12088 @var{type} should be promoted when they appear in expressions,
12089 analogous to the integer promotions, or @code{NULL_TREE} to use the
12090 front end's normal promotion rules. This hook is useful when there are
12091 target-specific types with special promotion rules.
12092 This is currently used only by the C and C++ front ends.
12095 @deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr})
12096 If defined, this hook returns the result of converting @var{expr} to
12097 @var{type}. It should return the converted expression,
12098 or @code{NULL_TREE} to apply the front end's normal conversion rules.
12099 This hook is useful when there are target-specific types with special
12101 This is currently used only by the C and C++ front ends.
12104 @deftypefn {Target Hook} bool TARGET_VERIFY_TYPE_CONTEXT (location_t @var{loc}, type_context_kind @var{context}, const_tree @var{type}, bool @var{silent_p})
12105 If defined, this hook returns false if there is a target-specific reason
12106 why type @var{type} cannot be used in the source language context described
12107 by @var{context}. When @var{silent_p} is false, the hook also reports an
12108 error against @var{loc} for invalid uses of @var{type}.
12110 Calls to this hook should be made through the global function
12111 @code{verify_type_context}, which makes the @var{silent_p} parameter
12112 default to false and also handles @code{error_mark_node}.
12114 The default implementation always returns true.
12118 This macro determines the size of the objective C jump buffer for the
12119 NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value.
12122 @defmac LIBGCC2_UNWIND_ATTRIBUTE
12123 Define this macro if any target-specific attributes need to be attached
12124 to the functions in @file{libgcc} that provide low-level support for
12125 call stack unwinding. It is used in declarations in @file{unwind-generic.h}
12126 and the associated definitions of those functions.
12129 @deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void)
12130 Define this macro to update the current function stack boundary if
12134 @deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void)
12135 This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a
12136 different argument pointer register is needed to access the function's
12137 argument list due to stack realignment. Return @code{NULL} if no DRAP
12141 @deftypefn {Target Hook} HARD_REG_SET TARGET_ZERO_CALL_USED_REGS (HARD_REG_SET @var{selected_regs})
12142 This target hook emits instructions to zero the subset of @var{selected_regs}
12143 that could conceivably contain values that are useful to an attacker.
12144 Return the set of registers that were actually cleared.
12146 For most targets, the returned set of registers is a subset of
12147 @var{selected_regs}, however, for some of the targets (for example MIPS),
12148 clearing some registers that are in the @var{selected_regs} requires
12149 clearing other call used registers that are not in the @var{selected_regs},
12150 under such situation, the returned set of registers must be a subset of all
12151 call used registers.
12153 The default implementation uses normal move instructions to zero
12154 all the registers in @var{selected_regs}. Define this hook if the
12155 target has more efficient ways of zeroing certain registers,
12156 or if you believe that certain registers would never contain
12157 values that are useful to an attacker.
12160 @deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void)
12161 When optimization is disabled, this hook indicates whether or not
12162 arguments should be allocated to stack slots. Normally, GCC allocates
12163 stacks slots for arguments when not optimizing in order to make
12164 debugging easier. However, when a function is declared with
12165 @code{__attribute__((naked))}, there is no stack frame, and the compiler
12166 cannot safely move arguments from the registers in which they are passed
12167 to the stack. Therefore, this hook should return true in general, but
12168 false for naked functions. The default implementation always returns true.
12171 @deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR
12172 On some architectures it can take multiple instructions to synthesize
12173 a constant. If there is another constant already in a register that
12174 is close enough in value then it is preferable that the new constant
12175 is computed from this register using immediate addition or
12176 subtraction. We accomplish this through CSE. Besides the value of
12177 the constant we also add a lower and an upper constant anchor to the
12178 available expressions. These are then queried when encountering new
12179 constants. The anchors are computed by rounding the constant up and
12180 down to a multiple of the value of @code{TARGET_CONST_ANCHOR}.
12181 @code{TARGET_CONST_ANCHOR} should be the maximum positive value
12182 accepted by immediate-add plus one. We currently assume that the
12183 value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on
12184 MIPS, where add-immediate takes a 16-bit signed value,
12185 @code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value
12186 is zero, which disables this optimization.
12189 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_ASAN_SHADOW_OFFSET (void)
12190 Return the offset bitwise ored into shifted address to get corresponding
12191 Address Sanitizer shadow memory address. NULL if Address Sanitizer is not
12192 supported by the target. May return 0 if Address Sanitizer is not supported
12196 @deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_MEMMODEL_CHECK (unsigned HOST_WIDE_INT @var{val})
12197 Validate target specific memory model mask bits. When NULL no target specific
12198 memory model bits are allowed.
12201 @deftypevr {Target Hook} {unsigned char} TARGET_ATOMIC_TEST_AND_SET_TRUEVAL
12202 This value should be set if the result written by
12203 @code{atomic_test_and_set} is not exactly 1, i.e.@: the
12204 @code{bool} @code{true}.
12207 @deftypefn {Target Hook} bool TARGET_HAS_IFUNC_P (void)
12208 It returns true if the target supports GNU indirect functions.
12209 The support includes the assembler, linker and dynamic linker.
12210 The default value of this hook is based on target's libc.
12213 @deftypefn {Target Hook} bool TARGET_IFUNC_REF_LOCAL_OK (void)
12214 Return true if it is OK to reference indirect function resolvers
12215 locally. The default is to return false.
12218 @deftypefn {Target Hook} {unsigned int} TARGET_ATOMIC_ALIGN_FOR_MODE (machine_mode @var{mode})
12219 If defined, this function returns an appropriate alignment in bits for an
12220 atomic object of machine_mode @var{mode}. If 0 is returned then the
12221 default alignment for the specified mode is used.
12224 @deftypefn {Target Hook} void TARGET_ATOMIC_ASSIGN_EXPAND_FENV (tree *@var{hold}, tree *@var{clear}, tree *@var{update})
12225 ISO C11 requires atomic compound assignments that may raise floating-point
12226 exceptions to raise exceptions corresponding to the arithmetic operation
12227 whose result was successfully stored in a compare-and-exchange sequence.
12228 This requires code equivalent to calls to @code{feholdexcept},
12229 @code{feclearexcept} and @code{feupdateenv} to be generated at
12230 appropriate points in the compare-and-exchange sequence. This hook should
12231 set @code{*@var{hold}} to an expression equivalent to the call to
12232 @code{feholdexcept}, @code{*@var{clear}} to an expression equivalent to
12233 the call to @code{feclearexcept} and @code{*@var{update}} to an expression
12234 equivalent to the call to @code{feupdateenv}. The three expressions are
12235 @code{NULL_TREE} on entry to the hook and may be left as @code{NULL_TREE}
12236 if no code is required in a particular place. The default implementation
12237 leaves all three expressions as @code{NULL_TREE}. The
12238 @code{__atomic_feraiseexcept} function from @code{libatomic} may be of use
12239 as part of the code generated in @code{*@var{update}}.
12242 @deftypefn {Target Hook} void TARGET_RECORD_OFFLOAD_SYMBOL (tree)
12243 Used when offloaded functions are seen in the compilation unit and no named
12244 sections are available. It is called once for each symbol that must be
12245 recorded in the offload function and variable table.
12248 @deftypefn {Target Hook} {char *} TARGET_OFFLOAD_OPTIONS (void)
12249 Used when writing out the list of options into an LTO file. It should
12250 translate any relevant target-specific options (such as the ABI in use)
12251 into one of the @option{-foffload} options that exist as a common interface
12252 to express such options. It should return a string containing these options,
12253 separated by spaces, which the caller will free.
12257 @defmac TARGET_SUPPORTS_WIDE_INT
12259 On older ports, large integers are stored in @code{CONST_DOUBLE} rtl
12260 objects. Newer ports define @code{TARGET_SUPPORTS_WIDE_INT} to be nonzero
12261 to indicate that large integers are stored in
12262 @code{CONST_WIDE_INT} rtl objects. The @code{CONST_WIDE_INT} allows
12263 very large integer constants to be represented. @code{CONST_DOUBLE}
12264 is limited to twice the size of the host's @code{HOST_WIDE_INT}
12267 Converting a port mostly requires looking for the places where
12268 @code{CONST_DOUBLE}s are used with @code{VOIDmode} and replacing that
12269 code with code that accesses @code{CONST_WIDE_INT}s. @samp{"grep -i
12270 const_double"} at the port level gets you to 95% of the changes that
12271 need to be made. There are a few places that require a deeper look.
12275 There is no equivalent to @code{hval} and @code{lval} for
12276 @code{CONST_WIDE_INT}s. This would be difficult to express in the md
12277 language since there are a variable number of elements.
12279 Most ports only check that @code{hval} is either 0 or -1 to see if the
12280 value is small. As mentioned above, this will no longer be necessary
12281 since small constants are always @code{CONST_INT}. Of course there
12282 are still a few exceptions, the alpha's constraint used by the zap
12283 instruction certainly requires careful examination by C code.
12284 However, all the current code does is pass the hval and lval to C
12285 code, so evolving the c code to look at the @code{CONST_WIDE_INT} is
12286 not really a large change.
12289 Because there is no standard template that ports use to materialize
12290 constants, there is likely to be some futzing that is unique to each
12294 The rtx costs may have to be adjusted to properly account for larger
12295 constants that are represented as @code{CONST_WIDE_INT}.
12298 All and all it does not take long to convert ports that the
12299 maintainer is familiar with.
12303 @deftypefn {Target Hook} bool TARGET_HAVE_SPECULATION_SAFE_VALUE (bool @var{active})
12304 This hook is used to determine the level of target support for
12305 @code{__builtin_speculation_safe_value}. If called with an argument
12306 of false, it returns true if the target has been modified to support
12307 this builtin. If called with an argument of true, it returns true
12308 if the target requires active mitigation execution might be speculative.
12310 The default implementation returns false if the target does not define
12311 a pattern named @code{speculation_barrier}. Else it returns true
12312 for the first case and whether the pattern is enabled for the current
12313 compilation for the second case.
12315 For targets that have no processors that can execute instructions
12316 speculatively an alternative implemenation of this hook is available:
12317 simply redefine this hook to @code{speculation_safe_value_not_needed}
12318 along with your other target hooks.
12321 @deftypefn {Target Hook} rtx TARGET_SPECULATION_SAFE_VALUE (machine_mode @var{mode}, rtx @var{result}, rtx @var{val}, rtx @var{failval})
12322 This target hook can be used to generate a target-specific code
12323 sequence that implements the @code{__builtin_speculation_safe_value}
12324 built-in function. The function must always return @var{val} in
12325 @var{result} in mode @var{mode} when the cpu is not executing
12326 speculatively, but must never return that when speculating until it
12327 is known that the speculation will not be unwound. The hook supports
12328 two primary mechanisms for implementing the requirements. The first
12329 is to emit a speculation barrier which forces the processor to wait
12330 until all prior speculative operations have been resolved; the second
12331 is to use a target-specific mechanism that can track the speculation
12332 state and to return @var{failval} if it can determine that
12333 speculation must be unwound at a later time.
12335 The default implementation simply copies @var{val} to @var{result} and
12336 emits a @code{speculation_barrier} instruction if that is defined.
12339 @deftypefn {Target Hook} void TARGET_RUN_TARGET_SELFTESTS (void)
12340 If selftests are enabled, run any selftests for this target.
12343 @deftypefn {Target Hook} bool TARGET_MEMTAG_CAN_TAG_ADDRESSES ()
12344 True if the backend architecture naturally supports ignoring some region
12345 of pointers. This feature means that @option{-fsanitize=hwaddress} can
12348 At preset, this feature does not support address spaces. It also requires
12349 @code{Pmode} to be the same as @code{ptr_mode}.
12352 @deftypefn {Target Hook} uint8_t TARGET_MEMTAG_TAG_SIZE ()
12353 Return the size of a tag (in bits) for this platform.
12355 The default returns 8.
12358 @deftypefn {Target Hook} uint8_t TARGET_MEMTAG_GRANULE_SIZE ()
12359 Return the size in real memory that each byte in shadow memory refers to.
12360 I.e. if a variable is @var{X} bytes long in memory, then this hook should
12361 return the value @var{Y} such that the tag in shadow memory spans
12362 @var{X}/@var{Y} bytes.
12364 Most variables will need to be aligned to this amount since two variables
12365 that are neighbors in memory and share a tag granule would need to share
12368 The default returns 16.
12371 @deftypefn {Target Hook} rtx TARGET_MEMTAG_INSERT_RANDOM_TAG (rtx @var{untagged}, rtx @var{target})
12372 Return an RTX representing the value of @var{untagged} but with a
12373 (possibly) random tag in it.
12374 Put that value into @var{target} if it is convenient to do so.
12375 This function is used to generate a tagged base for the current stack frame.
12378 @deftypefn {Target Hook} rtx TARGET_MEMTAG_ADD_TAG (rtx @var{base}, poly_int64 @var{addr_offset}, uint8_t @var{tag_offset})
12379 Return an RTX that represents the result of adding @var{addr_offset} to
12380 the address in pointer @var{base} and @var{tag_offset} to the tag in pointer
12382 The resulting RTX must either be a valid memory address or be able to get
12383 put into an operand with @code{force_operand}.
12385 Unlike other memtag hooks, this must return an expression and not emit any
12389 @deftypefn {Target Hook} rtx TARGET_MEMTAG_SET_TAG (rtx @var{untagged_base}, rtx @var{tag}, rtx @var{target})
12390 Return an RTX representing @var{untagged_base} but with the tag @var{tag}.
12391 Try and store this in @var{target} if convenient.
12392 @var{untagged_base} is required to have a zero tag when this hook is called.
12393 The default of this hook is to set the top byte of @var{untagged_base} to
12397 @deftypefn {Target Hook} rtx TARGET_MEMTAG_EXTRACT_TAG (rtx @var{tagged_pointer}, rtx @var{target})
12398 Return an RTX representing the tag stored in @var{tagged_pointer}.
12399 Store the result in @var{target} if it is convenient.
12400 The default represents the top byte of the original pointer.
12403 @deftypefn {Target Hook} rtx TARGET_MEMTAG_UNTAGGED_POINTER (rtx @var{tagged_pointer}, rtx @var{target})
12404 Return an RTX representing @var{tagged_pointer} with its tag set to zero.
12405 Store the result in @var{target} if convenient.
12406 The default clears the top byte of the original pointer.
12409 @deftypefn {Target Hook} HOST_WIDE_INT TARGET_GCOV_TYPE_SIZE (void)
12410 Returns the gcov type size in bits. This type is used for example for
12411 counters incremented by profiling and code-coverage events. The default
12412 value is 64, if the type size of long long is greater than 32, otherwise the
12413 default value is 32. A 64-bit type is recommended to avoid overflows of the
12414 counters. If the @option{-fprofile-update=atomic} is used, then the
12415 counters are incremented using atomic operations. Targets not supporting
12416 64-bit atomic operations may override the default value and request a 32-bit
12420 @deftypevr {Target Hook} bool TARGET_HAVE_SHADOW_CALL_STACK
12421 This value is true if the target platform supports
12422 @option{-fsanitize=shadow-call-stack}. The default value is false.