1 @c Copyright (C) 2009-2020 Free Software Foundation, Inc.
2 @c Contributed by ARM Ltd.
3 @c This is part of the GAS manual.
4 @c For copying conditions, see the file as.texinfo.
9 @node AArch64-Dependent
10 @chapter AArch64 Dependent Features
14 @node Machine Dependencies
15 @chapter AArch64 Dependent Features
18 @cindex AArch64 support
20 * AArch64 Options:: Options
21 * AArch64 Extensions:: Extensions
22 * AArch64 Syntax:: Syntax
23 * AArch64 Floating Point:: Floating Point
24 * AArch64 Directives:: AArch64 Machine Directives
25 * AArch64 Opcodes:: Opcodes
26 * AArch64 Mapping Symbols:: Mapping Symbols
31 @cindex AArch64 options (none)
32 @cindex options for AArch64 (none)
37 @cindex @option{-EB} command-line option, AArch64
39 This option specifies that the output generated by the assembler should
40 be marked as being encoded for a big-endian processor.
42 @cindex @option{-EL} command-line option, AArch64
44 This option specifies that the output generated by the assembler should
45 be marked as being encoded for a little-endian processor.
47 @cindex @option{-mabi=} command-line option, AArch64
49 Specify which ABI the source code uses. The recognized arguments
50 are: @code{ilp32} and @code{lp64}, which decides the generated object
51 file in ELF32 and ELF64 format respectively. The default is @code{lp64}.
53 @cindex @option{-mcpu=} command-line option, AArch64
54 @item -mcpu=@var{processor}[+@var{extension}@dots{}]
55 This option specifies the target processor. The assembler will issue an error
56 message if an attempt is made to assemble an instruction which will not execute
57 on the target processor. The following processor names are recognized:
89 The special name @code{all} may be used to allow the assembler to accept
90 instructions valid for any supported processor, including all optional
93 In addition to the basic instruction set, the assembler can be told to
94 accept, or restrict, various extension mnemonics that extend the
95 processor. @xref{AArch64 Extensions}.
97 If some implementations of a particular processor can have an
98 extension, then then those extensions are automatically enabled.
99 Consequently, you will not normally have to specify any additional
102 @cindex @option{-march=} command-line option, AArch64
103 @item -march=@var{architecture}[+@var{extension}@dots{}]
104 This option specifies the target architecture. The assembler will
105 issue an error message if an attempt is made to assemble an
106 instruction which will not execute on the target architecture. The
107 following architecture names are recognized: @code{armv8-a},
108 @code{armv8.1-a}, @code{armv8.2-a}, @code{armv8.3-a}, @code{armv8.4-a}
109 @code{armv8.5-a}, @code{armv8.6-a}, and @code{armv8-r}.
111 If both @option{-mcpu} and @option{-march} are specified, the
112 assembler will use the setting for @option{-mcpu}. If neither are
113 specified, the assembler will default to @option{-mcpu=all}.
115 The architecture option can be extended with the same instruction set
116 extension options as the @option{-mcpu} option. Unlike
117 @option{-mcpu}, extensions are not always enabled by default,
118 @xref{AArch64 Extensions}.
120 @cindex @code{-mverbose-error} command-line option, AArch64
121 @item -mverbose-error
122 This option enables verbose error messages for AArch64 gas. This option
123 is enabled by default.
125 @cindex @code{-mno-verbose-error} command-line option, AArch64
126 @item -mno-verbose-error
127 This option disables verbose error messages in AArch64 gas.
132 @node AArch64 Extensions
133 @section Architecture Extensions
135 The table below lists the permitted architecture extensions that are
136 supported by the assembler and the conditions under which they are
137 automatically enabled.
139 Multiple extensions may be specified, separated by a @code{+}.
140 Extension mnemonics may also be removed from those the assembler
141 accepts. This is done by prepending @code{no} to the option that adds
142 the extension. Extensions that are removed must be listed after all
143 extensions that have been added.
145 Enabling an extension that requires other extensions will
146 automatically cause those extensions to be enabled. Similarly,
147 disabling an extension that is required by other extensions will
148 automatically cause those extensions to be disabled.
150 @multitable @columnfractions .12 .17 .17 .54
151 @headitem Extension @tab Minimum Architecture @tab Enabled by default
153 @item @code{i8mm} @tab ARMv8.2-A @tab ARMv8.6-A or later
154 @tab Enable Int8 Matrix Multiply extension.
155 @item @code{f32mm} @tab ARMv8.2-A @tab No
156 @tab Enable F32 Matrix Multiply extension.
157 @item @code{f64mm} @tab ARMv8.2-A @tab No
158 @tab Enable F64 Matrix Multiply extension.
159 @item @code{bf16} @tab ARMv8.2-A @tab ARMv8.6-A or later
160 @tab Enable BFloat16 extension.
161 @item @code{compnum} @tab ARMv8.2-A @tab ARMv8.3-A or later
162 @tab Enable the complex number SIMD extensions. This implies
163 @code{fp16} and @code{simd}.
164 @item @code{crc} @tab ARMv8-A @tab ARMv8.1-A or later
165 @tab Enable CRC instructions.
166 @item @code{crypto} @tab ARMv8-A @tab No
167 @tab Enable cryptographic extensions. This implies @code{fp}, @code{simd}, @code{aes} and @code{sha2}.
168 @item @code{aes} @tab ARMv8-A @tab No
169 @tab Enable the AES cryptographic extensions. This implies @code{fp} and @code{simd}.
170 @item @code{sha2} @tab ARMv8-A @tab No
171 @tab Enable the SHA2 cryptographic extensions. This implies @code{fp} and @code{simd}.
172 @item @code{sha3} @tab ARMv8.2-A @tab No
173 @tab Enable the ARMv8.2-A SHA2 and SHA3 cryptographic extensions. This implies @code{fp}, @code{simd} and @code{sha2}.
174 @item @code{sm4} @tab ARMv8.2-A @tab No
175 @tab Enable the ARMv8.2-A SM3 and SM4 cryptographic extensions. This implies @code{fp} and @code{simd}.
176 @item @code{fp} @tab ARMv8-A @tab ARMv8-A or later
177 @tab Enable floating-point extensions.
178 @item @code{fp16} @tab ARMv8.2-A @tab ARMv8.2-A or later
179 @tab Enable ARMv8.2 16-bit floating-point support. This implies
181 @item @code{lor} @tab ARMv8-A @tab ARMv8.1-A or later
182 @tab Enable Limited Ordering Regions extensions.
183 @item @code{lse} @tab ARMv8-A @tab ARMv8.1-A or later
184 @tab Enable Large System extensions.
185 @item @code{pan} @tab ARMv8-A @tab ARMv8.1-A or later
186 @tab Enable Privileged Access Never support.
187 @item @code{profile} @tab ARMv8.2-A @tab No
188 @tab Enable statistical profiling extensions.
189 @item @code{ras} @tab ARMv8-A @tab ARMv8.2-A or later
190 @tab Enable the Reliability, Availability and Serviceability
192 @item @code{rcpc} @tab ARMv8.2-A @tab ARMv8.3-A or later
193 @tab Enable the weak release consistency extension.
194 @item @code{rdma} @tab ARMv8-A @tab ARMv8.1-A or later
195 @tab Enable ARMv8.1 Advanced SIMD extensions. This implies @code{simd}.
196 @item @code{simd} @tab ARMv8-A @tab ARMv8-A or later
197 @tab Enable Advanced SIMD extensions. This implies @code{fp}.
198 @item @code{sve} @tab ARMv8.2-A @tab No
199 @tab Enable the Scalable Vector Extensions. This implies @code{fp16},
200 @code{simd} and @code{compnum}.
201 @item @code{dotprod} @tab ARMv8.2-A @tab ARMv8.4-A or later
202 @tab Enable the Dot Product extension. This implies @code{simd}.
203 @item @code{fp16fml} @tab ARMv8.2-A @tab ARMv8.4-A or later
204 @tab Enable ARMv8.2 16-bit floating-point multiplication variant support.
205 This implies @code{fp16}.
206 @item @code{sb} @tab ARMv8-A @tab ARMv8.5-A or later
207 @tab Enable the speculation barrier instruction sb.
208 @item @code{predres} @tab ARMv8-A @tab ARMv8.5-A or later
209 @tab Enable the Execution and Data and Prediction instructions.
210 @item @code{rng} @tab ARMv8.5-A @tab No
211 @tab Enable ARMv8.5-A random number instructions.
212 @item @code{ssbs} @tab ARMv8-A @tab ARMv8.5-A or later
213 @tab Enable Speculative Store Bypassing Safe state read and write.
214 @item @code{memtag} @tab ARMv8.5-A @tab No
215 @tab Enable ARMv8.5-A Memory Tagging Extensions.
216 @item @code{tme} @tab ARMv8-A @tab No
217 @tab Enable Transactional Memory Extensions.
218 @item @code{sve2} @tab ARMv8-A @tab No
219 @tab Enable the SVE2 Extension.
220 @item @code{sve2-bitperm} @tab ARMv8-A @tab No
221 @tab Enable SVE2 BITPERM Extension.
222 @item @code{sve2-sm4} @tab ARMv8-A @tab No
223 @tab Enable SVE2 SM4 Extension.
224 @item @code{sve2-aes} @tab ARMv8-A @tab No
225 @tab Enable SVE2 AES Extension. This also enables the .Q->.B form of the
226 @code{pmullt} and @code{pmullb} instructions.
227 @item @code{sve2-sha3} @tab ARMv8-A @tab No
228 @tab Enable SVE2 SHA3 Extension.
234 * AArch64-Chars:: Special Characters
235 * AArch64-Regs:: Register Names
236 * AArch64-Relocations:: Relocations
240 @subsection Special Characters
242 @cindex line comment character, AArch64
243 @cindex AArch64 line comment character
244 The presence of a @samp{//} on a line indicates the start of a comment
245 that extends to the end of the current line. If a @samp{#} appears as
246 the first character of a line, the whole line is treated as a comment.
248 @cindex line separator, AArch64
249 @cindex statement separator, AArch64
250 @cindex AArch64 line separator
251 The @samp{;} character can be used instead of a newline to separate
254 @cindex immediate character, AArch64
255 @cindex AArch64 immediate character
256 The @samp{#} can be optionally used to indicate immediate operands.
259 @subsection Register Names
261 @cindex AArch64 register names
262 @cindex register names, AArch64
263 Please refer to the section @samp{4.4 Register Names} of
264 @samp{ARMv8 Instruction Set Overview}, which is available at
265 @uref{http://infocenter.arm.com}.
267 @node AArch64-Relocations
268 @subsection Relocations
270 @cindex relocations, AArch64
271 @cindex AArch64 relocations
272 @cindex MOVN, MOVZ and MOVK group relocations, AArch64
273 Relocations for @samp{MOVZ} and @samp{MOVK} instructions can be generated
274 by prefixing the label with @samp{#:abs_g2:} etc.
275 For example to load the 48-bit absolute address of @var{foo} into x0:
278 movz x0, #:abs_g2:foo // bits 32-47, overflow check
279 movk x0, #:abs_g1_nc:foo // bits 16-31, no overflow check
280 movk x0, #:abs_g0_nc:foo // bits 0-15, no overflow check
283 @cindex ADRP, ADD, LDR/STR group relocations, AArch64
284 Relocations for @samp{ADRP}, and @samp{ADD}, @samp{LDR} or @samp{STR}
285 instructions can be generated by prefixing the label with
286 @samp{:pg_hi21:} and @samp{#:lo12:} respectively.
288 For example to use 33-bit (+/-4GB) pc-relative addressing to
289 load the address of @var{foo} into x0:
292 adrp x0, :pg_hi21:foo
293 add x0, x0, #:lo12:foo
296 Or to load the value of @var{foo} into x0:
299 adrp x0, :pg_hi21:foo
300 ldr x0, [x0, #:lo12:foo]
303 Note that @samp{:pg_hi21:} is optional.
312 adrp x0, :pg_hi21:foo
315 @node AArch64 Floating Point
316 @section Floating Point
318 @cindex floating point, AArch64 (@sc{ieee})
319 @cindex AArch64 floating point (@sc{ieee})
320 The AArch64 architecture uses @sc{ieee} floating-point numbers.
322 @node AArch64 Directives
323 @section AArch64 Machine Directives
325 @cindex machine directives, AArch64
326 @cindex AArch64 machine directives
329 @c AAAAAAAAAAAAAAAAAAAAAAAAA
331 @cindex @code{.arch} directive, AArch64
332 @item .arch @var{name}
333 Select the target architecture. Valid values for @var{name} are the same as
334 for the @option{-march} command-line option.
336 Specifying @code{.arch} clears any previously selected architecture
339 @cindex @code{.arch_extension} directive, AArch64
340 @item .arch_extension @var{name}
341 Add or remove an architecture extension to the target architecture. Valid
342 values for @var{name} are the same as those accepted as architectural
343 extensions by the @option{-mcpu} command-line option.
345 @code{.arch_extension} may be used multiple times to add or remove extensions
346 incrementally to the architecture being compiled for.
348 @c BBBBBBBBBBBBBBBBBBBBBBBBBB
350 @cindex @code{.bss} directive, AArch64
352 This directive switches to the @code{.bss} section.
354 @c CCCCCCCCCCCCCCCCCCCCCCCCCC
356 @cindex @code{.cpu} directive, AArch64
357 @item .cpu @var{name}
358 Set the target processor. Valid values for @var{name} are the same as
359 those accepted by the @option{-mcpu=} command-line option.
361 @c DDDDDDDDDDDDDDDDDDDDDDDDDD
363 @cindex @code{.dword} directive, AArch64
364 @item .dword @var{expressions}
365 The @code{.dword} directive produces 64 bit values.
367 @c EEEEEEEEEEEEEEEEEEEEEEEEEE
369 @cindex @code{.even} directive, AArch64
371 The @code{.even} directive aligns the output on the next even byte
374 @c FFFFFFFFFFFFFFFFFFFFFFFFFF
376 @cindex @code{.float16} directive, AArch64
377 @item .float16 @var{value [,...,value_n]}
378 Place the half precision floating point representation of one or more
379 floating-point values into the current section.
380 The format used to encode the floating point values is always the
381 IEEE 754-2008 half precision floating point format.
383 @c GGGGGGGGGGGGGGGGGGGGGGGGGG
384 @c HHHHHHHHHHHHHHHHHHHHHHHHHH
385 @c IIIIIIIIIIIIIIIIIIIIIIIIII
387 @cindex @code{.inst} directive, AArch64
388 @item .inst @var{expressions}
389 Inserts the expressions into the output as if they were instructions,
392 @c JJJJJJJJJJJJJJJJJJJJJJJJJJ
393 @c KKKKKKKKKKKKKKKKKKKKKKKKKK
394 @c LLLLLLLLLLLLLLLLLLLLLLLLLL
396 @cindex @code{.ltorg} directive, AArch64
398 This directive causes the current contents of the literal pool to be
399 dumped into the current section (which is assumed to be the .text
400 section) at the current location (aligned to a word boundary).
401 GAS maintains a separate literal pool for each section and each
402 sub-section. The @code{.ltorg} directive will only affect the literal
403 pool of the current section and sub-section. At the end of assembly
404 all remaining, un-empty literal pools will automatically be dumped.
406 Note - older versions of GAS would dump the current literal
407 pool any time a section change occurred. This is no longer done, since
408 it prevents accurate control of the placement of literal pools.
410 @c MMMMMMMMMMMMMMMMMMMMMMMMMM
412 @c NNNNNNNNNNNNNNNNNNNNNNNNNN
413 @c OOOOOOOOOOOOOOOOOOOOOOOOOO
415 @c PPPPPPPPPPPPPPPPPPPPPPPPPP
417 @cindex @code{.pool} directive, AArch64
419 This is a synonym for .ltorg.
421 @c QQQQQQQQQQQQQQQQQQQQQQQQQQ
422 @c RRRRRRRRRRRRRRRRRRRRRRRRRR
424 @cindex @code{.req} directive, AArch64
425 @item @var{name} .req @var{register name}
426 This creates an alias for @var{register name} called @var{name}. For
433 ip0, ip1, lr and fp are automatically defined to
434 alias to X16, X17, X30 and X29 respectively.
436 @c SSSSSSSSSSSSSSSSSSSSSSSSSS
438 @c TTTTTTTTTTTTTTTTTTTTTTTTTT
440 @cindex @code{.tlsdescadd} directive, AArch64
441 @item @code{.tlsdescadd}
442 Emits a TLSDESC_ADD reloc on the next instruction.
444 @cindex @code{.tlsdesccall} directive, AArch64
445 @item @code{.tlsdesccall}
446 Emits a TLSDESC_CALL reloc on the next instruction.
448 @cindex @code{.tlsdescldr} directive, AArch64
449 @item @code{.tlsdescldr}
450 Emits a TLSDESC_LDR reloc on the next instruction.
452 @c UUUUUUUUUUUUUUUUUUUUUUUUUU
454 @cindex @code{.unreq} directive, AArch64
455 @item .unreq @var{alias-name}
456 This undefines a register alias which was previously defined using the
457 @code{req} directive. For example:
464 An error occurs if the name is undefined. Note - this pseudo op can
465 be used to delete builtin in register name aliases (eg 'w0'). This
466 should only be done if it is really necessary.
468 @c VVVVVVVVVVVVVVVVVVVVVVVVVV
470 @cindex @code{.variant_pcs} directive, AArch64
471 @item .variant_pcs @var{symbol}
472 This directive marks @var{symbol} referencing a function that may
473 follow a variant procedure call standard with different register
474 usage convention from the base procedure call standard.
476 @c WWWWWWWWWWWWWWWWWWWWWWWWWW
477 @c XXXXXXXXXXXXXXXXXXXXXXXXXX
479 @cindex @code{.xword} directive, AArch64
480 @item .xword @var{expressions}
481 The @code{.xword} directive produces 64 bit values. This is the same
482 as the @code{.dword} directive.
484 @c YYYYYYYYYYYYYYYYYYYYYYYYYY
485 @c ZZZZZZZZZZZZZZZZZZZZZZZZZZ
487 @cindex @code{.cfi_b_key_frame} directive, AArch64
488 @item @code{.cfi_b_key_frame}
489 The @code{.cfi_b_key_frame} directive inserts a 'B' character into the CIE
490 corresponding to the current frame's FDE, meaning that its return address has
491 been signed with the B-key. If two frames are signed with differing keys then
492 they will not share the same CIE. This information is intended to be used by
493 the stack unwinder in order to properly authenticate return addresses.
497 @node AArch64 Opcodes
500 @cindex AArch64 opcodes
501 @cindex opcodes for AArch64
502 GAS implements all the standard AArch64 opcodes. It also
503 implements several pseudo opcodes, including several synthetic load
508 @cindex @code{LDR reg,=<expr>} pseudo op, AArch64
511 ldr <register> , =<expression>
514 The constant expression will be placed into the nearest literal pool (if it not
515 already there) and a PC-relative LDR instruction will be generated.
519 For more information on the AArch64 instruction set and assembly language
520 notation, see @samp{ARMv8 Instruction Set Overview} available at
521 @uref{http://infocenter.arm.com}.
524 @node AArch64 Mapping Symbols
525 @section Mapping Symbols
527 The AArch64 ELF specification requires that special symbols be inserted
528 into object files to mark certain features:
534 At the start of a region of code containing AArch64 instructions.
538 At the start of a region of data.