1
@c Copyright 2001, 2002, 2003, 2006, 2011 Free Software Foundation, Inc.
2
@c This is part of the GAS manual.
3
@c For copying conditions, see the file as.texinfo.
4
@c MMIX description by Hans-Peter Nilsson, hp@bitrange.com
8
@chapter MMIX Dependent Features
11
@node Machine Dependencies
12
@chapter MMIX Dependent Features
17
* MMIX-Opts:: Command-line Options
18
* MMIX-Expand:: Instruction expansion
19
* MMIX-Syntax:: Syntax
20
* MMIX-mmixal:: Differences to @code{mmixal} syntax and semantics
24
@section Command-line Options
28
The MMIX version of @code{@value{AS}} has some machine-dependent options.
30
@cindex @samp{--fixed-special-register-names} command line option, MMIX
31
When @samp{--fixed-special-register-names} is specified, only the register
32
names specified in @ref{MMIX-Regs} are recognized in the instructions
33
@code{PUT} and @code{GET}.
35
@cindex @samp{--globalize-symbols} command line option, MMIX
36
You can use the @samp{--globalize-symbols} to make all symbols global.
37
This option is useful when splitting up a @code{mmixal} program into
40
@cindex @samp{--gnu-syntax} command line option, MMIX
41
The @samp{--gnu-syntax} turns off most syntax compatibility with
42
@code{mmixal}. Its usability is currently doubtful.
44
@cindex @samp{--relax} command line option, MMIX
45
The @samp{--relax} option is not fully supported, but will eventually make
46
the object file prepared for linker relaxation.
48
@cindex @samp{--no-predefined-syms} command line option, MMIX
49
If you want to avoid inadvertently calling a predefined symbol and would
50
rather get an error, for example when using @code{@value{AS}} with a
51
compiler or other machine-generated code, specify
52
@samp{--no-predefined-syms}. This turns off built-in predefined
53
definitions of all such symbols, including rounding-mode symbols, segment
54
symbols, @samp{BIT} symbols, and @code{TRAP} symbols used in @code{mmix}
55
``system calls''. It also turns off predefined special-register names,
56
except when used in @code{PUT} and @code{GET} instructions.
58
@cindex @samp{--no-expand} command line option, MMIX
59
By default, some instructions are expanded to fit the size of the operand
60
or an external symbol (@pxref{MMIX-Expand}). By passing
61
@samp{--no-expand}, no such expansion will be done, instead causing errors
62
at link time if the operand does not fit.
64
@cindex @samp{--no-merge-gregs} command line option, MMIX
65
The @code{mmixal} documentation (@pxref{mmixsite}) specifies that global
66
registers allocated with the @samp{GREG} directive (@pxref{MMIX-greg}) and
67
initialized to the same non-zero value, will refer to the same global
68
register. This isn't strictly enforceable in @code{@value{AS}} since the
69
final addresses aren't known until link-time, but it will do an effort
70
unless the @samp{--no-merge-gregs} option is specified. (Register merging
71
isn't yet implemented in @code{@value{LD}}.)
73
@cindex @samp{-x} command line option, MMIX
74
@code{@value{AS}} will warn every time it expands an instruction to fit an
75
operand unless the option @samp{-x} is specified. It is believed that
76
this behaviour is more useful than just mimicking @code{mmixal}'s
77
behaviour, in which instructions are only expanded if the @samp{-x} option
78
is specified, and assembly fails otherwise, when an instruction needs to
79
be expanded. It needs to be kept in mind that @code{mmixal} is both an
80
assembler and linker, while @code{@value{AS}} will expand instructions
81
that at link stage can be contracted. (Though linker relaxation isn't yet
82
implemented in @code{@value{LD}}.) The option @samp{-x} also imples
83
@samp{--linker-allocated-gregs}.
85
@cindex @samp{--no-pushj-stubs} command line option, MMIX
86
@cindex @samp{--no-stubs} command line option, MMIX
87
If instruction expansion is enabled, @code{@value{AS}} can expand a
88
@samp{PUSHJ} instruction into a series of instructions. The shortest
89
expansion is to not expand it, but just mark the call as redirectable to a
90
stub, which @code{@value{LD}} creates at link-time, but only if the
91
original @samp{PUSHJ} instruction is found not to reach the target. The
92
stub consists of the necessary instructions to form a jump to the target.
93
This happens if @code{@value{AS}} can assert that the @samp{PUSHJ}
94
instruction can reach such a stub. The option @samp{--no-pushj-stubs}
95
disables this shorter expansion, and the longer series of instructions is
96
then created at assembly-time. The option @samp{--no-stubs} is a synonym,
97
intended for compatibility with future releases, where generation of stubs
98
for other instructions may be implemented.
100
@cindex @samp{--linker-allocated-gregs} command line option, MMIX
101
Usually a two-operand-expression (@pxref{GREG-base}) without a matching
102
@samp{GREG} directive is treated as an error by @code{@value{AS}}. When
103
the option @samp{--linker-allocated-gregs} is in effect, they are instead
104
passed through to the linker, which will allocate as many global registers
108
@section Instruction expansion
110
@cindex instruction expansion, MMIX
111
When @code{@value{AS}} encounters an instruction with an operand that is
112
either not known or does not fit the operand size of the instruction,
113
@code{@value{AS}} (and @code{@value{LD}}) will expand the instruction into
114
a sequence of instructions semantically equivalent to the operand fitting
115
the instruction. Expansion will take place for the following
120
Expands to a sequence of four instructions: @code{SETL}, @code{INCML},
121
@code{INCMH} and @code{INCH}. The operand must be a multiple of four.
122
@item Conditional branches
123
A branch instruction is turned into a branch with the complemented
124
condition and prediction bit over five instructions; four instructions
125
setting @code{$255} to the operand value, which like with @code{GETA} must
126
be a multiple of four, and a final @code{GO $255,$255,0}.
128
Similar to expansion for conditional branches; four instructions set
129
@code{$255} to the operand value, followed by a @code{PUSHGO $255,$255,0}.
131
Similar to conditional branches and @code{PUSHJ}. The final instruction
132
is @code{GO $255,$255,0}.
135
The linker @code{@value{LD}} is expected to shrink these expansions for
136
code assembled with @samp{--relax} (though not currently implemented).
141
The assembly syntax is supposed to be upward compatible with that
142
described in Sections 1.3 and 1.4 of @samp{The Art of Computer
143
Programming, Volume 1}. Draft versions of those chapters as well as other
144
MMIX information is located at
145
@anchor{mmixsite}@url{http://www-cs-faculty.stanford.edu/~knuth/mmix-news.html}.
146
Most code examples from the mmixal package located there should work
147
unmodified when assembled and linked as single files, with a few
148
noteworthy exceptions (@pxref{MMIX-mmixal}).
150
Before an instruction is emitted, the current location is aligned to the
151
next four-byte boundary. If a label is defined at the beginning of the
152
line, its value will be the aligned value.
154
In addition to the traditional hex-prefix @samp{0x}, a hexadecimal number
155
can also be specified by the prefix character @samp{#}.
157
After all operands to an MMIX instruction or directive have been
158
specified, the rest of the line is ignored, treated as a comment.
161
* MMIX-Chars:: Special Characters
162
* MMIX-Symbols:: Symbols
163
* MMIX-Regs:: Register Names
164
* MMIX-Pseudos:: Assembler Directives
168
@subsection Special Characters
169
@cindex line comment characters, MMIX
170
@cindex MMIX line comment characters
172
The characters @samp{*} and @samp{#} are line comment characters; each
173
start a comment at the beginning of a line, but only at the beginning of a
174
line. A @samp{#} prefixes a hexadecimal number if found elsewhere on a
175
line. If a @samp{#} appears at the start of a line the whole line is
176
treated as a comment, but the line can also act as a logical line
177
number directive (@pxref{Comments}) or a preprocessor control command
178
(@pxref{Preprocessing}).
180
Two other characters, @samp{%} and @samp{!}, each start a comment anywhere
181
on the line. Thus you can't use the @samp{modulus} and @samp{not}
182
operators in expressions normally associated with these two characters.
184
A @samp{;} is a line separator, treated as a new-line, so separate
185
instructions can be specified on a single line.
189
The character @samp{:} is permitted in identifiers. There are two
190
exceptions to it being treated as any other symbol character: if a symbol
191
begins with @samp{:}, it means that the symbol is in the global namespace
192
and that the current prefix should not be prepended to that symbol
193
(@pxref{MMIX-prefix}). The @samp{:} is then not considered part of the
194
symbol. For a symbol in the label position (first on a line), a @samp{:}
195
at the end of a symbol is silently stripped off. A label is permitted,
196
but not required, to be followed by a @samp{:}, as with many other
199
The character @samp{@@} in an expression, is a synonym for @samp{.}, the
202
In addition to the common forward and backward local symbol formats
203
(@pxref{Symbol Names}), they can be specified with upper-case @samp{B} and
204
@samp{F}, as in @samp{8B} and @samp{9F}. A local label defined for the
205
current position is written with a @samp{H} appended to the number:
209
This and traditional local-label formats cannot be mixed: a label must be
210
defined and referred to using the same format.
212
There's a minor caveat: just as for the ordinary local symbols, the local
213
symbols are translated into ordinary symbols using control characters are
214
to hide the ordinal number of the symbol. Unfortunately, these symbols
215
are not translated back in error messages. Thus you may see confusing
216
error messages when local symbols are used. Control characters
217
@samp{\003} (control-C) and @samp{\004} (control-D) are used for the
218
MMIX-specific local-symbol syntax.
220
The symbol @samp{Main} is handled specially; it is always global.
222
By defining the symbols @samp{__.MMIX.start..text} and
223
@samp{__.MMIX.start..data}, the address of respectively the @samp{.text}
224
and @samp{.data} segments of the final program can be defined, though when
225
linking more than one object file, the code or data in the object file
226
containing the symbol is not guaranteed to be start at that position; just
227
the final executable. @xref{MMIX-loc}.
230
@subsection Register names
231
@cindex register names, MMIX
232
@cindex MMIX register names
234
Local and global registers are specified as @samp{$0} to @samp{$255}.
235
The recognized special register names are @samp{rJ}, @samp{rA}, @samp{rB},
236
@samp{rC}, @samp{rD}, @samp{rE}, @samp{rF}, @samp{rG}, @samp{rH},
237
@samp{rI}, @samp{rK}, @samp{rL}, @samp{rM}, @samp{rN}, @samp{rO},
238
@samp{rP}, @samp{rQ}, @samp{rR}, @samp{rS}, @samp{rT}, @samp{rU},
239
@samp{rV}, @samp{rW}, @samp{rX}, @samp{rY}, @samp{rZ}, @samp{rBB},
240
@samp{rTT}, @samp{rWW}, @samp{rXX}, @samp{rYY} and @samp{rZZ}. A leading
241
@samp{:} is optional for special register names.
243
Local and global symbols can be equated to register names and used in
244
place of ordinary registers.
246
Similarly for special registers, local and global symbols can be used.
247
Also, symbols equated from numbers and constant expressions are allowed in
248
place of a special register, except when either of the options
249
@code{--no-predefined-syms} and @code{--fixed-special-register-names} are
250
specified. Then only the special register names above are allowed for the
251
instructions having a special register operand; @code{GET} and @code{PUT}.
254
@subsection Assembler Directives
255
@cindex assembler directives, MMIX
256
@cindex pseudo-ops, MMIX
257
@cindex MMIX assembler directives
258
@cindex MMIX pseudo-ops
262
@cindex assembler directive LOC, MMIX
263
@cindex pseudo-op LOC, MMIX
264
@cindex MMIX assembler directive LOC
265
@cindex MMIX pseudo-op LOC
268
The @code{LOC} directive sets the current location to the value of the
269
operand field, which may include changing sections. If the operand is a
270
constant, the section is set to either @code{.data} if the value is
271
@code{0x2000000000000000} or larger, else it is set to @code{.text}.
272
Within a section, the current location may only be changed to
273
monotonically higher addresses. A LOC expression must be a previously
274
defined symbol or a ``pure'' constant.
276
An example, which sets the label @var{prev} to the current location, and
277
updates the current location to eight bytes forward:
282
When a LOC has a constant as its operand, a symbol
283
@code{__.MMIX.start..text} or @code{__.MMIX.start..data} is defined
284
depending on the address as mentioned above. Each such symbol is
285
interpreted as special by the linker, locating the section at that
286
address. Note that if multiple files are linked, the first object file
287
with that section will be mapped to that address (not necessarily the file
288
with the LOC definition).
291
@cindex assembler directive LOCAL, MMIX
292
@cindex pseudo-op LOCAL, MMIX
293
@cindex MMIX assembler directive LOCAL
294
@cindex MMIX pseudo-op LOCAL
299
LOCAL external_symbol
304
This directive-operation generates a link-time assertion that the operand
305
does not correspond to a global register. The operand is an expression
306
that at link-time resolves to a register symbol or a number. A number is
307
treated as the register having that number. There is one restriction on
308
the use of this directive: the pseudo-directive must be placed in a
309
section with contents, code or data.
312
@cindex assembler directive IS, MMIX
313
@cindex pseudo-op IS, MMIX
314
@cindex MMIX assembler directive IS
315
@cindex MMIX pseudo-op IS
318
The @code{IS} directive:
320
asymbol IS an_expression
322
sets the symbol @samp{asymbol} to @samp{an_expression}. A symbol may not
323
be set more than once using this directive. Local labels may be set using
324
this directive, for example:
330
@cindex assembler directive GREG, MMIX
331
@cindex pseudo-op GREG, MMIX
332
@cindex MMIX assembler directive GREG
333
@cindex MMIX pseudo-op GREG
336
This directive reserves a global register, gives it an initial value and
337
optionally gives it a symbolic name. Some examples:
343
.greg creg, another_data_value
346
The symbolic register name can be used in place of a (non-special)
347
register. If a value isn't provided, it defaults to zero. Unless the
348
option @samp{--no-merge-gregs} is specified, non-zero registers allocated
349
with this directive may be eliminated by @code{@value{AS}}; another
350
register with the same value used in its place.
351
Any of the instructions
382
can have a value nearby @anchor{GREG-base}an initial value in place of its
383
second and third operands. Here, ``nearby'' is defined as within the
384
range 0@dots{}255 from the initial value of such an allocated register.
387
buffer1 BYTE 0,0,0,0,0
388
buffer2 BYTE 0,0,0,0,0
393
In the example above, the @samp{Y} field of the @code{LDOUI} instruction
394
(LDOU with a constant Z) will be replaced with the global register
395
allocated for @samp{buffer1}, and the @samp{Z} field will have the value
396
5, the offset from @samp{buffer1} to @samp{buffer2}. The result is
397
equivalent to this code:
399
buffer1 BYTE 0,0,0,0,0
400
buffer2 BYTE 0,0,0,0,0
403
LDOU $42,tmpreg,(buffer2-buffer1)
406
Global registers allocated with this directive are allocated in order
407
higher-to-lower within a file. Other than that, the exact order of
408
register allocation and elimination is undefined. For example, the order
409
is undefined when more than one file with such directives are linked
410
together. With the options @samp{-x} and @samp{--linker-allocated-gregs},
411
@samp{GREG} directives for two-operand cases like the one mentioned above
412
can be omitted. Sufficient global registers will then be allocated by the
416
@cindex assembler directive BYTE, MMIX
417
@cindex pseudo-op BYTE, MMIX
418
@cindex MMIX assembler directive BYTE
419
@cindex MMIX pseudo-op BYTE
422
The @samp{BYTE} directive takes a series of operands separated by a comma.
423
If an operand is a string (@pxref{Strings}), each character of that string
424
is emitted as a byte. Other operands must be constant expressions without
425
forward references, in the range 0@dots{}255. If you need operands having
426
expressions with forward references, use @samp{.byte} (@pxref{Byte}). An
427
operand can be omitted, defaulting to a zero value.
432
@cindex assembler directive WYDE, MMIX
433
@cindex pseudo-op WYDE, MMIX
434
@cindex MMIX assembler directive WYDE
435
@cindex MMIX pseudo-op WYDE
436
@cindex assembler directive TETRA, MMIX
437
@cindex pseudo-op TETRA, MMIX
438
@cindex MMIX assembler directive TETRA
439
@cindex MMIX pseudo-op TETRA
440
@cindex assembler directive OCTA, MMIX
441
@cindex pseudo-op OCTA, MMIX
442
@cindex MMIX assembler directive OCTA
443
@cindex MMIX pseudo-op OCTA
445
@anchor{MMIX-constants}
446
The directives @samp{WYDE}, @samp{TETRA} and @samp{OCTA} emit constants of
447
two, four and eight bytes size respectively. Before anything else happens
448
for the directive, the current location is aligned to the respective
449
constant-size boundary. If a label is defined at the beginning of the
450
line, its value will be that after the alignment. A single operand can be
451
omitted, defaulting to a zero value emitted for the directive. Operands
452
can be expressed as strings (@pxref{Strings}), in which case each
453
character in the string is emitted as a separate constant of the size
454
indicated by the directive.
457
@cindex assembler directive PREFIX, MMIX
458
@cindex pseudo-op PREFIX, MMIX
459
@cindex MMIX assembler directive PREFIX
460
@cindex MMIX pseudo-op PREFIX
463
The @samp{PREFIX} directive sets a symbol name prefix to be prepended to
464
all symbols (except local symbols, @pxref{MMIX-Symbols}), that are not
465
prefixed with @samp{:}, until the next @samp{PREFIX} directive. Such
466
prefixes accumulate. For example,
472
defines a symbol @samp{abc} with the value 0.
476
@cindex assembler directive BSPEC, MMIX
477
@cindex pseudo-op BSPEC, MMIX
478
@cindex MMIX assembler directive BSPEC
479
@cindex MMIX pseudo-op BSPEC
480
@cindex assembler directive ESPEC, MMIX
481
@cindex pseudo-op ESPEC, MMIX
482
@cindex MMIX assembler directive ESPEC
483
@cindex MMIX pseudo-op ESPEC
486
A pair of @samp{BSPEC} and @samp{ESPEC} directives delimit a section of
487
special contents (without specified semantics). Example:
493
The single operand to @samp{BSPEC} must be number in the range
494
0@dots{}255. The @samp{BSPEC} number 80 is used by the GNU binutils
499
@section Differences to @code{mmixal}
500
@cindex mmixal differences
501
@cindex differences, mmixal
503
The binutils @code{@value{AS}} and @code{@value{LD}} combination has a few
504
differences in function compared to @code{mmixal} (@pxref{mmixsite}).
506
The replacement of a symbol with a GREG-allocated register
507
(@pxref{GREG-base}) is not handled the exactly same way in
508
@code{@value{AS}} as in @code{mmixal}. This is apparent in the
509
@code{mmixal} example file @code{inout.mms}, where different registers
510
with different offsets, eventually yielding the same address, are used in
511
the first instruction. This type of difference should however not affect
512
the function of any program unless it has specific assumptions about the
513
allocated register number.
515
Line numbers (in the @samp{mmo} object format) are currently not
518
Expression operator precedence is not that of mmixal: operator precedence
519
is that of the C programming language. It's recommended to use
520
parentheses to explicitly specify wanted operator precedence whenever more
521
than one type of operators are used.
523
The serialize unary operator @code{&}, the fractional division operator
524
@samp{//}, the logical not operator @code{!} and the modulus operator
525
@samp{%} are not available.
527
Symbols are not global by default, unless the option
528
@samp{--globalize-symbols} is passed. Use the @samp{.global} directive to
529
globalize symbols (@pxref{Global}).
531
Operand syntax is a bit stricter with @code{@value{AS}} than
532
@code{mmixal}. For example, you can't say @code{addu 1,2,3}, instead you
533
must write @code{addu $1,$2,3}.
535
You can't LOC to a lower address than those already visited
536
(i.e., ``backwards'').
538
A LOC directive must come before any emitted code.
540
Predefined symbols are visible as file-local symbols after use. (In the
541
ELF file, that is---the linked mmo file has no notion of a file-local
544
Some mapping of constant expressions to sections in LOC expressions is
545
attempted, but that functionality is easily confused and should be avoided
546
unless compatibility with @code{mmixal} is required. A LOC expression to
547
@samp{0x2000000000000000} or higher, maps to the @samp{.data} section and
548
lower addresses map to the @samp{.text} section (@pxref{MMIX-loc}).
550
The code and data areas are each contiguous. Sparse programs with
551
far-away LOC directives will take up the same amount of space as a
552
contiguous program with zeros filled in the gaps between the LOC
553
directives. If you need sparse programs, you might try and get the wanted
554
effect with a linker script and splitting up the code parts into sections
555
(@pxref{Section}). Assembly code for this, to be compatible with
556
@code{mmixal}, would look something like:
564
@code{@value{AS}} will not execute the LOC directive and @code{mmixal}
565
ignores the lines with @code{.}. This construct can be used generally to
568
Symbols can't be defined twice--not even to the same value.
570
Instruction mnemonics are recognized case-insensitive, though the
571
@samp{IS} and @samp{GREG} pseudo-operations must be specified in
572
upper-case characters.
574
There's no unicode support.
576
The following is a list of programs in @samp{mmix.tar.gz}, available at
577
@url{http://www-cs-faculty.stanford.edu/~knuth/mmix-news.html}, last
578
checked with the version dated 2001-08-25 (md5sum
579
c393470cfc86fac040487d22d2bf0172) that assemble with @code{mmixal} but do
580
not assemble with @code{@value{AS}}:
584
LOC to a previous address.
586
Redefines symbol @samp{Done}.
588
Uses the serial operator @samp{&}.