Psellos
Life So Short, the Craft So Long to Learn

Convert Linux ARM Assembly Code for iOS (Update 3)

August 28, 2012

While getting OCaml 4.00.0 working on iOS, I’ve learned quite a bit about the differences between the Linux ARM assembler and the iOS ARM assembler. The two are related: the iOS assembler is like a cousin of the Linux assembler whose ancestors left the home country a few generations back. However, the differences between the two were one of the main problems I had to solve.

This information was hard won: documentation for the iOS assembler seems rare to nonexistent (email me if you know where to find it). So I spent many hours reading the source code of as(1) at Apple’s open source site. To help others avoid duplicating the effort, I’m publishing here the latest version of my Python script arm-as-to-ios that converts ARM assembly code from the Linux format to the required iOS format. This script contains all the wisdom I’ve accumulated so far (including some that’s not required for the OCaml-on-iOS project).

Note: For this third update, I added a transformation to work around a problem with incremental linking of generated object files. You can read the details of the problem and the workaround in my blog post OCaml 4.00.0 on iOS: Progress Report.

I wrote the script specifically to convert the hand-written ARM assembly files of the OCaml runtime to work on iOS. The advantages of using a script are that the conversion is done consistently, and the script might still be useful if the assembly code is rewritten in the future.

Another advantage is that the script might help other people who need to port ARM assembly code from Linux to iOS. Granted, there probably aren’t a lot of people doing this. But if you are, maybe the script will be a useful starting point or at least a careful list of differences between the assemblers.

The output of the script is upward compatible. By this I mean that a converted assembly file works in its originally supported Linux environments as well as in iOS. This means that the converted assembly files can be sent “upstream” to maintainers (if they’re comfortable with supporting iOS and/or the extra complexity of the code).

The current version of arm-as-to-ios is 1.4.0. In its current form, it does the following conversions:

  • Declare the ARM architecture for iOS with a .machine pseudo-op. The possibilities are armv6 and armv7.

  • Specify that armv7 code should use the more space-efficient Thumb encoding in iOS.

  • Generate declarations for functions, similar to the .type pseudo-op for Linux assemblers. It appears that Apple requires .thumb_func declarations only for Thumb functions. Other symbols are apparently assumed to be ARM functions. To make this work upward-compatibly, I define an assembly macro named .funtype that is expanded properly in the different environments.

  • Make sure that global symbols have the right format. For Linux, they look like “abc”. For iOS, they look like “_abc” (with a leading underscore). To support upward compatibility, this is handled by a cpp macro named Glo() that generates the proper symbol form for each environment.

  • Make sure that assembler-local symbols have the right format. For Linux assemblers, they look like “.Lxxx”. For the iOS assembler, they look like “Lxxx”. To support upward compatibly, this is handled by a cpp macro named Loc() that generates the proper symbol form for each environment.

  • Replace uses of =value notation by explicit loads from memory. The Linux ARM assemblers interpret ldr rM, =value to mean that the value should be loaded into register M immediately (using mov) if possible, and loaded from memory (using ldr with PC-relative addressing) otherwise. The Apple assembler seems not to support this, so arm-as-to-ios replaces uses of =value with explicit memory loads, emitting the pool of values into the .text segment at the end of the file.

  • Convert jump tables (for the tbh instruction) to a form that works with the iOS assembler. The form commonly used in Linux generates a bad jump table under iOS—apparently the iOS assembler interprets the expressions differently.

  • Convert “dot relative” expressions to a form that the iOS assembler correctly treats as assembly-time constants. This is a workaround for what looks very much like an iOS assembler bug that prevents incremental linking of the generated object files. (Or possibly it’s a bug in the iOS linker.) Details of the problem and the workaround are given in my blog post OCaml 4.00.0 on iOS: Progress Report. You can also see an example of this conversion below.

  • Remove uses of two pseudo-ops, .type and .size, when assembling for iOS. They aren’t supported by Apple’s assembler. This is done by defining null assembly macros for them.

  • Define a macro cbz when using ARM encodings for iOS. The cbz instruction is Thumb-only. The definition replaces it with a pair of ARM instructions.

You can download the script here:

The full text of the script is also included at the end of this post.

This is the fourth version of arm-as-to-ios, and there may well be a few more changes when I work on the armv6 architecture. As I make changes, I’ll keep the linked script up to date. If there are significant changes I’ll make another post about them.

If you want to try out arm-as-to-ios, copy and paste the lines from the end of this post into a file named arm-as-to-ios, or download it from the above link. Mark it as a script with chmod:

$ chmod +x arm-as-to-ios

To use the script, specify the name of an ARM assembly file. If no files are given, the script processes its standard input.

The following small example demonstrates the translations that arm-as-to-ios performs. Here is a small file of Linux ARM assembly code:

        .syntax unified
        .text
        .align  2
        .globl  example
        .type example, %function
example:
        sub     r10, r10, 8
        cmp     r10, r11
        bcc     1f
        bx      lr
1:
        ldr     r7, =last_return_address
        str     lr, [r7]
        bl      .Lcall_gc
        ldr     lr, [r7]
        b       example

.Lcall_gc:
        ldr     r12, =bottom_of_stack
        str     sp, [r12]
        bl      garbage_collection
        bx      lr

/* Jump table fragment */

.Ljump:
        tbh     [pc, r4, lsl #1]
        .short  (.Ltest1-.)/2+0
        .short  (.Ltest2-.)/2+1
        .short  (.Ltest3-.)/2+2
.Ltest1:
        ldr     r4, [r4]
.Ltest2:
        str     r6, [sp, 8]
.Ltest3:
        str     r8, [r12]

/* Dot relative (frame table fragment) */

        .word   .L200000 - . + 0xfc000000
        .long   0x1c9080
        .thumb_func     example
        .word   example
        .short  8
        .short  0
.L200000:
        .asciz  "example.ml"

If you run arm-as-to-ios on this file, you get the following output that works for both Linux and iOS assemblers:

        .syntax unified

/* Apple compatibility macros */
#if defined(SYS_macosx)
#define Glo(s) _##s
#define Loc(s) L##s
#if defined(MODEL_armv6)
        .machine  armv6
        .macro  .funtype
        .endm
        .macro  cbz
        cmp     $0, #0
        beq     $1
        .endm
#else
        .machine  armv7
        .thumb
        .macro  .funtype
        .thumb_func $0
        .endm
#endif
        .macro  .type
        .endm
        .macro  .size
        .endm
#else
#define Glo(s) s
#define Loc(s) .L##s
        .macro  .funtype symbol
        .type  \symbol, %function
        .endm
#endif
/* End Apple compatibility macros */

        .text
        .align  2
        .globl  Glo(example)
        .funtype  Glo(example)
Glo(example):
        sub     r10, r10, 8
        cmp     r10, r11
        bcc     1f
        bx      lr
1:
        ldr     r7, Loc(Plast_return_address)
        str     lr, [r7]
        bl      Loc(call_gc)
        ldr     lr, [r7]
        b       Glo(example)

Loc(call_gc):
        ldr     r12, Loc(Pbottom_of_stack)
        str     sp, [r12]
        bl      Glo(garbage_collection)
        bx      lr

/* Jump table fragment */

Loc(jump):
        tbh     [pc, r4, lsl #1]
Loc(B27):
        .short  (Loc(test1)-Loc(B27))/2
        .short  (Loc(test2)-Loc(B27))/2
        .short  (Loc(test3)-Loc(B27))/2
Loc(test1):
        ldr     r4, [r4]
Loc(test2):
        str     r6, [sp, 8]
Loc(test3):
        str     r8, [r12]

/* Dot relative (frame table fragment) */

DR40 =   Loc(200000) - . + 0xfc000000
        .word DR40
        .long   0x1c9080
        .thumb_func     Glo(example)
        .word   Glo(example)
        .short  8
        .short  0
Loc(200000):
        .asciz  "example.ml"

/* Pool of addresses loaded into registers */

        .text
        .align 2
Loc(Plast_return_address):
        .long Glo(last_return_address)
Loc(Pbottom_of_stack):
        .long Glo(bottom_of_stack)

The output consists of a fixed prefix followed by the translation of the input file, followed by the pool of values to be loaded into registers.

The following shows a successful assembly for iOS of the arm.S file from OCaml 4.00.0.

$ PLT=/Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform
$ PLTBIN=$PLT/Developer/usr/bin
$ arm-as-to-ios asmrun/arm.S > armcompat.S
$ $PLTBIN/gcc -c -DSYS_macosx -DMODEL_armv7 -o armcompat.o armcompat.S
$ file armcompat.o
armcompat.o: Mach-O object arm
$ otool -tv armcompat.o | head
armcompat.o:
(__TEXT,__text) section
caml_call_gc:
00000000        f8dfc1e0        ldr.w   ip, [pc, #480]  @ 0x1e4
00000004        f8cce000        str.w   lr, [ip]
00000008        f8dfc1dc        ldr.w   ip, [pc, #476]  @ 0x1e8
0000000c        f8ccd000        str.w   sp, [ip]
00000010        ed2d0b10        vstmdb  sp!, {d0-d7}
00000014        e92d50ff        stmdb   sp!, {r0, r1, r2, r3, r4, r5, r6, r7, ip, lr}
00000018        f8dfc1d0        ldr.w   ip, [pc, #464]  @ 0x1ec

The following shows the compatible file being generated and assembled under Debian ARMEL (ARM-based Linux system). Note that the generated file armcompat.S is identical to the above—the output of arm-as-to-ios is a single file that works in both environments.

$ ./arm-as-to-ios asmrun/arm.S > armcompat.S
$ gcc -c -DSYS_linux_eabihf -o armcompat.o armcompat.S 
$ file armcompat.o
armcompat.o: ELF 32-bit LSB relocatable, ARM, version 1 (SYSV), not stripped
$ objdump -d armcompat.o | grep -v '^$' | head
armcompat.o:     file format elf32-littlearm
Disassembly of section .text:
00000000 <caml_call_gc>:
   0:   f8df c1e0       ldr.w   ip, [pc, #480]  ; 1e4 <caml_system__code_end>
   4:   f8cc e000       str.w   lr, [ip]
   8:   f8df c1dc       ldr.w   ip, [pc, #476]  ; 1e8 <caml_system__code_end+0x4>
   c:   f8cc d000       str.w   sp, [ip]
  10:   ed2d 0b10       vpush   {d0-d7}
  14:   e92d 50ff       stmdb   sp!, {r0, r1, r2, r3, r4, r5, r6, r7, ip, lr}
  18:   f8df c1d0       ldr.w   ip, [pc, #464]  ; 1ec <caml_system__code_end+0x8>

If you have any corrections, improvements, or other comments, leave them below or email me at jeffsco@psellos.com. I’d be very pleased to hear if the script has been helpful to anyone.

Posted by: Jeffrey

Appendix

#!/usr/bin/env python
#
# arm-as-to-ios     Modify ARM assembly code for the iOS assembler
#
# Copyright (c) 2012 Psellos   http://psellos.com/
# Licensed under the MIT License:
#     http://www.opensource.org/licenses/mit-license.php
#
# Resources for running OCaml on iOS: http://psellos.com/ocaml/
#
import sys
import re

VERSION = '1.4.0'

# Character classes for expression lexing.
#
g_ccid0 = '[$.A-Z_a-z\x80-\xff]'      # Beginning of id
g_ccid =  '[$.0-9A-Z_a-z\x80-\xff]'   # Later in id
def ccc(cc):                          # Complement the class
    if cc[1] == '^':
        return cc[0] + cc[2:]
    return cc[0] + '^' + cc[1:]
def ccce(cc):                         # Complement the class, include EOL
    return '(?:' + ccc(cc) + '|$)'

# Prefixes for pooled symbol labels and jump table base labels.  They're
# in the space of Linux assembler local symbols.  Later rules will
# modify them to the Loc() form.
#
g_poolpfx = '.LP'
g_basepfx = '.LB'


def exists(p, l):
    for l1 in l:
        if p(l1):
            return True
    return False


def forall(p, l):
    for l1 in l:
        if not p(l1):
            return False
    return True


def add_prefix(instrs):
    # Add compatibility macros for all systems, plus hardware
    # definitions and compatibility macros for iOS.
    #
    # All systems:
    #
    # Glo()     cpp macro for making global symbols (xxx vs _xxx)
    # Loc()     cpp macro for making local symbols (.Lxxx vs Lxxx)
    # .funtype  Expands to .thumb_func for iOS armv7 (null for armv6)
    #           Expands to .type %function for others
    #
    # iOS:
    #
    # .machine  armv6/armv7
    # .thumb    (for armv7)
    # cbz       Expands to cmp/beq for armv6 (Thumb-only instr)
    # .type     Not supported by Apple assembler
    # .size     Not supported by Apple assembler
    #
    defre = '#[ \t]*if.*def.*SYS'  # Add new defs near first existing ones
    skipre = '$|\.syntax[ \t]'     # Skip comment lines (and .syntax)

    for i in range(len(instrs)):
        if re.match(defre, instrs[i][1]):
            break
    else:
        i = 0
    for i in range(i, len(instrs)):
        if not re.match(skipre, instrs[i][1]):
            break
    instrs[i:0] = [
        ('', '', '\n'),
        ('/* Apple compatibility macros */', '', '\n'),
        ('', '#if defined(SYS_macosx)', '\n'),
        ('', '#define Glo(s) _##s', '\n'),
        ('', '#define Loc(s) L##s', '\n'),
        ('', '#if defined(MODEL_armv6)', '\n'),
        ('        ', '.machine  armv6', '\n'),
        ('        ', '.macro  .funtype', '\n'),
        ('        ', '.endm', '\n'),
        ('        ', '.macro  cbz', '\n'),
        ('        ', 'cmp     $0, #0', '\n'),
        ('        ', 'beq     $1', '\n'),
        ('        ', '.endm', '\n'),
        ('', '#else', '\n'),
        ('        ', '.machine  armv7', '\n'),
        ('        ', '.thumb', '\n'),
        ('        ', '.macro  .funtype', '\n'),
        ('        ', '.thumb_func $0', '\n'),
        ('        ', '.endm', '\n'),
        ('', '#endif', '\n'),
        ('        ', '.macro  .type', '\n'),
        ('        ', '.endm', '\n'),
        ('        ', '.macro  .size', '\n'),
        ('        ', '.endm', '\n'),
        ('', '#else', '\n'),
        ('', '#define Glo(s) s', '\n'),
        ('', '#define Loc(s) .L##s', '\n'),
        ('        ', '.macro  .funtype symbol', '\n'),
        ('        ', '.type  \\symbol, %function', '\n'),
        ('        ', '.endm', '\n'),
        ('', '#endif', '\n'),
        ('/* End Apple compatibility macros */', '', '\n'),
        ('', '', '\n')
    ]
    return instrs


# Regular expression for modified ldr lines
#
g_ldre = '(ldr[ \t][^,]*,[ \t]*)=(([^ \t\n@,/]|/(?!\*))*)(.*)'


def explicit_address_loads(instrs):
    # Linux assemblers allow the following:
    #
    #     ldr rM, =symbol
    #
    # which loads rM with [mov] (immediately) if possible, or creates an
    # entry in memory for the symbol value and loads it PC-relatively
    # with [ldr].
    #
    # The Apple assembler doesn't seem to support this notation.  If the
    # value is a suitable constant, it emits a valid [mov].  Otherwise
    # it seems to emit an invalid [ldr] that always generates an error.
    # (At least I have not been able to make it work).  So, change uses
    # of =symbol to explicit PC-relative loads.
    #
    # This requires a pool containing the addresses to be loaded.  For
    # now, we just keep track of it ourselves and emit it into the text
    # segment at the end of the file.
    #
    syms = {}
    result = []

    def repl1((syms, result), (a, b, c)):
        global g_poolpfx
        global g_ldre
        (b1, b2, b3) = parse_iparts(b)
        mo = re.match(g_ldre, b3, re.DOTALL)
        if mo:
            if mo.group(2) not in syms:
                syms[mo.group(2)] = len(syms)
            psym = mo.group(2)
            if psym[0:2] == '.L':
                psym = psym[2:]
            newb3 = mo.group(1) + g_poolpfx + psym + mo.group(4)
            result.append((a, b1 + b2 + newb3, c))
        else:
            result.append((a, b, c))
        return (syms, result)

    def pool1(result, s):
        global g_poolpfx
        psym = s
        if psym[0:2] == '.L':
            psym = psym[2:]
        result.append(('', g_poolpfx + psym + ':', '\n'))
        result.append(('        ', '.long ' + s, '\n'))
        return result

    reduce(repl1, instrs, (syms, result))
    if len(syms) > 0:
        result.append(('', '', '\n'))
        result.append(('/* Pool of addresses loaded into registers */',
                        '', '\n'))
        result.append(('', '', '\n'))
        result.append(('        ', '.text', '\n'))
        result.append(('        ', '.align 2', '\n'))
        reduce(pool1, sorted(syms, key=syms.get), result)
    return result


def global_symbols(instrs):
    # The form of a global symbol differs between Linux assemblers and
    # the Apple assember:
    #
    # Linux: xxx
    # Apple: _xxx
    #
    # Change occurrences of global symbols to use the Glo() cpp macro
    # defined in our prefix.
    #
    # We consider a symbol to be global if:
    #
    # a.  It appears in a .globl declaration; or
    # b.  It is referenced, has global form, and is not defined
    #
    glosyms = set()
    refsyms = set()
    defsyms = set()
    result = []

    def findglo1 (glosyms, (a, b, c)):
        if re.match('#', b):
            # Preprocessor line; nothing to do
            return glosyms
        (b1, b2, b3) = parse_iparts(b)
        mo = re.match('(\.globl)' + ccce(g_ccid), b3)
        if mo:
            tokens = parse_expr(b3[len(mo.group(1)):])
            if forall(lambda t: token_type(t) in ['space', 'id', ','], tokens):
                for t in tokens:
                    if token_type(t) == 'id':
                        glosyms.add(t)
        return glosyms

    def findref1 ((refsyms, skipct), (a, b, c)):

        def looksglobal(s):
            if re.match('(r|a|v|p|c|cr|f|s|d|q|mvax|wcgr)[0-9]+$', s, re.I):
                return False # numbered registers
            if re.match('(wr|sb|sl|fp|ip|sp|lr|pc)$', s, re.I):
                return False # named registers
            if re.match('(fpsid|fpscr|fpexc|mvfr1|mvfr0)$', s, re.I):
                return False # more named registers
            if re.match('(mvf|mvd|mvfx|mvdx|dspsc)$', s, re.I):
                return False # even more named registers
            if re.match('(wcid|wcon|wcssf|wcasf|acc)$', s, re.I):
                return False # even more named registers
            if re.match('\.$|\.L|[0-9]|#', s):
                return False # dot, local symbol, or number
            if re.match('(asl|lsl|lsr|asr|ror|rrx)$', s, re.I):
                return False # shift names
            return True

        if re.match('#', b):
            # Preprocessor line; nothing to do
            return (refsyms, skipct)

        # Track nesting of .macro/.endm.  For now, we don't look for
        # global syms in macro defs.  (Avoiding scoping probs etc.)
        #
        if skipct > 0 and re.match('\.(endm|endmacro)' + ccce(g_ccid), b):
            return (refsyms, skipct - 1)
        if re.match('\.macro' + ccce(g_ccid), b):
            return (refsyms, skipct + 1)
        if skipct > 0:
            return (refsyms, skipct)
        if re.match('\.(type|size|syntax|arch|fpu)' + ccce(g_ccid), b):
            return (refsyms, skipct)

        (b1, b2, b3) = parse_iparts(b)
        rtokens = parse_rexpr(b3)
        if len(rtokens) > 1 and rtokens[1] == '.req':
            # .req has atypical syntax; no symbol refs there anyway
            return (refsyms, skipct)
        for t in rtokens[1:]:
            if token_type(t) == 'id' and looksglobal(t):
                refsyms.add(t)
        return (refsyms, skipct)

    def finddef1(defsyms, (a, b, c)):
        if re.match('#', b):
            # Preprocessor line
            return defsyms
        (b1, b2, b3) = parse_iparts(b)
        rtokens = parse_rexpr(b3)
        if b1 != '':
            defsyms.add(b1)
        if len(rtokens) > 1 and rtokens[1] == '.req':
            defsyms.add(rtokens[0])
        return defsyms

    def repl1((glosyms, result), (a, b, c)):
        if re.match('#', b):
            # Preprocessor line
            result.append((a, b, c))
            return (glosyms, result)
        toglo = lambda s: 'Glo(' + s + ')'
        (b1, b2, b3) = parse_iparts(b)
        tokens = parse_expr(b3)

        if b1 in glosyms:
            b1 = toglo(b1)
        for i in range(len(tokens)):
            if token_type(tokens[i]) == 'id' and tokens[i] in glosyms:
                tokens[i] = toglo(tokens[i])
        result.append((a, b1 + b2 + ''.join(tokens), c))
        return (glosyms, result)

    reduce(findglo1, instrs, glosyms)
    reduce(findref1, instrs, (refsyms, 0))
    reduce(finddef1, instrs, defsyms)
    glosyms |= (refsyms - defsyms)
    reduce(repl1, instrs, (glosyms, result))
    return result


def local_symbols(instrs):
    # The form of a local symbol differs between Linux assemblers and
    # the Apple assember:
    #
    # Linux: .Lxxx
    # Apple: Lxxx
    #
    # Change occurrences of local symbols to use the Loc() cpp macro
    # defined in our prefix.
    #
    lsyms = set()
    result = []

    def find1 (lsyms, (a, b, c)):
        mo = re.match('(\.L[^ \t:]*)[ \t]*:', b)
        if mo:
            lsyms.add(mo.group(1))
        return lsyms

    def repl1((lsyms, result), (a, b, c)):
        matches = list(re.finditer('\.L[^ \t@:,+*/\-()]+', b))
        if matches != []:
            matches.reverse()
            newb = b
            for mo in matches:
                if mo.group() in lsyms:
                    newb = newb[0:mo.start()] + \
                            'Loc(' + mo.group()[2:] + ')' + \
                            newb[mo.end():]
            result.append((a, newb, c))
        else:
            result.append((a, b, c))
        return (lsyms, result)

    reduce(find1, instrs, lsyms)
    reduce(repl1, instrs, (lsyms, result))
    return result


def funtypes(instrs):
    # Linux assemblers accept declarations like this:
    #
    #     .type  symbol, %function
    #
    # For Thumb functions, the Apple assembler wants to see:
    #
    #     .thumb_func symbol
    #
    # Handle this by converting declarations to this:
    #
    #     .funtype symbol
    #
    # Our prefix defines an appropriate .funtype macro for each
    # environment.
    #
    result = []

    def repl1(result, (a, b, c)):
        mo = re.match('.type[ \t]+([^ \t,]*),[ \t]*%function', b)
        if mo:
            result.append((a, '.funtype  ' + mo.group(1), c))
        else:
            result.append((a, b, c))
        return result

    reduce(repl1, instrs, result)
    return result


def jump_tables(instrs):
    # Jump tables for Linux assemblers often look like this:
    #
    #     tbh [pc, rM, lsl #1]
    #     .short (.Labc-.)/2+0
    #     .short (.Ldef-.)/2+1
    #     .short (.Lghi-.)/2+2
    #
    # The Apple assembler disagrees about the meaning of this code,
    # producing jump tables that don't work.  Convert to the following:
    #
    #     tbh [pc, rM, lsl #1]
    # .LBxxx:
    #     .short (.Labc-.LBxxx)/2
    #     .short (.Ldef-.LBxxx)/2
    #     .short (.Lghi-.LBxxx)/2
    #
    # In fact we just convert sequences of .short pseudo-ops of the
    # right form.  There's no requirement that they follow a tbh
    # instruction.
    #
    baselabs = []
    result = []

    def short_match(seq, op):
        # Determine whether the op is a .short of the form that needs to
        # be converted: .short (symbol-.)/2+k.  If so, return a pair
        # containing the symbol and the value of k.  If not, return
        # None.  The short can only be converted if there were at least
        # k other .shorts in sequence before the current one.  A summary
        # of the previous .shorts is in seq.
        #
        # (A real parser would do a better job, but this was quick to
        # get working.)
        #
        sp = '([ \t]|/\*.*?\*/)*'              # space
        sp1 = '([ \t]|/\*.*?\*/)+'             # at least 1 space
        spe = '([ \t]|/\*.*?\*/|@[^\n]*)*$'    # end-of-instr space
        expr_re0 = (
            '\.short' + sp + '\(' + sp +       # .short (
            '([^ \t+\-*/@()]+)' + sp +         # symbol
            '-' + sp + '\.' + sp + '\)' + sp + # -.)
            '/' + sp + '2' + spe               # /2 END
        )
        expr_re1 = (
            '\.short' + sp + '\(' + sp +       # .short (
            '([^ \t+\-*/@()]+)' + sp +         # symbol
            '-' + sp + '\.' + sp + '\)' + sp + # -.)
            '/' + sp + '2' + sp +              # /2
            '\+' + sp +                        # +
            '((0[xX])?[0-9]+)' + spe           # k END
        )
        expr_re2 = (
            '\.short' + sp1 +                  # .short
            '((0[xX])?[0-9]+)' + sp +          # k
            '\+' + sp + '\(' + sp +            # +(
            '([^ \t+\-*/@()]+)' + sp +         # symbol
            '-' + sp + '\.' + sp + '\)' + sp + # -.)
            '/' + sp + '2' + spe               # /2 END
        )
        mo = re.match(expr_re0, op)
        if mo:
            return(mo.group(3), 0)
        mo = re.match(expr_re1, op)
        if mo:
            k = int(mo.group(11), 0)
            if k > len(seq):
                return None
            return (mo.group(3), k)
        mo = re.match(expr_re2, op)
        if mo:
            k = int(mo.group(2), 0)
            if k > len(seq):
                return None
            return (mo.group(7), k)
        return None

    def conv1 ((baselabs, shortseq, label, result), (a, b, c)):
        # Convert current instr (a,b,c) if it's a .short of the right
        # form that spans a previous sequence of .shorts.
        #
        (b1, b2, b3) = parse_iparts(b)

        if b3 == '':
            # No operation: just note label if present.
            result.append((a, b, c))
            if re.match('\.L.', b1):
                return (baselabs, shortseq, b1, result)
            return (baselabs, shortseq, label, result)

        if not re.match('.short[ \t]+[^ \t@]', b3):
            # Not a .short: clear shortseq and label
            result.append((a, b, c))
            return (baselabs, [], '', result)

        # We have a .short: figure out the label if any
        if re.match('\.L', b1):
            sl = b1
        else:
            sl = label

        mpair = short_match(shortseq, b3)
        if not mpair:
            # A .short, but not of right form
            shortseq.append((len(result), sl))
            result.append((a, b, c))
            return (baselabs, shortseq, '', result)

        # OK, we have a .short to convert!
        (sym, k) = mpair
        shortseq.append((len(result), sl))

        # Figure out base label (create one if necessary).
        bx = len(shortseq) - 1 - k
        bl = shortseq[bx][1]
        if bl == '':
            bl = g_basepfx + str(shortseq[bx][0])
            shortseq[bx] = (shortseq[bx][0], bl)
            baselabs.append(shortseq[bx])

        op = '.short\t(' + sym + '-' + bl + ')/2'

        result.append ((a, b1 + b2 + op, c))
        return (baselabs, shortseq, '', result)

    # Convert, accumulate result and new labels.
    reduce(conv1, instrs, (baselabs, [], '', result))

    # Add labels created here to the instruction stream.
    baselabs.reverse()
    for (ix, lab) in baselabs:
        result[ix:0] = [('', lab + ':', '\n')]

    # That does it
    return result


def dot_relative(instrs):
    # The Apple assembler (or possibly the linker) has trouble with code
    # that looks like this:
    #
    #     .word   .Label - . + 0x80000000
    #     .word   0x1966
    # .Label:
    #     .word   0x1967
    #
    # One way to describe the problem is that the assembler marks the
    # first .word for relocation when in fact it's an assembly-time
    # constant.  Translate to the following form, which doesn't generate
    # a relocation marking:
    #
    # DR0 =       .Label - . + 0x80000000
    #     .word   DR0
    #     .word   0x1966
    # .Label:
    #     .word   0x1967
    #
    prefix = 'DR'
    pseudos = '(\.byte|\.short|\.word|\.long|\.quad)'
    result = []

    def tok_ok(t):
        return t in ['.', '+', '-', '(', ')'] or \
            token_type(t) in ['space', 'locid', 'number']

    def dotrel_match(expr):
        # Determine whether the expression is one that needs to be
        # translated.
        tokens = parse_expr(expr)
        return forall(tok_ok, tokens) and \
            exists(lambda t: token_type(t) == 'locid', tokens) and \
            exists(lambda t: token_type(t) == 'number', tokens) and \
            exists(lambda t: t == '-', tokens) and \
            exists(lambda t: t == '.', tokens)

    def conv1(result, (a, b, c)):
        if re.match('#', b):
            # Preprocessor line
            result.append((a, b, c))
        else:
            (b1, b2, b3) = parse_iparts(b)
            mo = re.match(pseudos + ccce(g_ccid), b3)
            if mo:
                p = mo.group(1)
                expr = b3[len(p):]
                if dotrel_match(expr):
                    sym = prefix + str(len(result))
                    instr = sym + ' =' + expr
                    result.append(('', instr, '\n'))
                    result.append((a, b1 + b2 + p + ' ' + sym, c))
                else:
                    result.append((a, b, c))
            else:
                result.append((a, b, c))
        return result

    reduce(conv1, instrs, result)
    return result


def read_input():
    # Concatenate all the input files into a string.
    #
    def fnl(s):
        if s == '' or s[-1] == '\n':
            return s
        else:
            return s + '\n'

    if len(sys.argv) < 2:
        return fnl(sys.stdin.read())
    else:
        input = ""
        for f in sys.argv[1:]:
            try:
                fd = open(f)
                input = input + fnl(fd.read())
                fd.close()
            except:
                sys.stderr.write('arm-as-to-ios: cannot open ' + f + '\n')
        return input


def parse_instrs(s):
    # Parse the string into assembly instructions, also noting C
    # preprocessor lines.  Each instruction is represented as a triple:
    # (space/comments, instruction, end).  The end is either ';' or
    # '\n'.
    #
    def goodmo(mo):
        if mo == None:
            # Should never happen
            sys.stderr.write('arm-as-to-ios: internal parsing error\n')
            sys.exit(1)

    cpp_re = '([ \t]*)(#([^\n]*\\\\\n)*[^\n]*[^\\\\\n])\n'
    comment_re = '[ \t]*#[^\n]*'
    instr_re = (
        '(([ \t]|/\*.*?\*/|@[^\n]*)*)'  # Spaces & comments
        '(([ \t]|/\*.*?\*/|[^;\n])*)'   # "Instruction"
        '([;\n])'                       # End
    )
    instrs = []
    while s != '':
        if re.match('[ \t]*#[ \t]*(if|ifdef|elif|else|endif|define)', s):
            mo = re.match(cpp_re, s)
            goodmo(mo)
            instrs.append((mo.group(1), mo.group(2), '\n'))
        elif re.match('[ \t]*#', s):
            mo = re.match(comment_re, s)
            goodmo(mo)
            instrs.append((mo.group(0), '', '\n'))
        else:
            mo = re.match(instr_re, s, re.DOTALL)
            goodmo(mo)
            instrs.append((mo.group(1), mo.group(3), mo.group(5)))
        s = s[len(mo.group(0)):]
    return instrs


def parse_iparts(i):
    # Parse an instruction into smaller parts, returning a triple of
    # strings (label, colon, operation).  The colon part also contains
    # any surrounding spaces and comments (making the label and the
    # operation cleaner to process).
    #
    # (Caller warrants that the given string doesn't start with space or
    # a comment.  This is true for strings returned by the instruction
    # parser.)
    #
    lab_re = (
        '([^ \t:/@]+)'                  # Label
        '(([ \t]|/\*.*?\*/|@[^\n]*)*)'  # Spaces & comments
        ':'                             # Colon
        '(([ \t]|/\*.*?\*/|@[^\n]*)*)'  # Spaces & comments
        '([^\n]*)'                      # Operation
    )

    if len(i) > 0 and i[0] == '#':
        # C preprocessor line; treat as operation.
        return ('', '', i)
    mo = re.match(lab_re, i)
    if mo:
        return (mo.group(1), mo.group(2) + ':' + mo.group(4), mo.group(6))
    # No label, just an operation
    return ('', '', i)


def parse_expr(s):
    # Parse a string into a sequence of tokens.  A segment of white
    # space (including comments) is treated as a token, so that the
    # tokens can be reassembled into the string again.
    #
    result = []
    while s != '':
        mo = re.match('([ \t]|/\*.*?\*/|@.*)+', s)
        if not mo:
            # Glo(...) and Loc(...) are single tokens
            mo = re.match('(Glo|Loc)\([^()]*\)', s)
        if not mo:
            mo = re.match('"([^\\\\"]|\\\\.)*"', s)
        if not mo:
            mo = re.match(g_ccid0 + g_ccid + '*', s)
        if not mo:
            mo = re.match('[0-9]+[bf]', s)
        if not mo:
            mo = re.match('0[Xx][0-9a-fA-F]+|[0-9]+', s)
        if not mo:
            mo = re.match('.', s)
        result.append(mo.group(0))
        s = s[len(mo.group(0)):]
    return result


def parse_rexpr(s):
    # Like parse_expr(), but return only "real" tokens, not the
    # intervening space.
    #
    return filter(lambda t: token_type(t) != 'space', parse_expr(s))


def token_type(t):
    # Determine the type of a token.  Caller warrants that it was
    # returned by parse_expr() or parse_rexpr().
    #
    if re.match('[ \t]|/\*|@', t):
        return 'space'
    if re.match('Glo\(', t):
        return 'gloid'
    if re.match('Loc\(', t):
        return 'locid'
    if re.match('"', t):
        return 'string'
    if re.match(g_ccid0, t):
        return 'id'
    if re.match('[0-9]+[bf]', t):
        return 'label'
    if re.match('[0-9]', t):
        return 'number'
    return t # Sui generis


def debug_parse(a, b, c):
    # Show results of instuction stream parse.
    #
    (b1, b2, b3) = parse_iparts(b)
    newb = '{' + b1 + '}' + '{' + b2 + '}' + '{' + b3 + '}'
    sys.stdout.write('{' + a + '}' + newb + c)


def main():
    instrs = parse_instrs(read_input())
    instrs = explicit_address_loads(instrs)
    instrs = funtypes(instrs)
    instrs = jump_tables(instrs)
    instrs = global_symbols(instrs)
    instrs = local_symbols(instrs)
    instrs = dot_relative(instrs)
    instrs = add_prefix(instrs)
    for (a, b, c) in instrs:
       sys.stdout.write(a + b + c)


main()

Comments

blog comments powered by Disqus