3 # Ascetic x86_64 AT&T to MASM/NASM assembler translator by <appro>.
5 # Why AT&T to MASM and not vice versa? Several reasons. Because AT&T
6 # format is way easier to parse. Because it's simpler to "gear" from
7 # Unix ABI to Windows one [see cross-reference "card" at the end of
8 # file]. Because Linux targets were available first...
10 # In addition the script also "distills" code suitable for GNU
11 # assembler, so that it can be compiled with more rigid assemblers,
12 # such as Solaris /usr/ccs/bin/as.
14 # This translator is not designed to convert *arbitrary* assembler
15 # code from AT&T format to MASM one. It's designed to convert just
16 # enough to provide for dual-ABI OpenSSL modules development...
17 # There *are* limitations and you might have to modify your assembler
18 # code or this script to achieve the desired result...
20 # Currently recognized limitations:
22 # - can't use multiple ops per line;
24 # Dual-ABI styling rules.
26 # 1. Adhere to Unix register and stack layout [see cross-reference
27 # ABI "card" at the end for explanation].
28 # 2. Forget about "red zone," stick to more traditional blended
29 # stack frame allocation. If volatile storage is actually required
30 # that is. If not, just leave the stack as is.
31 # 3. Functions tagged with ".type name,@function" get crafted with
32 # unified Win64 prologue and epilogue automatically. If you want
33 # to take care of ABI differences yourself, tag functions as
34 # ".type name,@abi-omnipotent" instead.
35 # 4. To optimize the Win64 prologue you can specify number of input
36 # arguments as ".type name,@function,N." Keep in mind that if N is
37 # larger than 6, then you *have to* write "abi-omnipotent" code,
38 # because >6 cases can't be addressed with unified prologue.
39 # 5. Name local labels as .L*, do *not* use dynamic labels such as 1:
40 # (sorry about latter).
41 # 6. Don't use [or hand-code with .byte] "rep ret." "ret" mnemonic is
42 # required to identify the spots, where to inject Win64 epilogue!
43 # But on the pros, it's then prefixed with rep automatically:-)
44 # 7. Stick to explicit ip-relative addressing. If you have to use
45 # GOTPCREL addressing, stick to mov symbol@GOTPCREL(%rip),%r??.
46 # Both are recognized and translated to proper Win64 addressing
47 # modes. To support legacy code a synthetic directive, .picmeup,
48 # is implemented. It puts address of the *next* instruction into
49 # target register, e.g.:
52 # lea .Label-.(%rax),%rax
54 # 8. In order to provide for structured exception handling unified
55 # Win64 prologue copies %rsp value to %rax. For further details
56 # see SEH paragraph at the end.
57 # 9. .init segment is allowed to contain calls to functions only.
58 # a. If function accepts more than 4 arguments *and* >4th argument
59 # is declared as non 64-bit value, do clear its upper part.
63 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
65 { my ($stddev,$stdino,@junk)=stat(STDOUT);
66 my ($outdev,$outino,@junk)=stat($output);
68 open STDOUT,">$output" || die "can't open $output: $!"
69 if ($stddev!=$outdev || $stdino!=$outino);
72 my $gas=1; $gas=0 if ($output =~ /\.asm$/);
73 my $elf=1; $elf=0 if (!$gas);
78 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
85 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
86 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
89 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
90 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
91 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
93 { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
94 { $nasm = $1 + $2*0.01; $PTR=""; }
95 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
96 { $masm = $1 + $2*2**-16 + $4*2**-32; }
97 die "no assembler found on %PATH" if (!($nasm || $masm));
104 my $current_function;
107 { package opcode; # pick up opcodes
109 my $self = shift; # single instance in enough...
113 if ($line =~ /^([a-z][a-z0-9]*)/i) {
116 $line = substr($line,@+[0]); $line =~ s/^\s+//;
119 if ($self->{op} =~ /^(movz)b.*/) { # movz is pain...
122 } elsif ($self->{op} =~ /call|jmp/) {
124 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
126 } elsif ($self->{op} =~ /^v/) { # VEX
128 } elsif ($self->{op} =~ /movq/ && $line =~ /%xmm/) {
130 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
140 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
146 if ($self->{op} eq "movz") { # movz is pain...
147 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
148 } elsif ($self->{op} =~ /^set/) {
150 } elsif ($self->{op} eq "ret") {
152 if ($win64 && $current_function->{abi} eq "svr4") {
153 $epilogue = "movq 8(%rsp),%rdi\n\t" .
154 "movq 16(%rsp),%rsi\n\t";
156 $epilogue . ".byte 0xf3,0xc3";
157 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
158 ".p2align\t3\n\t.quad";
160 "$self->{op}$self->{sz}";
163 $self->{op} =~ s/^movz/movzx/;
164 if ($self->{op} eq "ret") {
166 if ($win64 && $current_function->{abi} eq "svr4") {
167 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t".
168 "mov rsi,QWORD${PTR}[16+rsp]\n\t";
170 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
171 } elsif ($self->{op} =~ /^(pop|push)f/) {
172 $self->{op} .= $self->{sz};
173 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
174 $self->{op} = "\tDQ";
182 $self->{op}=$op if (defined($op));
186 { package const; # pick up constants, which start with $
188 my $self = shift; # single instance in enough...
192 if ($line =~ /^\$([^,]+)/) {
195 $line = substr($line,@+[0]); $line =~ s/^\s+//;
203 # Solaris /usr/ccs/bin/as can't handle multiplications
205 $self->{value} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
206 $self->{value} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
207 sprintf "\$%s",$self->{value};
209 $self->{value} =~ s/(0b[0-1]+)/oct($1)/eig;
210 $self->{value} =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
211 sprintf "%s",$self->{value};
215 { package ea; # pick up effective addresses: expr(%reg,%reg,scale)
217 my $self = shift; # single instance in enough...
221 # optional * ---vvv--- appears in indirect jmp/call
222 if ($line =~ /^(\*?)([^\(,]*)\(([%\w,]+)\)/) {
223 $self->{asterisk} = $1;
225 ($self->{base},$self->{index},$self->{scale})=split(/,/,$3);
226 $self->{scale} = 1 if (!defined($self->{scale}));
228 $line = substr($line,@+[0]); $line =~ s/^\s+//;
230 if ($win64 && $self->{label} =~ s/\@GOTPCREL//) {
231 die if (opcode->mnemonic() ne "mov");
232 opcode->mnemonic("lea");
234 $self->{base} =~ s/^%//;
235 $self->{index} =~ s/^%// if (defined($self->{index}));
244 $self->{label} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
245 $self->{label} =~ s/\.L/$decor/g;
247 # Silently convert all EAs to 64-bit. This is required for
248 # elder GNU assembler and results in more compact code,
249 # *but* most importantly AES module depends on this feature!
250 $self->{index} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
251 $self->{base} =~ s/^[er](.?[0-9xpi])[d]?$/r\1/;
254 # Solaris /usr/ccs/bin/as can't handle multiplications
255 # in $self->{label}, new gas requires sign extension...
257 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
258 $self->{label} =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg;
259 $self->{label} =~ s/([0-9]+)/$1<<32>>32/eg;
260 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
262 if (defined($self->{index})) {
263 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk},
265 $self->{base}?"%$self->{base}":"",
266 $self->{index},$self->{scale};
268 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base};
271 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR", l=>"DWORD$PTR",
272 q=>"QWORD$PTR",o=>"OWORD$PTR",x=>"XMMWORD$PTR" );
274 $self->{label} =~ s/\./\$/g;
275 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
276 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
277 $sz="q" if ($self->{asterisk});
279 if (defined($self->{index})) {
280 sprintf "%s[%s%s*%d%s]",$szmap{$sz},
281 $self->{label}?"$self->{label}+":"",
282 $self->{index},$self->{scale},
283 $self->{base}?"+$self->{base}":"";
284 } elsif ($self->{base} eq "rip") {
285 sprintf "%s[%s]",$szmap{$sz},$self->{label};
287 sprintf "%s[%s%s]",$szmap{$sz},
288 $self->{label}?"$self->{label}+":"",
294 { package register; # pick up registers, which start with %.
296 my $class = shift; # muliple instances...
301 # optional * ---vvv--- appears in indirect jmp/call
302 if ($line =~ /^(\*?)%(\w+)/) {
304 $self->{asterisk} = $1;
307 $line = substr($line,@+[0]); $line =~ s/^\s+//;
315 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
316 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
317 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
318 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
319 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
320 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
321 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
322 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
328 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; }
329 else { $self->{value}; }
332 { package label; # pick up labels, which end with :
334 my $self = shift; # single instance is enough...
338 if ($line =~ /(^[\.\w]+)\:/) {
341 $line = substr($line,@+[0]); $line =~ s/^\s+//;
343 $self->{value} =~ s/^\.L/$decor/;
351 my $func = ($globals{$self->{value}} or $self->{value}) . ":";
353 $current_function->{name} eq $self->{value} &&
354 $current_function->{abi} eq "svr4") {
356 $func .= " movq %rdi,8(%rsp)\n";
357 $func .= " movq %rsi,16(%rsp)\n";
358 $func .= " movq %rsp,%rax\n";
359 $func .= "${decor}SEH_begin_$current_function->{name}:\n";
360 my $narg = $current_function->{narg};
361 $narg=6 if (!defined($narg));
362 $func .= " movq %rcx,%rdi\n" if ($narg>0);
363 $func .= " movq %rdx,%rsi\n" if ($narg>1);
364 $func .= " movq %r8,%rdx\n" if ($narg>2);
365 $func .= " movq %r9,%rcx\n" if ($narg>3);
366 $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
367 $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
370 } elsif ($self->{value} ne "$current_function->{name}") {
371 $self->{value} .= ":" if ($masm && $ret!~m/^\$/);
372 $self->{value} . ":";
373 } elsif ($win64 && $current_function->{abi} eq "svr4") {
374 my $func = "$current_function->{name}" .
375 ($nasm ? ":" : "\tPROC $current_function->{scope}") .
377 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n";
378 $func .= " mov QWORD${PTR}[16+rsp],rsi\n";
379 $func .= " mov rax,rsp\n";
380 $func .= "${decor}SEH_begin_$current_function->{name}:";
381 $func .= ":" if ($masm);
383 my $narg = $current_function->{narg};
384 $narg=6 if (!defined($narg));
385 $func .= " mov rdi,rcx\n" if ($narg>0);
386 $func .= " mov rsi,rdx\n" if ($narg>1);
387 $func .= " mov rdx,r8\n" if ($narg>2);
388 $func .= " mov rcx,r9\n" if ($narg>3);
389 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4);
390 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5);
393 "$current_function->{name}".
394 ($nasm ? ":" : "\tPROC $current_function->{scope}");
398 { package expr; # pick up expressioins
400 my $self = shift; # single instance is enough...
404 if ($line =~ /(^[^,]+)/) {
407 $line = substr($line,@+[0]); $line =~ s/^\s+//;
409 $self->{value} =~ s/\@PLT// if (!$elf);
410 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
411 $self->{value} =~ s/\.L/$decor/g;
417 if ($nasm && opcode->mnemonic()=~m/^j/) {
418 "NEAR ".$self->{value};
424 { package directive; # pick up directives, which start with .
426 my $self = shift; # single instance is enough...
430 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2:
431 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48,
432 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48,
433 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48,
434 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48,
435 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c,
436 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c,
437 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c,
438 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c );
440 if ($line =~ /^\s*(\.\w+)/) {
443 undef $self->{value};
444 $line = substr($line,@+[0]); $line =~ s/^\s+//;
447 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) {
449 $line=sprintf "0x%x,0x90000000",$opcode{$1};
453 /\.global|\.globl|\.extern/
454 && do { $globals{$line} = $prefix . $line;
455 $line = $globals{$line} if ($prefix);
458 /\.type/ && do { ($sym,$type,$narg) = split(',',$line);
459 if ($type eq "\@function") {
460 undef $current_function;
461 $current_function->{name} = $sym;
462 $current_function->{abi} = "svr4";
463 $current_function->{narg} = $narg;
464 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
465 } elsif ($type eq "\@abi-omnipotent") {
466 undef $current_function;
467 $current_function->{name} = $sym;
468 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
470 $line =~ s/\@abi\-omnipotent/\@function/;
471 $line =~ s/\@function.*/\@function/;
474 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) {
476 $line = join(",",unpack("C*",$1),0);
480 /\.rva|\.long|\.quad/
481 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
482 $line =~ s/\.L/$decor/g;
488 $self->{value} = $dir . "\t" . $line;
490 if ($dir =~ /\.extern/) {
491 $self->{value} = ""; # swallow extern
492 } elsif (!$elf && $dir =~ /\.type/) {
494 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
495 (defined($globals{$1})?".scl 2;":".scl 3;") .
496 "\t.type 32;\t.endef"
497 if ($win64 && $line =~ /([^,]+),\@function/);
498 } elsif (!$elf && $dir =~ /\.size/) {
500 if (defined($current_function)) {
501 $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
502 if ($win64 && $current_function->{abi} eq "svr4");
503 undef $current_function;
505 } elsif (!$elf && $dir =~ /\.align/) {
506 $self->{value} = ".p2align\t" . (log($line)/log(2));
507 } elsif ($dir eq ".section") {
508 $current_segment=$line;
509 if (!$elf && $current_segment eq ".init") {
510 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
511 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
513 } elsif ($dir =~ /\.(text|data)/) {
514 $current_segment=".$1";
515 } elsif ($dir =~ /\.hidden/) {
516 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; }
517 elsif ($flavour eq "mingw64") { $self->{value} = ""; }
518 } elsif ($dir =~ /\.comm/) {
519 $self->{value} = "$dir\t$prefix$line";
525 # non-gas case or nasm/masm
527 /\.text/ && do { my $v=undef;
529 $v="section .text code align=64\n";
531 $v="$current_segment\tENDS\n" if ($current_segment);
532 $current_segment = ".text\$";
533 $v.="$current_segment\tSEGMENT ";
534 $v.=$masm>=$masmref ? "ALIGN(64)" : "PAGE";
540 /\.data/ && do { my $v=undef;
542 $v="section .data data align=8\n";
544 $v="$current_segment\tENDS\n" if ($current_segment);
545 $current_segment = "_DATA";
546 $v.="$current_segment\tSEGMENT";
551 /\.section/ && do { my $v=undef;
552 $line =~ s/([^,]*).*/$1/;
553 $line = ".CRT\$XCU" if ($line eq ".init");
556 if ($line=~/\.([px])data/) {
558 $v.=$1 eq "p"? 4 : 8;
559 } elsif ($line=~/\.CRT\$/i) {
560 $v.=" rdata align=8";
563 $v="$current_segment\tENDS\n" if ($current_segment);
564 $v.="$line\tSEGMENT";
565 if ($line=~/\.([px])data/) {
567 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
568 } elsif ($line=~/\.CRT\$/i) {
569 $v.=" READONLY DWORD";
572 $current_segment = $line;
576 /\.extern/ && do { $self->{value} = "EXTERN\t".$line;
577 $self->{value} .= ":NEAR" if ($masm);
581 && do { $self->{value} = $masm?"PUBLIC":"global";
582 $self->{value} .= "\t".$line;
585 /\.size/ && do { if (defined($current_function)) {
586 undef $self->{value};
587 if ($current_function->{abi} eq "svr4") {
588 $self->{value}="${decor}SEH_end_$current_function->{name}:";
589 $self->{value}.=":\n" if($masm);
591 $self->{value}.="$current_function->{name}\tENDP" if($masm);
592 undef $current_function;
596 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; };
597 /\.(value|long|rva|quad)/
598 && do { my $sz = substr($1,0,1);
599 my @arr = split(/,\s*/,$line);
600 my $last = pop(@arr);
601 my $conv = sub { my $var=shift;
602 $var=~s/^(0b[0-1]+)/oct($1)/eig;
603 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
604 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
605 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
609 $sz =~ tr/bvlrq/BWDDQ/;
610 $self->{value} = "\tD$sz\t";
611 for (@arr) { $self->{value} .= &$conv($_).","; }
612 $self->{value} .= &$conv($last);
615 /\.byte/ && do { my @str=split(/,\s*/,$line);
616 map(s/(0b[0-1]+)/oct($1)/eig,@str);
617 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
619 $self->{value}.="DB\t"
620 .join(",",@str[0..15])."\n";
621 foreach (0..15) { shift @str; }
623 $self->{value}.="DB\t"
624 .join(",",@str) if (@str);
627 /\.comm/ && do { my @str=split(/,\s*/,$line);
630 $v.="common $prefix@str[0] @str[1]";
632 $v="$current_segment\tENDS\n" if ($current_segment);
633 $current_segment = ".data";
634 $v.="$current_segment\tSEGMENT\n";
635 $v.="COMM @str[0]:DWORD:".@str[1]/4;
654 my ($dst,$src,$rex)=@_;
656 $rex|=0x04 if($dst>=8);
657 $rex|=0x01 if($src>=8);
658 push @opcode,($rex|0x40) if ($rex);
661 # older gas and ml64 don't handle SSE>2 instructions
662 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
663 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 );
665 my $movq = sub { # elderly gas can't handle inter-register movq
668 if ($arg =~ /%xmm([0-9]+),%r(\w+)/) {
669 my ($src,$dst)=($1,$2);
670 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
671 rex(\@opcode,$src,$dst,0x8);
672 push @opcode,0x0f,0x7e;
673 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
675 } elsif ($arg =~ /%r(\w+),%xmm([0-9]+)/) {
676 my ($src,$dst)=($2,$1);
677 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
678 rex(\@opcode,$src,$dst,0x8);
679 push @opcode,0x0f,0x6e;
680 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
688 if (shift =~ /\$([0-9]+),%xmm([0-9]+),(%\w+)/) {
693 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; }
694 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; }
695 rex(\@opcode,$src,$dst);
696 push @opcode,0x0f,0x3a,0x16;
697 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
706 if (shift =~ /\$([0-9]+),(%\w+),%xmm([0-9]+)/) {
711 if ($src =~ /%r([0-9]+)/) { $src = $1; }
712 elsif ($src =~ /%e/) { $src = $regrm{$src}; }
713 rex(\@opcode,$dst,$src);
714 push @opcode,0x0f,0x3a,0x22;
715 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M
724 if (shift =~ /%xmm([0-9]+),%xmm([0-9]+)/) {
727 push @opcode,0x0f,0x38,0x00;
728 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
736 if (shift =~ /\$([0-9]+),%xmm([0-9]+),%xmm([0-9]+)/) {
739 push @opcode,0x0f,0x3a,0x0f;
740 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
748 my $pclmulqdq = sub {
749 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
752 push @opcode,0x0f,0x3a,0x44;
753 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
755 push @opcode,$c=~/^0/?oct($c):$c;
763 if (shift =~ /%[er](\w+)/) {
766 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
767 rex(\@opcode,0,$1,8);
768 push @opcode,0x0f,0xc7,0xf0|($dst&7);
789 $line =~ s|[#!].*$||; # get rid of asm-style comments...
790 $line =~ s|/\*.*\*/||; # ... and C-style comments...
791 $line =~ s|^\s+||; # ... and skip white spaces in beginning
797 if ($label=label->re(\$line)) { print $label->out(); }
799 if (directive->re(\$line)) {
800 printf "%s",directive->out();
801 } elsif ($opcode=opcode->re(\$line)) {
802 my $asm = eval("\$".$opcode->mnemonic());
805 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) {
806 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
810 ARGUMENT: while (1) {
813 if ($arg=register->re(\$line)) { opcode->size($arg->size()); }
814 elsif ($arg=const->re(\$line)) { }
815 elsif ($arg=ea->re(\$line)) { }
816 elsif ($arg=expr->re(\$line)) { }
817 else { last ARGUMENT; }
821 last ARGUMENT if ($line !~ /^,/);
828 my $sz=opcode->size();
831 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
832 @args = map($_->out($sz),@args);
833 printf "\t%s\t%s",$insn,join(",",@args);
835 $insn = $opcode->out();
838 # $insn.=$sz compensates for movq, pinsrw, ...
839 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
840 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; }
842 @args = reverse(@args);
843 undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
844 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
847 printf "\t%s",$opcode->out();
854 print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
855 print "END\n" if ($masm);
859 \f#################################################
860 # Cross-reference x86_64 ABI "card"
880 # (*) volatile register
881 # (-) preserved by callee
882 # (#) Nth argument, volatile
884 # In Unix terms top of stack is argument transfer area for arguments
885 # which could not be accomodated in registers. Or in other words 7th
886 # [integer] argument resides at 8(%rsp) upon function entry point.
887 # 128 bytes above %rsp constitute a "red zone" which is not touched
888 # by signal handlers and can be used as temporal storage without
889 # allocating a frame.
891 # In Win64 terms N*8 bytes on top of stack is argument transfer area,
892 # which belongs to/can be overwritten by callee. N is the number of
893 # arguments passed to callee, *but* not less than 4! This means that
894 # upon function entry point 5th argument resides at 40(%rsp), as well
895 # as that 32 bytes from 8(%rsp) can always be used as temporal
896 # storage [without allocating a frame]. One can actually argue that
897 # one can assume a "red zone" above stack pointer under Win64 as well.
898 # Point is that at apparently no occasion Windows kernel would alter
899 # the area above user stack pointer in true asynchronous manner...
901 # All the above means that if assembler programmer adheres to Unix
902 # register and stack layout, but disregards the "red zone" existense,
903 # it's possible to use following prologue and epilogue to "gear" from
904 # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
906 # omnipotent_function:
910 # movq %rcx,%rdi ; if 1st argument is actually present
911 # movq %rdx,%rsi ; if 2nd argument is actually ...
912 # movq %r8,%rdx ; if 3rd argument is ...
913 # movq %r9,%rcx ; if 4th argument ...
914 # movq 40(%rsp),%r8 ; if 5th ...
915 # movq 48(%rsp),%r9 ; if 6th ...
924 \f#################################################
925 # Win64 SEH, Structured Exception Handling.
927 # Unlike on Unix systems(*) lack of Win64 stack unwinding information
928 # has undesired side-effect at run-time: if an exception is raised in
929 # assembler subroutine such as those in question (basically we're
930 # referring to segmentation violations caused by malformed input
931 # parameters), the application is briskly terminated without invoking
932 # any exception handlers, most notably without generating memory dump
933 # or any user notification whatsoever. This poses a problem. It's
934 # possible to address it by registering custom language-specific
935 # handler that would restore processor context to the state at
936 # subroutine entry point and return "exception is not handled, keep
937 # unwinding" code. Writing such handler can be a challenge... But it's
938 # doable, though requires certain coding convention. Consider following
941 # .type function,@function
943 # movq %rsp,%rax # copy rsp to volatile register
944 # pushq %r15 # save non-volatile registers
948 # subq %rdi,%r11 # prepare [variable] stack frame
950 # movq %rax,0(%r11) # check for exceptions
951 # movq %r11,%rsp # allocate [variable] stack frame
952 # movq %rax,0(%rsp) # save original rsp value
955 # movq 0(%rsp),%rcx # pull original rsp value
956 # movq -24(%rcx),%rbp # restore non-volatile registers
957 # movq -16(%rcx),%rbx
959 # movq %rcx,%rsp # restore original rsp
961 # .size function,.-function
963 # The key is that up to magic_point copy of original rsp value remains
964 # in chosen volatile register and no non-volatile register, except for
965 # rsp, is modified. While past magic_point rsp remains constant till
966 # the very end of the function. In this case custom language-specific
967 # exception handler would look like this:
969 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
970 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
971 # { ULONG64 *rsp = (ULONG64 *)context->Rax;
972 # if (context->Rip >= magic_point)
973 # { rsp = ((ULONG64 **)context->Rsp)[0];
974 # context->Rbp = rsp[-3];
975 # context->Rbx = rsp[-2];
976 # context->R15 = rsp[-1];
978 # context->Rsp = (ULONG64)rsp;
979 # context->Rdi = rsp[1];
980 # context->Rsi = rsp[2];
982 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
983 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
984 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
985 # &disp->HandlerData,&disp->EstablisherFrame,NULL);
986 # return ExceptionContinueSearch;
989 # It's appropriate to implement this handler in assembler, directly in
990 # function's module. In order to do that one has to know members'
991 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
992 # values. Here they are:
1012 # sizeof(CONTEXT) 1232
1013 # DISPATCHER_CONTEXT.ControlPc 0
1014 # DISPATCHER_CONTEXT.ImageBase 8
1015 # DISPATCHER_CONTEXT.FunctionEntry 16
1016 # DISPATCHER_CONTEXT.EstablisherFrame 24
1017 # DISPATCHER_CONTEXT.TargetIp 32
1018 # DISPATCHER_CONTEXT.ContextRecord 40
1019 # DISPATCHER_CONTEXT.LanguageHandler 48
1020 # DISPATCHER_CONTEXT.HandlerData 56
1021 # UNW_FLAG_NHANDLER 0
1022 # ExceptionContinueSearch 1
1024 # In order to tie the handler to the function one has to compose
1025 # couple of structures: one for .xdata segment and one for .pdata.
1027 # UNWIND_INFO structure for .xdata segment would be
1029 # function_unwind_info:
1033 # This structure designates exception handler for a function with
1034 # zero-length prologue, no stack frame or frame register.
1036 # To facilitate composing of .pdata structures, auto-generated "gear"
1037 # prologue copies rsp value to rax and denotes next instruction with
1038 # .LSEH_begin_{function_name} label. This essentially defines the SEH
1039 # styling rule mentioned in the beginning. Position of this label is
1040 # chosen in such manner that possible exceptions raised in the "gear"
1041 # prologue would be accounted to caller and unwound from latter's frame.
1042 # End of function is marked with respective .LSEH_end_{function_name}
1043 # label. To summarize, .pdata segment would contain
1045 # .rva .LSEH_begin_function
1046 # .rva .LSEH_end_function
1047 # .rva function_unwind_info
1049 # Reference to functon_unwind_info from .xdata segment is the anchor.
1050 # In case you wonder why references are 32-bit .rvas and not 64-bit
1051 # .quads. References put into these two segments are required to be
1052 # *relative* to the base address of the current binary module, a.k.a.
1053 # image base. No Win64 module, be it .exe or .dll, can be larger than
1054 # 2GB and thus such relative references can be and are accommodated in
1057 # Having reviewed the example function code, one can argue that "movq
1058 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
1059 # rax would contain an undefined value. If this "offends" you, use
1060 # another register and refrain from modifying rax till magic_point is
1061 # reached, i.e. as if it was a non-volatile register. If more registers
1062 # are required prior [variable] frame setup is completed, note that
1063 # nobody says that you can have only one "magic point." You can
1064 # "liberate" non-volatile registers by denoting last stack off-load
1065 # instruction and reflecting it in finer grade unwind logic in handler.
1066 # After all, isn't it why it's called *language-specific* handler...
1068 # Attentive reader can notice that exceptions would be mishandled in
1069 # auto-generated "gear" epilogue. Well, exception effectively can't
1070 # occur there, because if memory area used by it was subject to
1071 # segmentation violation, then it would be raised upon call to the
1072 # function (and as already mentioned be accounted to caller, which is
1073 # not a problem). If you're still not comfortable, then define tail
1074 # "magic point" just prior ret instruction and have handler treat it...
1076 # (*) Note that we're talking about run-time, not debug-time. Lack of
1077 # unwind information makes debugging hard on both Windows and
1078 # Unix. "Unlike" referes to the fact that on Unix signal handler
1079 # will always be invoked, core dumped and appropriate exit code
1080 # returned to parent (for user notification).