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 $flavour; }
65 open STDOUT,">$output" || die "can't open $output: $!"
66 if (defined($output));
68 my $gas=1; $gas=0 if ($output =~ /\.asm$/);
69 my $elf=1; $elf=0 if (!$gas);
74 my $masmref=8 + 50727*2**-32; # 8.00.50727 shipped with VS2005
81 if ($flavour eq "mingw64") { $gas=1; $elf=0; $win64=1;
82 $prefix=`echo __USER_LABEL_PREFIX__ | $ENV{CC} -E -P -`;
85 elsif ($flavour eq "macosx") { $gas=1; $elf=0; $prefix="_"; $decor="L\$"; }
86 elsif ($flavour eq "masm") { $gas=0; $elf=0; $masm=$masmref; $win64=1; $decor="\$L\$"; }
87 elsif ($flavour eq "nasm") { $gas=0; $elf=0; $nasm=$nasmref; $win64=1; $decor="\$L\$"; $PTR=""; }
89 { if ($ENV{ASM} =~ m/nasm/ && `nasm -v` =~ m/version ([0-9]+)\.([0-9]+)/i)
90 { $nasm = $1 + $2*0.01; $PTR=""; }
91 elsif (`ml64 2>&1` =~ m/Version ([0-9]+)\.([0-9]+)(\.([0-9]+))?/)
92 { $masm = $1 + $2*2**-16 + $4*2**-32; }
93 die "no assembler found on %PATH" if (!($nasm || $masm));
100 my $current_function;
103 { package opcode; # pick up opcodes
105 my $self = shift; # single instance in enough...
109 if ($line =~ /^([a-z][a-z0-9]*)/i) {
112 $line = substr($line,@+[0]); $line =~ s/^\s+//;
115 if ($self->{op} =~ /^(movz)x?([bw]).*/) { # movz is pain...
118 } elsif ($self->{op} =~ /call|jmp/) {
120 } elsif ($self->{op} =~ /^p/ && $' !~ /^(ush|op|insrw)/) { # SSEn
122 } elsif ($self->{op} =~ /^v/) { # VEX
124 } elsif ($self->{op} =~ /mov[dq]/ && $line =~ /%xmm/) {
126 } elsif ($self->{op} =~ /([a-z]{3,})([qlwb])$/) {
136 $self->{sz} = $sz if (defined($sz) && !defined($self->{sz}));
142 if ($self->{op} eq "movz") { # movz is pain...
143 sprintf "%s%s%s",$self->{op},$self->{sz},shift;
144 } elsif ($self->{op} =~ /^set/) {
146 } elsif ($self->{op} eq "ret") {
148 if ($win64 && $current_function->{abi} eq "svr4") {
149 $epilogue = "movq 8(%rsp),%rdi\n\t" .
150 "movq 16(%rsp),%rsi\n\t";
152 $epilogue . ".byte 0xf3,0xc3";
153 } elsif ($self->{op} eq "call" && !$elf && $current_segment eq ".init") {
154 ".p2align\t3\n\t.quad";
156 "$self->{op}$self->{sz}";
159 $self->{op} =~ s/^movz/movzx/;
160 if ($self->{op} eq "ret") {
162 if ($win64 && $current_function->{abi} eq "svr4") {
163 $self->{op} = "mov rdi,QWORD${PTR}[8+rsp]\t;WIN64 epilogue\n\t".
164 "mov rsi,QWORD${PTR}[16+rsp]\n\t";
166 $self->{op} .= "DB\t0F3h,0C3h\t\t;repret";
167 } elsif ($self->{op} =~ /^(pop|push)f/) {
168 $self->{op} .= $self->{sz};
169 } elsif ($self->{op} eq "call" && $current_segment eq ".CRT\$XCU") {
170 $self->{op} = "\tDQ";
178 $self->{op}=$op if (defined($op));
182 { package const; # pick up constants, which start with $
184 my $self = shift; # single instance in enough...
188 if ($line =~ /^\$([^,]+)/) {
191 $line = substr($line,@+[0]); $line =~ s/^\s+//;
198 $self->{value} =~ s/\b(0b[0-1]+)/oct($1)/eig;
200 # Solaris /usr/ccs/bin/as can't handle multiplications
202 my $value = $self->{value};
203 $value =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
204 if ($value =~ s/([0-9]+\s*[\*\/\%]\s*[0-9]+)/eval($1)/eg) {
205 $self->{value} = $value;
207 sprintf "\$%s",$self->{value};
209 my $value = $self->{value};
210 $value =~ s/0x([0-9a-f]+)/0$1h/ig if ($masm);
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/;
253 # Solaris /usr/ccs/bin/as can't handle multiplications
254 # in $self->{label}...
256 $self->{label} =~ s/(?<![\w\$\.])(0x?[0-9a-f]+)/oct($1)/egi;
257 $self->{label} =~ s/\b([0-9]+\s*[\*\/\%]\s*[0-9]+)\b/eval($1)/eg;
259 # Some assemblers insist on signed presentation of 32-bit
260 # offsets, but sign extension is a tricky business in perl...
262 $self->{label} =~ s/\b([0-9]+)\b/$1<<32>>32/eg;
264 $self->{label} =~ s/\b([0-9]+)\b/$1>>0/eg;
267 if (!$self->{label} && $self->{index} && $self->{scale}==1 &&
268 $self->{base} =~ /(rbp|r13)/) {
269 $self->{base} = $self->{index}; $self->{index} = $1;
273 $self->{label} =~ s/^___imp_/__imp__/ if ($flavour eq "mingw64");
275 if (defined($self->{index})) {
276 sprintf "%s%s(%s,%%%s,%d)",$self->{asterisk},
278 $self->{base}?"%$self->{base}":"",
279 $self->{index},$self->{scale};
281 sprintf "%s%s(%%%s)", $self->{asterisk},$self->{label},$self->{base};
284 %szmap = ( b=>"BYTE$PTR", w=>"WORD$PTR",
285 l=>"DWORD$PTR", d=>"DWORD$PTR",
286 q=>"QWORD$PTR", o=>"OWORD$PTR",
287 x=>"XMMWORD$PTR", y=>"YMMWORD$PTR", z=>"ZMMWORD$PTR" );
289 $self->{label} =~ s/\./\$/g;
290 $self->{label} =~ s/(?<![\w\$\.])0x([0-9a-f]+)/0$1h/ig;
291 $self->{label} = "($self->{label})" if ($self->{label} =~ /[\*\+\-\/]/);
293 ($self->{asterisk}) && ($sz="q") ||
294 (opcode->mnemonic() =~ /^v?mov([qd])$/) && ($sz=$1) ||
295 (opcode->mnemonic() =~ /^v?pinsr([qdwb])$/) && ($sz=$1) ||
296 (opcode->mnemonic() =~ /^vpbroadcast([qdwb])$/) && ($sz=$1) ||
297 (opcode->mnemonic() =~ /^vinsert[fi]128$/) && ($sz="x");
299 if (defined($self->{index})) {
300 sprintf "%s[%s%s*%d%s]",$szmap{$sz},
301 $self->{label}?"$self->{label}+":"",
302 $self->{index},$self->{scale},
303 $self->{base}?"+$self->{base}":"";
304 } elsif ($self->{base} eq "rip") {
305 sprintf "%s[%s]",$szmap{$sz},$self->{label};
307 sprintf "%s[%s%s]",$szmap{$sz},
308 $self->{label}?"$self->{label}+":"",
314 { package register; # pick up registers, which start with %.
316 my $class = shift; # muliple instances...
321 # optional * ---vvv--- appears in indirect jmp/call
322 if ($line =~ /^(\*?)%(\w+)/) {
324 $self->{asterisk} = $1;
327 $line = substr($line,@+[0]); $line =~ s/^\s+//;
335 if ($self->{value} =~ /^r[\d]+b$/i) { $ret="b"; }
336 elsif ($self->{value} =~ /^r[\d]+w$/i) { $ret="w"; }
337 elsif ($self->{value} =~ /^r[\d]+d$/i) { $ret="l"; }
338 elsif ($self->{value} =~ /^r[\w]+$/i) { $ret="q"; }
339 elsif ($self->{value} =~ /^[a-d][hl]$/i){ $ret="b"; }
340 elsif ($self->{value} =~ /^[\w]{2}l$/i) { $ret="b"; }
341 elsif ($self->{value} =~ /^[\w]{2}$/i) { $ret="w"; }
342 elsif ($self->{value} =~ /^e[a-z]{2}$/i){ $ret="l"; }
348 if ($gas) { sprintf "%s%%%s",$self->{asterisk},$self->{value}; }
349 else { $self->{value}; }
352 { package label; # pick up labels, which end with :
354 my $self = shift; # single instance is enough...
358 if ($line =~ /(^[\.\w]+)\:/) {
361 $line = substr($line,@+[0]); $line =~ s/^\s+//;
363 $self->{value} =~ s/^\.L/$decor/;
371 my $func = ($globals{$self->{value}} or $self->{value}) . ":";
373 $current_function->{name} eq $self->{value} &&
374 $current_function->{abi} eq "svr4") {
376 $func .= " movq %rdi,8(%rsp)\n";
377 $func .= " movq %rsi,16(%rsp)\n";
378 $func .= " movq %rsp,%rax\n";
379 $func .= "${decor}SEH_begin_$current_function->{name}:\n";
380 my $narg = $current_function->{narg};
381 $narg=6 if (!defined($narg));
382 $func .= " movq %rcx,%rdi\n" if ($narg>0);
383 $func .= " movq %rdx,%rsi\n" if ($narg>1);
384 $func .= " movq %r8,%rdx\n" if ($narg>2);
385 $func .= " movq %r9,%rcx\n" if ($narg>3);
386 $func .= " movq 40(%rsp),%r8\n" if ($narg>4);
387 $func .= " movq 48(%rsp),%r9\n" if ($narg>5);
390 } elsif ($self->{value} ne "$current_function->{name}") {
391 $self->{value} .= ":" if ($masm && $ret!~m/^\$/);
392 $self->{value} . ":";
393 } elsif ($win64 && $current_function->{abi} eq "svr4") {
394 my $func = "$current_function->{name}" .
395 ($nasm ? ":" : "\tPROC $current_function->{scope}") .
397 $func .= " mov QWORD${PTR}[8+rsp],rdi\t;WIN64 prologue\n";
398 $func .= " mov QWORD${PTR}[16+rsp],rsi\n";
399 $func .= " mov rax,rsp\n";
400 $func .= "${decor}SEH_begin_$current_function->{name}:";
401 $func .= ":" if ($masm);
403 my $narg = $current_function->{narg};
404 $narg=6 if (!defined($narg));
405 $func .= " mov rdi,rcx\n" if ($narg>0);
406 $func .= " mov rsi,rdx\n" if ($narg>1);
407 $func .= " mov rdx,r8\n" if ($narg>2);
408 $func .= " mov rcx,r9\n" if ($narg>3);
409 $func .= " mov r8,QWORD${PTR}[40+rsp]\n" if ($narg>4);
410 $func .= " mov r9,QWORD${PTR}[48+rsp]\n" if ($narg>5);
413 "$current_function->{name}".
414 ($nasm ? ":" : "\tPROC $current_function->{scope}");
418 { package expr; # pick up expressions
420 my $self = shift; # single instance is enough...
424 if ($line =~ /(^[^,]+)/) {
427 $line = substr($line,@+[0]); $line =~ s/^\s+//;
429 $self->{value} =~ s/\@PLT// if (!$elf);
430 $self->{value} =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
431 $self->{value} =~ s/\.L/$decor/g;
437 if ($nasm && opcode->mnemonic()=~m/^j(?![re]cxz)/) {
438 "NEAR ".$self->{value};
444 { package directive; # pick up directives, which start with .
446 my $self = shift; # single instance is enough...
450 my %opcode = # lea 2f-1f(%rip),%dst; 1: nop; 2:
451 ( "%rax"=>0x01058d48, "%rcx"=>0x010d8d48,
452 "%rdx"=>0x01158d48, "%rbx"=>0x011d8d48,
453 "%rsp"=>0x01258d48, "%rbp"=>0x012d8d48,
454 "%rsi"=>0x01358d48, "%rdi"=>0x013d8d48,
455 "%r8" =>0x01058d4c, "%r9" =>0x010d8d4c,
456 "%r10"=>0x01158d4c, "%r11"=>0x011d8d4c,
457 "%r12"=>0x01258d4c, "%r13"=>0x012d8d4c,
458 "%r14"=>0x01358d4c, "%r15"=>0x013d8d4c );
460 if ($line =~ /^\s*(\.\w+)/) {
463 undef $self->{value};
464 $line = substr($line,@+[0]); $line =~ s/^\s+//;
467 /\.picmeup/ && do { if ($line =~ /(%r[\w]+)/i) {
469 $line=sprintf "0x%x,0x90000000",$opcode{$1};
473 /\.global|\.globl|\.extern/
474 && do { $globals{$line} = $prefix . $line;
475 $line = $globals{$line} if ($prefix);
478 /\.type/ && do { ($sym,$type,$narg) = split(',',$line);
479 if ($type eq "\@function") {
480 undef $current_function;
481 $current_function->{name} = $sym;
482 $current_function->{abi} = "svr4";
483 $current_function->{narg} = $narg;
484 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
485 } elsif ($type eq "\@abi-omnipotent") {
486 undef $current_function;
487 $current_function->{name} = $sym;
488 $current_function->{scope} = defined($globals{$sym})?"PUBLIC":"PRIVATE";
490 $line =~ s/\@abi\-omnipotent/\@function/;
491 $line =~ s/\@function.*/\@function/;
494 /\.asciz/ && do { if ($line =~ /^"(.*)"$/) {
496 $line = join(",",unpack("C*",$1),0);
500 /\.rva|\.long|\.quad/
501 && do { $line =~ s/([_a-z][_a-z0-9]*)/$globals{$1} or $1/gei;
502 $line =~ s/\.L/$decor/g;
508 $self->{value} = $dir . "\t" . $line;
510 if ($dir =~ /\.extern/) {
511 $self->{value} = ""; # swallow extern
512 } elsif (!$elf && $dir =~ /\.type/) {
514 $self->{value} = ".def\t" . ($globals{$1} or $1) . ";\t" .
515 (defined($globals{$1})?".scl 2;":".scl 3;") .
516 "\t.type 32;\t.endef"
517 if ($win64 && $line =~ /([^,]+),\@function/);
518 } elsif (!$elf && $dir =~ /\.size/) {
520 if (defined($current_function)) {
521 $self->{value} .= "${decor}SEH_end_$current_function->{name}:"
522 if ($win64 && $current_function->{abi} eq "svr4");
523 undef $current_function;
525 } elsif (!$elf && $dir =~ /\.align/) {
526 $self->{value} = ".p2align\t" . (log($line)/log(2));
527 } elsif ($dir eq ".section") {
528 $current_segment=$line;
529 if (!$elf && $current_segment eq ".init") {
530 if ($flavour eq "macosx") { $self->{value} = ".mod_init_func"; }
531 elsif ($flavour eq "mingw64") { $self->{value} = ".section\t.ctors"; }
533 } elsif ($dir =~ /\.(text|data)/) {
534 $current_segment=".$1";
535 } elsif ($dir =~ /\.hidden/) {
536 if ($flavour eq "macosx") { $self->{value} = ".private_extern\t$prefix$line"; }
537 elsif ($flavour eq "mingw64") { $self->{value} = ""; }
538 } elsif ($dir =~ /\.comm/) {
539 $self->{value} = "$dir\t$prefix$line";
540 $self->{value} =~ s|,([0-9]+),([0-9]+)$|",$1,".log($2)/log(2)|e if ($flavour eq "macosx");
546 # non-gas case or nasm/masm
548 /\.text/ && do { my $v=undef;
550 $v="section .text code align=64\n";
552 $v="$current_segment\tENDS\n" if ($current_segment);
553 $current_segment = ".text\$";
554 $v.="$current_segment\tSEGMENT ";
555 $v.=$masm>=$masmref ? "ALIGN(256)" : "PAGE";
561 /\.data/ && do { my $v=undef;
563 $v="section .data data align=8\n";
565 $v="$current_segment\tENDS\n" if ($current_segment);
566 $current_segment = "_DATA";
567 $v.="$current_segment\tSEGMENT";
572 /\.section/ && do { my $v=undef;
573 $line =~ s/([^,]*).*/$1/;
574 $line = ".CRT\$XCU" if ($line eq ".init");
577 if ($line=~/\.([px])data/) {
579 $v.=$1 eq "p"? 4 : 8;
580 } elsif ($line=~/\.CRT\$/i) {
581 $v.=" rdata align=8";
584 $v="$current_segment\tENDS\n" if ($current_segment);
585 $v.="$line\tSEGMENT";
586 if ($line=~/\.([px])data/) {
588 $v.=" ALIGN(".($1 eq "p" ? 4 : 8).")" if ($masm>=$masmref);
589 } elsif ($line=~/\.CRT\$/i) {
591 $v.=$masm>=$masmref ? "ALIGN(8)" : "DWORD";
594 $current_segment = $line;
598 /\.extern/ && do { $self->{value} = "EXTERN\t".$line;
599 $self->{value} .= ":NEAR" if ($masm);
603 && do { $self->{value} = $masm?"PUBLIC":"global";
604 $self->{value} .= "\t".$line;
607 /\.size/ && do { if (defined($current_function)) {
608 undef $self->{value};
609 if ($current_function->{abi} eq "svr4") {
610 $self->{value}="${decor}SEH_end_$current_function->{name}:";
611 $self->{value}.=":\n" if($masm);
613 $self->{value}.="$current_function->{name}\tENDP" if($masm && $current_function->{name});
614 undef $current_function;
618 /\.align/ && do { $self->{value} = "ALIGN\t".$line; last; };
619 /\.(value|long|rva|quad)/
620 && do { my $sz = substr($1,0,1);
621 my @arr = split(/,\s*/,$line);
622 my $last = pop(@arr);
623 my $conv = sub { my $var=shift;
624 $var=~s/^(0b[0-1]+)/oct($1)/eig;
625 $var=~s/^0x([0-9a-f]+)/0$1h/ig if ($masm);
626 if ($sz eq "D" && ($current_segment=~/.[px]data/ || $dir eq ".rva"))
627 { $var=~s/([_a-z\$\@][_a-z0-9\$\@]*)/$nasm?"$1 wrt ..imagebase":"imagerel $1"/egi; }
631 $sz =~ tr/bvlrq/BWDDQ/;
632 $self->{value} = "\tD$sz\t";
633 for (@arr) { $self->{value} .= &$conv($_).","; }
634 $self->{value} .= &$conv($last);
637 /\.byte/ && do { my @str=split(/,\s*/,$line);
638 map(s/(0b[0-1]+)/oct($1)/eig,@str);
639 map(s/0x([0-9a-f]+)/0$1h/ig,@str) if ($masm);
641 $self->{value}.="DB\t"
642 .join(",",@str[0..15])."\n";
643 foreach (0..15) { shift @str; }
645 $self->{value}.="DB\t"
646 .join(",",@str) if (@str);
649 /\.comm/ && do { my @str=split(/,\s*/,$line);
652 $v.="common $prefix@str[0] @str[1]";
654 $v="$current_segment\tENDS\n" if ($current_segment);
655 $current_segment = "_DATA";
656 $v.="$current_segment\tSEGMENT\n";
657 $v.="COMM @str[0]:DWORD:".@str[1]/4;
676 my ($dst,$src,$rex)=@_;
678 $rex|=0x04 if($dst>=8);
679 $rex|=0x01 if($src>=8);
680 push @opcode,($rex|0x40) if ($rex);
683 # older gas and ml64 don't handle SSE>2 instructions
684 my %regrm = ( "%eax"=>0, "%ecx"=>1, "%edx"=>2, "%ebx"=>3,
685 "%esp"=>4, "%ebp"=>5, "%esi"=>6, "%edi"=>7 );
687 my $movq = sub { # elderly gas can't handle inter-register movq
690 if ($arg =~ /%xmm([0-9]+),\s*%r(\w+)/) {
691 my ($src,$dst)=($1,$2);
692 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
693 rex(\@opcode,$src,$dst,0x8);
694 push @opcode,0x0f,0x7e;
695 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
697 } elsif ($arg =~ /%r(\w+),\s*%xmm([0-9]+)/) {
698 my ($src,$dst)=($2,$1);
699 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
700 rex(\@opcode,$src,$dst,0x8);
701 push @opcode,0x0f,0x6e;
702 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
710 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*(%\w+)/) {
715 if ($dst =~ /%r([0-9]+)d/) { $dst = $1; }
716 elsif ($dst =~ /%e/) { $dst = $regrm{$dst}; }
717 rex(\@opcode,$src,$dst);
718 push @opcode,0x0f,0x3a,0x16;
719 push @opcode,0xc0|(($src&7)<<3)|($dst&7); # ModR/M
728 if (shift =~ /\$([0-9]+),\s*(%\w+),\s*%xmm([0-9]+)/) {
733 if ($src =~ /%r([0-9]+)/) { $src = $1; }
734 elsif ($src =~ /%e/) { $src = $regrm{$src}; }
735 rex(\@opcode,$dst,$src);
736 push @opcode,0x0f,0x3a,0x22;
737 push @opcode,0xc0|(($dst&7)<<3)|($src&7); # ModR/M
746 if (shift =~ /%xmm([0-9]+),\s*%xmm([0-9]+)/) {
749 push @opcode,0x0f,0x38,0x00;
750 push @opcode,0xc0|($1&7)|(($2&7)<<3); # ModR/M
758 if (shift =~ /\$([0-9]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
761 push @opcode,0x0f,0x3a,0x0f;
762 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
770 my $pclmulqdq = sub {
771 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
774 push @opcode,0x0f,0x3a,0x44;
775 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
777 push @opcode,$c=~/^0/?oct($c):$c;
785 if (shift =~ /%[er](\w+)/) {
788 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
789 rex(\@opcode,0,$1,8);
790 push @opcode,0x0f,0xc7,0xf0|($dst&7);
798 if (shift =~ /%[er](\w+)/) {
801 if ($dst !~ /[0-9]+/) { $dst = $regrm{"%e$dst"}; }
802 rex(\@opcode,0,$1,8);
803 push @opcode,0x0f,0xc7,0xf8|($dst&7);
812 my ($dst,$src1,$src2,$rxb)=@_;
815 $rxb&=~(0x04<<5) if($dst>=8);
816 $rxb&=~(0x01<<5) if($src1>=8);
817 $rxb&=~(0x02<<5) if($src2>=8);
822 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
824 rxb(\@opcode,$3,$2,-1,0x08);
825 push @opcode,0x78,0xc2;
826 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
828 push @opcode,$c=~/^0/?oct($c):$c;
836 if (shift =~ /\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
838 rxb(\@opcode,$3,$2,-1,0x08);
839 push @opcode,0x78,0xc3;
840 push @opcode,0xc0|($2&7)|(($3&7)<<3); # ModR/M
842 push @opcode,$c=~/^0/?oct($c):$c;
865 $line =~ s|[#!].*$||; # get rid of asm-style comments...
866 $line =~ s|/\*.*\*/||; # ... and C-style comments...
867 $line =~ s|^\s+||; # ... and skip white spaces in beginning
868 $line =~ s|\s+$||; # ... and at the end
874 if ($label=label->re(\$line)) { print $label->out(); }
876 if (directive->re(\$line)) {
877 printf "%s",directive->out();
878 } elsif ($opcode=opcode->re(\$line)) {
879 my $asm = eval("\$".$opcode->mnemonic());
882 if ((ref($asm) eq 'CODE') && scalar(@bytes=&$asm($line))) {
883 print $gas?".byte\t":"DB\t",join(',',@bytes),"\n";
887 ARGUMENT: while (1) {
890 if ($arg=register->re(\$line)) { opcode->size($arg->size()); }
891 elsif ($arg=const->re(\$line)) { }
892 elsif ($arg=ea->re(\$line)) { }
893 elsif ($arg=expr->re(\$line)) { }
894 else { last ARGUMENT; }
898 last ARGUMENT if ($line !~ /^,/);
905 my $sz=opcode->size();
908 $insn = $opcode->out($#args>=1?$args[$#args]->size():$sz);
909 @args = map($_->out($sz),@args);
910 printf "\t%s\t%s",$insn,join(",",@args);
912 $insn = $opcode->out();
915 # $insn.=$sz compensates for movq, pinsrw, ...
916 if ($arg =~ /^xmm[0-9]+$/) { $insn.=$sz; $sz="x" if(!$sz); last; }
917 if ($arg =~ /^ymm[0-9]+$/) { $insn.=$sz; $sz="y" if(!$sz); last; }
918 if ($arg =~ /^zmm[0-9]+$/) { $insn.=$sz; $sz="z" if(!$sz); last; }
919 if ($arg =~ /^mm[0-9]+$/) { $insn.=$sz; $sz="q" if(!$sz); last; }
921 @args = reverse(@args);
922 undef $sz if ($nasm && $opcode->mnemonic() eq "lea");
923 printf "\t%s\t%s",$insn,join(",",map($_->out($sz),@args));
926 printf "\t%s",$opcode->out();
933 print "\n$current_segment\tENDS\n" if ($current_segment && $masm);
934 print "END\n" if ($masm);
938 \f#################################################
939 # Cross-reference x86_64 ABI "card"
959 # (*) volatile register
960 # (-) preserved by callee
961 # (#) Nth argument, volatile
963 # In Unix terms top of stack is argument transfer area for arguments
964 # which could not be accomodated in registers. Or in other words 7th
965 # [integer] argument resides at 8(%rsp) upon function entry point.
966 # 128 bytes above %rsp constitute a "red zone" which is not touched
967 # by signal handlers and can be used as temporal storage without
968 # allocating a frame.
970 # In Win64 terms N*8 bytes on top of stack is argument transfer area,
971 # which belongs to/can be overwritten by callee. N is the number of
972 # arguments passed to callee, *but* not less than 4! This means that
973 # upon function entry point 5th argument resides at 40(%rsp), as well
974 # as that 32 bytes from 8(%rsp) can always be used as temporal
975 # storage [without allocating a frame]. One can actually argue that
976 # one can assume a "red zone" above stack pointer under Win64 as well.
977 # Point is that at apparently no occasion Windows kernel would alter
978 # the area above user stack pointer in true asynchronous manner...
980 # All the above means that if assembler programmer adheres to Unix
981 # register and stack layout, but disregards the "red zone" existence,
982 # it's possible to use following prologue and epilogue to "gear" from
983 # Unix to Win64 ABI in leaf functions with not more than 6 arguments.
985 # omnipotent_function:
989 # movq %rcx,%rdi ; if 1st argument is actually present
990 # movq %rdx,%rsi ; if 2nd argument is actually ...
991 # movq %r8,%rdx ; if 3rd argument is ...
992 # movq %r9,%rcx ; if 4th argument ...
993 # movq 40(%rsp),%r8 ; if 5th ...
994 # movq 48(%rsp),%r9 ; if 6th ...
1003 \f#################################################
1004 # Win64 SEH, Structured Exception Handling.
1006 # Unlike on Unix systems(*) lack of Win64 stack unwinding information
1007 # has undesired side-effect at run-time: if an exception is raised in
1008 # assembler subroutine such as those in question (basically we're
1009 # referring to segmentation violations caused by malformed input
1010 # parameters), the application is briskly terminated without invoking
1011 # any exception handlers, most notably without generating memory dump
1012 # or any user notification whatsoever. This poses a problem. It's
1013 # possible to address it by registering custom language-specific
1014 # handler that would restore processor context to the state at
1015 # subroutine entry point and return "exception is not handled, keep
1016 # unwinding" code. Writing such handler can be a challenge... But it's
1017 # doable, though requires certain coding convention. Consider following
1020 # .type function,@function
1022 # movq %rsp,%rax # copy rsp to volatile register
1023 # pushq %r15 # save non-volatile registers
1027 # subq %rdi,%r11 # prepare [variable] stack frame
1029 # movq %rax,0(%r11) # check for exceptions
1030 # movq %r11,%rsp # allocate [variable] stack frame
1031 # movq %rax,0(%rsp) # save original rsp value
1034 # movq 0(%rsp),%rcx # pull original rsp value
1035 # movq -24(%rcx),%rbp # restore non-volatile registers
1036 # movq -16(%rcx),%rbx
1037 # movq -8(%rcx),%r15
1038 # movq %rcx,%rsp # restore original rsp
1040 # .size function,.-function
1042 # The key is that up to magic_point copy of original rsp value remains
1043 # in chosen volatile register and no non-volatile register, except for
1044 # rsp, is modified. While past magic_point rsp remains constant till
1045 # the very end of the function. In this case custom language-specific
1046 # exception handler would look like this:
1048 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
1049 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
1050 # { ULONG64 *rsp = (ULONG64 *)context->Rax;
1051 # if (context->Rip >= magic_point)
1052 # { rsp = ((ULONG64 **)context->Rsp)[0];
1053 # context->Rbp = rsp[-3];
1054 # context->Rbx = rsp[-2];
1055 # context->R15 = rsp[-1];
1057 # context->Rsp = (ULONG64)rsp;
1058 # context->Rdi = rsp[1];
1059 # context->Rsi = rsp[2];
1061 # memcpy (disp->ContextRecord,context,sizeof(CONTEXT));
1062 # RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase,
1063 # dips->ControlPc,disp->FunctionEntry,disp->ContextRecord,
1064 # &disp->HandlerData,&disp->EstablisherFrame,NULL);
1065 # return ExceptionContinueSearch;
1068 # It's appropriate to implement this handler in assembler, directly in
1069 # function's module. In order to do that one has to know members'
1070 # offsets in CONTEXT and DISPATCHER_CONTEXT structures and some constant
1071 # values. Here they are:
1091 # sizeof(CONTEXT) 1232
1092 # DISPATCHER_CONTEXT.ControlPc 0
1093 # DISPATCHER_CONTEXT.ImageBase 8
1094 # DISPATCHER_CONTEXT.FunctionEntry 16
1095 # DISPATCHER_CONTEXT.EstablisherFrame 24
1096 # DISPATCHER_CONTEXT.TargetIp 32
1097 # DISPATCHER_CONTEXT.ContextRecord 40
1098 # DISPATCHER_CONTEXT.LanguageHandler 48
1099 # DISPATCHER_CONTEXT.HandlerData 56
1100 # UNW_FLAG_NHANDLER 0
1101 # ExceptionContinueSearch 1
1103 # In order to tie the handler to the function one has to compose
1104 # couple of structures: one for .xdata segment and one for .pdata.
1106 # UNWIND_INFO structure for .xdata segment would be
1108 # function_unwind_info:
1112 # This structure designates exception handler for a function with
1113 # zero-length prologue, no stack frame or frame register.
1115 # To facilitate composing of .pdata structures, auto-generated "gear"
1116 # prologue copies rsp value to rax and denotes next instruction with
1117 # .LSEH_begin_{function_name} label. This essentially defines the SEH
1118 # styling rule mentioned in the beginning. Position of this label is
1119 # chosen in such manner that possible exceptions raised in the "gear"
1120 # prologue would be accounted to caller and unwound from latter's frame.
1121 # End of function is marked with respective .LSEH_end_{function_name}
1122 # label. To summarize, .pdata segment would contain
1124 # .rva .LSEH_begin_function
1125 # .rva .LSEH_end_function
1126 # .rva function_unwind_info
1128 # Reference to functon_unwind_info from .xdata segment is the anchor.
1129 # In case you wonder why references are 32-bit .rvas and not 64-bit
1130 # .quads. References put into these two segments are required to be
1131 # *relative* to the base address of the current binary module, a.k.a.
1132 # image base. No Win64 module, be it .exe or .dll, can be larger than
1133 # 2GB and thus such relative references can be and are accommodated in
1136 # Having reviewed the example function code, one can argue that "movq
1137 # %rsp,%rax" above is redundant. It is not! Keep in mind that on Unix
1138 # rax would contain an undefined value. If this "offends" you, use
1139 # another register and refrain from modifying rax till magic_point is
1140 # reached, i.e. as if it was a non-volatile register. If more registers
1141 # are required prior [variable] frame setup is completed, note that
1142 # nobody says that you can have only one "magic point." You can
1143 # "liberate" non-volatile registers by denoting last stack off-load
1144 # instruction and reflecting it in finer grade unwind logic in handler.
1145 # After all, isn't it why it's called *language-specific* handler...
1147 # Attentive reader can notice that exceptions would be mishandled in
1148 # auto-generated "gear" epilogue. Well, exception effectively can't
1149 # occur there, because if memory area used by it was subject to
1150 # segmentation violation, then it would be raised upon call to the
1151 # function (and as already mentioned be accounted to caller, which is
1152 # not a problem). If you're still not comfortable, then define tail
1153 # "magic point" just prior ret instruction and have handler treat it...
1155 # (*) Note that we're talking about run-time, not debug-time. Lack of
1156 # unwind information makes debugging hard on both Windows and
1157 # Unix. "Unlike" referes to the fact that on Unix signal handler
1158 # will always be invoked, core dumped and appropriate exit code
1159 # returned to parent (for user notification).