r10,%rbx		# context->Rip<prologue label
	jb	.Lcommon_seh_tail

	mov	152($context),%rax	# pull context->Rsp

	mov	4(%r11),%r10d		# HandlerData[1]
	lea	(%rsi,%r10),%r10	# epilogue label
	cmp	%r10,%rbx		# context->Rip>=epilogue label
	jae	.Lcommon_seh_tail

	mov	208($context),%rax	# pull context->R11

	lea	-0xa8(%rax),%rsi	# %xmm save area
	lea	512($context),%rdi	# & context.Xmm6
	mov	\$20,%ecx		# 10*sizeof(%xmm0)/sizeof(%rax)
	.long	0xa548f3fc		# cld; rep movsq

	mov	-8(%rax),%rbp		# restore saved %rbp
	mov	%rbp,160($context)	# restore context->Rbp


.Lcommon_seh_tail:
	mov	8(%rax),%rdi
	mov	16(%rax),%rsi
	mov	%rax,152($context)	# restore context->Rsp
	mov	%rsi,168($context)	# restore context->Rsi
	mov	%rdi,176($context)	# restore context->Rdi

	mov	40($disp),%rdi		# disp->ContextRecord
	mov	$context,%rsi		# context
	mov	\$154,%ecx		# sizeof(CONTEXT)
	.long	0xa548f3fc		# cld; rep movsq

	mov	$disp,%rsi
	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
	mov	0(%rsi),%r8		# arg3, disp->ControlPc
	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
	mov	40(%rsi),%r10		# disp->ContextRecord
	lea	56(%rsi),%r11		# &disp->HandlerData
	lea	24(%rsi),%r12		# &disp->EstablisherFrame
	mov	%r10,32(%rsp)		# arg5
	mov	%r11,40(%rsp)		# arg6
	mov	%r12,48(%rsp)		# arg7
	mov	%rcx,56(%rsp)		# arg8, (NULL)
	call	*__imp_RtlVirtualUnwind(%rip)

	mov	\$1,%eax		# ExceptionContinueSearch
	add	\$64,%rsp
	popfq
	pop	%r15
	pop	%r14
	pop	%r13
	pop	%r12
	pop	%rbp
	pop	%rbx
	pop	%rdi
	pop	%rsi
	ret
.size	ctr_xts_se_handler,.-ctr_xts_se_handler

.section	.pdata
.align	4
___
$code.=<<___ if ($PREFIX eq "aes_hw");
	.rva	.LSEH_begin_${PREFIX}_ctr32_encrypt_blocks
	.rva	.LSEH_end_${PREFIX}_ctr32_encrypt_blocks
	.rva	.LSEH_info_ctr32
___
$code.=<<___;
.section	.xdata
.align	8
___
$code.=<<___ if ($PREFIX eq "aes_hw");
.LSEH_info_ctr32:
	.byte	9,0,0,0
	.rva	ctr_xts_se_handler
	.rva	.Lctr32_body,.Lctr32_epilogue		# HandlerData[]
___
}

sub rex {
  local *opcode=shift;
  my ($dst,$src)=@_;
  my $rex=0;

    $rex|=0x04			if($dst>=8);
    $rex|=0x01			if($src>=8);
    push @opcode,$rex|0x40	if($rex);
}

sub aesni {
  my $line=shift;
  my @opcode=(0x66);

    if ($line=~/(aeskeygenassist)\s+\$([x0-9a-f]+),\s*%xmm([0-9]+),\s*%xmm([0-9]+)/) {
	rex(\@opcode,$4,$3);
	push @opcode,0x0f,0x3a,0xdf;
	push @opcode,0xc0|($3&7)|(($4&7)<<3);	# ModR/M
	my $c=$2;
	push @opcode,$c=~/^0/?oct($c):$c;
	return ".byte\t".join(',',@opcode);
    }
    elsif ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
	my %opcodelet = (
		"aesimc" => 0xdb,
		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
		"aesdec" => 0xde,	"aesdeclast" => 0xdf
	);
	return undef if (!defined($opcodelet{$1}));
	rex(\@opcode,$3,$2);
	push @opcode,0x0f,0x38,$opcodelet{$1};
	push @opcode,0xc0|($2&7)|(($3&7)<<3);	# ModR/M
	return ".byte\t".join(',',@opcode);
    }
    elsif ($line=~/(aes[a-z]+)\s+([0x1-9a-fA-F]*)\(%rsp\),\s*%xmm([0-9]+)/) {
	my %opcodelet = (
		"aesenc" => 0xdc,	"aesenclast" => 0xdd,
		"aesdec" => 0xde,	"aesdeclast" => 0xdf
	);
	return undef if (!defined($opcodelet{$1}));
	my $off = $2;
	push @opcode,0x44 if ($3>=8);
	push @opcode,0x0f,0x38,$opcodelet{$1};
	push @opcode,0x44|(($3&7)<<3),0x24;	# ModR/M
	push @opcode,($off=~/^0/?oct($off):$off)&0xff;
	return ".byte\t".join(',',@opcode);
    }
    return $line;
}

sub movbe {
	".byte	0x0f,0x38,0xf1,0x44,0x24,".shift;
}

$code =~ s/\`([^\`]*)\`/eval($1)/gem;
$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;
#$code =~ s/\bmovbe\s+%eax/bswap %eax; mov %eax/gm;	# debugging artefact
$code =~ s/\bmovbe\s+%eax,\s*([0-9]+)\(%rsp\)/movbe($1)/gem;

print $code;

close STDOUT or die "error closing STDOUT: $!";