aesni-sha1-x86_64.pl 31 KB
Newer Older

	&Xloop_avx(\&body_20_39);
	&Xloop_avx(\&body_20_39);
	&Xloop_avx(\&body_20_39);

$code.=<<___;
	vmovups	$iv,48($out,$in0)		# write output
	lea	64($in0),$in0

	add	0($ctx),$A			# update context
	add	4($ctx),@T[0]
	add	8($ctx),$C
	add	12($ctx),$D
	mov	$A,0($ctx)
	add	16($ctx),$E
	mov	@T[0],4($ctx)
	mov	@T[0],$B			# magic seed
	mov	$C,8($ctx)
	mov	$D,12($ctx)
	mov	$E,16($ctx)
	jmp	.Loop_avx

.align	16
.Ldone_avx:
___
				$jj=$j=$saved_j; @V=@saved_V;
				$r=$saved_r;     @rndkey=@saved_rndkey;

	&Xtail_avx(\&body_20_39);
	&Xtail_avx(\&body_20_39);
	&Xtail_avx(\&body_20_39);

$code.=<<___;
	vmovups	$iv,48($out,$in0)		# write output
	mov	88(%rsp),$ivp			# restore $ivp

	add	0($ctx),$A			# update context
	add	4($ctx),@T[0]
	add	8($ctx),$C
	mov	$A,0($ctx)
	add	12($ctx),$D
	mov	@T[0],4($ctx)
	add	16($ctx),$E
	mov	$C,8($ctx)
	mov	$D,12($ctx)
	mov	$E,16($ctx)
	vmovups	$iv,($ivp)			# write IV
	vzeroall
___
$code.=<<___ if ($win64);
	movaps	96+0(%rsp),%xmm6
	movaps	96+16(%rsp),%xmm7
	movaps	96+32(%rsp),%xmm8
	movaps	96+48(%rsp),%xmm9
	movaps	96+64(%rsp),%xmm10
	movaps	96+80(%rsp),%xmm11
	movaps	96+96(%rsp),%xmm12
	movaps	96+112(%rsp),%xmm13
	movaps	96+128(%rsp),%xmm14
	movaps	96+144(%rsp),%xmm15
___
$code.=<<___;
	lea	`104+($win64?10*16:0)`(%rsp),%rsi
	mov	0(%rsi),%r15
	mov	8(%rsi),%r14
	mov	16(%rsi),%r13
	mov	24(%rsi),%r12
	mov	32(%rsi),%rbp
	mov	40(%rsi),%rbx
	lea	48(%rsi),%rsp
.Lepilogue_avx:
	ret
.size	aesni_cbc_sha1_enc_avx,.-aesni_cbc_sha1_enc_avx
___
}
$code.=<<___;
.align	64
K_XX_XX:
.long	0x5a827999,0x5a827999,0x5a827999,0x5a827999	# K_00_19
.long	0x6ed9eba1,0x6ed9eba1,0x6ed9eba1,0x6ed9eba1	# K_20_39
.long	0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc,0x8f1bbcdc	# K_40_59
.long	0xca62c1d6,0xca62c1d6,0xca62c1d6,0xca62c1d6	# K_60_79
.long	0x00010203,0x04050607,0x08090a0b,0x0c0d0e0f	# pbswap mask

.asciz	"AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
.align	64
___

# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
if ($win64) {
$rec="%rcx";
$frame="%rdx";
$context="%r8";
$disp="%r9";

$code.=<<___;
.extern	__imp_RtlVirtualUnwind
.type	ssse3_handler,\@abi-omnipotent
.align	16
ssse3_handler:
	push	%rsi
	push	%rdi
	push	%rbx
	push	%rbp
	push	%r12
	push	%r13
	push	%r14
	push	%r15
	pushfq
	sub	\$64,%rsp

	mov	120($context),%rax	# pull context->Rax
	mov	248($context),%rbx	# pull context->Rip

	mov	8($disp),%rsi		# disp->ImageBase
	mov	56($disp),%r11		# disp->HandlerData

	mov	0(%r11),%r10d		# HandlerData[0]
	lea	(%rsi,%r10),%r10	# prologue label
	cmp	%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

	lea	96(%rax),%rsi
	lea	512($context),%rdi	# &context.Xmm6
	mov	\$20,%ecx
	.long	0xa548f3fc		# cld; rep movsq
	lea	`104+10*16`(%rax),%rax	# adjust stack pointer

	mov	0(%rax),%r15
	mov	8(%rax),%r14
	mov	16(%rax),%r13
	mov	24(%rax),%r12
	mov	32(%rax),%rbp
	mov	40(%rax),%rbx
	lea	48(%rax),%rax
	mov	%rbx,144($context)	# restore context->Rbx
	mov	%rbp,160($context)	# restore context->Rbp
	mov	%r12,216($context)	# restore context->R12
	mov	%r13,224($context)	# restore context->R13
	mov	%r14,232($context)	# restore context->R14
	mov	%r15,240($context)	# restore context->R15

.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	ssse3_handler,.-ssse3_handler

.section	.pdata
.align	4
	.rva	.LSEH_begin_aesni_cbc_sha1_enc_ssse3
	.rva	.LSEH_end_aesni_cbc_sha1_enc_ssse3
	.rva	.LSEH_info_aesni_cbc_sha1_enc_ssse3
___
$code.=<<___ if ($avx);
	.rva	.LSEH_begin_aesni_cbc_sha1_enc_avx
	.rva	.LSEH_end_aesni_cbc_sha1_enc_avx
	.rva	.LSEH_info_aesni_cbc_sha1_enc_avx
___
$code.=<<___;
.section	.xdata
.align	8
.LSEH_info_aesni_cbc_sha1_enc_ssse3:
	.byte	9,0,0,0
	.rva	ssse3_handler
	.rva	.Lprologue_ssse3,.Lepilogue_ssse3	# HandlerData[]
___
$code.=<<___ if ($avx);
.LSEH_info_aesni_cbc_sha1_enc_avx:
	.byte	9,0,0,0
	.rva	ssse3_handler
	.rva	.Lprologue_avx,.Lepilogue_avx		# 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 ($line=~/(aes[a-z]+)\s+%xmm([0-9]+),\s*%xmm([0-9]+)/) {
	my %opcodelet = (
		"aesenc" => 0xdc,	"aesenclast" => 0xdd
	);
	return undef if (!defined($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);
    }
    return $line;
}

$code =~ s/\`([^\`]*)\`/eval($1)/gem;
$code =~ s/\b(aes.*%xmm[0-9]+).*$/aesni($1)/gem;

print $code;
close STDOUT;