des_enc.m4

OpenSSL 0.9.8k 最新版OpenSSL

	ld	[out3+local0], local0
	xor	$2, local4, $2

	and	local2, 252, local2
	ld	[global1+local1], local1
	xor	$2, local5, $2            ! 6 finished local5 used

	ld	[global3+local2], local2
	srl	$1, 3, local3
	xor	$2, local0, $2

	ifelse($6,{}, {}, {ld	[$6+4], out1}) ! key next encryption/decryption
	sll	$1, 29, local4
	xor	$2, local1, $2

	ifelse($7,{}, {}, {retl})
	xor	$2, local2, $2
})


! {fp_macro}
!
!  parameter 1   right (original left)
!  parameter 2   left (original right)
!  parameter 3   1 for optional store to [in0]
!  parameter 4   1 for load input/output address to local5/7
!
!  The final permutation logic switches the halfes, meaning that
!  left and right ends up the the registers originally used.

define(fp_macro, {

! {fp_macro}
! $1 $2 $3 $4 $5 $6 $7 $8 $9

	! initially undo the rotate 3 left done after initial permutation
	! original left is received shifted 3 right and 29 left in local3/4

	sll	$2, 29, local1
	or	local3, local4, $1

	srl	$2, 3, $2
	sethi	%hi(0x55555555), local2

	or	$2, local1, $2
	or	local2, %lo(0x55555555), local2

	srl	$2, 1, local3
	sethi	%hi(0x00ff00ff), local1
	xor	local3, $1, local3
	or	local1, %lo(0x00ff00ff), local1
	and	local3, local2, local3
	sethi	%hi(0x33333333), local4
	sll	local3, 1, local2

	xor	$1, local3, $1

	srl	$1, 8, local3
	xor	$2, local2, $2
	xor	local3, $2, local3
	or	local4, %lo(0x33333333), local4
	and	local3, local1, local3
	sethi	%hi(0x0000ffff), local1
	sll	local3, 8, local2

	xor	$2, local3, $2

	srl	$2, 2, local3
	xor	$1, local2, $1
	xor	local3, $1, local3
	or	local1, %lo(0x0000ffff), local1
	and	local3, local4, local3
	sethi	%hi(0x0f0f0f0f), local4
	sll	local3, 2, local2

	ifelse($4,1, {LDPTR INPUT, local5})
	xor	$1, local3, $1

	ifelse($4,1, {LDPTR OUTPUT, local7})
	srl	$1, 16, local3
	xor	$2, local2, $2
	xor	local3, $2, local3
	or	local4, %lo(0x0f0f0f0f), local4
	and	local3, local1, local3
	sll	local3, 16, local2

	xor	$2, local3, local1

	srl	local1, 4, local3
	xor	$1, local2, $1
	xor	local3, $1, local3
	and	local3, local4, local3
	sll	local3, 4, local2

	xor	$1, local3, $1

	! optional store:

	ifelse($3,1, {st $1, [in0]})

	xor	local1, local2, $2

	ifelse($3,1, {st $2, [in0+4]})

})


! {fp_ip_macro}
!
! Does initial permutation for next block mixed with
! final permutation for current block.
!
! parameter 1   original left
! parameter 2   original right
! parameter 3   left ip
! parameter 4   right ip
! parameter 5   1: load ks1/ks2 to in3/in4, add 120 to in4
!                2: mov in4 to in3
!
! also adds -8 to length in2 and loads loop counter to out4

define(fp_ip_macro, {

! {fp_ip_macro}
! $1 $2 $3 $4 $5 $6 $7 $8 $9

	define({temp1},{out4})
	define({temp2},{local3})

	define({ip1},{local1})
	define({ip2},{local2})
	define({ip4},{local4})
	define({ip5},{local5})

	! $1 in local3, local4

	ld	[out2+256], ip1
	sll	out5, 29, temp1
	or	local3, local4, $1

	srl	out5, 3, $2
	ifelse($5,2,{mov in4, in3})

	ld	[out2+272], ip5
	srl	$4, 4, local0
	or	$2, temp1, $2

	srl	$2, 1, temp1
	xor	temp1, $1, temp1

	and	temp1, ip5, temp1
	xor	local0, $3, local0

	sll	temp1, 1, temp2
	xor	$1, temp1, $1

	and	local0, ip1, local0
	add	in2, -8, in2

	sll	local0, 4, local7
	xor	$3, local0, $3

	ld	[out2+268], ip4
	srl	$1, 8, temp1
	xor	$2, temp2, $2
	ld	[out2+260], ip2
	srl	$3, 16, local0
	xor	$4, local7, $4
	xor	temp1, $2, temp1
	xor	local0, $4, local0
	and	temp1, ip4, temp1
	and	local0, ip2, local0
	sll	temp1, 8, temp2
	xor	$2, temp1, $2
	sll	local0, 16, local7
	xor	$4, local0, $4

	srl	$2, 2, temp1
	xor	$1, temp2, $1

	ld	[out2+264], temp2         ! ip3
	srl	$4, 2, local0
	xor	$3, local7, $3
	xor	temp1, $1, temp1
	xor	local0, $3, local0
	and	temp1, temp2, temp1
	and	local0, temp2, local0
	sll	temp1, 2, temp2
	xor	$1, temp1, $1
	sll	local0, 2, local7
	xor	$3, local0, $3

	srl	$1, 16, temp1
	xor	$2, temp2, $2
	srl	$3, 8, local0
	xor	$4, local7, $4
	xor	temp1, $2, temp1
	xor	local0, $4, local0
	and	temp1, ip2, temp1
	and	local0, ip4, local0
	sll	temp1, 16, temp2
	xor	$2, temp1, local4
	sll	local0, 8, local7
	xor	$4, local0, $4

	srl	$4, 1, local0
	xor	$3, local7, $3

	srl	local4, 4, temp1
	xor	local0, $3, local0

	xor	$1, temp2, $1
	and	local0, ip5, local0

	sll	local0, 1, local7
	xor	temp1, $1, temp1

	xor	$3, local0, $3
	xor	$4, local7, $4

	sll	$3, 3, local5
	and	temp1, ip1, temp1

	sll	temp1, 4, temp2
	xor	$1, temp1, $1

	ifelse($5,1,{LDPTR	KS2, in4})
	sll	$4, 3, local2
	xor	local4, temp2, $2

	! reload since used as temporar:

	ld	[out2+280], out4          ! loop counter

	srl	$3, 29, local0
	ifelse($5,1,{add in4, 120, in4})

	ifelse($5,1,{LDPTR	KS1, in3})
	srl	$4, 29, local7

	or	local0, local5, $4
	or	local2, local7, $3

})



! {load_little_endian}
!
! parameter 1  address
! parameter 2  destination left
! parameter 3  destination right
! parameter 4  temporar
! parameter 5  label

define(load_little_endian, {

! {load_little_endian}
! $1 $2 $3 $4 $5 $6 $7 $8 $9

	! first in memory to rightmost in register

#ifdef OPENSSL_SYSNAME_ULTRASPARC
	andcc	$1, 3, global0
	bne,pn	%icc, $5
	nop

	lda	[$1] 0x88, $2
	add	$1, 4, $4

	ba,pt	%icc, $5a
	lda	[$4] 0x88, $3
#endif

$5:
	ldub	[$1+3], $2

	ldub	[$1+2], $4
	sll	$2, 8, $2
	or	$2, $4, $2

	ldub	[$1+1], $4
	sll	$2, 8, $2
	or	$2, $4, $2

	ldub	[$1+0], $4
	sll	$2, 8, $2
	or	$2, $4, $2


	ldub	[$1+3+4], $3

	ldub	[$1+2+4], $4
	sll	$3, 8, $3
	or	$3, $4, $3

	ldub	[$1+1+4], $4
	sll	$3, 8, $3
	or	$3, $4, $3

	ldub	[$1+0+4], $4
	sll	$3, 8, $3
	or	$3, $4, $3
$5a:

})


! {load_little_endian_inc}
!
! parameter 1  address
! parameter 2  destination left
! parameter 3  destination right
! parameter 4  temporar
! parameter 4  label
!
! adds 8 to address

define(load_little_endian_inc, {

! {load_little_endian_inc}
! $1 $2 $3 $4 $5 $6 $7 $8 $9

	! first in memory to rightmost in register

#ifdef OPENSSL_SYSNAME_ULTRASPARC
	andcc	$1, 3, global0
	bne,pn	%icc, $5
	nop

	lda	[$1] 0x88, $2
	add	$1, 4, $1

	lda	[$1] 0x88, $3
	ba,pt	%icc, $5a
	add	$1, 4, $1
#endif

$5:
	ldub	[$1+3], $2

	ldub	[$1+2], $4
	sll	$2, 8, $2
	or	$2, $4, $2

	ldub	[$1+1], $4
	sll	$2, 8, $2
	or	$2, $4, $2

	ldub	[$1+0], $4
	sll	$2, 8, $2
	or	$2, $4, $2

	ldub	[$1+3+4], $3
	add	$1, 8, $1

	ldub	[$1+2+4-8], $4
	sll	$3, 8, $3
	or	$3, $4, $3

	ldub	[$1+1+4-8], $4
	sll	$3, 8, $3
	or	$3, $4, $3

	ldub	[$1+0+4-8], $4
	sll	$3, 8, $3
	or	$3, $4, $3
$5a:

})


! {load_n_bytes}
!
! Loads 1 to 7 bytes little endian
! Remaining bytes are zeroed.
!
! parameter 1  address
! parameter 2  length
! parameter 3  destination register left
! parameter 4  destination register right
! parameter 5  temp
! parameter 6  temp2
! parameter 7  label
! parameter 8  return label

define(load_n_bytes, {

! {load_n_bytes}
! $1 $2 $5 $6 $7 $8 $7 $8 $9

$7.0:	call	.+8
	sll	$2, 2, $6

	add	%o7,$7.jmp.table-$7.0,$5

	add	$5, $6, $5
	mov	0, $4

	ld	[$5], $5

	jmp	%o7+$5
	mov	0, $3

$7.7:
	ldub	[$1+6], $5
	sll	$5, 16, $5
	or	$3, $5, $3
$7.6:
	ldub	[$1+5], $5
	sll	$5, 8, $5
	or	$3, $5, $3
$7.5:
	ldub	[$1+4], $5
	or	$3, $5, $3
$7.4:
	ldub	[$1+3], $5
	sll	$5, 24, $5
	or	$4, $5, $4
$7.3:
	ldub	[$1+2], $5
	sll	$5, 16, $5
	or	$4, $5, $4
$7.2:
	ldub	[$1+1], $5
	sll	$5, 8, $5
	or	$4, $5, $4
$7.1:
	ldub	[$1+0], $5
	ba	$8
	or	$4, $5, $4

	.align 4

$7.jmp.table:
	.word	0
	.word	$7.1-$7.0
	.word	$7.2-$7.0
	.word	$7.3-$7.0
	.word	$7.4-$7.0
	.word	$7.5-$7.0
	.word	$7.6-$7.0
	.word	$7.7-$7.0
})


! {store_little_endian}
!
! parameter 1  address
! parameter 2  source left
! parameter 3  source right
! parameter 4  temporar

define(store_little_endian, {

! {store_little_endian}
! $1 $2 $3 $4 $5 $6 $7 $8 $9

	! rightmost in register to first in memory

#ifdef OPENSSL_SYSNAME_ULTRASPARC
	andcc	$1, 3, global0
	bne,pn	%icc, $5
	nop

	sta	$2, [$1] 0x88
	add	$1, 4, $4

	ba,pt	%icc, $5a
	sta	$3, [$4] 0x88
#endif

$5:
	and	$2, 255, $4
	stub	$4, [$1+0]

	srl	$2, 8, $4
	and	$4, 255, $4
	stub	$4, [$1+1]

	srl	$2, 16, $4
	and	$4, 255, $4
	stub	$4, [$1+2]

	srl	$2, 24, $4
	stub	$4, [$1+3]


	and	$3, 255, $4
	stub	$4, [$1+0+4]

	srl	$3, 8, $4
	and	$4, 255, $4
	stub	$4, [$1+1+4]

	srl	$3, 16, $4
	and	$4, 255, $4
	stub	$4, [$1+2+4]

	srl	$3, 24, $4
	stub	$4, [$1+3+4]

$5a:

})


! {store_n_bytes}
!
! Stores 1 to 7 bytes little endian
!
! parameter 1  address
! parameter 2  length
! parameter 3  source register left
! parameter 4  source register right
! parameter 5  temp
! parameter 6  temp2
! parameter 7  label
! parameter 8  return label

define(store_n_bytes, {

! {store_n_bytes}
! $1 $2 $5 $6 $7 $8 $7 $8 $9