Changed AES macros, to handle a table register.

Also take more of the used registers as argument.

Rev: src/nettle/x86/machine.m4:1.9

x86/machine.m4

-dnl AES_LOAD(key, src)
+dnl AES_LOAD(a, b, c, d, src, key)
 dnl Loads the next block of data from src, and add the subkey pointed
 dnl to by key.
 dnl Note that x86 allows unaligned accesses.
 dnl Would it be preferable to interleave the loads and stores?
 define(<AES_LOAD>, <
-	movl	($2),%eax
-	movl	4($2),%ebx
-	movl	8($2),%ecx
-	movl	12($2),%edx
+	movl	($5),$1
+	movl	4($5),$2
+	movl	8($5),$3
+	movl	12($5),$4
 	
-	xorl	($1),%eax
-	xorl	4($1),%ebx
-	xorl	8($1),%ecx
-	xorl	12($1),%edx>)dnl
+	xorl	($6),$1
+	xorl	4($6),$2
+	xorl	8($6),$3
+	xorl	12($6),$4>)dnl
 
-dnl AES_STORE(key, dst)
-dnl Adds the subkey pointed to by %esi to %eax-%edx,
-dnl and stores the result in the area pointed to by %edi.
+dnl AES_STORE(a, b, c, d, key, dst)
+dnl Adds the subkey to a, b, c, d,
+dnl and stores the result in the area pointed to by dst.
 dnl Note that x86 allows unaligned accesses.
 dnl Would it be preferable to interleave the loads and stores?
 define(<AES_STORE>, <
-	xorl	($1),%eax
-	xorl	4($1),%ebx
-	xorl	8($1),%ecx
-	xorl	12($1),%edx
+	xorl	($5),$1
+	xorl	4($5),$2
+	xorl	8($5),$3
+	xorl	12($5),$4
 
-	movl	%eax,($2)
-	movl	%ebx,4($2)
-	movl	%ecx,8($2)
-	movl	%edx,12($2)>)dnl
+	movl	$1,($6)
+	movl	$2,4($6)
+	movl	$3,8($6)
+	movl	$4,12($6)>)dnl
 
-dnl AES_ROUND(table,a,b,c,d)
-dnl Computes one word of the AES round. Leaves result in %edi.
+dnl AES_ROUND(table,a,b,c,d,out,tmp)
+dnl Computes one word of the AES round. Leaves result in $6.
 define(<AES_ROUND>, <
-	movl	%e<>$2<>x, %esi
-	andl	<$>0xff, %esi
-	shll	<$>2,%esi		C  index in table
-	movl	AES_TABLE0 + $1 (%esi),%edi
-	movl	%e<>$3<>x, %esi
-	shrl	<$>6,%esi
-	andl	<$>0x000003fc,%esi C  clear all but offset bytes
-	xorl	AES_TABLE1 + $1 (%esi),%edi
-	movl	%e<>$4<>x,%esi	C  third one
-	shrl	<$>14,%esi
-	andl	<$>0x000003fc,%esi
-	xorl	AES_TABLE2 + $1 (%esi),%edi
-	movl	%e<>$5<>x,%esi	C  fourth one
-	shrl	<$>22,%esi
-	andl	<$>0x000003fc,%esi
-	xorl	AES_TABLE3 + $1 (%esi),%edi>)dnl
+	movl	$2, $7
+	andl	<$>0xff, $7
+	movl	AES_TABLE0 ($1, $7,4),$6
+	movl	$3, $7
+	shrl	<$>6,$7
+	andl	<$>0x000003fc,$7	C  clear all but offset bytes
+	xorl	AES_TABLE1 ($1, $7),$6
+	movl	$4,$7			C  third one
+	shrl	<$>14,$7
+	andl	<$>0x000003fc,$7
+	xorl	AES_TABLE2 ($1, $7),$6
+	movl	$5,$7			C  fourth one
+	shrl	<$>22,$7
+	andl	<$>0x000003fc,$7
+	xorl	AES_TABLE3 ($1, $7),$6>)dnl
 
-dnl AES_FINAL_ROUND(a, b, c, d)
+dnl AES_FINAL_ROUND(a, b, c, d, out, tmp)
 dnl Computes one word of the final round. Leaves result in %edi.
 dnl Note that we have to quote $ in constants.
 define(<AES_FINAL_ROUND>, <
 	C FIXME: Perform substitution on least significant byte here,
 	C to save work later.
-	movl	%e<>$1<>x,%edi
-	andl	<$>0x000000ff,%edi
-	movl	%e<>$2<>x,%ebp
-	andl	<$>0x0000ff00,%ebp
-	orl	%ebp,%edi
-	movl	%e<>$3<>x,%ebp
-	andl	<$>0x00ff0000,%ebp
-	orl	%ebp,%edi
-	movl	%e<>$4<>x,%ebp
-	andl	<$>0xff000000,%ebp
-	orl	%ebp,%edi>)dnl
+	movl	$1,$5
+	andl	<$>0x000000ff,$5
+	movl	$2,$6
+	andl	<$>0x0000ff00,$6
+	orl	$6, $5
+	movl	$3,$6
+	andl	<$>0x00ff0000,$6
+	orl	$6, $5
+	movl	$4,$6
+	andl	<$>0xff000000,$6
+	orl	$6, $5>)dnl
 
-dnl AES_SUBST_BYTE(table)
+dnl AES_SUBST_BYTE(table, tmp)
 dnl Substitutes the least significant byte of
 dnl each of eax, ebx, ecx and edx, and also rotates
 dnl the words one byte to the left.
+dnl FIXME: AES_SBOX is zero. Any win by deleting the offset?
 define(<AES_SUBST_BYTE>, <
-	movl	%eax,%ebp
-	andl	<$>0x000000ff,%ebp
-	movb	AES_SBOX + $1 (%ebp),%al
+	movl	%eax,$2
+	andl	<$>0x000000ff,$2
+	movb	AES_SBOX ($1, $2),%al
 	roll	<$>8,%eax
 
-	movl	%ebx,%ebp
-	andl	<$>0x000000ff,%ebp
-	movb	AES_SBOX + $1 (%ebp),%bl
+	movl	%ebx,$2
+	andl	<$>0x000000ff,$2
+	movb	AES_SBOX ($1, $2),%bl
 	roll	<$>8,%ebx
 
-	movl	%ecx,%ebp
-	andl	<$>0x000000ff,%ebp
-	movb	AES_SBOX + $1 (%ebp),%cl
+	movl	%ecx,$2
+	andl	<$>0x000000ff,$2
+	movb	AES_SBOX ($1, $2),%cl
 	roll	<$>8,%ecx
 
-	movl	%edx,%ebp
-	andl	<$>0x000000ff,%ebp
-	movb	AES_SBOX + $1 (%ebp),%dl
+	movl	%edx,$2
+	andl	<$>0x000000ff,$2
+	movb	AES_SBOX ($1, $2),%dl
 	roll	<$>8,%edx>)dnl
 
 C OFFSET(i)