Further micro optimizations.

Rev: nettle/x86/aes.m4:1.4

x86/aes.m4

-dnl BYTEREG(reg) gives the 8-bit register corresponding to the given 32-bit register.
+dnl LREG(reg) gives the 8-bit register corresponding to the given 32-bit register.
 dnl Use in AES_SUBST_BYTE below, and is used by both the x86 and the x86_64 assembler.
-define(<BYTEREG>,<ifelse(
+define(<LREG>,<ifelse(
 	$1, %eax, %al,
 	$1, %ebx, %bl,
 	$1, %ecx, %cl,
-	$1, %edx, %dl,
-	dnl The rest are x86_64 only	
-	$1, %esi, %sil,
-	$1, %edi, %dil,
-	$1, %ebp, %bpl,
-	$1, %esp, %spl,
-	$1, %r8d, %r8b,
-	$1, %r9d, %r9b,
-	$1, %r10d, %r10b,
-	$1, %r11d, %r11b,
-	$1, %r12d, %r12b,
-	$1, %r13d, %r13b,
-	$1, %r14d, %r14b,
-	$1, %r15d, %r15b)>)dnl
+	$1, %edx, %dl)>)dnl
+
+define(<HREG>,<ifelse(
+	$1, %eax, %ah,
+	$1, %ebx, %bh,
+	$1, %ecx, %ch,
+	$1, %edx, %dh)>)dnl
 
 dnl AES_LOAD(a, b, c, d, src, key)
 dnl Loads the next block of data from src, and add the subkey pointed
@@ -51,20 +44,18 @@ define(<AES_STORE>, <
 	movl	$3,8($6)
 	movl	$4,12($6)>)dnl
 
-dnl AES_ROUND(table,a,b,c,d,out,tmp)
+dnl AES_ROUND(table,a,b,c,d,out,ptr)
 dnl Computes one word of the AES round. Leaves result in $6.
 define(<AES_ROUND>, <
-	movzbl	BYTEREG($2), $7
-	movl	AES_TABLE0 ($1, $7,4),$6
-	movl	$3, $7
-	shrl	<$>8,$7
-	andl	<$>0xff,$7
+	movzbl	LREG($2), $7
+	movl	AES_TABLE0 ($1, $7, 4),$6
+	movzbl	HREG($3), $7
 	xorl	AES_TABLE1 ($1, $7, 4),$6
-	movl	$4,$7			C  third one
+	movl	$4,$7
 	shrl	<$>16,$7
 	andl	<$>0xff,$7
 	xorl	AES_TABLE2 ($1, $7, 4),$6
-	movl	$5,$7			C  fourth one
+	movl	$5,$7
 	shrl	<$>24,$7
 	xorl	AES_TABLE3 ($1, $7, 4),$6>)dnl
 
@@ -74,8 +65,7 @@ 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	$1,$5
-	andl	<$>0x000000ff,$5
+	movzbl	LREG($1),$5
 	movl	$2,$6
 	andl	<$>0x0000ff00,$6
 	orl	$6, $5
@@ -86,24 +76,24 @@ define(<AES_FINAL_ROUND>, <
 	andl	<$>0xff000000,$6
 	orl	$6, $5>)dnl
 
-dnl AES_SUBST_BYTE(A, B, C, D, table, tmp)
+dnl AES_SUBST_BYTE(A, B, C, D, table, ptr)
 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 Uses that AES_SBOX == 0
 define(<AES_SUBST_BYTE>, <
-	movzbl  BYTEREG($1),$6
-	movb	($5, $6),BYTEREG($1)
+	movzbl  LREG($1),$6
+	movb	($5, $6),LREG($1)
 	roll	<$>8,$1
 
-	movzbl  BYTEREG($2),$6
-	movb	($5, $6),BYTEREG($2)
+	movzbl  LREG($2),$6
+	movb	($5, $6),LREG($2)
 	roll	<$>8,$2
 
-	movzbl  BYTEREG($3),$6
-	movb	($5, $6),BYTEREG($3)
+	movzbl  LREG($3),$6
+	movb	($5, $6),LREG($3)
 	roll	<$>8,$3
 
-	movzbl  BYTEREG($4),$6
-	movb	($5, $6),BYTEREG($4)
+	movzbl  LREG($4),$6
+	movb	($5, $6),LREG($4)
 	roll	<$>8,$4>)dnl