Add basic libmp support for the x86.

9 files changed, 349 insertions, 15 deletions

diff --git a/src/libmp/386/mkfile b/src/libmp/386/mkfile
new file mode 100644
index 00000000..06a0a988
--- /dev/null
+++ b/src/libmp/386/mkfile
@@ -0,0 +1,19 @@
+<$PLAN9/src/mkhdr
+
+LIB=libmp.a
+SFILES=\
+	mpdigdiv.s\
+	mpvecadd.s\
+	mpvecdigmuladd.s\
+	mpvecdigmulsub.s\
+	mpvecsub.s\
+
+HFILES=$PLAN9/include/mp.h ../port/dat.h
+
+OFILES=${SFILES:%.s=%.$O}
+
+UPDATE=mkfile\
+	$HFILES\
+	$SFILES\
+
+<$PLAN9/src/mksyslib
diff --git a/src/libmp/386/mpdigdiv.s b/src/libmp/386/mpdigdiv.s
new file mode 100644
index 00000000..f02e1616
--- /dev/null
+++ b/src/libmp/386/mpdigdiv.s
@@ -0,0 +1,39 @@
+.text
+
+.p2align 2,0x90
+.globl mpdigdiv
+	.type mpdigdiv, @function
+mpdigdiv:
+	/* Prelude */
+	pushl %ebp
+	movl %ebx, -4(%esp)		/* save on stack */
+
+	movl	8(%esp), %ebx
+	movl	(%ebx), %eax
+	movl	4(%ebx), %edx
+
+	movl	12(%esp), %ebx
+	movl	16(%esp), %ebp
+	xorl	%ecx, %ecx
+	cmpl	%ebx, %edx		/* dividend >= 2^32 * divisor */
+	jae	divovfl
+	cmpl	%ecx, %ebx		/* divisor == 1 */
+	je	divovfl
+	divl	%ebx		/* AX = DX:AX/BX */
+	movl	%eax, (%ebp)
+	jmp done
+
+	/* return all 1's */
+divovfl:
+	notl	%ecx
+	movl	%ecx, (%ebp)
+
+done:
+	/* Postlude */
+	movl -4(%esp), %ebx		/* restore from stack */
+	movl %esp, %ebp
+	leave
+	ret
+
+.endmpdigdiv:
+	.size mpdigdiv,.endmpdigdiv-mpdigdiv
diff --git a/src/libmp/386/mpvecadd.s b/src/libmp/386/mpvecadd.s
new file mode 100644
index 00000000..1f23dcd7
--- /dev/null
+++ b/src/libmp/386/mpvecadd.s
@@ -0,0 +1,73 @@
+/*
+ *	mpvecadd(mpdigit *a, int alen, mpdigit *b, int blen, mpdigit *sum)
+ *
+ *		sum[0:alen] = a[0:alen-1] + b[0:blen-1]
+ *
+ *	prereq: alen >= blen, sum has room for alen+1 digits
+ */
+
+.text
+
+.p2align 2,0x90
+.globl mpvecadd
+	.type mpvecadd, @function
+mpvecadd:
+	/* Prelude */
+	pushl %ebp
+	movl %ebx, -4(%esp)		/* save on stack */
+	movl %esi, -8(%esp)
+	movl %edi, -12(%esp)
+
+	movl	12(%esp), %edx		/* alen */
+	movl	20(%esp), %ecx		/* blen */
+	movl	8(%esp), %esi		/* a */
+	movl	16(%esp), %ebx		/* b */
+	subl	%ecx, %edx
+	movl	24(%esp), %edi		/* sum */
+	xorl	%ebp, %ebp		/* this also sets carry to 0 */
+
+	/* skip addition if b is zero */
+	testl	%ecx,%ecx
+	je	_add1
+
+	/* sum[0:blen-1],carry = a[0:blen-1] + b[0:blen-1] */
+_addloop1:
+	movl	(%esi, %ebp, 4), %eax
+	adcl	(%ebx, %ebp, 4), %eax
+	movl	%eax, (%edi, %ebp, 4)
+	incl	%ebp
+	loop	_addloop1
+
+_add1:
+	/* jump if alen > blen */
+	incl	%edx
+	movl	%edx, %ecx
+	loop	_addloop2
+
+	/* sum[alen] = carry */
+_addend:
+	jb	_addcarry
+	movl	$0, (%edi, %ebp, 4)
+	jmp done
+
+_addcarry:
+	movl	$1, (%edi, %ebp, 4)
+	jmp done
+
+	/* sum[blen:alen-1],carry = a[blen:alen-1] + 0 */
+_addloop2:
+	movl	(%esi, %ebp, 4),%eax
+	adcl	$0, %eax
+	movl	%eax, (%edi, %ebp, 4)
+	incl	%ebp
+	loop	_addloop2
+	jmp	_addend
+
+done:
+	/* Postlude */
+	movl -4(%esp), %ebx		/* restore from stack */
+	movl -8(%esp), %esi
+	movl -12(%esp), %edi
+	movl %esp, %ebp
+	leave
+	ret
diff --git a/src/libmp/386/mpvecdigmuladd.s b/src/libmp/386/mpvecdigmuladd.s
new file mode 100644
index 00000000..987c6330
--- /dev/null
+++ b/src/libmp/386/mpvecdigmuladd.s
@@ -0,0 +1,69 @@
+/*
+ *	mpvecdigmul(mpdigit *b, int n, mpdigit m, mpdigit *p)
+ *
+ *	p += b*m
+ *
+ *	each step look like:
+ *		hi,lo = m*b[i]
+ *		lo += oldhi + carry
+ *		hi += carry
+ *		p[i] += lo
+ *		oldhi = hi
+ *
+ *	the registers are:
+ *		hi = DX		- constrained by hardware
+ *		lo = AX		- constrained by hardware
+ *		b+n = SI	- can't be BP
+ *		p+n = DI	- can't be BP
+ *		i-n = BP
+ *		m = BX
+ *		oldhi = CX
+ *		
+ */
+.text
+
+.p2align 2,0x90
+.globl mpvecdigmuladd
+	.type mpvecdigmuladd, @function
+mpvecdigmuladd:
+	/* Prelude */
+	pushl %ebp
+	movl %ebx, -4(%esp)		/* save on stack */
+	movl %esi, -8(%esp)
+	movl %edi, -12(%esp)
+
+	movl	8(%esp), %esi		/* b */
+	movl	12(%esp), %ecx		/* n */
+	movl	16(%esp), %ebx		/* m */
+	movl	20(%esp), %edi		/* p */
+	movl	%ecx, %ebp
+	negl	%ebp			/* BP = -n */
+	shll	$2, %ecx
+	addl	%ecx, %esi		/* SI = b + n */
+	addl	%ecx, %edi		/* DI = p + n */
+	xorl	%ecx, %ecx
+_muladdloop:
+	movl	(%esi, %ebp, 4), %eax	/* lo = b[i] */
+	mull	%ebx			/* hi, lo = b[i] * m */
+	addl	%ecx,%eax		/* lo += oldhi */
+	jae	_muladdnocarry1
+	incl	%edx			/* hi += carry */
+_muladdnocarry1:
+	addl	%eax, (%edi, %ebp, 4)	/* p[i] += lo */
+	jae	_muladdnocarry2
+	incl	%edx			/* hi += carry */
+_muladdnocarry2:
+	movl	%edx, %ecx		/* oldhi = hi */
+	incl	%ebp			/* i++ */
+	jnz	_muladdloop
+	xorl	%eax, %eax
+	addl	%ecx, (%edi, %ebp, 4)	/* p[n] + oldhi */
+	adcl	%eax, %eax		/* return carry out of p[n] */
+
+	/* Postlude */
+	movl -4(%esp), %ebx		/* restore from stack */
+	movl -8(%esp), %esi
+	movl -12(%esp), %edi
+	movl %esp, %ebp
+	leave
+	ret
diff --git a/src/libmp/386/mpvecdigmulsub.s b/src/libmp/386/mpvecdigmulsub.s
new file mode 100644
index 00000000..47089c87
--- /dev/null
+++ b/src/libmp/386/mpvecdigmulsub.s
@@ -0,0 +1,78 @@
+/*
+ *	mpvecdigmulsub(mpdigit *b, int n, mpdigit m, mpdigit *p)
+ *
+ *	p -= b*m
+ *
+ *	each step look like:
+ *		hi,lo = m*b[i]
+ *		lo += oldhi + carry
+ *		hi += carry
+ *		p[i] += lo
+ *		oldhi = hi
+ *
+ *	the registers are:
+ *		hi = DX		- constrained by hardware
+ *		lo = AX		- constrained by hardware
+ *		b = SI		- can't be BP
+ *		p = DI		- can't be BP
+ *		i = BP
+ *		n = CX		- constrained by LOOP instr
+ *		m = BX
+ *		oldhi = EX
+ *		
+ */
+.text
+
+/* XXX: had to use "-4(%esp)" kludge to get around inability to
+ *      push/pop without first adjusting %esp.  This may not be
+ *      as fast as using push/pop (and accessing pushed element
+ *      with "(%esp)".)
+ */
+
+.p2align 2,0x90
+.globl mpvecdigmulsub
+	.type mpvecdigmulsub, @function
+mpvecdigmulsub:
+	/* Prelude */
+	pushl %ebp
+	movl %ebx, -8(%esp)		/* save on stack */
+	movl %esi, -12(%esp)
+	movl %edi, -16(%esp)
+
+	movl	8(%esp), %esi		/* b */
+	movl	12(%esp), %ecx		/* n */
+	movl	16(%esp), %ebx		/* m */
+	movl	20(%esp), %edi		/* p */
+	xorl	%ebp, %ebp
+	movl	%ebp, -4(%esp)
+_mulsubloop:
+	movl	(%esi, %ebp, 4),%eax	/* lo = b[i] */
+	mull	%ebx			/* hi, lo = b[i] * m */
+	addl	-4(%esp), %eax		/* lo += oldhi */
+	jae	_mulsubnocarry1
+	incl	%edx			/* hi += carry */
+_mulsubnocarry1:
+	subl	%eax, (%edi, %ebp, 4)
+	jae	_mulsubnocarry2
+	incl	%edx			/* hi += carry */
+_mulsubnocarry2:
+	movl	%edx, -4(%esp)
+	incl	%ebp
+	loop	_mulsubloop
+	movl	-4(%esp), %eax
+	subl	%eax, (%edi, %ebp, 4)
+	jae	_mulsubnocarry3
+	movl	$-1, %eax
+	jmp done
+
+_mulsubnocarry3:
+	movl	$1, %eax
+
+done:
+	/* Postlude */
+	movl -8(%esp), %ebx		/* restore from stack */
+	movl -12(%esp), %esi
+	movl -16(%esp), %edi
+	movl %esp, %ebp
+	leave
+	ret
diff --git a/src/libmp/386/mpvecsub.s b/src/libmp/386/mpvecsub.s
new file mode 100644
index 00000000..a56b4968
--- /dev/null
+++ b/src/libmp/386/mpvecsub.s
@@ -0,0 +1,62 @@
+/*
+ *	mpvecsub(mpdigit *a, int alen, mpdigit *b, int blen, mpdigit *diff)
+ *
+ *		diff[0:alen-1] = a[0:alen-1] - b[0:blen-1]
+ *
+ *	prereq: alen >= blen, diff has room for alen digits
+ */
+.text
+
+.p2align 2,0x90
+.globl mpvecsub
+	.type mpvecsub, @function
+mpvecsub:
+	/* Prelude */
+	pushl %ebp
+	movl %ebx, -4(%esp)		/* save on stack */
+	movl %esi, -8(%esp)
+	movl %edi, -12(%esp)
+
+	movl	8(%esp), %esi		/* a */
+	movl	16(%esp), %ebx		/* b */
+	movl	12(%esp), %edx		/* alen */
+	movl	20(%esp), %ecx		/* blen */
+	movl	24(%esp), %edi		/* diff */
+	subl	%ecx,%edx
+	xorl	%ebp,%ebp			/* this also sets carry to 0 */
+
+	/* skip subraction if b is zero */
+	testl	%ecx,%ecx
+	jz	_sub1
+
+	/* diff[0:blen-1],borrow = a[0:blen-1] - b[0:blen-1] */
+_subloop1:
+	movl	(%esi, %ebp, 4), %eax
+	sbbl	(%ebx, %ebp, 4), %eax
+	movl	%eax, (%edi, %ebp, 4)
+	incl	%ebp
+	loop	_subloop1
+
+_sub1:
+	incl	%edx
+	movl	%edx,%ecx
+	loop	_subloop2
+	jmp done
+
+	/* diff[blen:alen-1] = a[blen:alen-1] - 0 */
+_subloop2:
+	movl	(%esi, %ebp, 4), %eax
+	sbbl	$0, %eax
+	movl	%eax, (%edi, %ebp, 4)
+	INCL	%ebp
+	LOOP	_subloop2
+
+done:
+	/* Postlude */
+	movl -4(%esp), %ebx		/* restore from stack */
+	movl -8(%esp), %esi
+	movl -12(%esp), %edi
+	movl %esp, %ebp
+	leave
+	ret
+
diff --git a/src/libmp/mkfile b/src/libmp/mkfile
index aab57462..417804df 100644
--- a/src/libmp/mkfile
+++ b/src/libmp/mkfile
@@ -2,6 +2,6 @@
 
 DIRS=\
 	port\
-#	$systype-$objtype\
+	$objtype\
 
 <$PLAN9/src/mkdirs
diff --git a/src/libmp/port/mkfile b/src/libmp/port/mkfile
index a9176dd3..682c6763 100644
--- a/src/libmp/port/mkfile
+++ b/src/libmp/port/mkfile
@@ -33,10 +33,9 @@ FILES=\
 	mptov\
 	mptouv\
 
-OFILES=${FILES:%=%.$O}
-
+ALLOFILES=${FILES:%=%.$O}
 # cull things in the per-machine directories from this list
-# OFILES=	`{sh ./reduce $O $objtype $ALLOFILES}
+OFILES=	`{sh ./reduce $O $objtype $ALLOFILES}
 
 HFILES=\
 	$PLAN9/include/lib9.h\
@@ -44,3 +43,6 @@ HFILES=\
 	dat.h\
 
 <$PLAN9/src/mksyslib
+
+poot:
+	echo $OFILES
diff --git a/src/libmp/port/reduce b/src/libmp/port/reduce
index a857a28c..b3980756 100644
--- a/src/libmp/port/reduce
+++ b/src/libmp/port/reduce
@@ -3,14 +3,6 @@ shift
 objtype=$1
 shift
 
-ls -p ../$objtype/*.[cs] >[2]/dev/null | sed 's/..$//' > /tmp/reduce.$pid
-#
-#	if empty directory, just return the input files
-#
-if (! ~ $status '|') {
-	echo $*
-	rm /tmp/reduce.$pid
-	exit 0
-}
-echo $* | tr ' ' \012 | grep -v -f /tmp/reduce.$pid | tr \012 ' '
-rm /tmp/reduce.$pid
+reduce="("`ls -p ../$objtype/*.[cs] 2>/dev/null | sed 's/..$//' | sed 's/^.*\///' | tr '\012' '|' | sed 's/.$//'`")"
+
+echo $* | tr ' ' '\012' | egrep -v $reduce | tr \012 ' '