1 files changed, 78 insertions, 0 deletions
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