$NetBSD: patch-ae,v 1.1 2002/09/29 15:03:33 wiz Exp $

--- src/runtime/mach-dep/SPARC.prim.asm.orig	Fri Jun 15 15:05:19 2001
+++ src/runtime/mach-dep/SPARC.prim.asm
@@ -307,7 +307,7 @@ pending_sigs:	/* there are pending signa
 	mov	ALLOCPTR,LIMITPTR	    /* (delay slot) */
 
 
-#if defined(OPSYS_SUNOS) || defined(OPSYS_NEXTSTEP)
+#if defined(OPSYS_SUNOS) || defined(OPSYS_NEXTSTEP) || defined(OPSYS_NETBSD)
 /* ZeroLimitPtr:
  *
  * Zero the heap limit pointer so that a trap will be generated on the next limit
@@ -622,13 +622,62 @@ _ml_mul:
 	mov	%g1,%l1			  /* save %g1 which may get trashed */
 	mov	%g2,%l2
 	mov	%g3,%l3
+#ifdef OPSYS_NETBSD
+			/* NetBSDs .mul trashes these */
+	mov     %i4,%l4
+	mov     %i5,%l5
+	mov	%i0,%l6
+	mov	%i1,%l7
+#endif
 	mov	%i2,%o0
 	call	.mul
 	mov	%i3,%o1			  /* (delay slot) */
 	mov	%l1,%g1			  /* restore %g1 */
 	mov	%l2,%g2
 	mov	%l3,%g3
+#ifdef OPSYS_NETBSD
+	mov     %l4,%i4
+	mov	%l5,%i5
+	mov	%l6,%i0
+	mov	%l7,%i1
+			/* the code below is a hack:
+			 * NetBSDs .mul does a 32x32->64 bit signed multiply
+			 * The SML code assumes that the multiply
+			 * overflowed (over 32bit) if the Z flag is
+			 * clear on return from .mul 
+			 * NetBSDs .mul doesn't do that, so we have to check
+			 * for overflow. This is done in the following way:
+			 * - If the upper 32 bits are clear:
+	 		 *    + if bit 31 is set => overflow
+		         *    + if bit 31 is clear => OK
+			 * - If the upper 32 bits are != -1 => overflow
+			 * - If the upper 32 bits are == -1:
+			 *    + If bit 31 is clear => overflow
+			 *    + If bit 31 is set   => OK
+			 *
+		   	 * I should be shot for this code ...
+			 */
+	cmp %o1,0
+	bnz 4f
+	nop
+	/* is bit 31 of %o0 set ? */
+	addcc %o0,%o0,%o1
+	bcc 2f
+	nop
+5:      restore %o0,0,%o2                 /* result in %o2 (delay slot) */
+        t       ST_INT_OVERFLOW         /* generate overflow trap */
+
+	
+4:	cmp %o1,-1			/* upper 32 bits == -1 ? */
+	bnz 5
+	nop
+	addcc %o0,%o0,%o1		/* bit 31 clear ? */
+	bcc 2f
+	nop
+#else
 	bnz	1f			  /* if z is clear, then overflow */
+#endif
+2:
 	restore %o0,0,%o2		  /* result in %o2 (delay slot) */
 	retl
 	nop
@@ -642,7 +691,15 @@ _ml_mul:
  * locals of the new window, since .div is a leaf routine.
  */
 _ml_div:
-	save	%sp,-SA(WINDOWSIZE),%sp
+#ifdef OPSYS_NETBSD
+		/* hack time again: NetBSDs .div trashes too many registers
+		 * we have to store them in the stack frame, so make room
+		 * for six registers (o0-o5) 
+		 */
+	save	%sp,-SA(WINDOWSIZE+24),%sp
+#else
+        save    %sp,-SA(WINDOWSIZE+24),%sp
+#endif
 	addcc	%i3,%g0,%o1		/* %o1 is divisor (and check for zero) */
 	bz	1f
 				    /* save %g1, %g2 and %g3 (using new window) */
@@ -650,9 +707,30 @@ _ml_div:
 	mov	%g1,%l1			/* (delay slot) */
 	mov	%g2,%l2
 	mov	%g3,%l3
+#ifdef OPSYS_NETBSD
+	/* save g6,g7,o0-o5, they get trashed. 
+	   Note that %o0-%o5 are now %i0-%i5, since we did a 'save'
+	   Since %g6 is the global AllocPtr for SML, this 
+	   is rather bad :-)
+	 */  
+	mov %g6,%l4
+	mov %g7,%l5
+	std	%i0,[%fp-16]
+	std	%i2,[%fp-24]
+	std	%i4,[%fp-32]
+#endif
+
 	call	.div
 	mov	%i2,%o0			/* (delay slot) */
 				    /* restore %g1, %g2 and %g3 */
+#ifdef OPSYS_NETBSD
+	mov %l4,%g6
+	mov %l5,%g7
+	ldd	[%fp-32],%i4
+	ldd	[%fp-24],%i2
+	ldd	[%fp-16],%i0
+#endif
+
 	mov	%l3,%g3
 	mov	%l2,%g2
 	mov	%l1,%g1
@@ -679,11 +757,24 @@ _ml_umul:
 	mov	%g2,%l2
 	mov	%g3,%l3
 	mov	%i2,%o0
+#ifdef OPSYS_NETBSD
+		/* Save what might be trashed by NetBSDs .umul */
+	mov	%i4,%l4
+	mov	%i5,%l5
+	mov	%i0,%l6
+	mov 	%i1,%l7
+#endif
 	call	.umul
 	mov	%i3,%o1			  /* (delay slot) */
 	mov	%l1,%g1			  /* restore %g1 */
 	mov	%l2,%g2
 	mov	%l3,%g3
+#ifdef OPSYS_NETBSD
+	mov	%l4,%i4
+	mov	%l5,%i5
+	mov	%l6,%i0
+	mov	%l7,%i1
+#endif
 	ret
 	restore %o0,0,%o2		  /* result in %o2 (delay slot) */
 
@@ -694,7 +785,12 @@ _ml_umul:
  * locals of the new window, since .div is a leaf routine.
  */
 _ml_udiv:
-	save	%sp,-SA(WINDOWSIZE),%sp
+#ifdef OPSYS_NETBSD
+		/* see comment for _ml__div_ */
+	save	%sp,-SA(WINDOWSIZE+24),%sp
+#else
+        save    %sp,-SA(WINDOWSIZE+24),%sp 
+#endif
 	addcc	%i3,%g0,%o1		/* %o1 is divisor (and check for zero) */
 	bz	1f
 				    /* save %g1, %g2 and %g3 (using new window) */
@@ -702,9 +798,23 @@ _ml_udiv:
 	mov	%g1,%l1			/* (delay slot) */
 	mov	%g2,%l2
 	mov	%g3,%l3
+#ifdef OPSYS_NETBSD
+	mov	%g6,%l4
+	mov 	%g7,%l5
+        std     %i0,[%fp-16]
+        std     %i2,[%fp-24]
+        std     %i4,[%fp-32]
+#endif
 	call	.udiv
 	mov	%i2,%o0			/* (delay slot) */
 				    /* restore %g1, %g2 and %g3 */
+#ifdef OPSYS_NETBSD
+	mov 	%l4,%g6
+	mov 	%l5,%g7
+        ldd     [%fp-32],%i4
+        ldd     [%fp-24],%i2
+        ldd     [%fp-16],%i0
+#endif
 	mov	%l3,%g3
 	mov	%l2,%g2
 	mov	%l1,%g1
