Reorganisation of the source tree

Most of the other users of the fptools build system have migrated to
Cabal, and with the move to darcs we can now flatten the source tree
without losing history, so here goes.

The main change is that the ghc/ subdir is gone, and most of what it
contained is now at the top level.  The build system now makes no
pretense at being multi-project, it is just the GHC build system.

No doubt this will break many things, and there will be a period of
instability while we fix the dependencies.  A straightforward build
should work, but I haven't yet fixed binary/source distributions.
Changes to the Building Guide will follow, too.

20 files changed, 1993 insertions, 0 deletions

diff --git a/rts/gmp/mpn/hppa/README b/rts/gmp/mpn/hppa/README
new file mode 100644
index 0000000000..97e7abe011
--- /dev/null
+++ b/rts/gmp/mpn/hppa/README
@@ -0,0 +1,91 @@
+This directory contains mpn functions for various HP PA-RISC chips.  Code
+that runs faster on the PA7100 and later implementations, is in the pa7100
+directory.
+
+RELEVANT OPTIMIZATION ISSUES
+
+  Load and Store timing
+
+On the PA7000 no memory instructions can issue the two cycles after a store.
+For the PA7100, this is reduced to one cycle.
+
+The PA7100 has a lookup-free cache, so it helps to schedule loads and the
+dependent instruction really far from each other.
+
+STATUS
+
+1. mpn_mul_1 could be improved to 6.5 cycles/limb on the PA7100, using the
+   instructions below (but some sw pipelining is needed to avoid the
+   xmpyu-fstds delay):
+
+	fldds	s1_ptr
+
+	xmpyu
+	fstds	N(%r30)
+	xmpyu
+	fstds	N(%r30)
+
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+
+	addc
+	stws	res_ptr
+	addc
+	stws	res_ptr
+
+	addib	Loop
+
+2. mpn_addmul_1 could be improved from the current 10 to 7.5 cycles/limb
+   (asymptotically) on the PA7100, using the instructions below.  With proper
+   sw pipelining and the unrolling level below, the speed becomes 8
+   cycles/limb.
+
+	fldds	s1_ptr
+	fldds	s1_ptr
+
+	xmpyu
+	fstds	N(%r30)
+	xmpyu
+	fstds	N(%r30)
+	xmpyu
+	fstds	N(%r30)
+	xmpyu
+	fstds	N(%r30)
+
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	ldws	N(%r30)
+	addc
+	addc
+	addc
+	addc
+	addc	%r0,%r0,cy-limb
+
+	ldws	res_ptr
+	ldws	res_ptr
+	ldws	res_ptr
+	ldws	res_ptr
+	add
+	stws	res_ptr
+	addc
+	stws	res_ptr
+	addc
+	stws	res_ptr
+	addc
+	stws	res_ptr
+
+	addib
+
+3. For the PA8000 we have to stick to using 32-bit limbs before compiler
+   support emerges.  But we want to use 64-bit operations whenever possible,
+   in particular for loads and stores.  It is possible to handle mpn_add_n
+   efficiently by rotating (when s1/s2 are aligned), masking+bit field
+   inserting when (they are not).  The speed should double compared to the
+   code used today.
diff --git a/rts/gmp/mpn/hppa/add_n.s b/rts/gmp/mpn/hppa/add_n.s
new file mode 100644
index 0000000000..c53b2f71b3
--- /dev/null
+++ b/rts/gmp/mpn/hppa/add_n.s
@@ -0,0 +1,58 @@
+; HP-PA  __gmpn_add_n -- Add two limb vectors of the same length > 0 and store
+; sum in a third limb vector.
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+; One might want to unroll this as for other processors, but it turns
+; out that the data cache contention after a store makes such
+; unrolling useless.  We can't come under 5 cycles/limb anyway.
+
+	.code
+	.export		__gmpn_add_n
+__gmpn_add_n
+	.proc
+	.callinfo	frame=0,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+
+	addib,=		-1,%r23,L$end	; check for (SIZE == 1)
+	 add		%r20,%r19,%r28	; add first limbs ignoring cy
+
+L$loop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,<>	-1,%r23,L$loop
+	 addc		%r20,%r19,%r28
+
+L$end	stws		%r28,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r0,%r0,%r28
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/gmp-mparam.h b/rts/gmp/mpn/hppa/gmp-mparam.h
new file mode 100644
index 0000000000..98b6d9ce3c
--- /dev/null
+++ b/rts/gmp/mpn/hppa/gmp-mparam.h
@@ -0,0 +1,63 @@
+/* gmp-mparam.h -- Compiler/machine parameter header file.
+
+Copyright (C) 1991, 1993, 1994, 1999, 2000 Free Software Foundation, Inc.
+
+This file is part of the GNU MP Library.
+
+The GNU MP Library is free software; you can redistribute it and/or modify
+it under the terms of the GNU Lesser General Public License as published by
+the Free Software Foundation; either version 2.1 of the License, or (at your
+option) any later version.
+
+The GNU MP Library is distributed in the hope that it will be useful, but
+WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+License for more details.
+
+You should have received a copy of the GNU Lesser General Public License
+along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+MA 02111-1307, USA. */
+
+#define BITS_PER_MP_LIMB 32
+#define BYTES_PER_MP_LIMB 4
+#define BITS_PER_LONGINT 32
+#define BITS_PER_INT 32
+#define BITS_PER_SHORTINT 16
+#define BITS_PER_CHAR 8
+
+/* These values are for the PA7100 using GCC.  */
+/* Generated by tuneup.c, 2000-07-25. */
+
+#ifndef KARATSUBA_MUL_THRESHOLD
+#define KARATSUBA_MUL_THRESHOLD   30
+#endif
+#ifndef TOOM3_MUL_THRESHOLD
+#define TOOM3_MUL_THRESHOLD      172
+#endif
+
+#ifndef KARATSUBA_SQR_THRESHOLD
+#define KARATSUBA_SQR_THRESHOLD   59
+#endif
+#ifndef TOOM3_SQR_THRESHOLD
+#define TOOM3_SQR_THRESHOLD      185
+#endif
+
+#ifndef BZ_THRESHOLD
+#define BZ_THRESHOLD              96
+#endif
+
+#ifndef FIB_THRESHOLD
+#define FIB_THRESHOLD            122
+#endif
+
+#ifndef POWM_THRESHOLD
+#define POWM_THRESHOLD            18
+#endif
+
+#ifndef GCD_ACCEL_THRESHOLD
+#define GCD_ACCEL_THRESHOLD       46
+#endif
+#ifndef GCDEXT_THRESHOLD
+#define GCDEXT_THRESHOLD          33
+#endif
diff --git a/rts/gmp/mpn/hppa/hppa1_1/addmul_1.s b/rts/gmp/mpn/hppa/hppa1_1/addmul_1.s
new file mode 100644
index 0000000000..c7d218f922
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/addmul_1.s
@@ -0,0 +1,102 @@
+; HP-PA-1.1 __gmpn_addmul_1 -- Multiply a limb vector with a limb and
+; add the result to a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	r26
+; s1_ptr	r25
+; size		r24
+; s2_limb	r23
+
+; This runs at 11 cycles/limb on a PA7000.  With the used instructions, it
+; can not become faster due to data cache contention after a store.  On the
+; PA7100 it runs at 10 cycles/limb, and that can not be improved either,
+; since only the xmpyu does not need the integer pipeline, so the only
+; dual-issue we will get are addc+xmpyu.  Unrolling could gain a cycle/limb
+; on the PA7100.
+
+; There are some ideas described in mul_1.s that applies to this code too.
+
+	.code
+	.export		__gmpn_addmul_1
+__gmpn_addmul_1
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	ldo		64(%r30),%r30
+	fldws,ma	4(%r25),%fr5
+	stw		%r23,-16(%r30)		; move s2_limb ...
+	addib,=		-1,%r24,L$just_one_limb
+	 fldws		-16(%r30),%fr4		; ... into fr4
+	add		%r0,%r0,%r0		; clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds		%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r19		; least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+; Main loop
+L$loop	ldws		0(%r26),%r29
+	fldws,ma	4(%r25),%fr5
+	add		%r29,%r19,%r19
+	stws,ma		%r19,4(%r26)
+	addc		%r28,%r1,%r19
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addc		%r0,%r28,%r28
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end	ldw		0(%r26),%r29
+	add		%r29,%r19,%r19
+	stws,ma		%r19,4(%r26)
+	addc		%r28,%r1,%r19
+	ldw		-16(%r30),%r28
+	ldws		0(%r26),%r29
+	addc		%r0,%r28,%r28
+	add		%r29,%r19,%r19
+	stws,ma		%r19,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+L$just_one_limb
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		0(%r26),%r29
+	fstds		%fr6,-16(%r30)
+	ldw		-12(%r30),%r1
+	ldw		-16(%r30),%r28
+	add		%r29,%r1,%r19
+	stw		%r19,0(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/mul_1.s b/rts/gmp/mpn/hppa/hppa1_1/mul_1.s
new file mode 100644
index 0000000000..4512fddec9
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/mul_1.s
@@ -0,0 +1,98 @@
+; HP-PA-1.1 __gmpn_mul_1 -- Multiply a limb vector with a limb and store
+; the result in a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	r26
+; s1_ptr	r25
+; size		r24
+; s2_limb	r23
+
+; This runs at 9 cycles/limb on a PA7000.  With the used instructions, it can
+; not become faster due to data cache contention after a store.  On the
+; PA7100 it runs at 7 cycles/limb, and that can not be improved either, since
+; only the xmpyu does not need the integer pipeline, so the only dual-issue
+; we will get are addc+xmpyu.  Unrolling would not help either CPU.
+
+; We could use fldds to read two limbs at a time from the S1 array, and that
+; could bring down the times to 8.5 and 6.5 cycles/limb for the PA7000 and
+; PA7100, respectively.  We don't do that since it does not seem worth the
+; (alignment) troubles...
+
+; At least the PA7100 is rumored to be able to deal with cache-misses
+; without stalling instruction issue.  If this is true, and the cache is
+; actually also lockup-free, we should use a deeper software pipeline, and
+; load from S1 very early!  (The loads and stores to -12(sp) will surely be
+; in the cache.)
+
+	.code
+	.export		__gmpn_mul_1
+__gmpn_mul_1
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	ldo		64(%r30),%r30
+	fldws,ma	4(%r25),%fr5
+	stw		%r23,-16(%r30)		; move s2_limb ...
+	addib,=		-1,%r24,L$just_one_limb
+	 fldws		-16(%r30),%fr4		; ... into fr4
+	add		%r0,%r0,%r0		; clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds	 	%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r19		; least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+; Main loop
+L$loop	fldws,ma	4(%r25),%fr5
+	stws,ma		%r19,4(%r26)
+	addc		%r28,%r1,%r19
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end	stws,ma		%r19,4(%r26)
+	addc		%r28,%r1,%r19
+	ldw		-16(%r30),%r28
+	stws,ma		%r19,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+L$just_one_limb
+	xmpyu		%fr4,%fr5,%fr6
+	fstds		%fr6,-16(%r30)
+	ldw		-16(%r30),%r28
+	ldo		-64(%r30),%r30
+	bv		0(%r2)
+	 fstws		%fr6R,0(%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/add_n.s b/rts/gmp/mpn/hppa/hppa1_1/pa7100/add_n.s
new file mode 100644
index 0000000000..4f4be08b37
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/add_n.s
@@ -0,0 +1,75 @@
+; HP-PA  __gmpn_add_n -- Add two limb vectors of the same length > 0 and store
+; sum in a third limb vector.
+; This is optimized for the PA7100, where is runs at 4.25 cycles/limb
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+	.code
+	.export		__gmpn_add_n
+__gmpn_add_n
+	.proc
+	.callinfo	frame=0,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+
+	addib,<=	-5,%r23,L$rest
+	 add		%r20,%r19,%r28	; add first limbs ignoring cy
+
+L$loop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addc		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addc		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addc		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,>		-4,%r23,L$loop
+	addc		%r20,%r19,%r28
+
+L$rest	addib,=		4,%r23,L$end
+	nop
+L$eloop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,>		-1,%r23,L$eloop
+	addc		%r20,%r19,%r28
+
+L$end	stws		%r28,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r0,%r0,%r28
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/addmul_1.S b/rts/gmp/mpn/hppa/hppa1_1/pa7100/addmul_1.S
new file mode 100644
index 0000000000..04db06822e
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/addmul_1.S
@@ -0,0 +1,189 @@
+; HP-PA 7100/7200 __gmpn_addmul_1 -- Multiply a limb vector with a limb and
+; add the result to a second limb vector.
+
+; Copyright (C) 1995, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+; INPUT PARAMETERS
+#define res_ptr	%r26
+#define s1_ptr	%r25
+#define size	%r24
+#define s2_limb	%r23
+
+#define cylimb	%r28
+#define s0	%r19
+#define s1	%r20
+#define s2	%r3
+#define s3	%r4
+#define lo0	%r21
+#define lo1	%r5
+#define lo2	%r6
+#define lo3	%r7
+#define hi0	%r22
+#define hi1	%r23				/* safe to reuse */
+#define hi2	%r29
+#define hi3	%r1
+
+	.code
+	.export		__gmpn_addmul_1
+__gmpn_addmul_1
+	.proc
+	.callinfo	frame=128,no_calls
+	.entry
+
+	ldo	128(%r30),%r30
+	stws	s2_limb,-16(%r30)
+	add	 %r0,%r0,cylimb			; clear cy and cylimb
+	addib,<	-4,size,L$few_limbs
+	fldws	-16(%r30),%fr31R
+
+	ldo	-112(%r30),%r31
+	stw	%r3,-96(%r30)
+	stw	%r4,-92(%r30)
+	stw	%r5,-88(%r30)
+	stw	%r6,-84(%r30)
+	stw	%r7,-80(%r30)
+
+	bb,>=,n	 s1_ptr,29,L$0
+
+	fldws,ma 4(s1_ptr),%fr4
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4,%fr31R,%fr5
+	fstds	 %fr5,-16(%r31)
+	ldws	-16(%r31),cylimb
+	ldws	-12(%r31),lo0
+	add	 s0,lo0,s0
+	addib,< -1,size,L$few_limbs
+	stws,ma	 s0,4(res_ptr)
+
+; start software pipeline ----------------------------------------------------
+L$0	fldds,ma 8(s1_ptr),%fr4
+	fldds,ma 8(s1_ptr),%fr8
+
+	xmpyu	 %fr4L,%fr31R,%fr5
+	xmpyu	 %fr4R,%fr31R,%fr6
+	xmpyu	 %fr8L,%fr31R,%fr9
+	xmpyu	 %fr8R,%fr31R,%fr10
+
+	fstds	 %fr5,-16(%r31)
+	fstds	 %fr6,-8(%r31)
+	fstds	 %fr9,0(%r31)
+	fstds	 %fr10,8(%r31)
+
+	ldws   -16(%r31),hi0
+	ldws   -12(%r31),lo0
+	ldws	-8(%r31),hi1
+	ldws	-4(%r31),lo1
+	ldws	 0(%r31),hi2
+	ldws	 4(%r31),lo2
+	ldws	 8(%r31),hi3
+	ldws	12(%r31),lo3
+
+	addc	 lo0,cylimb,lo0
+	addc	 lo1,hi0,lo1
+	addc	 lo2,hi1,lo2
+	addc	 lo3,hi2,lo3
+
+	addib,<	 -4,size,L$end
+	addc	 %r0,hi3,cylimb			; propagate carry into cylimb
+; main loop ------------------------------------------------------------------
+L$loop	fldds,ma 8(s1_ptr),%fr4
+	fldds,ma 8(s1_ptr),%fr8
+
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4L,%fr31R,%fr5
+	ldws	 4(res_ptr),s1
+	xmpyu	 %fr4R,%fr31R,%fr6
+	ldws	 8(res_ptr),s2
+	xmpyu	 %fr8L,%fr31R,%fr9
+	ldws	12(res_ptr),s3
+	xmpyu	 %fr8R,%fr31R,%fr10
+
+	fstds	 %fr5,-16(%r31)
+	add	 s0,lo0,s0
+	fstds	 %fr6,-8(%r31)
+	addc	 s1,lo1,s1
+	fstds	 %fr9,0(%r31)
+	addc	 s2,lo2,s2
+	fstds	 %fr10,8(%r31)
+	addc	 s3,lo3,s3
+
+	ldws   -16(%r31),hi0
+	ldws   -12(%r31),lo0
+	ldws	-8(%r31),hi1
+	ldws	-4(%r31),lo1
+	ldws	 0(%r31),hi2
+	ldws	 4(%r31),lo2
+	ldws	 8(%r31),hi3
+	ldws	12(%r31),lo3
+
+	addc	 lo0,cylimb,lo0
+	stws,ma	 s0,4(res_ptr)
+	addc	 lo1,hi0,lo1
+	stws,ma	 s1,4(res_ptr)
+	addc	 lo2,hi1,lo2
+	stws,ma	 s2,4(res_ptr)
+	addc	 lo3,hi2,lo3
+	stws,ma	 s3,4(res_ptr)
+
+	addib,>= -4,size,L$loop
+	addc	 %r0,hi3,cylimb			; propagate carry into cylimb
+; finish software pipeline ---------------------------------------------------
+L$end	ldws	 0(res_ptr),s0
+	ldws	 4(res_ptr),s1
+	ldws	 8(res_ptr),s2
+	ldws	12(res_ptr),s3
+
+	add	 s0,lo0,s0
+	stws,ma	 s0,4(res_ptr)
+	addc	 s1,lo1,s1
+	stws,ma	 s1,4(res_ptr)
+	addc	 s2,lo2,s2
+	stws,ma	 s2,4(res_ptr)
+	addc	 s3,lo3,s3
+	stws,ma	 s3,4(res_ptr)
+
+; restore callee-saves registers ---------------------------------------------
+	ldw	-96(%r30),%r3
+	ldw	-92(%r30),%r4
+	ldw	-88(%r30),%r5
+	ldw	-84(%r30),%r6
+	ldw	-80(%r30),%r7
+
+L$few_limbs
+	addib,=,n 4,size,L$ret
+L$loop2	fldws,ma 4(s1_ptr),%fr4
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4,%fr31R,%fr5
+	fstds	 %fr5,-16(%r30)
+	ldws	-16(%r30),hi0
+	ldws	-12(%r30),lo0
+	addc	 lo0,cylimb,lo0
+	addc	 %r0,hi0,cylimb
+	add	 s0,lo0,s0
+	stws,ma	 s0,4(res_ptr)
+	addib,<> -1,size,L$loop2
+	nop
+
+L$ret	addc	 %r0,cylimb,cylimb
+	bv	 0(%r2)
+	ldo	 -128(%r30),%r30
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/lshift.s b/rts/gmp/mpn/hppa/hppa1_1/pa7100/lshift.s
new file mode 100644
index 0000000000..31669b1a55
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/lshift.s
@@ -0,0 +1,83 @@
+; HP-PA  __gmpn_lshift --
+; This is optimized for the PA7100, where is runs at 3.25 cycles/limb
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s_ptr		gr25
+; size		gr24
+; cnt		gr23
+
+	.code
+	.export		__gmpn_lshift
+__gmpn_lshift
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	sh2add		%r24,%r25,%r25
+	sh2add		%r24,%r26,%r26
+	ldws,mb		-4(0,%r25),%r22
+	subi		32,%r23,%r1
+	mtsar		%r1
+	addib,=		-1,%r24,L$0004
+	vshd		%r0,%r22,%r28		; compute carry out limb
+	ldws,mb		-4(0,%r25),%r29
+	addib,<=	-5,%r24,L$rest
+	vshd		%r22,%r29,%r20
+
+L$loop	ldws,mb		-4(0,%r25),%r22
+	stws,mb		%r20,-4(0,%r26)
+	vshd		%r29,%r22,%r20
+	ldws,mb		-4(0,%r25),%r29
+	stws,mb		%r20,-4(0,%r26)
+	vshd		%r22,%r29,%r20
+	ldws,mb		-4(0,%r25),%r22
+	stws,mb		%r20,-4(0,%r26)
+	vshd		%r29,%r22,%r20
+	ldws,mb		-4(0,%r25),%r29
+	stws,mb		%r20,-4(0,%r26)
+	addib,>		-4,%r24,L$loop
+	vshd		%r22,%r29,%r20
+
+L$rest	addib,=		4,%r24,L$end1
+	nop
+L$eloop	ldws,mb		-4(0,%r25),%r22
+	stws,mb		%r20,-4(0,%r26)
+	addib,<=	-1,%r24,L$end2
+	vshd		%r29,%r22,%r20
+	ldws,mb		-4(0,%r25),%r29
+	stws,mb		%r20,-4(0,%r26)
+	addib,>		-1,%r24,L$eloop
+	vshd		%r22,%r29,%r20
+
+L$end1	stws,mb		%r20,-4(0,%r26)
+	vshd		%r29,%r0,%r20
+	bv		0(%r2)
+	stw		%r20,-4(0,%r26)
+L$end2	stws,mb		%r20,-4(0,%r26)
+L$0004	vshd		%r22,%r0,%r20
+	bv		0(%r2)
+	stw		%r20,-4(0,%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/rshift.s b/rts/gmp/mpn/hppa/hppa1_1/pa7100/rshift.s
new file mode 100644
index 0000000000..d32b10b4b1
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/rshift.s
@@ -0,0 +1,80 @@
+; HP-PA  __gmpn_rshift --
+; This is optimized for the PA7100, where is runs at 3.25 cycles/limb
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s_ptr		gr25
+; size		gr24
+; cnt		gr23
+
+	.code
+	.export		__gmpn_rshift
+__gmpn_rshift
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r22
+	mtsar		%r23
+	addib,=		-1,%r24,L$0004
+	vshd		%r22,%r0,%r28		; compute carry out limb
+	ldws,ma		4(0,%r25),%r29
+	addib,<=	-5,%r24,L$rest
+	vshd		%r29,%r22,%r20
+
+L$loop	ldws,ma		4(0,%r25),%r22
+	stws,ma		%r20,4(0,%r26)
+	vshd		%r22,%r29,%r20
+	ldws,ma		4(0,%r25),%r29
+	stws,ma		%r20,4(0,%r26)
+	vshd		%r29,%r22,%r20
+	ldws,ma		4(0,%r25),%r22
+	stws,ma		%r20,4(0,%r26)
+	vshd		%r22,%r29,%r20
+	ldws,ma		4(0,%r25),%r29
+	stws,ma		%r20,4(0,%r26)
+	addib,>		-4,%r24,L$loop
+	vshd		%r29,%r22,%r20
+
+L$rest	addib,=		4,%r24,L$end1
+	nop
+L$eloop	ldws,ma		4(0,%r25),%r22
+	stws,ma		%r20,4(0,%r26)
+	addib,<=	-1,%r24,L$end2
+	vshd		%r22,%r29,%r20
+	ldws,ma		4(0,%r25),%r29
+	stws,ma		%r20,4(0,%r26)
+	addib,>		-1,%r24,L$eloop
+	vshd		%r29,%r22,%r20
+
+L$end1	stws,ma		%r20,4(0,%r26)
+	vshd		%r0,%r29,%r20
+	bv		0(%r2)
+	stw		%r20,0(0,%r26)
+L$end2	stws,ma		%r20,4(0,%r26)
+L$0004	vshd		%r0,%r22,%r20
+	bv		0(%r2)
+	stw		%r20,0(0,%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/sub_n.s b/rts/gmp/mpn/hppa/hppa1_1/pa7100/sub_n.s
new file mode 100644
index 0000000000..0eec41c4b3
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/sub_n.s
@@ -0,0 +1,76 @@
+; HP-PA  __gmpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+; store difference in a third limb vector.
+; This is optimized for the PA7100, where is runs at 4.25 cycles/limb
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+	.code
+	.export		__gmpn_sub_n
+__gmpn_sub_n
+	.proc
+	.callinfo	frame=0,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+
+	addib,<=	-5,%r23,L$rest
+	 sub		%r20,%r19,%r28	; subtract first limbs ignoring cy
+
+L$loop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	subb		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	subb		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	subb		%r20,%r19,%r28
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,>		-4,%r23,L$loop
+	subb		%r20,%r19,%r28
+
+L$rest	addib,=		4,%r23,L$end
+	nop
+L$eloop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,>		-1,%r23,L$eloop
+	subb		%r20,%r19,%r28
+
+L$end	stws		%r28,0(0,%r26)
+	addc		%r0,%r0,%r28
+	bv		0(%r2)
+	 subi		1,%r28,%r28
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/pa7100/submul_1.S b/rts/gmp/mpn/hppa/hppa1_1/pa7100/submul_1.S
new file mode 100644
index 0000000000..0fba21dcef
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/pa7100/submul_1.S
@@ -0,0 +1,195 @@
+; HP-PA 7100/7200 __gmpn_submul_1 -- Multiply a limb vector with a limb and
+; subtract the result from a second limb vector.
+
+; Copyright (C) 1995, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+; INPUT PARAMETERS
+#define res_ptr	%r26
+#define s1_ptr	%r25
+#define size	%r24
+#define s2_limb	%r23
+
+#define cylimb	%r28
+#define s0	%r19
+#define s1	%r20
+#define s2	%r3
+#define s3	%r4
+#define lo0	%r21
+#define lo1	%r5
+#define lo2	%r6
+#define lo3	%r7
+#define hi0	%r22
+#define hi1	%r23				/* safe to reuse */
+#define hi2	%r29
+#define hi3	%r1
+
+	.code
+	.export		__gmpn_submul_1
+__gmpn_submul_1
+	.proc
+	.callinfo	frame=128,no_calls
+	.entry
+
+	ldo	128(%r30),%r30
+	stws	s2_limb,-16(%r30)
+	add	 %r0,%r0,cylimb			; clear cy and cylimb
+	addib,<	-4,size,L$few_limbs
+	fldws	-16(%r30),%fr31R
+
+	ldo	-112(%r30),%r31
+	stw	%r3,-96(%r30)
+	stw	%r4,-92(%r30)
+	stw	%r5,-88(%r30)
+	stw	%r6,-84(%r30)
+	stw	%r7,-80(%r30)
+
+	bb,>=,n	 s1_ptr,29,L$0
+
+	fldws,ma 4(s1_ptr),%fr4
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4,%fr31R,%fr5
+	fstds	 %fr5,-16(%r31)
+	ldws	-16(%r31),cylimb
+	ldws	-12(%r31),lo0
+	sub	 s0,lo0,s0
+	add	 s0,lo0,%r0			; invert cy
+	addib,< -1,size,L$few_limbs
+	stws,ma	 s0,4(res_ptr)
+
+; start software pipeline ----------------------------------------------------
+L$0	fldds,ma 8(s1_ptr),%fr4
+	fldds,ma 8(s1_ptr),%fr8
+
+	xmpyu	 %fr4L,%fr31R,%fr5
+	xmpyu	 %fr4R,%fr31R,%fr6
+	xmpyu	 %fr8L,%fr31R,%fr9
+	xmpyu	 %fr8R,%fr31R,%fr10
+
+	fstds	 %fr5,-16(%r31)
+	fstds	 %fr6,-8(%r31)
+	fstds	 %fr9,0(%r31)
+	fstds	 %fr10,8(%r31)
+
+	ldws   -16(%r31),hi0
+	ldws   -12(%r31),lo0
+	ldws	-8(%r31),hi1
+	ldws	-4(%r31),lo1
+	ldws	 0(%r31),hi2
+	ldws	 4(%r31),lo2
+	ldws	 8(%r31),hi3
+	ldws	12(%r31),lo3
+
+	addc	 lo0,cylimb,lo0
+	addc	 lo1,hi0,lo1
+	addc	 lo2,hi1,lo2
+	addc	 lo3,hi2,lo3
+
+	addib,<	 -4,size,L$end
+	addc	 %r0,hi3,cylimb			; propagate carry into cylimb
+; main loop ------------------------------------------------------------------
+L$loop	fldds,ma 8(s1_ptr),%fr4
+	fldds,ma 8(s1_ptr),%fr8
+
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4L,%fr31R,%fr5
+	ldws	 4(res_ptr),s1
+	xmpyu	 %fr4R,%fr31R,%fr6
+	ldws	 8(res_ptr),s2
+	xmpyu	 %fr8L,%fr31R,%fr9
+	ldws	12(res_ptr),s3
+	xmpyu	 %fr8R,%fr31R,%fr10
+
+	fstds	 %fr5,-16(%r31)
+	sub	 s0,lo0,s0
+	fstds	 %fr6,-8(%r31)
+	subb	 s1,lo1,s1
+	fstds	 %fr9,0(%r31)
+	subb	 s2,lo2,s2
+	fstds	 %fr10,8(%r31)
+	subb	 s3,lo3,s3
+	subb	 %r0,%r0,lo0			; these two insns ...
+	add	 lo0,lo0,%r0			; ... just invert cy
+
+	ldws   -16(%r31),hi0
+	ldws   -12(%r31),lo0
+	ldws	-8(%r31),hi1
+	ldws	-4(%r31),lo1
+	ldws	 0(%r31),hi2
+	ldws	 4(%r31),lo2
+	ldws	 8(%r31),hi3
+	ldws	12(%r31),lo3
+
+	addc	 lo0,cylimb,lo0
+	stws,ma	 s0,4(res_ptr)
+	addc	 lo1,hi0,lo1
+	stws,ma	 s1,4(res_ptr)
+	addc	 lo2,hi1,lo2
+	stws,ma	 s2,4(res_ptr)
+	addc	 lo3,hi2,lo3
+	stws,ma	 s3,4(res_ptr)
+
+	addib,>= -4,size,L$loop
+	addc	 %r0,hi3,cylimb			; propagate carry into cylimb
+; finish software pipeline ---------------------------------------------------
+L$end	ldws	 0(res_ptr),s0
+	ldws	 4(res_ptr),s1
+	ldws	 8(res_ptr),s2
+	ldws	12(res_ptr),s3
+
+	sub	 s0,lo0,s0
+	stws,ma	 s0,4(res_ptr)
+	subb	 s1,lo1,s1
+	stws,ma	 s1,4(res_ptr)
+	subb	 s2,lo2,s2
+	stws,ma	 s2,4(res_ptr)
+	subb	 s3,lo3,s3
+	stws,ma	 s3,4(res_ptr)
+	subb	 %r0,%r0,lo0			; these two insns ...
+	add	 lo0,lo0,%r0			; ... invert cy
+
+; restore callee-saves registers ---------------------------------------------
+	ldw	-96(%r30),%r3
+	ldw	-92(%r30),%r4
+	ldw	-88(%r30),%r5
+	ldw	-84(%r30),%r6
+	ldw	-80(%r30),%r7
+
+L$few_limbs
+	addib,=,n 4,size,L$ret
+L$loop2	fldws,ma 4(s1_ptr),%fr4
+	ldws	 0(res_ptr),s0
+	xmpyu	 %fr4,%fr31R,%fr5
+	fstds	 %fr5,-16(%r30)
+	ldws	-16(%r30),hi0
+	ldws	-12(%r30),lo0
+	addc	 lo0,cylimb,lo0
+	addc	 %r0,hi0,cylimb
+	sub	 s0,lo0,s0
+	add	 s0,lo0,%r0			; invert cy
+	stws,ma	 s0,4(res_ptr)
+	addib,<> -1,size,L$loop2
+	nop
+
+L$ret	addc	 %r0,cylimb,cylimb
+	bv	 0(%r2)
+	ldo	 -128(%r30),%r30
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/submul_1.s b/rts/gmp/mpn/hppa/hppa1_1/submul_1.s
new file mode 100644
index 0000000000..20a5b5ce0a
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/submul_1.s
@@ -0,0 +1,111 @@
+; HP-PA-1.1 __gmpn_submul_1 -- Multiply a limb vector with a limb and
+; subtract the result from a second limb vector.
+
+; Copyright (C) 1992, 1993, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	r26
+; s1_ptr	r25
+; size		r24
+; s2_limb	r23
+
+; This runs at 12 cycles/limb on a PA7000.  With the used instructions, it
+; can not become faster due to data cache contention after a store.  On the
+; PA7100 it runs at 11 cycles/limb, and that can not be improved either,
+; since only the xmpyu does not need the integer pipeline, so the only
+; dual-issue we will get are addc+xmpyu.  Unrolling could gain a cycle/limb
+; on the PA7100.
+
+; There are some ideas described in mul_1.s that applies to this code too.
+
+; It seems possible to make this run as fast as __gmpn_addmul_1, if we use
+; 	sub,>>=	%r29,%r19,%r22
+;	addi	1,%r28,%r28
+; but that requires reworking the hairy software pipeline...
+
+	.code
+	.export		__gmpn_submul_1
+__gmpn_submul_1
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	ldo		64(%r30),%r30
+	fldws,ma	4(%r25),%fr5
+	stw		%r23,-16(%r30)		; move s2_limb ...
+	addib,=		-1,%r24,L$just_one_limb
+	 fldws		-16(%r30),%fr4		; ... into fr4
+	add		%r0,%r0,%r0		; clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds		%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r19		; least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+; Main loop
+L$loop	ldws		0(%r26),%r29
+	fldws,ma	4(%r25),%fr5
+	sub		%r29,%r19,%r22
+	add		%r22,%r19,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r28,%r1,%r19
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addc		%r0,%r28,%r28
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end	ldw		0(%r26),%r29
+	sub		%r29,%r19,%r22
+	add		%r22,%r19,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r28,%r1,%r19
+	ldw		-16(%r30),%r28
+	ldws		0(%r26),%r29
+	addc		%r0,%r28,%r28
+	sub		%r29,%r19,%r22
+	add		%r22,%r19,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+L$just_one_limb
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		0(%r26),%r29
+	fstds		%fr6,-16(%r30)
+	ldw		-12(%r30),%r1
+	ldw		-16(%r30),%r28
+	sub		%r29,%r1,%r22
+	add		%r22,%r1,%r0
+	stw		%r22,0(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/udiv_qrnnd.S b/rts/gmp/mpn/hppa/hppa1_1/udiv_qrnnd.S
new file mode 100644
index 0000000000..b83d6f4dd2
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/udiv_qrnnd.S
@@ -0,0 +1,80 @@
+; HP-PA  __udiv_qrnnd division support, used from longlong.h.
+; This version runs fast on PA 7000 and later.
+
+; Copyright (C) 1993, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; rem_ptr	gr26
+; n1		gr25
+; n0		gr24
+; d		gr23
+
+	.code
+L$0000	.word		0x43f00000		; 2^64
+	.word		0x0
+	.export		__gmpn_udiv_qrnnd
+__gmpn_udiv_qrnnd
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+	ldo		64(%r30),%r30
+
+	stws		%r25,-16(0,%r30)	; n_hi
+	stws		%r24,-12(0,%r30)	; n_lo
+#ifdef PIC
+	addil		LT%L$0000,%r19
+	ldo		RT%L$0000(%r1),%r19
+#else
+	ldil		L%L$0000,%r19
+	ldo		R%L$0000(%r19),%r19
+#endif	
+	fldds		-16(0,%r30),%fr5
+	stws		%r23,-12(0,%r30)
+	comib,<=	0,%r25,L$1
+	fcnvxf,dbl,dbl	%fr5,%fr5
+	fldds		0(0,%r19),%fr4
+	fadd,dbl	%fr4,%fr5,%fr5
+L$1
+	fcpy,sgl	%fr0,%fr6L
+	fldws		-12(0,%r30),%fr6R
+	fcnvxf,dbl,dbl	%fr6,%fr4
+
+	fdiv,dbl	%fr5,%fr4,%fr5
+
+	fcnvfx,dbl,dbl	%fr5,%fr4
+	fstws		%fr4R,-16(%r30)
+	xmpyu		%fr4R,%fr6R,%fr6
+	ldws		-16(%r30),%r28
+	fstds		%fr6,-16(0,%r30)
+	ldws		-12(0,%r30),%r21
+	ldws		-16(0,%r30),%r20
+	sub		%r24,%r21,%r22
+	subb		%r25,%r20,%r19
+	comib,=		0,%r19,L$2
+	ldo		-64(%r30),%r30
+
+	add		%r22,%r23,%r22
+	ldo		-1(%r28),%r28
+L$2	bv		0(%r2)
+	stws		%r22,0(0,%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa1_1/umul.s b/rts/gmp/mpn/hppa/hppa1_1/umul.s
new file mode 100644
index 0000000000..1f1300ac9b
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa1_1/umul.s
@@ -0,0 +1,42 @@
+; Copyright (C) 1999 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+	.code
+	.export		__umul_ppmm
+	.align 4
+__umul_ppmm
+	.proc
+	.callinfo frame=64,no_calls
+	.entry
+
+	ldo 64(%r30),%r30
+	stw %r25,-16(0,%r30)
+	fldws -16(0,%r30),%fr22R
+	stw %r24,-16(0,%r30)
+	fldws -16(0,%r30),%fr22L
+	xmpyu %fr22R,%fr22L,%fr22
+	fstds %fr22,-16(0,%r30)
+	ldw -16(0,%r30),%r28
+	ldw -12(0,%r30),%r29
+	stw %r29,0(0,%r26)
+	bv 0(%r2)
+	ldo -64(%r30),%r30
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa2_0/add_n.s b/rts/gmp/mpn/hppa/hppa2_0/add_n.s
new file mode 100644
index 0000000000..6e97278a39
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa2_0/add_n.s
@@ -0,0 +1,88 @@
+; HP-PA 2.0 32-bit __gmpn_add_n -- Add two limb vectors of the same length > 0
+; and store sum in a third limb vector.
+
+; Copyright (C) 1997, 1998, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+; This runs at 2 cycles/limb on PA8000.
+
+	.code
+	.export	__gmpn_add_n
+__gmpn_add_n
+	.proc
+	.callinfo frame=0,no_calls
+	.entry
+
+	sub		%r0,%r23,%r22
+	zdep		%r22,30,3,%r28		; r28 = 2 * (-n & 7)
+	zdep		%r22,29,3,%r22		; r22 = 4 * (-n & 7)
+	sub		%r25,%r22,%r25		; offset s1_ptr
+	sub		%r24,%r22,%r24		; offset s2_ptr
+	sub		%r26,%r22,%r26		; offset res_ptr
+	blr		%r28,%r0		; branch into loop
+	add		%r0,%r0,%r0		; reset carry
+
+L$loop	ldw		0(%r25),%r20
+	ldw		0(%r24),%r31
+	addc		%r20,%r31,%r20
+	stw		%r20,0(%r26)
+L$7	ldw		4(%r25),%r21
+	ldw		4(%r24),%r19
+	addc		%r21,%r19,%r21
+	stw		%r21,4(%r26)
+L$6	ldw		8(%r25),%r20
+	ldw		8(%r24),%r31
+	addc		%r20,%r31,%r20
+	stw		%r20,8(%r26)
+L$5	ldw		12(%r25),%r21
+	ldw		12(%r24),%r19
+	addc		%r21,%r19,%r21
+	stw		%r21,12(%r26)
+L$4	ldw		16(%r25),%r20
+	ldw		16(%r24),%r31
+	addc		%r20,%r31,%r20
+	stw		%r20,16(%r26)
+L$3	ldw		20(%r25),%r21
+	ldw		20(%r24),%r19
+	addc		%r21,%r19,%r21
+	stw		%r21,20(%r26)
+L$2	ldw		24(%r25),%r20
+	ldw		24(%r24),%r31
+	addc		%r20,%r31,%r20
+	stw		%r20,24(%r26)
+L$1	ldw		28(%r25),%r21
+	ldo		32(%r25),%r25
+	ldw		28(%r24),%r19
+	addc		%r21,%r19,%r21
+	stw		%r21,28(%r26)
+	ldo		32(%r24),%r24
+	addib,>		-8,%r23,L$loop
+	ldo		32(%r26),%r26
+
+	bv		(%r2)
+	.exit
+	addc		%r0,%r0,%r28
+	.procend
diff --git a/rts/gmp/mpn/hppa/hppa2_0/sub_n.s b/rts/gmp/mpn/hppa/hppa2_0/sub_n.s
new file mode 100644
index 0000000000..7d9b50fc27
--- /dev/null
+++ b/rts/gmp/mpn/hppa/hppa2_0/sub_n.s
@@ -0,0 +1,88 @@
+; HP-PA 2.0 32-bit __gmpn_sub_n -- Subtract two limb vectors of the same
+; length > 0 and store difference in a third limb vector.
+
+; Copyright (C) 1997, 1998, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+; This runs at 2 cycles/limb on PA8000.
+
+	.code
+	.export	__gmpn_sub_n
+__gmpn_sub_n
+	.proc
+	.callinfo frame=0,no_calls
+	.entry
+
+	sub		%r0,%r23,%r22
+	zdep		%r22,30,3,%r28		; r28 = 2 * (-n & 7)
+	zdep		%r22,29,3,%r22		; r22 = 4 * (-n & 7)
+	sub		%r25,%r22,%r25		; offset s1_ptr
+	sub		%r24,%r22,%r24		; offset s2_ptr
+	blr		%r28,%r0		; branch into loop
+	sub		%r26,%r22,%r26		; offset res_ptr and set carry
+
+L$loop	ldw		0(%r25),%r20
+	ldw		0(%r24),%r31
+	subb		%r20,%r31,%r20
+	stw		%r20,0(%r26)
+L$7	ldw		4(%r25),%r21
+	ldw		4(%r24),%r19
+	subb		%r21,%r19,%r21
+	stw		%r21,4(%r26)
+L$6	ldw		8(%r25),%r20
+	ldw		8(%r24),%r31
+	subb		%r20,%r31,%r20
+	stw		%r20,8(%r26)
+L$5	ldw		12(%r25),%r21
+	ldw		12(%r24),%r19
+	subb		%r21,%r19,%r21
+	stw		%r21,12(%r26)
+L$4	ldw		16(%r25),%r20
+	ldw		16(%r24),%r31
+	subb		%r20,%r31,%r20
+	stw		%r20,16(%r26)
+L$3	ldw		20(%r25),%r21
+	ldw		20(%r24),%r19
+	subb		%r21,%r19,%r21
+	stw		%r21,20(%r26)
+L$2	ldw		24(%r25),%r20
+	ldw		24(%r24),%r31
+	subb		%r20,%r31,%r20
+	stw		%r20,24(%r26)
+L$1	ldw		28(%r25),%r21
+	ldo		32(%r25),%r25
+	ldw		28(%r24),%r19
+	subb		%r21,%r19,%r21
+	stw		%r21,28(%r26)
+	ldo		32(%r24),%r24
+	addib,>		-8,%r23,L$loop
+	ldo		32(%r26),%r26
+
+	addc		%r0,%r0,%r28
+	bv		(%r2)
+	.exit
+	subi		1,%r28,%r28
+	.procend
diff --git a/rts/gmp/mpn/hppa/lshift.s b/rts/gmp/mpn/hppa/lshift.s
new file mode 100644
index 0000000000..f5a2daad60
--- /dev/null
+++ b/rts/gmp/mpn/hppa/lshift.s
@@ -0,0 +1,66 @@
+; HP-PA  __gmpn_lshift --
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s_ptr		gr25
+; size		gr24
+; cnt		gr23
+
+	.code
+	.export		__gmpn_lshift
+__gmpn_lshift
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	sh2add		%r24,%r25,%r25
+	sh2add		%r24,%r26,%r26
+	ldws,mb		-4(0,%r25),%r22
+	subi		32,%r23,%r1
+	mtsar		%r1
+	addib,=		-1,%r24,L$0004
+	vshd		%r0,%r22,%r28		; compute carry out limb
+	ldws,mb		-4(0,%r25),%r29
+	addib,=		-1,%r24,L$0002
+	vshd		%r22,%r29,%r20
+
+L$loop	ldws,mb		-4(0,%r25),%r22
+	stws,mb		%r20,-4(0,%r26)
+	addib,=		-1,%r24,L$0003
+	vshd		%r29,%r22,%r20
+	ldws,mb		-4(0,%r25),%r29
+	stws,mb		%r20,-4(0,%r26)
+	addib,<>	-1,%r24,L$loop
+	vshd		%r22,%r29,%r20
+
+L$0002	stws,mb		%r20,-4(0,%r26)
+	vshd		%r29,%r0,%r20
+	bv		0(%r2)
+	stw		%r20,-4(0,%r26)
+L$0003	stws,mb		%r20,-4(0,%r26)
+L$0004	vshd		%r22,%r0,%r20
+	bv		0(%r2)
+	stw		%r20,-4(0,%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/rshift.s b/rts/gmp/mpn/hppa/rshift.s
new file mode 100644
index 0000000000..e05e2f10b5
--- /dev/null
+++ b/rts/gmp/mpn/hppa/rshift.s
@@ -0,0 +1,63 @@
+; HP-PA  __gmpn_rshift -- 
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s_ptr		gr25
+; size		gr24
+; cnt		gr23
+
+	.code
+	.export		__gmpn_rshift
+__gmpn_rshift
+	.proc
+	.callinfo	frame=64,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r22
+	mtsar		%r23
+	addib,=		-1,%r24,L$0004
+	vshd		%r22,%r0,%r28		; compute carry out limb
+	ldws,ma		4(0,%r25),%r29
+	addib,=		-1,%r24,L$0002
+	vshd		%r29,%r22,%r20
+
+L$loop	ldws,ma		4(0,%r25),%r22
+	stws,ma		%r20,4(0,%r26)
+	addib,=		-1,%r24,L$0003
+	vshd		%r22,%r29,%r20
+	ldws,ma		4(0,%r25),%r29
+	stws,ma		%r20,4(0,%r26)
+	addib,<>	-1,%r24,L$loop
+	vshd		%r29,%r22,%r20
+
+L$0002	stws,ma		%r20,4(0,%r26)
+	vshd		%r0,%r29,%r20
+	bv		0(%r2)
+	stw		%r20,0(0,%r26)
+L$0003	stws,ma		%r20,4(0,%r26)
+L$0004	vshd		%r0,%r22,%r20
+	bv		0(%r2)
+	stw		%r20,0(0,%r26)
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/sub_n.s b/rts/gmp/mpn/hppa/sub_n.s
new file mode 100644
index 0000000000..8f770ad1ad
--- /dev/null
+++ b/rts/gmp/mpn/hppa/sub_n.s
@@ -0,0 +1,59 @@
+; HP-PA  __gmpn_sub_n -- Subtract two limb vectors of the same length > 0 and
+; store difference in a third limb vector.
+
+; Copyright (C) 1992, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; res_ptr	gr26
+; s1_ptr	gr25
+; s2_ptr	gr24
+; size		gr23
+
+; One might want to unroll this as for other processors, but it turns
+; out that the data cache contention after a store makes such
+; unrolling useless.  We can't come under 5 cycles/limb anyway.
+
+	.code
+	.export		__gmpn_sub_n
+__gmpn_sub_n
+	.proc
+	.callinfo	frame=0,no_calls
+	.entry
+
+	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+
+	addib,=		-1,%r23,L$end	; check for (SIZE == 1)
+	 sub		%r20,%r19,%r28	; subtract first limbs ignoring cy
+
+L$loop	ldws,ma		4(0,%r25),%r20
+	ldws,ma		4(0,%r24),%r19
+	stws,ma		%r28,4(0,%r26)
+	addib,<>	-1,%r23,L$loop
+	 subb		%r20,%r19,%r28
+
+L$end	stws		%r28,0(0,%r26)
+	addc		%r0,%r0,%r28
+	bv		0(%r2)
+	 subi		1,%r28,%r28
+
+	.exit
+	.procend
diff --git a/rts/gmp/mpn/hppa/udiv_qrnnd.s b/rts/gmp/mpn/hppa/udiv_qrnnd.s
new file mode 100644
index 0000000000..9aa3b8a830
--- /dev/null
+++ b/rts/gmp/mpn/hppa/udiv_qrnnd.s
@@ -0,0 +1,286 @@
+; HP-PA  __udiv_qrnnd division support, used from longlong.h.
+; This version runs fast on pre-PA7000 CPUs.
+
+; Copyright (C) 1993, 1994, 2000 Free Software Foundation, Inc.
+
+; This file is part of the GNU MP Library.
+
+; The GNU MP Library is free software; you can redistribute it and/or modify
+; it under the terms of the GNU Lesser General Public License as published by
+; the Free Software Foundation; either version 2.1 of the License, or (at your
+; option) any later version.
+
+; The GNU MP Library is distributed in the hope that it will be useful, but
+; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
+; License for more details.
+
+; You should have received a copy of the GNU Lesser General Public License
+; along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
+; the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
+; MA 02111-1307, USA.
+
+
+; INPUT PARAMETERS
+; rem_ptr	gr26
+; n1		gr25
+; n0		gr24
+; d		gr23
+
+; The code size is a bit excessive.  We could merge the last two ds;addc
+; sequences by simply moving the "bb,< Odd" instruction down.  The only
+; trouble is the FFFFFFFF code that would need some hacking.
+
+	.code
+	.export		__gmpn_udiv_qrnnd
+__gmpn_udiv_qrnnd
+	.proc
+	.callinfo	frame=0,no_calls
+	.entry
+
+	comb,<		%r23,0,L$largedivisor
+	 sub		%r0,%r23,%r1		; clear cy as side-effect
+	ds		%r0,%r1,%r0
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r23,%r25
+	addc		%r24,%r24,%r28
+	ds		%r25,%r23,%r25
+	comclr,>=	%r25,%r0,%r0
+	addl		%r25,%r23,%r25
+	stws		%r25,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r28,%r28,%r28
+
+L$largedivisor
+	extru		%r24,31,1,%r19		; r19 = n0 & 1
+	bb,<		%r23,31,L$odd
+	 extru		%r23,30,31,%r22		; r22 = d >> 1
+	shd		%r25,%r24,1,%r24	; r24 = new n0
+	extru		%r25,30,31,%r25		; r25 = new n1
+	sub		%r0,%r22,%r21
+	ds		%r0,%r21,%r0
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	comclr,>=	%r25,%r0,%r0
+	addl		%r25,%r22,%r25
+	sh1addl		%r25,%r19,%r25
+	stws		%r25,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r24,%r24,%r28
+
+L$odd	addib,sv,n	1,%r22,L$FF..		; r22 = (d / 2 + 1)
+	shd		%r25,%r24,1,%r24	; r24 = new n0
+	extru		%r25,30,31,%r25		; r25 = new n1
+	sub		%r0,%r22,%r21
+	ds		%r0,%r21,%r0
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r24
+	ds		%r25,%r22,%r25
+	addc		%r24,%r24,%r28
+	comclr,>=	%r25,%r0,%r0
+	addl		%r25,%r22,%r25
+	sh1addl		%r25,%r19,%r25
+; We have computed (n1,,n0) / (d + 1), q' = r28, r' = r25
+	add,nuv		%r28,%r25,%r25
+	addl		%r25,%r1,%r25
+	addc		%r0,%r28,%r28
+	sub,<<		%r25,%r23,%r0
+	addl		%r25,%r1,%r25
+	stws		%r25,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r0,%r28,%r28
+
+; This is just a special case of the code above.
+; We come here when d == 0xFFFFFFFF
+L$FF..	add,uv		%r25,%r24,%r24
+	sub,<<		%r24,%r23,%r0
+	ldo		1(%r24),%r24
+	stws		%r24,0(0,%r26)
+	bv		0(%r2)
+	 addc		%r0,%r25,%r28
+
+	.exit
+	.procend