(CFLAGS-tst-align.c): Add -mpreferred-stack-boundary=4.

5 files changed, 399 insertions, 0 deletions

diff --git a/sysdeps/hppa/hppa1.1/Implies b/sysdeps/hppa/hppa1.1/Implies
new file mode 100644
index 0000000000..5f935a299c
--- /dev/null
+++ b/sysdeps/hppa/hppa1.1/Implies
@@ -0,0 +1,4 @@
+wordsize-32
+ieee754/flt-32
+ieee754/dbl-64
+ieee754/ldbl-128
diff --git a/sysdeps/hppa/hppa1.1/addmul_1.s b/sysdeps/hppa/hppa1.1/addmul_1.s
new file mode 100644
index 0000000000..a1fb083a83
--- /dev/null
+++ b/sysdeps/hppa/hppa1.1/addmul_1.s
@@ -0,0 +1,104 @@
+;! HP-PA-1.1 __mpn_addmul_1 -- Multiply a limb vector with a limb and
+;! add the result to a second limb vector.
+
+;! Copyright (C) 1992, 1993, 1994 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.
+
+	.text
+	.export		__mpn_addmul_1
+__mpn_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$just_one_limb
+	 fldws		-16(%r30),%fr4		;! ... into fr4
+	add		%r0,%r0,%r0		;! clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds		%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r20		;! least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+;! Main loop
+L$loop:	
+	ldws		0(%r26),%r29
+	fldws,ma	4(%r25),%fr5
+	add		%r29,%r20,%r20
+	stws,ma		%r20,4(%r26)
+	addc		%r28,%r1,%r20
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addc		%r0,%r28,%r28
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end:		
+	ldw		0(%r26),%r29
+	add		%r29,%r20,%r20
+	stws,ma		%r20,4(%r26)
+	addc		%r28,%r1,%r20
+	ldw		-16(%r30),%r28
+	ldws		0(%r26),%r29
+	addc		%r0,%r28,%r28
+	add		%r29,%r20,%r20
+	stws,ma		%r20,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	ldo		-64(%r30),%r30
+
+L$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,%r20
+	stw		%r20,0(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+	.exit
+	.procend
diff --git a/sysdeps/hppa/hppa1.1/mul_1.s b/sysdeps/hppa/hppa1.1/mul_1.s
new file mode 100644
index 0000000000..00c770f272
--- /dev/null
+++ b/sysdeps/hppa/hppa1.1/mul_1.s
@@ -0,0 +1,100 @@
+;! HP-PA-1.1 __mpn_mul_1 -- Multiply a limb vector with a limb and store
+;! the result in a second limb vector.
+
+;! Copyright (C) 1992, 1993, 1994 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.)
+
+	.text
+	.export		__mpn_mul_1
+__mpn_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$just_one_limb
+	 fldws		-16(%r30),%fr4		;! ... into fr4
+	add		%r0,%r0,%r0		;! clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds	 	%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r20		;! least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+;! Main loop
+L$loop:	
+	fldws,ma	4(%r25),%fr5
+	stws,ma		%r20,4(%r26)
+	addc		%r28,%r1,%r20
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end:	
+	stws,ma		%r20,4(%r26)
+	addc		%r28,%r1,%r20
+	ldw		-16(%r30),%r28
+	stws,ma		%r20,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+L$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/sysdeps/hppa/hppa1.1/submul_1.s b/sysdeps/hppa/hppa1.1/submul_1.s
new file mode 100644
index 0000000000..997bd6d521
--- /dev/null
+++ b/sysdeps/hppa/hppa1.1/submul_1.s
@@ -0,0 +1,113 @@
+;! HP-PA-1.1 __mpn_submul_1 -- Multiply a limb vector with a limb and
+;! subtract the result from a second limb vector.
+
+;! Copyright (C) 1992, 1993, 1994 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 __mpn_addmul_1, if we use
+;! 	sub,>>=	%r29,%r20,%r22
+;!	addi	1,%r28,%r28
+;! but that requires reworking the hairy software pipeline...
+
+	.text
+	.export		__mpn_submul_1
+__mpn_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$just_one_limb
+	 fldws		-16(%r30),%fr4		;! ... into fr4
+	add		%r0,%r0,%r0		;! clear carry
+	xmpyu		%fr4,%fr5,%fr6
+	fldws,ma	4(%r25),%fr7
+	fstds		%fr6,-16(%r30)
+	xmpyu		%fr4,%fr7,%fr8
+	ldw		-12(%r30),%r20		;! least significant limb in product
+	ldw		-16(%r30),%r28
+
+	fstds		%fr8,-16(%r30)
+	addib,=		-1,%r24,L$end
+	 ldw		-12(%r30),%r1
+
+;! Main loop
+L$loop:	
+	ldws		0(%r26),%r29
+	fldws,ma	4(%r25),%fr5
+	sub		%r29,%r20,%r22
+	add		%r22,%r20,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r28,%r1,%r20
+	xmpyu		%fr4,%fr5,%fr6
+	ldw		-16(%r30),%r28
+	fstds		%fr6,-16(%r30)
+	addc		%r0,%r28,%r28
+	addib,<>	-1,%r24,L$loop
+	 ldw		-12(%r30),%r1
+
+L$end:	
+	ldw		0(%r26),%r29
+	sub		%r29,%r20,%r22
+	add		%r22,%r20,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r28,%r1,%r20
+	ldw		-16(%r30),%r28
+	ldws		0(%r26),%r29
+	addc		%r0,%r28,%r28
+	sub		%r29,%r20,%r22
+	add		%r22,%r20,%r0
+	stws,ma		%r22,4(%r26)
+	addc		%r0,%r28,%r28
+	bv		0(%r2)
+	 ldo		-64(%r30),%r30
+
+L$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/sysdeps/hppa/hppa1.1/udiv_qrnnd.s b/sysdeps/hppa/hppa1.1/udiv_qrnnd.s
new file mode 100644
index 0000000000..fdc63e59e5
--- /dev/null
+++ b/sysdeps/hppa/hppa1.1/udiv_qrnnd.s
@@ -0,0 +1,78 @@
+;! HP-PA  __udiv_qrnnd division support, used from longlong.h.
+;! This version runs fast on PA 7000 and later.
+
+;! Copyright (C) 1993, 1994 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
+
+	.text
+L$0000:	
+	.word		0x43f00000
+	.word		0x0
+	.export		__udiv_qrnnd
+__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
+	b,l		L$0,%r1
+	ldo		L$0000-L$0(%r1),%r1
+L$0:
+	fldds		-16(0,%r30),%fr5
+	stws		%r23,-12(0,%r30)
+	comib,<=	0,%r25,L$1
+	fcnvxf,dbl,dbl	%fr5,%fr5
+	fldds		0(0,%r1),%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,%r1
+	comib,=		0,%r1,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