PPC assembly pack: add ChaCha20 and Poly1305 modules.

Reviewed-by: Richard Levitte <levitte@openssl.org>

4 files changed, 1376 insertions, 0 deletions

diff --git a/crypto/poly1305/Makefile.in b/crypto/poly1305/Makefile.in
index 0984a5221c..d027eedc78 100644
--- a/crypto/poly1305/Makefile.in
+++ b/crypto/poly1305/Makefile.in
@@ -42,6 +42,10 @@ poly1305-x86.s:		asm/poly1305-x86.pl
 	$(PERL) asm/poly1305-x86.pl $(PERLASM_SCHEME) $(CFLAGS) $(PROCESSOR) > $@
 poly1305-x86_64.s:	asm/poly1305-x86_64.pl
 	$(PERL) asm/poly1305-x86_64.pl $(PERLASM_SCHEME) > $@
+poly1305-ppc.s:		asm/poly1305-ppc.pl
+	$(PERL) asm/poly1305-ppc.pl $(PERLASM_SCHEME) $@
+poly1305-ppcfp.s:	asm/poly1305-ppcfp.pl
+	$(PERL) asm/poly1305-ppcfp.pl $(PERLASM_SCHEME) $@
 
 poly1305-%.S:	asm/poly1305-%.pl;	$(PERL) $< $(PERLASM_SCHEME) $@
 
diff --git a/crypto/poly1305/asm/poly1305-ppc.pl b/crypto/poly1305/asm/poly1305-ppc.pl
new file mode 100755
index 0000000000..46130c9327
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-ppc.pl
@@ -0,0 +1,636 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements Poly1305 hash for PowerPC.
+#
+# June 2015
+#
+# Numbers are cycles per processed byte with poly1305_blocks alone,
+# and improvement coefficients relative to gcc-generated code.
+#
+#			-m32		-m64
+#
+# Freescale e300	14.8/+80%	-
+# PPC74x0		7.40/+60%	-
+# PPC970		7.20/+114%	3.51/+205%
+# POWER6		3.96/+250%	2.02/+170%
+# POWER7		3.67/+260%	1.87/+100%
+# POWER8		-		2.13/+200%
+#
+# Do we need floating-point implementation for PPC? Results presented
+# in poly1305_ieee754.c are tricky to compare to, because they are for
+# compiler-generated code. On the other hand it's known that floating-
+# point performance can be dominated by FPU latency, which means that
+# there is limit even for ideally optimized (and even vectorized) code.
+# And this limit is estimated to be higher than above -m64 results. Or
+# in other words floating-point implementation can be meaningful to
+# consider only in 32-bit application context. We probably have to
+# recognize that 32-bit builds are getting less popular on high-end
+# systems and therefore tend to target embedded ones, which might not
+# even have FPU...
+#
+# On side note, Power ISA 2.07 enables vector base 2^26 implementation,
+# and POWER8 might have capacity to break 1.0 cycle per byte barrier...
+
+$flavour = shift;
+
+if ($flavour =~ /64/) {
+	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
+	$UCMP	="cmpld";
+	$STU	="stdu";
+	$POP	="ld";
+	$PUSH	="std";
+} elsif ($flavour =~ /32/) {
+	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
+	$UCMP	="cmplw";
+	$STU	="stwu";
+	$POP	="lwz";
+	$PUSH	="stw";
+} else { die "nonsense $flavour"; }
+
+# Define endianess based on flavour
+# i.e.: linux64le
+$LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
+
+$FRAME=24*$SIZE_T;
+
+$sp="r1";
+my ($ctx,$inp,$len,$padbit) = map("r$_",(3..6));
+my ($mac,$nonce)=($inp,$len);
+my $mask = "r0";
+
+$code=<<___;
+.machine	"any"
+.text
+___
+							if ($flavour =~ /64/) {
+###############################################################################
+# base 2^64 implementation
+
+my ($h0,$h1,$h2,$d0,$d1,$d2, $r0,$r1,$s1, $t0,$t1) = map("r$_",(7..12,27..31));
+
+$code.=<<___;
+.globl	.poly1305_init_int
+.align	4
+.poly1305_init_int:
+	xor	r0,r0,r0
+	std	r0,0($ctx)		# zero hash value
+	std	r0,8($ctx)
+	std	r0,16($ctx)
+
+	$UCMP	$inp,r0
+	beq-	Lno_key
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	ld	$d0,0($inp)		# load key material
+	ld	$d1,8($inp)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	li	$h0,4
+	lwbrx	$d0,0,$inp		# load key material
+	li	$d1,8
+	lwbrx	$h0,$h0,$inp
+	li	$h1,12
+	lwbrx	$d1,$d1,$inp
+	lwbrx	$h1,$h1,$inp
+	insrdi	$d0,$h0,32,0
+	insrdi	$d1,$h1,32,0
+___
+$code.=<<___;
+	lis	$h1,0xfff		# 0x0fff0000
+	ori	$h1,$h1,0xfffc		# 0x0ffffffc
+	insrdi	$h1,$h1,32,0		# 0x0ffffffc0ffffffc
+	ori	$h0,$h1,3		# 0x0ffffffc0fffffff
+
+	and	$d0,$d0,$h0
+	and	$d1,$d1,$h1
+
+	std	$d0,32($ctx)		# store key
+	std	$d1,40($ctx)
+
+Lno_key:
+	xor	r3,r3,r3
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+.size	.poly1305_init_int,.-.poly1305_init_int
+
+.globl	.poly1305_blocks
+.align	4
+.poly1305_blocks:
+	srdi.	$len,$len,4
+	beq-	Labort
+
+	$STU	$sp,-$FRAME($sp)
+	mflr	r0
+	$PUSH	r27,`$FRAME-$SIZE_T*5`($sp)
+	$PUSH	r28,`$FRAME-$SIZE_T*4`($sp)
+	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
+	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
+	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
+
+	ld	$r0,32($ctx)		# load key
+	ld	$r1,40($ctx)
+
+	ld	$h0,0($ctx)		# load hash value
+	ld	$h1,8($ctx)
+	ld	$h2,16($ctx)
+
+	srdi	$s1,$r1,2
+	mtctr	$len
+	add	$s1,$s1,$r1		# s1 = r1 + r1>>2
+	li	$mask,3
+	b	Loop
+
+.align	4
+Loop:
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	ld	$t0,0($inp)		# load input
+	ld	$t1,8($inp)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	li	$d0,4
+	lwbrx	$t0,0,$inp		# load input
+	li	$t1,8
+	lwbrx	$d0,$d0,$inp
+	li	$d1,12
+	lwbrx	$t1,$t1,$inp
+	lwbrx	$d1,$d1,$inp
+	insrdi	$t0,$d0,32,0
+	insrdi	$t1,$d1,32,0
+___
+$code.=<<___;
+	addi	$inp,$inp,16
+
+	addc	$h0,$h0,$t0		# accumulate input
+	adde	$h1,$h1,$t1
+
+	mulld	$d0,$h0,$r0		# h0*r0
+	mulhdu	$d1,$h0,$r0
+	adde	$h2,$h2,$padbit
+
+	mulld	$t0,$h1,$s1		# h1*5*r1
+	mulhdu	$t1,$h1,$s1
+	addc	$d0,$d0,$t0
+	adde	$d1,$d1,$t1
+
+	mulld	$t0,$h0,$r1		# h0*r1
+	mulhdu	$d2,$h0,$r1
+	addc	$d1,$d1,$t0
+	addze	$d2,$d2
+
+	mulld	$t0,$h1,$r0		# h1*r0
+	mulhdu	$t1,$h1,$r0
+	addc	$d1,$d1,$t0
+	adde	$d2,$d2,$t1
+
+	mulld	$t0,$h2,$s1		# h2*5*r1
+	mulld	$t1,$h2,$r0		# h2*r0
+	addc	$d1,$d1,$t0
+	adde	$d2,$d2,$t1
+
+	andc	$t0,$d2,$mask		# final reduction step
+	and	$h2,$d2,$mask
+	srdi	$t1,$t0,2
+	add	$t0,$t0,$t1
+	addc	$h0,$d0,$t0
+	addze	$h1,$d1
+
+	bdnz	Loop
+
+	std	$h0,0($ctx)		# store hash value
+	std	$h1,8($ctx)
+	std	$h2,16($ctx)
+
+	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
+	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
+	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
+	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
+	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
+	addi	$sp,$sp,$FRAME
+Labort:
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,5,4,0
+.size	.poly1305_blocks,.-.poly1305_blocks
+
+.globl	.poly1305_emit
+.align	4
+.poly1305_emit:
+	ld	$h0,0($ctx)		# load hash
+	ld	$h1,8($ctx)
+	ld	$h2,16($ctx)
+	ld	$padbit,0($nonce)	# load nonce
+	ld	$nonce,8($nonce)
+
+	addic	$d0,$h0,5		# compare to modulus
+	addze	$d1,$h1
+	addze	$d2,$h2
+
+	srdi	$mask,$d2,2		# did it carry/borrow?
+	neg	$mask,$mask
+
+	andc	$h0,$h0,$mask
+	and	$d0,$d0,$mask
+	andc	$h1,$h1,$mask
+	and	$d1,$d1,$mask
+	or	$h0,$h0,$d0
+	or	$h1,$h1,$d1
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	rotldi	$padbit,$padbit,32	# flip nonce words
+	rotldi	$nonce,$nonce,32
+___
+$code.=<<___;
+	addc	$h0,$h0,$padbit		# accumulate nonce
+	adde	$h1,$h1,$nonce
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	std	$h0,0($mac)		# write result
+	std	$h1,8($mac)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	extrdi	r0,$h0,32,0
+	li	$d0,4
+	stwbrx	$h0,0,$mac		# write result
+	extrdi	$h0,$h1,32,0
+	li	$d1,8
+	stwbrx	r0,$d0,$mac
+	li	$d2,12
+	stwbrx	$h1,$d1,$mac
+	stwbrx	$h0,$d2,$mac
+___
+$code.=<<___;
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,3,0
+.size	.poly1305_emit,.-.poly1305_emit
+___
+							} else {
+###############################################################################
+# base 2^32 implementation
+
+my ($h0,$h1,$h2,$h3,$h4, $r0,$r1,$r2,$r3, $s1,$s2,$s3,
+    $t0,$t1,$t2,$t3, $D0,$D1,$D2,$D3, $d0,$d1,$d2,$d3
+   ) = map("r$_",(7..12,14..31));
+
+$code.=<<___;
+.globl	.poly1305_init_int
+.align	4
+.poly1305_init_int:
+	xor	r0,r0,r0
+	stw	r0,0($ctx)		# zero hash value
+	stw	r0,4($ctx)
+	stw	r0,8($ctx)
+	stw	r0,12($ctx)
+	stw	r0,16($ctx)
+
+	$UCMP	$inp,r0
+	beq-	Lno_key
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	lw	$h0,0($inp)		# load key material
+	lw	$h1,4($inp)
+	lw	$h2,8($inp)
+	lw	$h3,12($inp)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	li	$h1,4
+	lwbrx	$h0,0,$inp		# load key material
+	li	$h2,8
+	lwbrx	$h1,$h1,$inp
+	li	$h3,12
+	lwbrx	$h2,$h2,$inp
+	lwbrx	$h3,$h3,$inp
+___
+$code.=<<___;
+	lis	$mask,0xf000		# 0xf0000000
+	li	$r0,-4
+	andc	$r0,$r0,$mask		# 0x0ffffffc
+
+	andc	$h0,$h0,$mask
+	and	$h1,$h1,$r0
+	and	$h2,$h2,$r0
+	and	$h3,$h3,$r0
+
+	stw	$h0,32($ctx)		# store key
+	stw	$h1,36($ctx)
+	stw	$h2,40($ctx)
+	stw	$h3,44($ctx)
+
+Lno_key:
+	xor	r3,r3,r3
+	blr
+	.long	0
+	.byte	0,12,0x14,0,0,0,2,0
+.size	.poly1305_init_int,.-.poly1305_init_int
+
+.globl	.poly1305_blocks
+.align	4
+.poly1305_blocks:
+	srwi.	$len,$len,4
+	beq-	Labort
+
+	$STU	$sp,-$FRAME($sp)
+	mflr	r0
+	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
+	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
+	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
+	$PUSH	r17,`$FRAME-$SIZE_T*15`($sp)
+	$PUSH	r18,`$FRAME-$SIZE_T*14`($sp)
+	$PUSH	r19,`$FRAME-$SIZE_T*13`($sp)
+	$PUSH	r20,`$FRAME-$SIZE_T*12`($sp)
+	$PUSH	r21,`$FRAME-$SIZE_T*11`($sp)
+	$PUSH	r22,`$FRAME-$SIZE_T*10`($sp)
+	$PUSH	r23,`$FRAME-$SIZE_T*9`($sp)
+	$PUSH	r24,`$FRAME-$SIZE_T*8`($sp)
+	$PUSH	r25,`$FRAME-$SIZE_T*7`($sp)
+	$PUSH	r26,`$FRAME-$SIZE_T*6`($sp)
+	$PUSH	r27,`$FRAME-$SIZE_T*5`($sp)
+	$PUSH	r28,`$FRAME-$SIZE_T*4`($sp)
+	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
+	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
+	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
+
+	lwz	$r0,32($ctx)		# load key
+	lwz	$r1,36($ctx)
+	lwz	$r2,40($ctx)
+	lwz	$r3,44($ctx)
+
+	lwz	$h0,0($ctx)		# load hash value
+	lwz	$h1,4($ctx)
+	lwz	$h2,8($ctx)
+	lwz	$h3,12($ctx)
+	lwz	$h4,16($ctx)
+
+	srwi	$s1,$r1,2
+	srwi	$s2,$r2,2
+	srwi	$s3,$r3,2
+	add	$s1,$s1,$r1		# si = ri + ri>>2
+	add	$s2,$s2,$r2
+	add	$s3,$s3,$r3
+	mtctr	$len
+	li	$mask,3
+	b	Loop
+
+.align	4
+Loop:
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	lwz	$d0,0($inp)		# load input
+	lwz	$d1,4($inp)
+	lwz	$d2,8($inp)
+	lwz	$d3,12($inp)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	li	$d1,4
+	lwbrx	$d0,0,$inp		# load input
+	li	$d2,8
+	lwbrx	$d1,$d1,$inp
+	li	$d3,12
+	lwbrx	$d2,$d2,$inp
+	lwbrx	$d3,$d3,$inp
+___
+$code.=<<___;
+	addi	$inp,$inp,16
+
+	addc	$h0,$h0,$d0		# accumulate input
+	adde	$h1,$h1,$d1
+	adde	$h2,$h2,$d2
+
+	mullw	$d0,$h0,$r0		# h0*r0
+	mulhwu	$D0,$h0,$r0
+
+	mullw	$d1,$h0,$r1		# h0*r1
+	mulhwu	$D1,$h0,$r1
+
+	mullw	$d2,$h0,$r2		# h0*r2
+	mulhwu	$D2,$h0,$r2
+
+	 adde	$h3,$h3,$d3
+	 adde	$h4,$h4,$padbit
+
+	mullw	$d3,$h0,$r3		# h0*r3
+	mulhwu	$D3,$h0,$r3
+
+	mullw	$t0,$h1,$s3		# h1*s3
+	mulhwu	$t1,$h1,$s3
+
+	mullw	$t2,$h1,$r0		# h1*r0
+	mulhwu	$t3,$h1,$r0
+	 addc	$d0,$d0,$t0
+	 adde	$D0,$D0,$t1
+
+	mullw	$t0,$h1,$r1		# h1*r1
+	mulhwu	$t1,$h1,$r1
+	 addc	$d1,$d1,$t2
+	 adde	$D1,$D1,$t3
+
+	mullw	$t2,$h1,$r2		# h1*r2
+	mulhwu	$t3,$h1,$r2
+	 addc	$d2,$d2,$t0
+	 adde	$D2,$D2,$t1
+
+	mullw	$t0,$h2,$s2		# h2*s2
+	mulhwu	$t1,$h2,$s2
+	 addc	$d3,$d3,$t2
+	 adde	$D3,$D3,$t3
+
+	mullw	$t2,$h2,$s3		# h2*s3
+	mulhwu	$t3,$h2,$s3
+	 addc	$d0,$d0,$t0
+	 adde	$D0,$D0,$t1
+
+	mullw	$t0,$h2,$r0		# h2*r0
+	mulhwu	$t1,$h2,$r0
+	 addc	$d1,$d1,$t2
+	 adde	$D1,$D1,$t3
+
+	mullw	$t2,$h2,$r1		# h2*r1
+	mulhwu	$t3,$h2,$r1
+	 addc	$d2,$d2,$t0
+	 adde	$D2,$D2,$t1
+
+	mullw	$t0,$h3,$s1		# h3*s1
+	mulhwu	$t1,$h3,$s1
+	 addc	$d3,$d3,$t2
+	 adde	$D3,$D3,$t3
+
+	mullw	$t2,$h3,$s2		# h3*s2
+	mulhwu	$t3,$h3,$s2
+	 addc	$d0,$d0,$t0
+	 adde	$D0,$D0,$t1
+
+	mullw	$t0,$h3,$s3		# h3*s3
+	mulhwu	$t1,$h3,$s3
+	 addc	$d1,$d1,$t2
+	 adde	$D1,$D1,$t3
+
+	mullw	$t2,$h3,$r0		# h3*r0
+	mulhwu	$t3,$h3,$r0
+	 addc	$d2,$d2,$t0
+	 adde	$D2,$D2,$t1
+
+	mullw	$t0,$h4,$s1		# h4*s1
+	 addc	$d3,$d3,$t2
+	 adde	$D3,$D3,$t3
+	addc	$d1,$d1,$t0
+
+	mullw	$t1,$h4,$s2		# h4*s2
+	 addze	$D1,$D1
+	addc	$d2,$d2,$t1
+	addze	$D2,$D2
+
+	mullw	$t2,$h4,$s3		# h4*s3
+	addc	$d3,$d3,$t2
+	addze	$D3,$D3
+
+	mullw	$h4,$h4,$r0		# h4*r0
+
+	addc	$h1,$d1,$D0
+	adde	$h2,$d2,$D1
+	adde	$h3,$d3,$D2
+	adde	$h4,$h4,$D3
+
+	andc	$D0,$h4,$mask		# final reduction step
+	and	$h4,$h4,$mask
+	srwi	$D1,$D0,2
+	add	$D0,$D0,$D1
+	addc	$h0,$d0,$D0
+	addze	$h1,$h1
+	addze	$h2,$h2
+	addze	$h3,$h3
+
+	bdnz	Loop
+
+	stw	$h0,0($ctx)		# store hash value
+	stw	$h1,4($ctx)
+	stw	$h2,8($ctx)
+	stw	$h3,12($ctx)
+	stw	$h4,16($ctx)
+
+	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
+	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
+	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
+	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
+	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
+	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
+	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
+	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
+	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
+	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
+	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
+	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
+	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
+	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
+	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
+	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
+	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
+	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
+	addi	$sp,$sp,$FRAME
+Labort:
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,18,4,0
+.size	.poly1305_blocks,.-.poly1305_blocks
+
+.globl	.poly1305_emit
+.align	4
+.poly1305_emit:
+	$STU	$sp,-$FRAME($sp)
+	mflr	r0
+	$PUSH	r28,`$FRAME-$SIZE_T*4`($sp)
+	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
+	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
+	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
+
+	lwz	$h0,0($ctx)		# load hash
+	lwz	$h1,4($ctx)
+	lwz	$h2,8($ctx)
+	lwz	$h3,12($ctx)
+	lwz	$h4,16($ctx)
+
+	addic	$d0,$h0,5		# compare to modulus
+	addze	$d1,$h1
+	addze	$d2,$h2
+	addze	$d3,$h3
+	addze	$mask,$h4
+
+	srwi	$mask,$mask,2		# did it carry/borrow?
+	neg	$mask,$mask
+
+	andc	$h0,$h0,$mask
+	and	$d0,$d0,$mask
+	andc	$h1,$h1,$mask
+	and	$d1,$d1,$mask
+	or	$h0,$h0,$d0
+	lwz	$d0,0($nonce)		# load nonce
+	andc	$h2,$h2,$mask
+	and	$d2,$d2,$mask
+	or	$h1,$h1,$d1
+	lwz	$d1,4($nonce)
+	andc	$h3,$h3,$mask
+	and	$d3,$d3,$mask
+	or	$h2,$h2,$d2
+	lwz	$d2,8($nonce)
+	or	$h3,$h3,$d3
+	lwz	$d3,12($nonce)
+
+	addc	$h0,$h0,$d0		# accumulate nonce
+	adde	$h1,$h1,$d1
+	adde	$h2,$h2,$d2
+	adde	$h3,$h3,$d3
+___
+$code.=<<___	if ($LITTLE_ENDIAN);
+	stw	$h0,0($mac)		# write result
+	stw	$h1,4($mac)
+	stw	$h2,8($mac)
+	stw	$h3,12($mac)
+___
+$code.=<<___	if (!$LITTLE_ENDIAN);
+	li	$d1,4
+	stwbrx	$h0,0,$mac		# write result
+	li	$d2,8
+	stwbrx	$h1,$d1,$mac
+	li	$d3,12
+	stwbrx	$h2,$d2,$mac
+	stwbrx	$h3,$d3,$mac
+___
+$code.=<<___;
+	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
+	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
+	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
+	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
+	addi	$sp,$sp,$FRAME
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,4,3,0
+.size	.poly1305_emit,.-.poly1305_emit
+___
+							}
+$code.=<<___;
+.asciz	"Poly1305 for PPC, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval $1/gem;
+print $code;
+close STDOUT;
diff --git a/crypto/poly1305/asm/poly1305-ppcfp.pl b/crypto/poly1305/asm/poly1305-ppcfp.pl
new file mode 100755
index 0000000000..061a556377
--- /dev/null
+++ b/crypto/poly1305/asm/poly1305-ppcfp.pl
@@ -0,0 +1,732 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# This module implements Poly1305 hash for PowerPC FPU.
+#
+# June 2015
+#
+# Numbers are cycles per processed byte with poly1305_blocks alone,
+# and improvement coefficients relative to gcc-generated code.
+#
+# Freescale e300	9.78/+30%
+# PPC74x0		7.08/+50%
+# PPC970		6.24/+80%
+# POWER7		3.50/+30%
+# POWER8		3.75/+10%
+
+$flavour = shift;
+
+if ($flavour =~ /64/) {
+	$SIZE_T	=8;
+	$LRSAVE	=2*$SIZE_T;
+	$UCMP	="cmpld";
+	$STU	="stdu";
+	$POP	="ld";
+	$PUSH	="std";
+} elsif ($flavour =~ /32/) {
+	$SIZE_T	=4;
+	$LRSAVE	=$SIZE_T;
+	$UCMP	="cmplw";
+	$STU	="stwu";
+	$POP	="lwz";
+	$PUSH	="stw";
+} else { die "nonsense $flavour"; }
+
+$LITTLE_ENDIAN = ($flavour=~/le$/) ? 4 : 0;
+
+$LWXLE = $LITTLE_ENDIAN ? "lwzx" : "lwbrx";
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
+die "can't locate ppc-xlate.pl";
+
+open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
+
+$LOCALS=6*$SIZE_T;
+$FRAME=$LOCALS+6*8+18*8;
+
+my $sp="r1";
+
+my ($ctx,$inp,$len,$padbit) = map("r$_",(3..6));
+my ($in0,$in1,$in2,$in3,$i1,$i2,$i3) = map("r$_",(7..12,6));
+
+my ($h0lo,$h0hi,$h1lo,$h1hi,$h2lo,$h2hi,$h3lo,$h3hi,
+    $two0,$two32,$two64,$two96,$two130,$five_two130,
+    $r0lo,$r0hi,$r1lo,$r1hi,$r2lo,$r2hi,
+    $s2lo,$s2hi,$s3lo,$s3hi,
+    $c0lo,$c0hi,$c1lo,$c1hi,$c2lo,$c2hi,$c3lo,$c3hi) = map("f$_",(0..31));
+# borrowings
+my ($r3lo,$r3hi,$s1lo,$s1hi) = ($c0lo,$c0hi,$c1lo,$c1hi);
+my ($x0,$x1,$x2,$x3) = ($c2lo,$c2hi,$c3lo,$c3hi);
+my ($y0,$y1,$y2,$y3) = ($c3lo,$c3hi,$c1lo,$c1hi);
+
+$code.=<<___;
+.machine	"any"
+.text
+
+.globl	.poly1305_init_fpu
+.align	6
+.poly1305_init_fpu:
+	$STU	$sp,-$LOCALS($sp)		# minimal frame
+	mflr	$padbit
+	$PUSH	$padbit,`$LOCALS+$LRSAVE`($sp)
+
+	bl	LPICmeup
+
+	xor	r0,r0,r0
+	mtlr	$padbit				# restore lr
+
+	lfd	$two0,8*0($len)			# load constants
+	lfd	$two32,8*1($len)
+	lfd	$two64,8*2($len)
+	lfd	$two96,8*3($len)
+	lfd	$two130,8*4($len)
+	lfd	$five_two130,8*5($len)
+
+	stfd	$two0,8*0($ctx)			# initial hash value, biased 0
+	stfd	$two32,8*1($ctx)
+	stfd	$two64,8*2($ctx)
+	stfd	$two96,8*3($ctx)
+
+	$UCMP	$inp,r0
+	beq-	Lno_key
+
+	lfd	$h3lo,8*13($len)		# new fpscr
+	mffs	$h3hi				# old fpscr
+
+	stfd	$two0,8*4($ctx)			# key "template"
+	stfd	$two32,8*5($ctx)
+	stfd	$two64,8*6($ctx)
+	stfd	$two96,8*7($ctx)
+
+	li	$in1,4
+	li	$in2,8
+	li	$in3,12
+	$LWXLE	$in0,0,$inp			# load key
+	$LWXLE	$in1,$in1,$inp
+	$LWXLE	$in2,$in2,$inp
+	$LWXLE	$in3,$in3,$inp
+
+	lis	$i1,0xf000			#   0xf0000000
+	ori	$i2,$i1,3			#   0xf0000003
+	andc	$in0,$in0,$i1			# &=0x0fffffff
+	andc	$in1,$in1,$i2			# &=0x0ffffffc
+	andc	$in2,$in2,$i2
+	andc	$in3,$in3,$i2
+
+	stw	$in0,`8*4+(4^$LITTLE_ENDIAN)`($ctx)	# fill "template"
+	stw	$in1,`8*5+(4^$LITTLE_ENDIAN)`($ctx)
+	stw	$in2,`8*6+(4^$LITTLE_ENDIAN)`($ctx)
+	stw	$in3,`8*7+(4^$LITTLE_ENDIAN)`($ctx)
+
+	mtfsf	255,$h3lo			# fpscr
+	stfd	$two0,8*18($ctx)		# copy constants to context
+	stfd	$two32,8*19($ctx)
+	stfd	$two64,8*20($ctx)
+	stfd	$two96,8*21($ctx)
+	stfd	$two130,8*22($ctx)
+	stfd	$five_two130,8*23($ctx)
+
+	lfd	$h0lo,8*4($ctx)			# load [biased] key
+	lfd	$h1lo,8*5($ctx)
+	lfd	$h2lo,8*6($ctx)
+	lfd	$h3lo,8*7($ctx)
+
+	fsub	$h0lo,$h0lo,$two0		# r0
+	fsub	$h1lo,$h1lo,$two32		# r1
+	fsub	$h2lo,$h2lo,$two64		# r2
+	fsub	$h3lo,$h3lo,$two96		# r3
+
+	lfd	$two0,8*6($len)			# more constants
+	lfd	$two32,8*7($len)
+	lfd	$two64,8*8($len)
+	lfd	$two96,8*9($len)
+
+	fmul	$h1hi,$h1lo,$five_two130	# s1
+	fmul	$h2hi,$h2lo,$five_two130	# s2
+	 stfd	$h3hi,8*15($ctx)		# borrow slot for original fpscr
+	fmul	$h3hi,$h3lo,$five_two130	# s3
+
+	fadd	$h0hi,$h0lo,$two0
+	 stfd	$h1hi,8*12($ctx)		# put aside for now
+	fadd	$h1hi,$h1lo,$two32
+	 stfd	$h2hi,8*13($ctx)
+	fadd	$h2hi,$h2lo,$two64
+	 stfd	$h3hi,8*14($ctx)
+	fadd	$h3hi,$h3lo,$two96
+
+	fsub	$h0hi,$h0hi,$two0
+	fsub	$h1hi,$h1hi,$two32
+	fsub	$h2hi,$h2hi,$two64
+	fsub	$h3hi,$h3hi,$two96
+
+	lfd	$two0,8*10($len)		# more constants
+	lfd	$two32,8*11($len)
+	lfd	$two64,8*12($len)
+
+	fsub	$h0lo,$h0lo,$h0hi
+	fsub	$h1lo,$h1lo,$h1hi
+	fsub	$h2lo,$h2lo,$h2hi
+	fsub	$h3lo,$h3lo,$h3hi
+
+	stfd	$h0hi,8*5($ctx)			# r0hi
+	stfd	$h1hi,8*7($ctx)			# r1hi
+	stfd	$h2hi,8*9($ctx)			# r2hi
+	stfd	$h3hi,8*11($ctx)		# r3hi
+
+	stfd	$h0lo,8*4($ctx)			# r0lo
+	stfd	$h1lo,8*6($ctx)			# r1lo
+	stfd	$h2lo,8*8($ctx)			# r2lo
+	stfd	$h3lo,8*10($ctx)		# r3lo
+
+	lfd	$h1lo,8*12($ctx)		# s1
+	lfd	$h2lo,8*13($ctx)		# s2
+	lfd	$h3lo,8*14($ctx)		# s3
+	lfd	$h0lo,8*15($ctx)		# pull original fpscr
+
+	fadd	$h1hi,$h1lo,$two0
+	fadd	$h2hi,$h2lo,$two32
+	fadd	$h3hi,$h3lo,$two64
+
+	fsub	$h1hi,$h1hi,$two0
+	fsub	$h2hi,$h2hi,$two32
+	fsub	$h3hi,$h3hi,$two64
+
+	fsub	$h1lo,$h1lo,$h1hi
+	fsub	$h2lo,$h2lo,$h2hi
+	fsub	$h3lo,$h3lo,$h3hi
+
+	stfd	$h1hi,8*13($ctx)		# s1hi
+	stfd	$h2hi,8*15($ctx)		# s2hi
+	stfd	$h3hi,8*17($ctx)		# s3hi
+
+	stfd	$h1lo,8*12($ctx)		# s1lo
+	stfd	$h2lo,8*14($ctx)		# s2lo
+	stfd	$h3lo,8*16($ctx)		# s3lo
+
+	mtfsf	255,$h0lo			# restore fpscr
+Lno_key:
+	xor	r3,r3,r3
+	addi	$sp,$sp,$LOCALS
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,0,2,0
+.size	.poly1305_init_fpu,.-.poly1305_init_fpu
+
+.globl	.poly1305_blocks_fpu
+.align	4
+.poly1305_blocks_fpu:
+	srwi.	$len,$len,4
+	beq-	Labort
+
+	$STU	$sp,-$FRAME($sp)
+	mflr	r0
+	stfd	f14,`$FRAME-8*18`($sp)
+	stfd	f15,`$FRAME-8*17`($sp)
+	stfd	f16,`$FRAME-8*16`($sp)
+	stfd	f17,`$FRAME-8*15`($sp)
+	stfd	f18,`$FRAME-8*14`($sp)
+	stfd	f19,`$FRAME-8*13`($sp)
+	stfd	f20,`$FRAME-8*12`($sp)
+	stfd	f21,`$FRAME-8*11`($sp)
+	stfd	f22,`$FRAME-8*10`($sp)
+	stfd	f23,`$FRAME-8*9`($sp)
+	stfd	f24,`$FRAME-8*8`($sp)
+	stfd	f25,`$FRAME-8*7`($sp)
+	stfd	f26,`$FRAME-8*6`($sp)
+	stfd	f27,`$FRAME-8*5`($sp)
+	stfd	f28,`$FRAME-8*4`($sp)
+	stfd	f29,`$FRAME-8*3`($sp)
+	stfd	f30,`$FRAME-8*2`($sp)
+	stfd	f31,`$FRAME-8*1`($sp)
+	$PUSH	r0,`$FRAME+$LRSAVE`($sp)
+
+	xor	r0,r0,r0
+	li	$in3,1
+	mtctr	$len
+	neg	$len,$len
+	stw	r0,`$LOCALS+8*4+(0^$LITTLE_ENDIAN)`($sp)
+	stw	$in3,`$LOCALS+8*4+(4^$LITTLE_ENDIAN)`($sp)
+
+	lfd	$two0,8*18($ctx)		# load constants
+	lfd	$two32,8*19($ctx)
+	lfd	$two64,8*20($ctx)
+	lfd	$two96,8*21($ctx)
+	lfd	$two130,8*22($ctx)
+	lfd	$five_two130,8*23($ctx)
+
+	lfd	$h0lo,8*0($ctx)			# load [biased] hash value
+	lfd	$h1lo,8*1($ctx)
+	lfd	$h2lo,8*2($ctx)
+	lfd	$h3lo,8*3($ctx)
+
+	stfd	$two0,`$LOCALS+8*0`($sp)	# input "template"
+	oris	$in3,$padbit,`(1023+52+96)<<4`
+	stfd	$two32,`$LOCALS+8*1`($sp)
+	stfd	$two64,`$LOCALS+8*2`($sp)
+	stw	$in3,`$LOCALS+8*3+(0^$LITTLE_ENDIAN)`($sp)
+
+	li	$i1,4
+	li	$i2,8
+	li	$i3,12
+	$LWXLE	$in0,0,$inp			# load input
+	$LWXLE	$in1,$i1,$inp
+	$LWXLE	$in2,$i2,$inp
+	$LWXLE	$in3,$i3,$inp
+	addi	$inp,$inp,16
+
+	stw	$in0,`$LOCALS+8*0+(4^$LITTLE_ENDIAN)`($sp)	# fill "template"
+	stw	$in1,`$LOCALS+8*1+(4^$LITTLE_ENDIAN)`($sp)
+	stw	$in2,`$LOCALS+8*2+(4^$LITTLE_ENDIAN)`($sp)
+	stw	$in3,`$LOCALS+8*3+(4^$LITTLE_ENDIAN)`($sp)
+
+	mffs	$x0				# original fpscr
+	lfd	$x1,`$LOCALS+8*4`($sp)		# new fpscr
+	lfd	$r0lo,8*4($ctx)			# load key
+	lfd	$r0hi,8*5($ctx)
+	lfd	$r1lo,8*6($ctx)
+	lfd	$r1hi,8*7($ctx)
+	lfd	$r2lo,8*8($ctx)
+	lfd	$r2hi,8*9($ctx)
+	lfd	$r3lo,8*10($ctx)
+	lfd	$r3hi,8*11($ctx)
+	lfd	$s1lo,8*12($ctx)
+	lfd	$s1hi,8*13($ctx)
+	lfd	$s2lo,8*14($ctx)
+	lfd	$s2hi,8*15($ctx)
+	lfd	$s3lo,8*16($ctx)
+	lfd	$s3hi,8*17($ctx)
+
+	stfd	$x0,`$LOCALS+8*4`($sp)		# save original fpscr
+	mtfsf	255,$x1
+
+	addic	$len,$len,1
+	addze	r0,r0
+	slwi.	r0,r0,4
+	sub	$inp,$inp,r0			# conditional rewind
+
+	lfd	$x0,`$LOCALS+8*0`($sp)
+	lfd	$x1,`$LOCALS+8*1`($sp)
+	lfd	$x2,`$LOCALS+8*2`($sp)
+	lfd	$x3,`$LOCALS+8*3`($sp)
+
+	fsub	$h0lo,$h0lo,$two0		# de-bias hash value
+	 $LWXLE	$in0,0,$inp			# modulo-scheduled input load
+	fsub	$h1lo,$h1lo,$two32
+	 $LWXLE	$in1,$i1,$inp
+	fsub	$h2lo,$h2lo,$two64
+	 $LWXLE	$in2,$i2,$inp
+	fsub	$h3lo,$h3lo,$two96
+	 $LWXLE	$in3,$i3,$inp
+
+	fsub	$x0,$x0,$two0			# de-bias input
+	 addi	$inp,$inp,16
+	fsub	$x1,$x1,$two32
+	fsub	$x2,$x2,$two64
+	fsub	$x3,$x3,$two96
+
+	fadd	$x0,$x0,$h0lo			# accumulate input
+	 stw	$in0,`$LOCALS+8*0+(4^$LITTLE_ENDIAN)`($sp)
+	fadd	$x1,$x1,$h1lo
+	 stw	$in1,`$LOCALS+8*1+(4^$LITTLE_ENDIAN)`($sp)
+	fadd	$x2,$x2,$h2lo
+	 stw	$in2,`$LOCALS+8*2+(4^$LITTLE_ENDIAN)`($sp)
+	fadd	$x3,$x3,$h3lo
+	 stw	$in3,`$LOCALS+8*3+(4^$LITTLE_ENDIAN)`($sp)
+
+	b	Lentry
+
+.align	4
+Loop:
+	fsub	$y0,$y0,$two0			# de-bias input
+	 addic	$len,$len,1
+	fsub	$y1,$y1,$two32
+	 addze	r0,r0
+	fsub	$y2,$y2,$two64
+	 slwi.	r0,r0,4
+	fsub	$y3,$y3,$two96
+	 sub	$inp,$inp,r0			# conditional rewind
+
+	fadd	$h0lo,$h0lo,$y0			# accumulate input
+	fadd	$h0hi,$h0hi,$y1
+	fadd	$h2lo,$h2lo,$y2
+	fadd	$h2hi,$h2hi,$y3
+
+	######################################### base 2^48 -> base 2^32
+	fadd	$c1lo,$h1lo,$two64
+	 $LWXLE	$in0,0,$inp			# modulo-scheduled input load
+	fadd	$c1hi,$h1hi,$two64
+	 $LWXLE	$in1,$i1,$inp
+	fadd	$c3lo,$h3lo,$two130
+	 $LWXLE	$in2,$i2,$inp
+	fadd	$c3hi,$h3hi,$two130
+	 $LWXLE	$in3,$i3,$inp
+	fadd	$c0lo,$h0lo,$two32
+	 addi	$inp,$inp,16
+	fadd	$c0hi,$h0hi,$two32
+	fadd	$c2lo,$h2lo,$two96
+	fadd	$c2hi,$h2hi,$two96
+
+	fsub	$c1lo,$c1lo,$two64
+	 stw	$in0,`$LOCALS+8*0+(4^$LITTLE_ENDIAN)`($sp)	# fill "template"
+	fsub	$c1hi,$c1hi,$two64
+	 stw	$in1,`$LOCALS+8*1+(4^$LITTLE_ENDIAN)`($sp)
+	fsub	$c3lo,$c3lo,$two130
+	 stw	$in2,`$LOCALS+8*2+(4^$LITTLE_ENDIAN)`($sp)
+	fsub	$c3hi,$c3hi,$two130
+	 stw	$in3,`$LOCALS+8*3+(4^$LITTLE_ENDIAN)`($sp)
+	fsub	$c0lo,$c0lo,$two32
+	fsub	$c0hi,$c0hi,$two32
+	fsub	$c2lo,$c2lo,$two96
+	fsub	$c2hi,$c2hi,$two96
+
+	fsub	$h1lo,$h1lo,$c1lo
+	fsub	$h1hi,$h1hi,$c1hi
+	fsub	$h3lo,$h3lo,$c3lo
+	fsub	$h3hi,$h3hi,$c3hi
+	fsub	$h2lo,$h2lo,$c2lo
+	fsub	$h2hi,$h2hi,$c2hi
+	fsub	$h0lo,$h0lo,$c0lo
+	fsub	$h0hi,$h0hi,$c0hi
+
+	fadd	$h1lo,$h1lo,$c0lo
+	fadd	$h1hi,$h1hi,$c0hi
+	fadd	$h3lo,$h3lo,$c2lo
+	fadd	$h3hi,$h3hi,$c2hi
+	fadd	$h2lo,$h2lo,$c1lo
+	fadd	$h2hi,$h2hi,$c1hi
+	fmadd	$h0lo,$c3lo,$five_two130,$h0lo
+	fmadd	$h0hi,$c3hi,$five_two130,$h0hi
+
+	fadd	$x1,$h1lo,$h1hi
+	 lfd	$s1lo,8*12($ctx)		# reload constants
+	fadd	$x3,$h3lo,$h3hi
+	 lfd	$s1hi,8*13($ctx)
+	fadd	$x2,$h2lo,$h2hi
+	 lfd	$r3lo,8*10($ctx)
+	fadd	$x0,$h0lo,$h0hi
+	 lfd	$r3hi,8*11($ctx)
+Lentry:
+	fmul	$h0lo,$s3lo,$x1
+	fmul	$h0hi,$s3hi,$x1
+	fmul	$h2lo,$r1lo,$x1
+	fmul	$h2hi,$r1hi,$x1
+	fmul	$h1lo,$r0lo,$x1
+	fmul	$h1hi,$r0hi,$x1
+	fmul	$h3lo,$r2lo,$x1
+	fmul	$h3hi,$r2hi,$x1
+
+	fmadd	$h0lo,$s1lo,$x3,$h0lo
+	fmadd	$h0hi,$s1hi,$x3,$h0hi
+	fmadd	$h2lo,$s3lo,$x3,$h2lo
+	fmadd	$h2hi,$s3hi,$x3,$h2hi
+	fmadd	$h1lo,$s2lo,$x3,$h1lo
+	fmadd	$h1hi,$s2hi,$x3,$h1hi
+	fmadd	$h3lo,$r0lo,$x3,$h3lo
+	fmadd	$h3hi,$r0hi,$x3,$h3hi
+
+	fmadd	$h0lo,$s2lo,$x2,$h0lo
+	fmadd	$h0hi,$s2hi,$x2,$h0hi
+	fmadd	$h2lo,$r0lo,$x2,$h2lo
+	fmadd	$h2hi,$r0hi,$x2,$h2hi
+	fmadd	$h1lo,$s3lo,$x2,$h1lo
+	fmadd	$h1hi,$s3hi,$x2,$h1hi
+	fmadd	$h3lo,$r1lo,$x2,$h3lo
+	fmadd	$h3hi,$r1hi,$x2,$h3hi
+
+	fmadd	$h0lo,$r0lo,$x0,$h0lo
+	 lfd	$y0,`$LOCALS+8*0`($sp)		# load [biased] input
+	fmadd	$h0hi,$r0hi,$x0,$h0hi
+	 lfd	$y1,`$LOCALS+8*1`($sp)
+	fmadd	$h2lo,$r2lo,$x0,$h2lo
+	 lfd	$y2,`$LOCALS+8*2`($sp)
+	fmadd	$h2hi,$r2hi,$x0,$h2hi
+	 lfd	$y3,`$LOCALS+8*3`($sp)
+	fmadd	$h1lo,$r1lo,$x0,$h1lo
+	fmadd	$h1hi,$r1hi,$x0,$h1hi
+	fmadd	$h3lo,$r3lo,$x0,$h3lo
+	fmadd	$h3hi,$r3hi,$x0,$h3hi
+
+	bdnz	Loop
+
+	######################################### base 2^48 -> base 2^32
+	fadd	$c0lo,$h0lo,$two32
+	fadd	$c0hi,$h0hi,$two32
+	fadd	$c2lo,$h2lo,$two96
+	fadd	$c2hi,$h2hi,$two96
+	fadd	$c1lo,$h1lo,$two64
+	fadd	$c1hi,$h1hi,$two64
+	fadd	$c3lo,$h3lo,$two130
+	fadd	$c3hi,$h3hi,$two130
+
+	fsub	$c0lo,$c0lo,$two32
+	fsub	$c0hi,$c0hi,$two32
+	fsub	$c2lo,$c2lo,$two96
+	fsub	$c2hi,$c2hi,$two96
+	fsub	$c1lo,$c1lo,$two64
+	fsub	$c1hi,$c1hi,$two64
+	fsub	$c3lo,$c3lo,$two130
+	fsub	$c3hi,$c3hi,$two130
+
+	fsub	$h1lo,$h1lo,$c1lo
+	fsub	$h1hi,$h1hi,$c1hi
+	fsub	$h3lo,$h3lo,$c3lo
+	fsub	$h3hi,$h3hi,$c3hi
+	fsub	$h2lo,$h2lo,$c2lo
+	fsub	$h2hi,$h2hi,$c2hi
+	fsub	$h0lo,$h0lo,$c0lo
+	fsub	$h0hi,$h0hi,$c0hi
+
+	fadd	$h1lo,$h1lo,$c0lo
+	fadd	$h1hi,$h1hi,$c0hi
+	fadd	$h3lo,$h3lo,$c2lo
+	fadd	$h3hi,$h3hi,$c2hi
+	fadd	$h2lo,$h2lo,$c1lo
+	fadd	$h2hi,$h2hi,$c1hi
+	fmadd	$h0lo,$c3lo,$five_two130,$h0lo
+	fmadd	$h0hi,$c3hi,$five_two130,$h0hi
+
+	fadd	$x1,$h1lo,$h1hi
+	fadd	$x3,$h3lo,$h3hi
+	fadd	$x2,$h2lo,$h2hi
+	fadd	$x0,$h0lo,$h0hi
+
+	lfd	$h0lo,`$LOCALS+8*4`($sp)	# pull saved fpscr
+	fadd	$x1,$x1,$two32			# bias
+	fadd	$x3,$x3,$two96
+	fadd	$x2,$x2,$two64
+	fadd	$x0,$x0,$two0
+
+	stfd	$x1,8*1($ctx)			# store [biased] hash value
+	stfd	$x3,8*3($ctx)
+	stfd	$x2,8*2($ctx)
+	stfd	$x0,8*0($ctx)
+
+	mtfsf	255,$h0lo			# restore original fpscr
+	lfd	f14,`$FRAME-8*18`($sp)
+	lfd	f15,`$FRAME-8*17`($sp)
+	lfd	f16,`$FRAME-8*16`($sp)
+	lfd	f17,`$FRAME-8*15`($sp)
+	lfd	f18,`$FRAME-8*14`($sp)
+	lfd	f19,`$FRAME-8*13`($sp)
+	lfd	f20,`$FRAME-8*12`($sp)
+	lfd	f21,`$FRAME-8*11`($sp)
+	lfd	f22,`$FRAME-8*10`($sp)
+	lfd	f23,`$FRAME-8*9`($sp)
+	lfd	f24,`$FRAME-8*8`($sp)
+	lfd	f25,`$FRAME-8*7`($sp)
+	lfd	f26,`$FRAME-8*6`($sp)
+	lfd	f27,`$FRAME-8*5`($sp)
+	lfd	f28,`$FRAME-8*4`($sp)
+	lfd	f29,`$FRAME-8*3`($sp)
+	lfd	f30,`$FRAME-8*2`($sp)
+	lfd	f31,`$FRAME-8*1`($sp)
+	addi	$sp,$sp,$FRAME
+Labort:
+	blr
+	.long	0
+	.byte	0,12,4,1,0x80,0,4,0
+.size	.po