Add ec/asm/ecp_nistz256-sparcv9.pl.

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

2 files changed, 3048 insertions, 0 deletions

diff --git a/crypto/ec/Makefile b/crypto/ec/Makefile
index fa2fc4cbb2..423f60bb54 100644
--- a/crypto/ec/Makefile
+++ b/crypto/ec/Makefile
@@ -54,6 +54,9 @@ ecp_nistz256-x86_64.s: asm/ecp_nistz256-x86_64.pl
 ecp_nistz256-avx2.s:   asm/ecp_nistz256-avx2.pl
 	$(PERL) asm/ecp_nistz256-avx2.pl $(PERLASM_SCHEME) > $@
 
+ecp_nistz256-sparcv9.S:	asm/ecp_nistz256-sparcv9.pl
+	$(PERL) asm/ecp_nistz256-sparcv9.pl $(CFLAGS) > $@
+
 ecp_nistz256-%.S:	asm/ecp_nistz256-%.pl;  $(PERL) $< $(PERLASM_SCHEME) $@
 ecp_nistz256-armv4.o:	ecp_nistz256-armv4.S
 ecp_nistz256-armv8.o:	ecp_nistz256-armv8.S
diff --git a/crypto/ec/asm/ecp_nistz256-sparcv9.pl b/crypto/ec/asm/ecp_nistz256-sparcv9.pl
new file mode 100755
index 0000000000..5693b75e5b
--- /dev/null
+++ b/crypto/ec/asm/ecp_nistz256-sparcv9.pl
@@ -0,0 +1,3045 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# ECP_NISTZ256 module for SPARCv9.
+#
+# February 2015.
+#
+# Original ECP_NISTZ256 submission targeting x86_64 is detailed in
+# http://eprint.iacr.org/2013/816. In the process of adaptation
+# original .c module was made 32-bit savvy in order to make this
+# implementation possible.
+#
+#			with/without -DECP_NISTZ256_ASM
+# UltraSPARC III	+12-18%
+# SPARC T4		+99-550% (+66-150% on 32-bit Solaris)
+#
+# Ranges denote minimum and maximum improvement coefficients depending
+# on benchmark. Lower coefficients are for ECDSA sign, server-side
+# operation. Keep in mind that +200% means 3x improvement.
+
+$code.=<<___;
+#include "sparc_arch.h"
+
+#define LOCALS	(STACK_BIAS+STACK_FRAME)
+#ifdef	__arch64__
+.register	%g2,#scratch
+.register	%g3,#scratch
+# define STACK64_FRAME	STACK_FRAME
+# define LOCALS64	LOCALS
+#else
+# define STACK64_FRAME	(2047+192)
+# define LOCALS64	STACK64_FRAME
+#endif
+
+.section	".text",#alloc,#execinstr
+___
+########################################################################
+# Convert ecp_nistz256_table.c to layout expected by ecp_nistz_gather_w7
+#
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+open TABLE,"<ecp_nistz256_table.c"		or
+open TABLE,"<${dir}../ecp_nistz256_table.c"	or
+die "failed to open ecp_nistz256_table.c:",$!;
+
+use integer;
+
+foreach(<TABLE>) {
+	s/TOBN\(\s*(0x[0-9a-f]+),\s*(0x[0-9a-f]+)\s*\)/push @arr,hex($2),hex($1)/geo;
+}
+close TABLE;
+
+# See ecp_nistz256_table.c for explanation for why it's 64*16*37.
+# 64*16*37-1 is because $#arr returns last valid index or @arr, not
+# amount of elements.
+die "insane number of elements" if ($#arr != 64*16*37-1);
+
+$code.=<<___;
+.globl	ecp_nistz256_precomputed
+.align	4096
+ecp_nistz256_precomputed:
+___
+########################################################################
+# this conversion smashes P256_POINT_AFFINE by individual bytes with
+# 64 byte interval, similar to
+#	1111222233334444
+#	1234123412341234
+for(1..37) {
+	@tbl = splice(@arr,0,64*16);
+	for($i=0;$i<64;$i++) {
+		undef @line;
+		for($j=0;$j<64;$j++) {
+			push @line,(@tbl[$j*16+$i/4]>>(($i%4)*8))&0xff;
+		}
+		$code.=".byte\t";
+		$code.=join(',',map { sprintf "0x%02x",$_} @line);
+		$code.="\n";
+	}
+}
+
+{{{
+my ($rp,$ap,$bp)=map("%i$_",(0..2));
+my @acc=map("%l$_",(0..7));
+my ($t0,$t1,$t2,$t3,$t4,$t5,$t6,$t7)=(map("%o$_",(0..5)),"%g4","%g5");
+my ($bi,$a0,$mask,$carry)=(map("%i$_",(3..5)),"%g1");
+my ($rp_real,$ap_real)=("%g2","%g3");
+
+$code.=<<___;
+.size	ecp_nistz256_precomputed,.-ecp_nistz256_precomputed
+.align	64
+.LRR:	! 2^512 mod P precomputed for NIST P256 polynomial
+.long	0x00000003, 0x00000000, 0xffffffff, 0xfffffffb
+.long	0xfffffffe, 0xffffffff, 0xfffffffd, 0x00000004
+.Lone:
+.long	1,0,0,0,0,0,0,0
+.asciz	"ECP_NISTZ256 for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
+
+! void	ecp_nistz256_to_mont(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_to_mont
+.align	64
+ecp_nistz256_to_mont:
+	save	%sp,-STACK_FRAME,%sp
+	nop
+1:	call	.+8
+	add	%o7,.LRR-1b,$bp
+	call	__ecp_nistz256_mul_mont
+	nop
+	ret
+	restore
+.size	ecp_nistz256_to_mont,.-ecp_nistz256_to_mont
+
+! void	ecp_nistz256_from_mont(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_from_mont
+.align	32
+ecp_nistz256_from_mont:
+	save	%sp,-STACK_FRAME,%sp
+	nop
+1:	call	.+8
+	add	%o7,.Lone-1b,$bp
+	call	__ecp_nistz256_mul_mont
+	nop
+	ret
+	restore
+.size	ecp_nistz256_from_mont,.-ecp_nistz256_from_mont
+
+! void	ecp_nistz256_mul_mont(BN_ULONG %i0[8],const BN_ULONG %i1[8],
+!					      const BN_ULONG %i2[8]);
+.globl	ecp_nistz256_mul_mont
+.align	32
+ecp_nistz256_mul_mont:
+	save	%sp,-STACK_FRAME,%sp
+	nop
+	call	__ecp_nistz256_mul_mont
+	nop
+	ret
+	restore
+.size	ecp_nistz256_mul_mont,.-ecp_nistz256_mul_mont
+
+! void	ecp_nistz256_sqr_mont(BN_ULONG %i0[8],const BN_ULONG %i2[8]);
+.globl	ecp_nistz256_sqr_mont
+.align	32
+ecp_nistz256_sqr_mont:
+	save	%sp,-STACK_FRAME,%sp
+	mov	$ap,$bp
+	call	__ecp_nistz256_mul_mont
+	nop
+	ret
+	restore
+.size	ecp_nistz256_sqr_mont,.-ecp_nistz256_sqr_mont
+___
+
+########################################################################
+# Special thing to keep in mind is that $t0-$t7 hold 64-bit values,
+# while all others are meant to keep 32. "Meant to" means that additions
+# to @acc[0-7] do "contaminate" upper bits, but they are cleared before
+# they can affect outcome (follow 'and' with $mask). Also keep in mind
+# that addition with carry is addition with 32-bit carry, even though
+# CPU is 64-bit. [Addition with 64-bit carry was introduced in T3, see
+# below for VIS3 code paths.]
+
+$code.=<<___;
+.align	32
+__ecp_nistz256_mul_mont:
+	ld	[$bp+0],$bi		! b[0]
+	mov	-1,$mask
+	ld	[$ap+0],$a0
+	srl	$mask,0,$mask		! 0xffffffff
+	ld	[$ap+4],$t1
+	ld	[$ap+8],$t2
+	ld	[$ap+12],$t3
+	ld	[$ap+16],$t4
+	ld	[$ap+20],$t5
+	ld	[$ap+24],$t6
+	ld	[$ap+28],$t7
+	mulx	$a0,$bi,$t0		! a[0-7]*b[0], 64-bit results
+	mulx	$t1,$bi,$t1
+	mulx	$t2,$bi,$t2
+	mulx	$t3,$bi,$t3
+	mulx	$t4,$bi,$t4
+	mulx	$t5,$bi,$t5
+	mulx	$t6,$bi,$t6
+	mulx	$t7,$bi,$t7
+	srlx	$t0,32,@acc[1]		! extract high parts
+	srlx	$t1,32,@acc[2]
+	srlx	$t2,32,@acc[3]
+	srlx	$t3,32,@acc[4]
+	srlx	$t4,32,@acc[5]
+	srlx	$t5,32,@acc[6]
+	srlx	$t6,32,@acc[7]
+	srlx	$t7,32,@acc[0]		! "@acc[8]"
+	mov	0,$carry
+___
+for($i=1;$i<8;$i++) {
+$code.=<<___;
+	addcc	@acc[1],$t1,@acc[1]	! accumulate high parts
+	ld	[$bp+4*$i],$bi		! b[$i]
+	ld	[$ap+4],$t1		! re-load a[1-7]
+	addccc	@acc[2],$t2,@acc[2]
+	addccc	@acc[3],$t3,@acc[3]
+	ld	[$ap+8],$t2
+	ld	[$ap+12],$t3
+	addccc	@acc[4],$t4,@acc[4]
+	addccc	@acc[5],$t5,@acc[5]
+	ld	[$ap+16],$t4
+	ld	[$ap+20],$t5
+	addccc	@acc[6],$t6,@acc[6]
+	addccc	@acc[7],$t7,@acc[7]
+	ld	[$ap+24],$t6
+	ld	[$ap+28],$t7
+	addccc	@acc[0],$carry,@acc[0]	! "@acc[8]"
+	addc	%g0,%g0,$carry
+___
+	# Reduction iteration is normally performed by accumulating
+	# result of multiplication of modulus by "magic" digit [and
+	# omitting least significant word, which is guaranteed to
+	# be 0], but thanks to special form of modulus and "magic"
+	# digit being equal to least significant word, it can be
+	# performed with additions and subtractions alone. Indeed:
+	#
+	#        ffff.0001.0000.0000.0000.ffff.ffff.ffff
+	# *                                         abcd
+	# + xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
+	#
+	# Now observing that ff..ff*x = (2^n-1)*x = 2^n*x-x, we
+	# rewrite above as:
+	#
+	#   xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.abcd
+	# + abcd.0000.abcd.0000.0000.abcd.0000.0000.0000
+	# -      abcd.0000.0000.0000.0000.0000.0000.abcd
+	#
+	# or marking redundant operations:
+	#
+	#   xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.xxxx.----
+	# + abcd.0000.abcd.0000.0000.abcd.----.----.----
+	# -      abcd.----.----.----.----.----.----.----
+
+$code.=<<___;
+	! multiplication-less reduction
+	addcc	@acc[3],$t0,@acc[3]	! r[3]+=r[0]
+	addccc	@acc[4],%g0,@acc[4]	! r[4]+=0
+	 and	@acc[1],$mask,@acc[1]
+	 and	@acc[2],$mask,@acc[2]
+	addccc	@acc[5],%g0,@acc[5]	! r[5]+=0
+	addccc	@acc[6],$t0,@acc[6]	! r[6]+=r[0]
+	 and	@acc[3],$mask,@acc[3]
+	 and	@acc[4],$mask,@acc[4]
+	addccc	@acc[7],%g0,@acc[7]	! r[7]+=0
+	addccc	@acc[0],$t0,@acc[0]	! r[8]+=r[0]	"@acc[8]"
+	 and	@acc[5],$mask,@acc[5]
+	 and	@acc[6],$mask,@acc[6]
+	addc	$carry,%g0,$carry	! top-most carry
+	subcc	@acc[7],$t0,@acc[7]	! r[7]-=r[0]
+	subccc	@acc[0],%g0,@acc[0]	! r[8]-=0	"@acc[8]"
+	subc	$carry,%g0,$carry	! top-most carry
+	 and	@acc[7],$mask,@acc[7]
+	 and	@acc[0],$mask,@acc[0]	! "@acc[8]"
+___
+	push(@acc,shift(@acc));		# rotate registers to "omit" acc[0]
+$code.=<<___;
+	mulx	$a0,$bi,$t0		! a[0-7]*b[$i], 64-bit results
+	mulx	$t1,$bi,$t1
+	mulx	$t2,$bi,$t2
+	mulx	$t3,$bi,$t3
+	mulx	$t4,$bi,$t4
+	mulx	$t5,$bi,$t5
+	mulx	$t6,$bi,$t6
+	mulx	$t7,$bi,$t7
+	add	@acc[0],$t0,$t0		! accumulate low parts, can't overflow
+	add	@acc[1],$t1,$t1
+	srlx	$t0,32,@acc[1]		! extract high parts
+	add	@acc[2],$t2,$t2
+	srlx	$t1,32,@acc[2]
+	add	@acc[3],$t3,$t3
+	srlx	$t2,32,@acc[3]
+	add	@acc[4],$t4,$t4
+	srlx	$t3,32,@acc[4]
+	add	@acc[5],$t5,$t5
+	srlx	$t4,32,@acc[5]
+	add	@acc[6],$t6,$t6
+	srlx	$t5,32,@acc[6]
+	add	@acc[7],$t7,$t7
+	srlx	$t6,32,@acc[7]
+	srlx	$t7,32,@acc[0]		! "@acc[8]"
+___
+}
+$code.=<<___;
+	addcc	@acc[1],$t1,@acc[1]	! accumulate high parts
+	addccc	@acc[2],$t2,@acc[2]
+	addccc	@acc[3],$t3,@acc[3]
+	addccc	@acc[4],$t4,@acc[4]
+	addccc	@acc[5],$t5,@acc[5]
+	addccc	@acc[6],$t6,@acc[6]
+	addccc	@acc[7],$t7,@acc[7]
+	addccc	@acc[0],$carry,@acc[0]	! "@acc[8]"
+	addc	%g0,%g0,$carry
+
+	addcc	@acc[3],$t0,@acc[3]	! multiplication-less reduction
+	addccc	@acc[4],%g0,@acc[4]
+	addccc	@acc[5],%g0,@acc[5]
+	addccc	@acc[6],$t0,@acc[6]
+	addccc	@acc[7],%g0,@acc[7]
+	addccc	@acc[0],$t0,@acc[0]	! "@acc[8]"
+	addc	$carry,%g0,$carry
+	subcc	@acc[7],$t0,@acc[7]
+	subccc	@acc[0],%g0,@acc[0]	! "@acc[8]"
+	subc	$carry,%g0,$carry	! top-most carry
+___
+	push(@acc,shift(@acc));		# rotate registers to omit acc[0]
+$code.=<<___;
+	! Final step is "if result > mod, subtract mod", but we do it
+	! "other way around", namely subtract modulus from result
+	! and if it borrowed, add modulus back.
+
+	subcc	@acc[0],-1,@acc[0]	! subtract modulus
+	subccc	@acc[1],-1,@acc[1]
+	subccc	@acc[2],-1,@acc[2]
+	subccc	@acc[3],0,@acc[3]
+	subccc	@acc[4],0,@acc[4]
+	subccc	@acc[5],0,@acc[5]
+	subccc	@acc[6],1,@acc[6]
+	subccc	@acc[7],-1,@acc[7]
+	subc	$carry,0,$carry		! broadcast borrow bit
+
+	! Note that because mod has special form, i.e. consists of
+	! 0xffffffff, 1 and 0s, we can conditionally synthesize it by
+	! using value of broadcasted borrow and the borrow bit itself.
+	! To minimize dependency chain we first broadcast and then
+	! extract the bit by negating (follow $bi).
+
+	addcc	@acc[0],$carry,@acc[0]	! add modulus or zero
+	addccc	@acc[1],$carry,@acc[1]
+	neg	$carry,$bi
+	st	@acc[0],[$rp]
+	addccc	@acc[2],$carry,@acc[2]
+	st	@acc[1],[$rp+4]
+	addccc	@acc[3],0,@acc[3]
+	st	@acc[2],[$rp+8]
+	addccc	@acc[4],0,@acc[4]
+	st	@acc[3],[$rp+12]
+	addccc	@acc[5],0,@acc[5]
+	st	@acc[4],[$rp+16]
+	addccc	@acc[6],$bi,@acc[6]
+	st	@acc[5],[$rp+20]
+	addc	@acc[7],$carry,@acc[7]
+	st	@acc[6],[$rp+24]
+	retl
+	st	@acc[7],[$rp+28]
+.size	__ecp_nistz256_mul_mont,.-__ecp_nistz256_mul_mont
+
+! void	ecp_nistz256_add(BN_ULONG %i0[8],const BN_ULONG %i1[8],
+!					 const BN_ULONG %i2[8]);
+.globl	ecp_nistz256_add
+.align	32
+ecp_nistz256_add:
+	save	%sp,-STACK_FRAME,%sp
+	ld	[$ap],@acc[0]
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	call	__ecp_nistz256_add
+	ld	[$ap+28],@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_add,.-ecp_nistz256_add
+
+.align	32
+__ecp_nistz256_add:
+	ld	[$bp+0],$t0		! b[0]
+	ld	[$bp+4],$t1
+	ld	[$bp+8],$t2
+	ld	[$bp+12],$t3
+	addcc	@acc[0],$t0,@acc[0]
+	ld	[$bp+16],$t4
+	ld	[$bp+20],$t5
+	addccc	@acc[1],$t1,@acc[1]
+	ld	[$bp+24],$t6
+	ld	[$bp+28],$t7
+	addccc	@acc[2],$t2,@acc[2]
+	addccc	@acc[3],$t3,@acc[3]
+	addccc	@acc[4],$t4,@acc[4]
+	addccc	@acc[5],$t5,@acc[5]
+	addccc	@acc[6],$t6,@acc[6]
+	addccc	@acc[7],$t7,@acc[7]
+	subc	%g0,%g0,$carry		! broadcast carry bit
+
+.Lreduce_by_sub:
+
+	! if a+b carries, subtract modulus.
+	!
+	! Note that because mod has special form, i.e. consists of
+	! 0xffffffff, 1 and 0s, we can conditionally synthesize it by
+	! using value of broadcasted borrow and the borrow bit itself.
+	! To minimize dependency chain we first broadcast and then
+	! extract the bit by negating (follow $bi).
+
+	subcc	@acc[0],$carry,@acc[0]	! subtract synthesized modulus
+	subccc	@acc[1],$carry,@acc[1]
+	neg	$carry,$bi
+	st	@acc[0],[$rp]
+	subccc	@acc[2],$carry,@acc[2]
+	st	@acc[1],[$rp+4]
+	subccc	@acc[3],0,@acc[3]
+	st	@acc[2],[$rp+8]
+	subccc	@acc[4],0,@acc[4]
+	st	@acc[3],[$rp+12]
+	subccc	@acc[5],0,@acc[5]
+	st	@acc[4],[$rp+16]
+	subccc	@acc[6],$bi,@acc[6]
+	st	@acc[5],[$rp+20]
+	subc	@acc[7],$carry,@acc[7]
+	st	@acc[6],[$rp+24]
+	retl
+	st	@acc[7],[$rp+28]
+.size	__ecp_nistz256_add,.-__ecp_nistz256_add
+
+! void	ecp_nistz256_mul_by_2(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_mul_by_2
+.align	32
+ecp_nistz256_mul_by_2:
+	save	%sp,-STACK_FRAME,%sp
+	ld	[$ap],@acc[0]
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	call	__ecp_nistz256_mul_by_2
+	ld	[$ap+28],@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_mul_by_2,.-ecp_nistz256_mul_by_2
+
+.align	32
+__ecp_nistz256_mul_by_2:
+	addcc	@acc[0],@acc[0],@acc[0]	! a+a=2*a
+	addccc	@acc[1],@acc[1],@acc[1]
+	addccc	@acc[2],@acc[2],@acc[2]
+	addccc	@acc[3],@acc[3],@acc[3]
+	addccc	@acc[4],@acc[4],@acc[4]
+	addccc	@acc[5],@acc[5],@acc[5]
+	addccc	@acc[6],@acc[6],@acc[6]
+	addccc	@acc[7],@acc[7],@acc[7]
+	b	.Lreduce_by_sub
+	subc	%g0,%g0,$carry		! broadcast carry bit
+.size	__ecp_nistz256_mul_by_2,.-__ecp_nistz256_mul_by_2
+
+! void	ecp_nistz256_mul_by_3(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_mul_by_3
+.align	32
+ecp_nistz256_mul_by_3:
+	save	%sp,-STACK_FRAME,%sp
+	ld	[$ap],@acc[0]
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	call	__ecp_nistz256_mul_by_3
+	ld	[$ap+28],@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_mul_by_3,.-ecp_nistz256_mul_by_3
+
+.align	32
+__ecp_nistz256_mul_by_3:
+	addcc	@acc[0],@acc[0],$t0	! a+a=2*a
+	addccc	@acc[1],@acc[1],$t1
+	addccc	@acc[2],@acc[2],$t2
+	addccc	@acc[3],@acc[3],$t3
+	addccc	@acc[4],@acc[4],$t4
+	addccc	@acc[5],@acc[5],$t5
+	addccc	@acc[6],@acc[6],$t6
+	addccc	@acc[7],@acc[7],$t7
+	subc	%g0,%g0,$carry		! broadcast carry bit
+
+	subcc	$t0,$carry,$t0		! .Lreduce_by_sub but without stores
+	neg	$carry,$bi
+	subccc	$t1,$carry,$t1
+	subccc	$t2,$carry,$t2
+	subccc	$t3,0,$t3
+	subccc	$t4,0,$t4
+	subccc	$t5,0,$t5
+	subccc	$t6,$bi,$t6
+	subc	$t7,$carry,$t7
+
+	addcc	$t0,@acc[0],@acc[0]	! 2*a+a=3*a
+	addccc	$t1,@acc[1],@acc[1]
+	addccc	$t2,@acc[2],@acc[2]
+	addccc	$t3,@acc[3],@acc[3]
+	addccc	$t4,@acc[4],@acc[4]
+	addccc	$t5,@acc[5],@acc[5]
+	addccc	$t6,@acc[6],@acc[6]
+	addccc	$t7,@acc[7],@acc[7]
+	b	.Lreduce_by_sub
+	subc	%g0,%g0,$carry		! broadcast carry bit
+.size	__ecp_nistz256_mul_by_3,.-__ecp_nistz256_mul_by_3
+
+! void	ecp_nistz256_sub(BN_ULONG %i0[8],const BN_ULONG %i1[8],
+!				         const BN_ULONG %i2[8]);
+.globl	ecp_nistz256_sub
+.align	32
+ecp_nistz256_sub:
+	save	%sp,-STACK_FRAME,%sp
+	ld	[$ap],@acc[0]
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	call	__ecp_nistz256_sub_from
+	ld	[$ap+28],@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_sub,.-ecp_nistz256_sub
+
+! void	ecp_nistz256_neg(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_neg
+.align	32
+ecp_nistz256_neg:
+	save	%sp,-STACK_FRAME,%sp
+	mov	$ap,$bp
+	mov	0,@acc[0]
+	mov	0,@acc[1]
+	mov	0,@acc[2]
+	mov	0,@acc[3]
+	mov	0,@acc[4]
+	mov	0,@acc[5]
+	mov	0,@acc[6]
+	call	__ecp_nistz256_sub_from
+	mov	0,@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_neg,.-ecp_nistz256_neg
+
+.align	32
+__ecp_nistz256_sub_from:
+	ld	[$bp+0],$t0		! b[0]
+	ld	[$bp+4],$t1
+	ld	[$bp+8],$t2
+	ld	[$bp+12],$t3
+	subcc	@acc[0],$t0,@acc[0]
+	ld	[$bp+16],$t4
+	ld	[$bp+20],$t5
+	subccc	@acc[1],$t1,@acc[1]
+	subccc	@acc[2],$t2,@acc[2]
+	ld	[$bp+24],$t6
+	ld	[$bp+28],$t7
+	subccc	@acc[3],$t3,@acc[3]
+	subccc	@acc[4],$t4,@acc[4]
+	subccc	@acc[5],$t5,@acc[5]
+	subccc	@acc[6],$t6,@acc[6]
+	subccc	@acc[7],$t7,@acc[7]
+	subc	%g0,%g0,$carry		! broadcast borrow bit
+
+.Lreduce_by_add:
+
+	! if a-b borrows, add modulus.
+	!
+	! Note that because mod has special form, i.e. consists of
+	! 0xffffffff, 1 and 0s, we can conditionally synthesize it by
+	! using value of broadcasted borrow and the borrow bit itself.
+	! To minimize dependency chain we first broadcast and then
+	! extract the bit by negating (follow $bi).
+
+	addcc	@acc[0],$carry,@acc[0]	! add synthesized modulus
+	addccc	@acc[1],$carry,@acc[1]
+	neg	$carry,$bi
+	st	@acc[0],[$rp]
+	addccc	@acc[2],$carry,@acc[2]
+	st	@acc[1],[$rp+4]
+	addccc	@acc[3],0,@acc[3]
+	st	@acc[2],[$rp+8]
+	addccc	@acc[4],0,@acc[4]
+	st	@acc[3],[$rp+12]
+	addccc	@acc[5],0,@acc[5]
+	st	@acc[4],[$rp+16]
+	addccc	@acc[6],$bi,@acc[6]
+	st	@acc[5],[$rp+20]
+	addc	@acc[7],$carry,@acc[7]
+	st	@acc[6],[$rp+24]
+	retl
+	st	@acc[7],[$rp+28]
+.size	__ecp_nistz256_sub_from,.-__ecp_nistz256_sub_from
+
+.align	32
+__ecp_nistz256_sub_morf:
+	ld	[$bp+0],$t0		! b[0]
+	ld	[$bp+4],$t1
+	ld	[$bp+8],$t2
+	ld	[$bp+12],$t3
+	subcc	$t0,@acc[0],@acc[0]
+	ld	[$bp+16],$t4
+	ld	[$bp+20],$t5
+	subccc	$t1,@acc[1],@acc[1]
+	subccc	$t2,@acc[2],@acc[2]
+	ld	[$bp+24],$t6
+	ld	[$bp+28],$t7
+	subccc	$t3,@acc[3],@acc[3]
+	subccc	$t4,@acc[4],@acc[4]
+	subccc	$t5,@acc[5],@acc[5]
+	subccc	$t6,@acc[6],@acc[6]
+	subccc	$t7,@acc[7],@acc[7]
+	b	.Lreduce_by_add
+	subc	%g0,%g0,$carry		! broadcast borrow bit
+.size	__ecp_nistz256_sub_morf,.-__ecp_nistz256_sub_morf
+
+! void	ecp_nistz256_div_by_2(BN_ULONG %i0[8],const BN_ULONG %i1[8]);
+.globl	ecp_nistz256_div_by_2
+.align	32
+ecp_nistz256_div_by_2:
+	save	%sp,-STACK_FRAME,%sp
+	ld	[$ap],@acc[0]
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	call	__ecp_nistz256_div_by_2
+	ld	[$ap+28],@acc[7]
+	ret
+	restore
+.size	ecp_nistz256_div_by_2,.-ecp_nistz256_div_by_2
+
+.align	32
+__ecp_nistz256_div_by_2:
+	! ret = (a is odd ? a+mod : a) >> 1
+
+	and	@acc[0],1,$bi
+	neg	$bi,$carry
+	addcc	@acc[0],$carry,@acc[0]
+	addccc	@acc[1],$carry,@acc[1]
+	addccc	@acc[2],$carry,@acc[2]
+	addccc	@acc[3],0,@acc[3]
+	addccc	@acc[4],0,@acc[4]
+	addccc	@acc[5],0,@acc[5]
+	addccc	@acc[6],$bi,@acc[6]
+	addccc	@acc[7],$carry,@acc[7]
+	addc	%g0,%g0,$carry
+
+	! ret >>= 1
+
+	srl	@acc[0],1,@acc[0]
+	sll	@acc[1],31,$t0
+	srl	@acc[1],1,@acc[1]
+	or	@acc[0],$t0,@acc[0]
+	sll	@acc[2],31,$t1
+	srl	@acc[2],1,@acc[2]
+	or	@acc[1],$t1,@acc[1]
+	sll	@acc[3],31,$t2
+	st	@acc[0],[$rp]
+	srl	@acc[3],1,@acc[3]
+	or	@acc[2],$t2,@acc[2]
+	sll	@acc[4],31,$t3
+	st	@acc[1],[$rp+4]
+	srl	@acc[4],1,@acc[4]
+	or	@acc[3],$t3,@acc[3]
+	sll	@acc[5],31,$t4
+	st	@acc[2],[$rp+8]
+	srl	@acc[5],1,@acc[5]
+	or	@acc[4],$t4,@acc[4]
+	sll	@acc[6],31,$t5
+	st	@acc[3],[$rp+12]
+	srl	@acc[6],1,@acc[6]
+	or	@acc[5],$t5,@acc[5]
+	sll	@acc[7],31,$t6
+	st	@acc[4],[$rp+16]
+	srl	@acc[7],1,@acc[7]
+	or	@acc[6],$t6,@acc[6]
+	sll	$carry,31,$t7
+	st	@acc[5],[$rp+20]
+	or	@acc[7],$t7,@acc[7]
+	st	@acc[6],[$rp+24]
+	retl
+	st	@acc[7],[$rp+28]
+.size	__ecp_nistz256_div_by_2,.-__ecp_nistz256_div_by_2
+___
+
+########################################################################
+# following subroutines are "literal" implemetation of those found in
+# ecp_nistz256.c
+#
+########################################################################
+# void ecp_nistz256_point_double(P256_POINT *out,const P256_POINT *inp);
+#
+{
+my ($S,$M,$Zsqr,$tmp0)=map(32*$_,(0..3));
+# above map() describes stack layout with 4 temporary
+# 256-bit vectors on top.
+
+$code.=<<___;
+#ifdef __PIC__
+SPARC_PIC_THUNK(%g1)
+#endif
+
+.globl	ecp_nistz256_point_double
+.align	32
+ecp_nistz256_point_double:
+	SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
+	ld	[%g1],%g1		! OPENSSL_sparcv9cap_P[0]
+	and	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK),%g1
+	cmp	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK)
+	be	ecp_nistz256_point_double_vis3
+	nop
+
+	save	%sp,-STACK_FRAME-32*4,%sp
+
+	mov	$rp,$rp_real
+	mov	$ap,$ap_real
+
+	ld	[$ap+32],@acc[0]
+	ld	[$ap+32+4],@acc[1]
+	ld	[$ap+32+8],@acc[2]
+	ld	[$ap+32+12],@acc[3]
+	ld	[$ap+32+16],@acc[4]
+	ld	[$ap+32+20],@acc[5]
+	ld	[$ap+32+24],@acc[6]
+	ld	[$ap+32+28],@acc[7]
+	call	__ecp_nistz256_mul_by_2	! p256_mul_by_2(S, in_y);
+	add	%sp,LOCALS+$S,$rp
+
+	add	$ap_real,64,$bp
+	add	$ap_real,64,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(Zsqr, in_z);
+	add	%sp,LOCALS+$Zsqr,$rp
+
+	add	$ap_real,0,$bp
+	call	__ecp_nistz256_add	! p256_add(M, Zsqr, in_x);
+	add	%sp,LOCALS+$M,$rp
+
+	add	%sp,LOCALS+$S,$bp
+	add	%sp,LOCALS+$S,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(S, S);
+	add	%sp,LOCALS+$S,$rp
+
+	ld	[$ap_real],@acc[0]
+	add	%sp,LOCALS+$Zsqr,$bp
+	ld	[$ap_real+4],@acc[1]
+	ld	[$ap_real+8],@acc[2]
+	ld	[$ap_real+12],@acc[3]
+	ld	[$ap_real+16],@acc[4]
+	ld	[$ap_real+20],@acc[5]
+	ld	[$ap_real+24],@acc[6]
+	ld	[$ap_real+28],@acc[7]
+	call	__ecp_nistz256_sub_from	! p256_sub(Zsqr, in_x, Zsqr);
+	add	%sp,LOCALS+$Zsqr,$rp
+
+	add	$ap_real,32,$bp
+	add	$ap_real,64,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(tmp0, in_z, in_y);
+	add	%sp,LOCALS+$tmp0,$rp
+
+	call	__ecp_nistz256_mul_by_2	! p256_mul_by_2(res_z, tmp0);
+	add	$rp_real,64,$rp
+
+	add	%sp,LOCALS+$Zsqr,$bp
+	add	%sp,LOCALS+$M,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(M, M, Zsqr);
+	add	%sp,LOCALS+$M,$rp
+
+	call	__ecp_nistz256_mul_by_3	! p256_mul_by_3(M, M);
+	add	%sp,LOCALS+$M,$rp
+
+	add	%sp,LOCALS+$S,$bp
+	add	%sp,LOCALS+$S,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(tmp0, S);
+	add	%sp,LOCALS+$tmp0,$rp
+
+	call	__ecp_nistz256_div_by_2	! p256_div_by_2(res_y, tmp0);
+	add	$rp_real,32,$rp
+
+	add	$ap_real,0,$bp
+	add	%sp,LOCALS+$S,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S, S, in_x);
+	add	%sp,LOCALS+$S,$rp
+
+	call	__ecp_nistz256_mul_by_2	! p256_mul_by_2(tmp0, S);
+	add	%sp,LOCALS+$tmp0,$rp
+
+	add	%sp,LOCALS+$M,$bp
+	add	%sp,LOCALS+$M,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(res_x, M);
+	add	$rp_real,0,$rp
+
+	add	%sp,LOCALS+$tmp0,$bp
+	call	__ecp_nistz256_sub_from	! p256_sub(res_x, res_x, tmp0);
+	add	$rp_real,0,$rp
+
+	add	%sp,LOCALS+$S,$bp
+	call	__ecp_nistz256_sub_morf	! p256_sub(S, S, res_x);
+	add	%sp,LOCALS+$S,$rp
+
+	add	%sp,LOCALS+$M,$bp
+	add	%sp,LOCALS+$S,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S, S, M);
+	add	%sp,LOCALS+$S,$rp
+
+	add	$rp_real,32,$bp
+	call	__ecp_nistz256_sub_from	! p256_sub(res_y, S, res_y);
+	add	$rp_real,32,$rp
+
+	ret
+	restore
+.size	ecp_nistz256_point_double,.-ecp_nistz256_point_double
+___
+}
+
+########################################################################
+# void ecp_nistz256_point_add(P256_POINT *out,const P256_POINT *in1,
+#			      const P256_POINT *in2);
+{
+my ($res_x,$res_y,$res_z,
+    $H,$Hsqr,$R,$Rsqr,$Hcub,
+    $U1,$U2,$S1,$S2)=map(32*$_,(0..11));
+my ($Z1sqr, $Z2sqr) = ($Hsqr, $Rsqr);
+
+# above map() describes stack layout with 12 temporary
+# 256-bit vectors on top. Then we reserve some space for
+# !in1infty, !in2infty, result of check for zero and return pointer.
+
+my $bp_real=$rp_real;
+
+$code.=<<___;
+.globl	ecp_nistz256_point_add
+.align	32
+ecp_nistz256_point_add:
+	SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
+	ld	[%g1],%g1		! OPENSSL_sparcv9cap_P[0]
+	and	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK),%g1
+	cmp	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK)
+	be	ecp_nistz256_point_add_vis3
+	nop
+
+	save	%sp,-STACK_FRAME-32*12-32,%sp
+
+	stx	$rp,[%fp+STACK_BIAS-8]	! off-load $rp
+	mov	$ap,$ap_real
+	mov	$bp,$bp_real
+
+	ld	[$bp],@acc[0]		! in2_x
+	ld	[$bp+4],@acc[1]
+	ld	[$bp+8],@acc[2]
+	ld	[$bp+12],@acc[3]
+	ld	[$bp+16],@acc[4]
+	ld	[$bp+20],@acc[5]
+	ld	[$bp+24],@acc[6]
+	ld	[$bp+28],@acc[7]
+	ld	[$bp+32],$t0		! in2_y
+	ld	[$bp+32+4],$t1
+	ld	[$bp+32+8],$t2
+	ld	[$bp+32+12],$t3
+	ld	[$bp+32+16],$t4
+	ld	[$bp+32+20],$t5
+	ld	[$bp+32+24],$t6
+	ld	[$bp+32+28],$t7
+	or	@acc[1],@acc[0],@acc[0]
+	or	@acc[3],@acc[2],@acc[2]
+	or	@acc[5],@acc[4],@acc[4]
+	or	@acc[7],@acc[6],@acc[6]
+	or	@acc[2],@acc[0],@acc[0]
+	or	@acc[6],@acc[4],@acc[4]
+	or	@acc[4],@acc[0],@acc[0]
+	or	$t1,$t0,$t0
+	or	$t3,$t2,$t2
+	or	$t5,$t4,$t4
+	or	$t7,$t6,$t6
+	or	$t2,$t0,$t0
+	or	$t6,$t4,$t4
+	or	$t4,$t0,$t0
+	or	@acc[0],$t0,$t0		! !in2infty
+	movrnz	$t0,-1,$t0
+	st	$t0,[%fp+STACK_BIAS-12]
+
+	ld	[$ap],@acc[0]		! in1_x
+	ld	[$ap+4],@acc[1]
+	ld	[$ap+8],@acc[2]
+	ld	[$ap+12],@acc[3]
+	ld	[$ap+16],@acc[4]
+	ld	[$ap+20],@acc[5]
+	ld	[$ap+24],@acc[6]
+	ld	[$ap+28],@acc[7]
+	ld	[$ap+32],$t0		! in1_y
+	ld	[$ap+32+4],$t1
+	ld	[$ap+32+8],$t2
+	ld	[$ap+32+12],$t3
+	ld	[$ap+32+16],$t4
+	ld	[$ap+32+20],$t5
+	ld	[$ap+32+24],$t6
+	ld	[$ap+32+28],$t7
+	or	@acc[1],@acc[0],@acc[0]
+	or	@acc[3],@acc[2],@acc[2]
+	or	@acc[5],@acc[4],@acc[4]
+	or	@acc[7],@acc[6],@acc[6]
+	or	@acc[2],@acc[0],@acc[0]
+	or	@acc[6],@acc[4],@acc[4]
+	or	@acc[4],@acc[0],@acc[0]
+	or	$t1,$t0,$t0
+	or	$t3,$t2,$t2
+	or	$t5,$t4,$t4
+	or	$t7,$t6,$t6
+	or	$t2,$t0,$t0
+	or	$t6,$t4,$t4
+	or	$t4,$t0,$t0
+	or	@acc[0],$t0,$t0		! !in1infty
+	movrnz	$t0,-1,$t0
+	st	$t0,[%fp+STACK_BIAS-16]
+
+	add	$bp_real,64,$bp
+	add	$bp_real,64,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(Z2sqr, in2_z);
+	add	%sp,LOCALS+$Z2sqr,$rp
+
+	add	$ap_real,64,$bp
+	add	$ap_real,64,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(Z1sqr, in1_z);
+	add	%sp,LOCALS+$Z1sqr,$rp
+
+	add	$bp_real,64,$bp
+	add	%sp,LOCALS+$Z2sqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S1, Z2sqr, in2_z);
+	add	%sp,LOCALS+$S1,$rp
+
+	add	$ap_real,64,$bp
+	add	%sp,LOCALS+$Z1sqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S2, Z1sqr, in1_z);
+	add	%sp,LOCALS+$S2,$rp
+
+	add	$ap_real,32,$bp
+	add	%sp,LOCALS+$S1,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S1, S1, in1_y);
+	add	%sp,LOCALS+$S1,$rp
+
+	add	$bp_real,32,$bp
+	add	%sp,LOCALS+$S2,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S2, S2, in2_y);
+	add	%sp,LOCALS+$S2,$rp
+
+	add	%sp,LOCALS+$S1,$bp
+	call	__ecp_nistz256_sub_from	! p256_sub(R, S2, S1);
+	add	%sp,LOCALS+$R,$rp
+
+	or	@acc[1],@acc[0],@acc[0]	! see if result is zero
+	or	@acc[3],@acc[2],@acc[2]
+	or	@acc[5],@acc[4],@acc[4]
+	or	@acc[7],@acc[6],@acc[6]
+	or	@acc[2],@acc[0],@acc[0]
+	or	@acc[6],@acc[4],@acc[4]
+	or	@acc[4],@acc[0],@acc[0]
+	st	@acc[0],[%fp+STACK_BIAS-20]
+
+	add	$ap_real,0,$bp
+	add	%sp,LOCALS+$Z2sqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(U1, in1_x, Z2sqr);
+	add	%sp,LOCALS+$U1,$rp
+
+	add	$bp_real,0,$bp
+	add	%sp,LOCALS+$Z1sqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(U2, in2_x, Z1sqr);
+	add	%sp,LOCALS+$U2,$rp
+
+	add	%sp,LOCALS+$U1,$bp
+	call	__ecp_nistz256_sub_from	! p256_sub(H, U2, U1);
+	add	%sp,LOCALS+$H,$rp
+
+	or	@acc[1],@acc[0],@acc[0]	! see if result is zero
+	or	@acc[3],@acc[2],@acc[2]
+	or	@acc[5],@acc[4],@acc[4]
+	or	@acc[7],@acc[6],@acc[6]
+	or	@acc[2],@acc[0],@acc[0]
+	or	@acc[6],@acc[4],@acc[4]
+	orcc	@acc[4],@acc[0],@acc[0]
+
+	bne,pt	%icc,.Ladd_proceed	! is_equal(U1,U2)?
+	nop
+
+	ld	[%fp+STACK_BIAS-12],$t0
+	ld	[%fp+STACK_BIAS-16],$t1
+	ld	[%fp+STACK_BIAS-20],$t2
+	andcc	$t0,$t1,%g0
+	be,pt	%icc,.Ladd_proceed	! (in1infty || in2infty)?
+	nop
+	andcc	$t2,$t2,%g0
+	be,pt	%icc,.Ladd_proceed	! is_equal(S1,S2)?
+	nop
+
+	ldx	[%fp+STACK_BIAS-8],$rp
+	st	%g0,[$rp]
+	st	%g0,[$rp+4]
+	st	%g0,[$rp+8]
+	st	%g0,[$rp+12]
+	st	%g0,[$rp+16]
+	st	%g0,[$rp+20]
+	st	%g0,[$rp+24]
+	st	%g0,[$rp+28]
+	st	%g0,[$rp+32]
+	st	%g0,[$rp+32+4]
+	st	%g0,[$rp+32+8]
+	st	%g0,[$rp+32+12]
+	st	%g0,[$rp+32+16]
+	st	%g0,[$rp+32+20]
+	st	%g0,[$rp+32+24]
+	st	%g0,[$rp+32+28]
+	st	%g0,[$rp+64]
+	st	%g0,[$rp+64+4]
+	st	%g0,[$rp+64+8]
+	st	%g0,[$rp+64+12]
+	st	%g0,[$rp+64+16]
+	st	%g0,[$rp+64+20]
+	st	%g0,[$rp+64+24]
+	st	%g0,[$rp+64+28]
+	b	.Ladd_done
+	nop
+
+.align	16
+.Ladd_proceed:
+	add	%sp,LOCALS+$R,$bp
+	add	%sp,LOCALS+$R,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(Rsqr, R);
+	add	%sp,LOCALS+$Rsqr,$rp
+
+	add	$ap_real,64,$bp
+	add	%sp,LOCALS+$H,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(res_z, H, in1_z);
+	add	%sp,LOCALS+$res_z,$rp
+
+	add	%sp,LOCALS+$H,$bp
+	add	%sp,LOCALS+$H,$ap
+	call	__ecp_nistz256_mul_mont	! p256_sqr_mont(Hsqr, H);
+	add	%sp,LOCALS+$Hsqr,$rp
+
+	add	$bp_real,64,$bp
+	add	%sp,LOCALS+$res_z,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(res_z, res_z, in2_z);
+	add	%sp,LOCALS+$res_z,$rp
+
+	add	%sp,LOCALS+$H,$bp
+	add	%sp,LOCALS+$Hsqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(Hcub, Hsqr, H);
+	add	%sp,LOCALS+$Hcub,$rp
+
+	add	%sp,LOCALS+$U1,$bp
+	add	%sp,LOCALS+$Hsqr,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(U2, U1, Hsqr);
+	add	%sp,LOCALS+$U2,$rp
+
+	call	__ecp_nistz256_mul_by_2	! p256_mul_by_2(Hsqr, U2);
+	add	%sp,LOCALS+$Hsqr,$rp
+
+	add	%sp,LOCALS+$Rsqr,$bp
+	call	__ecp_nistz256_sub_morf	! p256_sub(res_x, Rsqr, Hsqr);
+	add	%sp,LOCALS+$res_x,$rp
+
+	add	%sp,LOCALS+$Hcub,$bp
+	call	__ecp_nistz256_sub_from	!  p256_sub(res_x, res_x, Hcub);
+	add	%sp,LOCALS+$res_x,$rp
+
+	add	%sp,LOCALS+$U2,$bp
+	call	__ecp_nistz256_sub_morf	! p256_sub(res_y, U2, res_x);
+	add	%sp,LOCALS+$res_y,$rp
+
+	add	%sp,LOCALS+$Hcub,$bp
+	add	%sp,LOCALS+$S1,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(S2, S1, Hcub);
+	add	%sp,LOCALS+$S2,$rp
+
+	add	%sp,LOCALS+$R,$bp
+	add	%sp,LOCALS+$res_y,$ap
+	call	__ecp_nistz256_mul_mont	! p256_mul_mont(res_y, res_y, R);
+	add	%sp,LOCALS+$res_y,$rp
+
+	add	%sp,LOCALS+$S2,$bp
+	call	__ecp_nistz256_sub_from	! p256_sub(res_y, res_y, S2);
+	add	%sp,LOCALS+$res_y,$rp
+
+	ld	[%fp+STACK_BIAS-16],$t1	! !in1infty
+	ld	[%fp+STACK_BIAS-12],$t2	! !in2infty
+	ldx	[%fp+STACK_BIAS-8],$rp
+___
+for($i=0;$i<96;$i+=8) {			# conditional moves
+$code.=<<___;
+	ld	[%sp+LOCALS+$i],@acc[0]		! res
+	ld	[%sp+LOCALS+$i+4],@acc[1]
+	ld	[$bp_real+$i],@acc[2]		! in2
+	ld	[$bp_real+$i+4],@acc[3]
+	ld	[$ap_real+$i],@acc[4]		! in1
+	ld	[$ap_real+$i+4],@acc[5]
+	movrz	$t1,@acc[2],@acc[0]
+	movrz	$t1,@acc[3],@acc[1]
+	movrz	$t2,@acc[4],@acc[0]
+	movrz	$t2,@acc[5],@acc[1]
+	st	@acc[0],[$rp+$i]
+	st	@acc[1],[$rp+$i+4]
+___
+}
+$code.=<<___;
+.Ladd_done:
+	ret
+	restore
+.size	ecp_nistz256_point_add,.-ecp_nistz256_point_add
+___
+}
+
+########################################################################
+# void ecp_nistz256_point_add_affine(P256_POINT *out,const P256_POINT *in1,
+#				     const P256_POINT_AFFINE *in2);
+{
+my ($res_x,$res_y,$res_z,
+    $U2,$S2,$H,$R,$Hsqr,$Hcub,$Rsqr)=map(32*$_,(0..9));
+my $Z1sqr = $S2;
+# above map() describes stack layout with 10 temporary
+# 256-bit vectors on top. Then we reserve some space for
+# !in1infty, !in2infty, result of check for zero and return pointer.
+
+my @ONE_mont=(1,0,0,-1,-1,-1,-2,0);
+my $bp_real=$rp_real;
+
+$code.=<<___;
+.globl	ecp_nistz256_point_add_affine
+.align	32
+ecp_nistz256_point_add_affine:
+	SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
+	ld	[%g1],%g1		! OPENSSL_sparcv9cap_P[0]
+	and	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK),%g1
+	cmp	%g1,(SPARCV9_VIS3|SPARCV9_64BIT_STACK)
+	be	ecp_ni