Initial import

1 files changed, 1558 insertions, 0 deletions

diff --git a/openssl/crypto/bn/asm/sparcv8plus.S b/openssl/crypto/bn/asm/sparcv8plus.S
new file mode 100644
index 00000000..63de1860
--- /dev/null
+++ b/openssl/crypto/bn/asm/sparcv8plus.S
@@ -0,0 +1,1558 @@
+.ident	"sparcv8plus.s, Version 1.4"
+.ident	"SPARC v9 ISA artwork by Andy Polyakov <appro@fy.chalmers.se>"
+
+/*
+ * ====================================================================
+ * Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+ * project.
+ *
+ * Rights for redistribution and usage in source and binary forms are
+ * granted according to the OpenSSL license. Warranty of any kind is
+ * disclaimed.
+ * ====================================================================
+ */
+
+/*
+ * This is my modest contributon to OpenSSL project (see
+ * http://www.openssl.org/ for more information about it) and is
+ * a drop-in UltraSPARC ISA replacement for crypto/bn/bn_asm.c
+ * module. For updates see http://fy.chalmers.se/~appro/hpe/.
+ *
+ * Questions-n-answers.
+ *
+ * Q. How to compile?
+ * A. With SC4.x/SC5.x:
+ *
+ *	cc -xarch=v8plus -c bn_asm.sparc.v8plus.S -o bn_asm.o
+ *
+ *    and with gcc:
+ *
+ *	gcc -mcpu=ultrasparc -c bn_asm.sparc.v8plus.S -o bn_asm.o
+ *
+ *    or if above fails (it does if you have gas installed):
+ *
+ *	gcc -E bn_asm.sparc.v8plus.S | as -xarch=v8plus /dev/fd/0 -o bn_asm.o
+ *
+ *    Quick-n-dirty way to fuse the module into the library.
+ *    Provided that the library is already configured and built
+ *    (in 0.9.2 case with no-asm option):
+ *
+ *	# cd crypto/bn
+ *	# cp /some/place/bn_asm.sparc.v8plus.S .
+ *	# cc -xarch=v8plus -c bn_asm.sparc.v8plus.S -o bn_asm.o
+ *	# make
+ *	# cd ../..
+ *	# make; make test
+ *
+ *    Quick-n-dirty way to get rid of it:
+ *
+ *	# cd crypto/bn
+ *	# touch bn_asm.c
+ *	# make
+ *	# cd ../..
+ *	# make; make test
+ *
+ * Q. V8plus achitecture? What kind of beast is that?
+ * A. Well, it's rather a programming model than an architecture...
+ *    It's actually v9-compliant, i.e. *any* UltraSPARC, CPU under
+ *    special conditions, namely when kernel doesn't preserve upper
+ *    32 bits of otherwise 64-bit registers during a context switch.
+ *
+ * Q. Why just UltraSPARC? What about SuperSPARC?
+ * A. Original release did target UltraSPARC only. Now SuperSPARC
+ *    version is provided along. Both version share bn_*comba[48]
+ *    implementations (see comment later in code for explanation).
+ *    But what's so special about this UltraSPARC implementation?
+ *    Why didn't I let compiler do the job? Trouble is that most of
+ *    available compilers (well, SC5.0 is the only exception) don't
+ *    attempt to take advantage of UltraSPARC's 64-bitness under
+ *    32-bit kernels even though it's perfectly possible (see next
+ *    question).
+ *
+ * Q. 64-bit registers under 32-bit kernels? Didn't you just say it
+ *    doesn't work?
+ * A. You can't adress *all* registers as 64-bit wide:-( The catch is
+ *    that you actually may rely upon %o0-%o5 and %g1-%g4 being fully
+ *    preserved if you're in a leaf function, i.e. such never calling
+ *    any other functions. All functions in this module are leaf and
+ *    10 registers is a handful. And as a matter of fact none-"comba"
+ *    routines don't require even that much and I could even afford to
+ *    not allocate own stack frame for 'em:-)
+ *
+ * Q. What about 64-bit kernels?
+ * A. What about 'em? Just kidding:-) Pure 64-bit version is currently
+ *    under evaluation and development...
+ *
+ * Q. What about shared libraries?
+ * A. What about 'em? Kidding again:-) Code does *not* contain any
+ *    code position dependencies and it's safe to include it into
+ *    shared library as is.
+ *
+ * Q. How much faster does it go?
+ * A. Do you have a good benchmark? In either case below is what I
+ *    experience with crypto/bn/expspeed.c test program:
+ *
+ *	v8plus module on U10/300MHz against bn_asm.c compiled with:
+ *
+ *	cc-5.0 -xarch=v8plus -xO5 -xdepend	+7-12%
+ *	cc-4.2 -xarch=v8plus -xO5 -xdepend	+25-35%
+ *	egcs-1.1.2 -mcpu=ultrasparc -O3		+35-45%
+ *
+ *	v8 module on SS10/60MHz against bn_asm.c compiled with:
+ *
+ *	cc-5.0 -xarch=v8 -xO5 -xdepend		+7-10%
+ *	cc-4.2 -xarch=v8 -xO5 -xdepend		+10%
+ *	egcs-1.1.2 -mv8 -O3			+35-45%
+ *
+ *    As you can see it's damn hard to beat the new Sun C compiler
+ *    and it's in first place GNU C users who will appreciate this
+ *    assembler implementation:-)	
+ */
+
+/*
+ * Revision history.
+ *
+ * 1.0	- initial release;
+ * 1.1	- new loop unrolling model(*);
+ *	- some more fine tuning;
+ * 1.2	- made gas friendly;
+ *	- updates to documentation concerning v9;
+ *	- new performance comparison matrix;
+ * 1.3	- fixed problem with /usr/ccs/lib/cpp;
+ * 1.4	- native V9 bn_*_comba[48] implementation (15% more efficient)
+ *	  resulting in slight overall performance kick;
+ *	- some retunes;
+ *	- support for GNU as added;
+ *
+ * (*)	Originally unrolled loop looked like this:
+ *	    for (;;) {
+ *		op(p+0); if (--n==0) break;
+ *		op(p+1); if (--n==0) break;
+ *		op(p+2); if (--n==0) break;
+ *		op(p+3); if (--n==0) break;
+ *		p+=4;
+ *	    }
+ *	I unroll according to following:
+ *	    while (n&~3) {
+ *		op(p+0); op(p+1); op(p+2); op(p+3);
+ *		p+=4; n=-4;
+ *	    }
+ *	    if (n) {
+ *		op(p+0); if (--n==0) return;
+ *		op(p+2); if (--n==0) return;
+ *		op(p+3); return;
+ *	    }
+ */
+
+#if defined(__SUNPRO_C) && defined(__sparcv9)
+  /* They've said -xarch=v9 at command line */
+  .register	%g2,#scratch
+  .register	%g3,#scratch
+# define	FRAME_SIZE	-192
+#elif defined(__GNUC__) && defined(__arch64__)
+  /* They've said -m64 at command line */
+  .register	%g2,#scratch
+  .register	%g3,#scratch
+# define	FRAME_SIZE	-192
+#else 
+# define	FRAME_SIZE	-96
+#endif 
+/*
+ * GNU assembler can't stand stuw:-(
+ */
+#define stuw st
+
+.section	".text",#alloc,#execinstr
+.file		"bn_asm.sparc.v8plus.S"
+
+.align	32
+
+.global bn_mul_add_words
+/*
+ * BN_ULONG bn_mul_add_words(rp,ap,num,w)
+ * BN_ULONG *rp,*ap;
+ * int num;
+ * BN_ULONG w;
+ */
+bn_mul_add_words:
+	sra	%o2,%g0,%o2	! signx %o2
+	brgz,a	%o2,.L_bn_mul_add_words_proceed
+	lduw	[%o1],%g2
+	retl
+	clr	%o0
+	nop
+	nop
+	nop
+
+.L_bn_mul_add_words_proceed:
+	srl	%o3,%g0,%o3	! clruw	%o3
+	andcc	%o2,-4,%g0
+	bz,pn	%icc,.L_bn_mul_add_words_tail
+	clr	%o5
+
+.L_bn_mul_add_words_loop:	! wow! 32 aligned!
+	lduw	[%o0],%g1
+	lduw	[%o1+4],%g3
+	mulx	%o3,%g2,%g2
+	add	%g1,%o5,%o4
+	nop
+	add	%o4,%g2,%o4
+	stuw	%o4,[%o0]
+	srlx	%o4,32,%o5
+
+	lduw	[%o0+4],%g1
+	lduw	[%o1+8],%g2
+	mulx	%o3,%g3,%g3
+	add	%g1,%o5,%o4
+	dec	4,%o2
+	add	%o4,%g3,%o4
+	stuw	%o4,[%o0+4]
+	srlx	%o4,32,%o5
+
+	lduw	[%o0+8],%g1
+	lduw	[%o1+12],%g3
+	mulx	%o3,%g2,%g2
+	add	%g1,%o5,%o4
+	inc	16,%o1
+	add	%o4,%g2,%o4
+	stuw	%o4,[%o0+8]
+	srlx	%o4,32,%o5
+
+	lduw	[%o0+12],%g1
+	mulx	%o3,%g3,%g3
+	add	%g1,%o5,%o4
+	inc	16,%o0
+	add	%o4,%g3,%o4
+	andcc	%o2,-4,%g0
+	stuw	%o4,[%o0-4]
+	srlx	%o4,32,%o5
+	bnz,a,pt	%icc,.L_bn_mul_add_words_loop
+	lduw	[%o1],%g2
+
+	brnz,a,pn	%o2,.L_bn_mul_add_words_tail
+	lduw	[%o1],%g2
+.L_bn_mul_add_words_return:
+	retl
+	mov	%o5,%o0
+
+.L_bn_mul_add_words_tail:
+	lduw	[%o0],%g1
+	mulx	%o3,%g2,%g2
+	add	%g1,%o5,%o4
+	dec	%o2
+	add	%o4,%g2,%o4
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_mul_add_words_return
+	stuw	%o4,[%o0]
+
+	lduw	[%o1+4],%g2
+	lduw	[%o0+4],%g1
+	mulx	%o3,%g2,%g2
+	add	%g1,%o5,%o4
+	dec	%o2
+	add	%o4,%g2,%o4
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_mul_add_words_return
+	stuw	%o4,[%o0+4]
+
+	lduw	[%o1+8],%g2
+	lduw	[%o0+8],%g1
+	mulx	%o3,%g2,%g2
+	add	%g1,%o5,%o4
+	add	%o4,%g2,%o4
+	stuw	%o4,[%o0+8]
+	retl
+	srlx	%o4,32,%o0
+
+.type	bn_mul_add_words,#function
+.size	bn_mul_add_words,(.-bn_mul_add_words)
+
+.align	32
+
+.global bn_mul_words
+/*
+ * BN_ULONG bn_mul_words(rp,ap,num,w)
+ * BN_ULONG *rp,*ap;
+ * int num;
+ * BN_ULONG w;
+ */
+bn_mul_words:
+	sra	%o2,%g0,%o2	! signx %o2
+	brgz,a	%o2,.L_bn_mul_words_proceeed
+	lduw	[%o1],%g2
+	retl
+	clr	%o0
+	nop
+	nop
+	nop
+
+.L_bn_mul_words_proceeed:
+	srl	%o3,%g0,%o3	! clruw	%o3
+	andcc	%o2,-4,%g0
+	bz,pn	%icc,.L_bn_mul_words_tail
+	clr	%o5
+
+.L_bn_mul_words_loop:		! wow! 32 aligned!
+	lduw	[%o1+4],%g3
+	mulx	%o3,%g2,%g2
+	add	%g2,%o5,%o4
+	nop
+	stuw	%o4,[%o0]
+	srlx	%o4,32,%o5
+
+	lduw	[%o1+8],%g2
+	mulx	%o3,%g3,%g3
+	add	%g3,%o5,%o4
+	dec	4,%o2
+	stuw	%o4,[%o0+4]
+	srlx	%o4,32,%o5
+
+	lduw	[%o1+12],%g3
+	mulx	%o3,%g2,%g2
+	add	%g2,%o5,%o4
+	inc	16,%o1
+	stuw	%o4,[%o0+8]
+	srlx	%o4,32,%o5
+
+	mulx	%o3,%g3,%g3
+	add	%g3,%o5,%o4
+	inc	16,%o0
+	stuw	%o4,[%o0-4]
+	srlx	%o4,32,%o5
+	andcc	%o2,-4,%g0
+	bnz,a,pt	%icc,.L_bn_mul_words_loop
+	lduw	[%o1],%g2
+	nop
+	nop
+
+	brnz,a,pn	%o2,.L_bn_mul_words_tail
+	lduw	[%o1],%g2
+.L_bn_mul_words_return:
+	retl
+	mov	%o5,%o0
+
+.L_bn_mul_words_tail:
+	mulx	%o3,%g2,%g2
+	add	%g2,%o5,%o4
+	dec	%o2
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_mul_words_return
+	stuw	%o4,[%o0]
+
+	lduw	[%o1+4],%g2
+	mulx	%o3,%g2,%g2
+	add	%g2,%o5,%o4
+	dec	%o2
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_mul_words_return
+	stuw	%o4,[%o0+4]
+
+	lduw	[%o1+8],%g2
+	mulx	%o3,%g2,%g2
+	add	%g2,%o5,%o4
+	stuw	%o4,[%o0+8]
+	retl
+	srlx	%o4,32,%o0
+
+.type	bn_mul_words,#function
+.size	bn_mul_words,(.-bn_mul_words)
+
+.align  32
+.global	bn_sqr_words
+/*
+ * void bn_sqr_words(r,a,n)
+ * BN_ULONG *r,*a;
+ * int n;
+ */
+bn_sqr_words:
+	sra	%o2,%g0,%o2	! signx %o2
+	brgz,a	%o2,.L_bn_sqr_words_proceeed
+	lduw	[%o1],%g2
+	retl
+	clr	%o0
+	nop
+	nop
+	nop
+
+.L_bn_sqr_words_proceeed:
+	andcc	%o2,-4,%g0
+	nop
+	bz,pn	%icc,.L_bn_sqr_words_tail
+	nop
+
+.L_bn_sqr_words_loop:		! wow! 32 aligned!
+	lduw	[%o1+4],%g3
+	mulx	%g2,%g2,%o4
+	stuw	%o4,[%o0]
+	srlx	%o4,32,%o5
+	stuw	%o5,[%o0+4]
+	nop
+
+	lduw	[%o1+8],%g2
+	mulx	%g3,%g3,%o4
+	dec	4,%o2
+	stuw	%o4,[%o0+8]
+	srlx	%o4,32,%o5
+	stuw	%o5,[%o0+12]
+
+	lduw	[%o1+12],%g3
+	mulx	%g2,%g2,%o4
+	srlx	%o4,32,%o5
+	stuw	%o4,[%o0+16]
+	inc	16,%o1
+	stuw	%o5,[%o0+20]
+
+	mulx	%g3,%g3,%o4
+	inc	32,%o0
+	stuw	%o4,[%o0-8]
+	srlx	%o4,32,%o5
+	andcc	%o2,-4,%g2
+	stuw	%o5,[%o0-4]
+	bnz,a,pt	%icc,.L_bn_sqr_words_loop
+	lduw	[%o1],%g2
+	nop
+
+	brnz,a,pn	%o2,.L_bn_sqr_words_tail
+	lduw	[%o1],%g2
+.L_bn_sqr_words_return:
+	retl
+	clr	%o0
+
+.L_bn_sqr_words_tail:
+	mulx	%g2,%g2,%o4
+	dec	%o2
+	stuw	%o4,[%o0]
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_sqr_words_return
+	stuw	%o5,[%o0+4]
+
+	lduw	[%o1+4],%g2
+	mulx	%g2,%g2,%o4
+	dec	%o2
+	stuw	%o4,[%o0+8]
+	srlx	%o4,32,%o5
+	brz,pt	%o2,.L_bn_sqr_words_return
+	stuw	%o5,[%o0+12]
+
+	lduw	[%o1+8],%g2
+	mulx	%g2,%g2,%o4
+	srlx	%o4,32,%o5
+	stuw	%o4,[%o0+16]
+	stuw	%o5,[%o0+20]
+	retl
+	clr	%o0
+
+.type	bn_sqr_words,#function
+.size	bn_sqr_words,(.-bn_sqr_words)
+
+.align	32
+.global bn_div_words
+/*
+ * BN_ULONG bn_div_words(h,l,d)
+ * BN_ULONG h,l,d;
+ */
+bn_div_words:
+	sllx	%o0,32,%o0
+	or	%o0,%o1,%o0
+	udivx	%o0,%o2,%o0
+	retl
+	srl	%o0,%g0,%o0	! clruw	%o0
+
+.type	bn_div_words,#function
+.size	bn_div_words,(.-bn_div_words)
+
+.align	32
+
+.global bn_add_words
+/*
+ * BN_ULONG bn_add_words(rp,ap,bp,n)
+ * BN_ULONG *rp,*ap,*bp;
+ * int n;
+ */
+bn_add_words:
+	sra	%o3,%g0,%o3	! signx %o3
+	brgz,a	%o3,.L_bn_add_words_proceed
+	lduw	[%o1],%o4
+	retl
+	clr	%o0
+
+.L_bn_add_words_proceed:
+	andcc	%o3,-4,%g0
+	bz,pn	%icc,.L_bn_add_words_tail
+	addcc	%g0,0,%g0	! clear carry flag
+
+.L_bn_add_words_loop:		! wow! 32 aligned!
+	dec	4,%o3
+	lduw	[%o2],%o5
+	lduw	[%o1+4],%g1
+	lduw	[%o2+4],%g2
+	lduw	[%o1+8],%g3
+	lduw	[%o2+8],%g4
+	addccc	%o5,%o4,%o5
+	stuw	%o5,[%o0]
+
+	lduw	[%o1+12],%o4
+	lduw	[%o2+12],%o5
+	inc	16,%o1
+	addccc	%g1,%g2,%g1
+	stuw	%g1,[%o0+4]
+	
+	inc	16,%o2
+	addccc	%g3,%g4,%g3
+	stuw	%g3,[%o0+8]
+
+	inc	16,%o0
+	addccc	%o5,%o4,%o5
+	stuw	%o5,[%o0-4]
+	and	%o3,-4,%g1
+	brnz,a,pt	%g1,.L_bn_add_words_loop
+	lduw	[%o1],%o4
+
+	brnz,a,pn	%o3,.L_bn_add_words_tail
+	lduw	[%o1],%o4
+.L_bn_add_words_return:
+	clr	%o0
+	retl
+	movcs	%icc,1,%o0
+	nop
+
+.L_bn_add_words_tail:
+	lduw	[%o2],%o5
+	dec	%o3
+	addccc	%o5,%o4,%o5
+	brz,pt	%o3,.L_bn_add_words_return
+	stuw	%o5,[%o0]
+
+	lduw	[%o1+4],%o4
+	lduw	[%o2+4],%o5
+	dec	%o3
+	addccc	%o5,%o4,%o5
+	brz,pt	%o3,.L_bn_add_words_return
+	stuw	%o5,[%o0+4]
+
+	lduw	[%o1+8],%o4
+	lduw	[%o2+8],%o5
+	addccc	%o5,%o4,%o5
+	stuw	%o5,[%o0+8]
+	clr	%o0
+	retl
+	movcs	%icc,1,%o0
+
+.type	bn_add_words,#function
+.size	bn_add_words,(.-bn_add_words)
+
+.global bn_sub_words
+/*
+ * BN_ULONG bn_sub_words(rp,ap,bp,n)
+ * BN_ULONG *rp,*ap,*bp;
+ * int n;
+ */
+bn_sub_words:
+	sra	%o3,%g0,%o3	! signx %o3
+	brgz,a	%o3,.L_bn_sub_words_proceed
+	lduw	[%o1],%o4
+	retl
+	clr	%o0
+
+.L_bn_sub_words_proceed:
+	andcc	%o3,-4,%g0
+	bz,pn	%icc,.L_bn_sub_words_tail
+	addcc	%g0,0,%g0	! clear carry flag
+
+.L_bn_sub_words_loop:		! wow! 32 aligned!
+	dec	4,%o3
+	lduw	[%o2],%o5
+	lduw	[%o1+4],%g1
+	lduw	[%o2+4],%g2
+	lduw	[%o1+8],%g3
+	lduw	[%o2+8],%g4
+	subccc	%o4,%o5,%o5
+	stuw	%o5,[%o0]
+
+	lduw	[%o1+12],%o4
+	lduw	[%o2+12],%o5
+	inc	16,%o1
+	subccc	%g1,%g2,%g2
+	stuw	%g2,[%o0+4]
+
+	inc	16,%o2
+	subccc	%g3,%g4,%g4
+	stuw	%g4,[%o0+8]
+
+	inc	16,%o0
+	subccc	%o4,%o5,%o5
+	stuw	%o5,[%o0-4]
+	and	%o3,-4,%g1
+	brnz,a,pt	%g1,.L_bn_sub_words_loop
+	lduw	[%o1],%o4
+
+	brnz,a,pn	%o3,.L_bn_sub_words_tail
+	lduw	[%o1],%o4
+.L_bn_sub_words_return:
+	clr	%o0
+	retl
+	movcs	%icc,1,%o0
+	nop
+
+.L_bn_sub_words_tail:		! wow! 32 aligned!
+	lduw	[%o2],%o5
+	dec	%o3
+	subccc	%o4,%o5,%o5
+	brz,pt	%o3,.L_bn_sub_words_return
+	stuw	%o5,[%o0]
+
+	lduw	[%o1+4],%o4
+	lduw	[%o2+4],%o5
+	dec	%o3
+	subccc	%o4,%o5,%o5
+	brz,pt	%o3,.L_bn_sub_words_return
+	stuw	%o5,[%o0+4]
+
+	lduw	[%o1+8],%o4
+	lduw	[%o2+8],%o5
+	subccc	%o4,%o5,%o5
+	stuw	%o5,[%o0+8]
+	clr	%o0
+	retl
+	movcs	%icc,1,%o0
+
+.type	bn_sub_words,#function
+.size	bn_sub_words,(.-bn_sub_words)
+
+/*
+ * Code below depends on the fact that upper parts of the %l0-%l7
+ * and %i0-%i7 are zeroed by kernel after context switch. In
+ * previous versions this comment stated that "the trouble is that
+ * it's not feasible to implement the mumbo-jumbo in less V9
+ * instructions:-(" which apparently isn't true thanks to
+ * 'bcs,a %xcc,.+8; inc %rd' pair. But the performance improvement
+ * results not from the shorter code, but from elimination of
+ * multicycle none-pairable 'rd %y,%rd' instructions.
+ *
+ *							Andy.
+ */
+
+/*
+ * Here is register usage map for *all* routines below.
+ */
+#define t_1	%o0
+#define	t_2	%o1
+#define c_12	%o2
+#define c_3	%o3
+
+#define ap(I)	[%i1+4*I]
+#define bp(I)	[%i2+4*I]
+#define rp(I)	[%i0+4*I]
+
+#define	a_0	%l0
+#define	a_1	%l1
+#define	a_2	%l2
+#define	a_3	%l3
+#define	a_4	%l4
+#define	a_5	%l5
+#define	a_6	%l6
+#define	a_7	%l7
+
+#define	b_0	%i3
+#define	b_1	%i4
+#define	b_2	%i5
+#define	b_3	%o4
+#define	b_4	%o5
+#define	b_5	%o7
+#define	b_6	%g1
+#define	b_7	%g4
+
+.align	32
+.global bn_mul_comba8
+/*
+ * void bn_mul_comba8(r,a,b)
+ * BN_ULONG *r,*a,*b;
+ */
+bn_mul_comba8:
+	save	%sp,FRAME_SIZE,%sp
+	mov	1,t_2
+	lduw	ap(0),a_0
+	sllx	t_2,32,t_2
+	lduw	bp(0),b_0	!=
+	lduw	bp(1),b_1
+	mulx	a_0,b_0,t_1	!mul_add_c(a[0],b[0],c1,c2,c3);
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(0)	!=!r[0]=c1;
+
+	lduw	ap(1),a_1
+	mulx	a_0,b_1,t_1	!mul_add_c(a[0],b[1],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3		!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(2),a_2
+	mulx	a_1,b_0,t_1	!=!mul_add_c(a[1],b[0],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(1)	!r[1]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,b_0,t_1	!mul_add_c(a[2],b[0],c3,c1,c2);
+	addcc	c_12,t_1,c_12	!=
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	bp(2),b_2	!=
+	mulx	a_1,b_1,t_1	!mul_add_c(a[1],b[1],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	lduw	bp(3),b_3
+	mulx	a_0,b_2,t_1	!mul_add_c(a[0],b[2],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(2)	!r[2]=c3;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_0,b_3,t_1	!mul_add_c(a[0],b[3],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_1,b_2,t_1	!=!mul_add_c(a[1],b[2],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	lduw	ap(3),a_3
+	mulx	a_2,b_1,t_1	!mul_add_c(a[2],b[1],c1,c2,c3);
+	addcc	c_12,t_1,c_12	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(4),a_4
+	mulx	a_3,b_0,t_1	!=!mul_add_c(a[3],b[0],c1,c2,c3);!=
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(3)	!r[3]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_4,b_0,t_1	!mul_add_c(a[4],b[0],c2,c3,c1);
+	addcc	c_12,t_1,c_12	!=
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,b_1,t_1	!=!mul_add_c(a[3],b[1],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_2,b_2,t_1	!=!mul_add_c(a[2],b[2],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	bp(4),b_4	!=
+	mulx	a_1,b_3,t_1	!mul_add_c(a[1],b[3],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	lduw	bp(5),b_5
+	mulx	a_0,b_4,t_1	!mul_add_c(a[0],b[4],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(4)	!r[4]=c2;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_0,b_5,t_1	!mul_add_c(a[0],b[5],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_1,b_4,t_1	!mul_add_c(a[1],b[4],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_2,b_3,t_1	!mul_add_c(a[2],b[3],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_3,b_2,t_1	!mul_add_c(a[3],b[2],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	lduw	ap(5),a_5
+	mulx	a_4,b_1,t_1	!mul_add_c(a[4],b[1],c3,c1,c2);
+	addcc	c_12,t_1,c_12	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(6),a_6
+	mulx	a_5,b_0,t_1	!=!mul_add_c(a[5],b[0],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(5)	!r[5]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_6,b_0,t_1	!mul_add_c(a[6],b[0],c1,c2,c3);
+	addcc	c_12,t_1,c_12	!=
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_5,b_1,t_1	!=!mul_add_c(a[5],b[1],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_4,b_2,t_1	!=!mul_add_c(a[4],b[2],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,b_3,t_1	!=!mul_add_c(a[3],b[3],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_2,b_4,t_1	!=!mul_add_c(a[2],b[4],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	bp(6),b_6	!=
+	mulx	a_1,b_5,t_1	!mul_add_c(a[1],b[5],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	lduw	bp(7),b_7
+	mulx	a_0,b_6,t_1	!mul_add_c(a[0],b[6],c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(6)	!r[6]=c1;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_0,b_7,t_1	!mul_add_c(a[0],b[7],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_1,b_6,t_1	!mul_add_c(a[1],b[6],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_2,b_5,t_1	!mul_add_c(a[2],b[5],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_3,b_4,t_1	!mul_add_c(a[3],b[4],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_4,b_3,t_1	!mul_add_c(a[4],b[3],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_5,b_2,t_1	!mul_add_c(a[5],b[2],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	lduw	ap(7),a_7
+	mulx	a_6,b_1,t_1	!=!mul_add_c(a[6],b[1],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_7,b_0,t_1	!=!mul_add_c(a[7],b[0],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(7)	!r[7]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_7,b_1,t_1	!=!mul_add_c(a[7],b[1],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_6,b_2,t_1	!mul_add_c(a[6],b[2],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_5,b_3,t_1	!mul_add_c(a[5],b[3],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_4,b_4,t_1	!mul_add_c(a[4],b[4],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_3,b_5,t_1	!mul_add_c(a[3],b[5],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_2,b_6,t_1	!mul_add_c(a[2],b[6],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_1,b_7,t_1	!mul_add_c(a[1],b[7],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(8)	!r[8]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,b_7,t_1	!=!mul_add_c(a[2],b[7],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	mulx	a_3,b_6,t_1	!mul_add_c(a[3],b[6],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_4,b_5,t_1	!mul_add_c(a[4],b[5],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_5,b_4,t_1	!mul_add_c(a[5],b[4],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_6,b_3,t_1	!mul_add_c(a[6],b[3],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_7,b_2,t_1	!mul_add_c(a[7],b[2],c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(9)	!r[9]=c1;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_7,b_3,t_1	!mul_add_c(a[7],b[3],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_6,b_4,t_1	!mul_add_c(a[6],b[4],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_5,b_5,t_1	!mul_add_c(a[5],b[5],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_4,b_6,t_1	!mul_add_c(a[4],b[6],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_3,b_7,t_1	!mul_add_c(a[3],b[7],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(10)	!r[10]=c2;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_4,b_7,t_1	!mul_add_c(a[4],b[7],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_5,b_6,t_1	!mul_add_c(a[5],b[6],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_6,b_5,t_1	!mul_add_c(a[6],b[5],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_7,b_4,t_1	!mul_add_c(a[7],b[4],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(11)	!r[11]=c3;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_7,b_5,t_1	!mul_add_c(a[7],b[5],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_6,b_6,t_1	!mul_add_c(a[6],b[6],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_5,b_7,t_1	!mul_add_c(a[5],b[7],c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(12)	!r[12]=c1;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_6,b_7,t_1	!mul_add_c(a[6],b[7],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_7,b_6,t_1	!mul_add_c(a[7],b[6],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	st	t_1,rp(13)	!r[13]=c2;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_7,b_7,t_1	!mul_add_c(a[7],b[7],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(14)	!r[14]=c3;
+	stuw	c_12,rp(15)	!r[15]=c1;
+
+	ret
+	restore	%g0,%g0,%o0	!=
+
+.type	bn_mul_comba8,#function
+.size	bn_mul_comba8,(.-bn_mul_comba8)
+
+.align	32
+
+.global bn_mul_comba4
+/*
+ * void bn_mul_comba4(r,a,b)
+ * BN_ULONG *r,*a,*b;
+ */
+bn_mul_comba4:
+	save	%sp,FRAME_SIZE,%sp
+	lduw	ap(0),a_0
+	mov	1,t_2
+	lduw	bp(0),b_0
+	sllx	t_2,32,t_2	!=
+	lduw	bp(1),b_1
+	mulx	a_0,b_0,t_1	!mul_add_c(a[0],b[0],c1,c2,c3);
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(0)	!=!r[0]=c1;
+
+	lduw	ap(1),a_1
+	mulx	a_0,b_1,t_1	!mul_add_c(a[0],b[1],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3		!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(2),a_2
+	mulx	a_1,b_0,t_1	!=!mul_add_c(a[1],b[0],c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(1)	!r[1]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,b_0,t_1	!mul_add_c(a[2],b[0],c3,c1,c2);
+	addcc	c_12,t_1,c_12	!=
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	bp(2),b_2	!=
+	mulx	a_1,b_1,t_1	!mul_add_c(a[1],b[1],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3	!=
+	lduw	bp(3),b_3
+	mulx	a_0,b_2,t_1	!mul_add_c(a[0],b[2],c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(2)	!r[2]=c3;
+	or	c_12,c_3,c_12	!=
+
+	mulx	a_0,b_3,t_1	!mul_add_c(a[0],b[3],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	mulx	a_1,b_2,t_1	!mul_add_c(a[1],b[2],c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8	!=
+	add	c_3,t_2,c_3
+	lduw	ap(3),a_3
+	mulx	a_2,b_1,t_1	!mul_add_c(a[2],b[1],c1,c2,c3);
+	addcc	c_12,t_1,c_12	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,b_0,t_1	!mul_add_c(a[3],b[0],c1,c2,c3);!=
+	addcc	c_12,t_1,t_1	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(3)	!=!r[3]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_3,b_1,t_1	!mul_add_c(a[3],b[1],c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3		!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_2,b_2,t_1	!mul_add_c(a[2],b[2],c2,c3,c1);
+	addcc	c_12,t_1,c_12	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_1,b_3,t_1	!mul_add_c(a[1],b[3],c2,c3,c1);
+	addcc	c_12,t_1,t_1	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(4)	!=!r[4]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,b_3,t_1	!mul_add_c(a[2],b[3],c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3		!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,b_2,t_1	!mul_add_c(a[3],b[2],c3,c1,c2);
+	addcc	c_12,t_1,t_1	!=
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(5)	!=!r[5]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_3,b_3,t_1	!mul_add_c(a[3],b[3],c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	srlx	t_1,32,c_12	!=
+	stuw	t_1,rp(6)	!r[6]=c1;
+	stuw	c_12,rp(7)	!r[7]=c2;
+	
+	ret
+	restore	%g0,%g0,%o0
+
+.type	bn_mul_comba4,#function
+.size	bn_mul_comba4,(.-bn_mul_comba4)
+
+.align	32
+
+.global bn_sqr_comba8
+bn_sqr_comba8:
+	save	%sp,FRAME_SIZE,%sp
+	mov	1,t_2
+	lduw	ap(0),a_0
+	sllx	t_2,32,t_2
+	lduw	ap(1),a_1
+	mulx	a_0,a_0,t_1	!sqr_add_c(a,0,c1,c2,c3);
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(0)	!r[0]=c1;
+
+	lduw	ap(2),a_2
+	mulx	a_0,a_1,t_1	!=!sqr_add_c2(a,1,0,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(1)	!r[1]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,a_0,t_1	!sqr_add_c2(a,2,0,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(3),a_3
+	mulx	a_1,a_1,t_1	!sqr_add_c(a,1,c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(2)	!r[2]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_0,a_3,t_1	!sqr_add_c2(a,3,0,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(4),a_4
+	mulx	a_1,a_2,t_1	!sqr_add_c2(a,2,1,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	st	t_1,rp(3)	!r[3]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_4,a_0,t_1	!sqr_add_c2(a,4,0,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,a_1,t_1	!sqr_add_c2(a,3,1,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(5),a_5
+	mulx	a_2,a_2,t_1	!sqr_add_c(a,2,c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(4)	!r[4]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_0,a_5,t_1	!sqr_add_c2(a,5,0,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_1,a_4,t_1	!sqr_add_c2(a,4,1,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(6),a_6
+	mulx	a_2,a_3,t_1	!sqr_add_c2(a,3,2,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(5)	!r[5]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_6,a_0,t_1	!sqr_add_c2(a,6,0,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_5,a_1,t_1	!sqr_add_c2(a,5,1,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_4,a_2,t_1	!sqr_add_c2(a,4,2,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(7),a_7
+	mulx	a_3,a_3,t_1	!=!sqr_add_c(a,3,c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(6)	!r[6]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_0,a_7,t_1	!sqr_add_c2(a,7,0,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_1,a_6,t_1	!sqr_add_c2(a,6,1,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_2,a_5,t_1	!sqr_add_c2(a,5,2,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,a_4,t_1	!sqr_add_c2(a,4,3,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(7)	!r[7]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_7,a_1,t_1	!sqr_add_c2(a,7,1,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_6,a_2,t_1	!sqr_add_c2(a,6,2,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_5,a_3,t_1	!sqr_add_c2(a,5,3,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_4,a_4,t_1	!sqr_add_c(a,4,c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(8)	!r[8]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,a_7,t_1	!sqr_add_c2(a,7,2,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_3,a_6,t_1	!sqr_add_c2(a,6,3,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_4,a_5,t_1	!sqr_add_c2(a,5,4,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(9)	!r[9]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_7,a_3,t_1	!sqr_add_c2(a,7,3,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_6,a_4,t_1	!sqr_add_c2(a,6,4,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_5,a_5,t_1	!sqr_add_c(a,5,c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(10)	!r[10]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_4,a_7,t_1	!sqr_add_c2(a,7,4,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_5,a_6,t_1	!sqr_add_c2(a,6,5,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(11)	!r[11]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_7,a_5,t_1	!sqr_add_c2(a,7,5,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_6,a_6,t_1	!sqr_add_c(a,6,c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(12)	!r[12]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_6,a_7,t_1	!sqr_add_c2(a,7,6,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(13)	!r[13]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_7,a_7,t_1	!sqr_add_c(a,7,c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(14)	!r[14]=c3;
+	stuw	c_12,rp(15)	!r[15]=c1;
+
+	ret
+	restore	%g0,%g0,%o0
+
+.type	bn_sqr_comba8,#function
+.size	bn_sqr_comba8,(.-bn_sqr_comba8)
+
+.align	32
+
+.global bn_sqr_comba4
+/*
+ * void bn_sqr_comba4(r,a)
+ * BN_ULONG *r,*a;
+ */
+bn_sqr_comba4:
+	save	%sp,FRAME_SIZE,%sp
+	mov	1,t_2
+	lduw	ap(0),a_0
+	sllx	t_2,32,t_2
+	lduw	ap(1),a_1
+	mulx	a_0,a_0,t_1	!sqr_add_c(a,0,c1,c2,c3);
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(0)	!r[0]=c1;
+
+	lduw	ap(2),a_2
+	mulx	a_0,a_1,t_1	!sqr_add_c2(a,1,0,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(1)	!r[1]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,a_0,t_1	!sqr_add_c2(a,2,0,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	lduw	ap(3),a_3
+	mulx	a_1,a_1,t_1	!sqr_add_c(a,1,c3,c1,c2);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(2)	!r[2]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_0,a_3,t_1	!sqr_add_c2(a,3,0,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_1,a_2,t_1	!sqr_add_c2(a,2,1,c1,c2,c3);
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(3)	!r[3]=c1;
+	or	c_12,c_3,c_12
+
+	mulx	a_3,a_1,t_1	!sqr_add_c2(a,3,1,c2,c3,c1);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,c_12
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	mulx	a_2,a_2,t_1	!sqr_add_c(a,2,c2,c3,c1);
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(4)	!r[4]=c2;
+	or	c_12,c_3,c_12
+
+	mulx	a_2,a_3,t_1	!sqr_add_c2(a,3,2,c3,c1,c2);
+	addcc	c_12,t_1,c_12
+	clr	c_3
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	addcc	c_12,t_1,t_1
+	bcs,a	%xcc,.+8
+	add	c_3,t_2,c_3
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(5)	!r[5]=c3;
+	or	c_12,c_3,c_12
+
+	mulx	a_3,a_3,t_1	!sqr_add_c(a,3,c1,c2,c3);
+	addcc	c_12,t_1,t_1
+	srlx	t_1,32,c_12
+	stuw	t_1,rp(6)	!r[6]=c1;
+	stuw	c_12,rp(7)	!r[7]=c2;
+	
+	ret
+	restore	%g0,%g0,%o0
+
+.type	bn_sqr_comba4,#function
+.size	bn_sqr_comba4,(.-bn_sqr_comba4)
+
+.align	32