xref: /freebsd/crypto/openssl/crypto/bn/asm/ppc.pl (revision 38f0b757fd84d17d0fc24739a7cda160c4516d81)
1#!/usr/bin/env perl
2#
3# Implemented as a Perl wrapper as we want to support several different
4# architectures with single file. We pick up the target based on the
5# file name we are asked to generate.
6#
7# It should be noted though that this perl code is nothing like
8# <openssl>/crypto/perlasm/x86*. In this case perl is used pretty much
9# as pre-processor to cover for platform differences in name decoration,
10# linker tables, 32-/64-bit instruction sets...
11#
12# As you might know there're several PowerPC ABI in use. Most notably
13# Linux and AIX use different 32-bit ABIs. Good news are that these ABIs
14# are similar enough to implement leaf(!) functions, which would be ABI
15# neutral. And that's what you find here: ABI neutral leaf functions.
16# In case you wonder what that is...
17#
18#       AIX performance
19#
20#	MEASUREMENTS WITH cc ON a 200 MhZ PowerPC 604e.
21#
22#	The following is the performance of 32-bit compiler
23#	generated code:
24#
25#	OpenSSL 0.9.6c 21 dec 2001
26#	built on: Tue Jun 11 11:06:51 EDT 2002
27#	options:bn(64,32) ...
28#compiler: cc -DTHREADS  -DAIX -DB_ENDIAN -DBN_LLONG -O3
29#                  sign    verify    sign/s verify/s
30#rsa  512 bits   0.0098s   0.0009s    102.0   1170.6
31#rsa 1024 bits   0.0507s   0.0026s     19.7    387.5
32#rsa 2048 bits   0.3036s   0.0085s      3.3    117.1
33#rsa 4096 bits   2.0040s   0.0299s      0.5     33.4
34#dsa  512 bits   0.0087s   0.0106s    114.3     94.5
35#dsa 1024 bits   0.0256s   0.0313s     39.0     32.0
36#
37#	Same bechmark with this assembler code:
38#
39#rsa  512 bits   0.0056s   0.0005s    178.6   2049.2
40#rsa 1024 bits   0.0283s   0.0015s     35.3    674.1
41#rsa 2048 bits   0.1744s   0.0050s      5.7    201.2
42#rsa 4096 bits   1.1644s   0.0179s      0.9     55.7
43#dsa  512 bits   0.0052s   0.0062s    191.6    162.0
44#dsa 1024 bits   0.0149s   0.0180s     67.0     55.5
45#
46#	Number of operations increases by at almost 75%
47#
48#	Here are performance numbers for 64-bit compiler
49#	generated code:
50#
51#	OpenSSL 0.9.6g [engine] 9 Aug 2002
52#	built on: Fri Apr 18 16:59:20 EDT 2003
53#	options:bn(64,64) ...
54#	compiler: cc -DTHREADS -D_REENTRANT -q64 -DB_ENDIAN -O3
55#                  sign    verify    sign/s verify/s
56#rsa  512 bits   0.0028s   0.0003s    357.1   3844.4
57#rsa 1024 bits   0.0148s   0.0008s     67.5   1239.7
58#rsa 2048 bits   0.0963s   0.0028s     10.4    353.0
59#rsa 4096 bits   0.6538s   0.0102s      1.5     98.1
60#dsa  512 bits   0.0026s   0.0032s    382.5    313.7
61#dsa 1024 bits   0.0081s   0.0099s    122.8    100.6
62#
63#	Same benchmark with this assembler code:
64#
65#rsa  512 bits   0.0020s   0.0002s    510.4   6273.7
66#rsa 1024 bits   0.0088s   0.0005s    114.1   2128.3
67#rsa 2048 bits   0.0540s   0.0016s     18.5    622.5
68#rsa 4096 bits   0.3700s   0.0058s      2.7    171.0
69#dsa  512 bits   0.0016s   0.0020s    610.7    507.1
70#dsa 1024 bits   0.0047s   0.0058s    212.5    173.2
71#
72#	Again, performance increases by at about 75%
73#
74#       Mac OS X, Apple G5 1.8GHz (Note this is 32 bit code)
75#       OpenSSL 0.9.7c 30 Sep 2003
76#
77#       Original code.
78#
79#rsa  512 bits   0.0011s   0.0001s    906.1  11012.5
80#rsa 1024 bits   0.0060s   0.0003s    166.6   3363.1
81#rsa 2048 bits   0.0370s   0.0010s     27.1    982.4
82#rsa 4096 bits   0.2426s   0.0036s      4.1    280.4
83#dsa  512 bits   0.0010s   0.0012s   1038.1    841.5
84#dsa 1024 bits   0.0030s   0.0037s    329.6    269.7
85#dsa 2048 bits   0.0101s   0.0127s     98.9     78.6
86#
87#       Same benchmark with this assembler code:
88#
89#rsa  512 bits   0.0007s   0.0001s   1416.2  16645.9
90#rsa 1024 bits   0.0036s   0.0002s    274.4   5380.6
91#rsa 2048 bits   0.0222s   0.0006s     45.1   1589.5
92#rsa 4096 bits   0.1469s   0.0022s      6.8    449.6
93#dsa  512 bits   0.0006s   0.0007s   1664.2   1376.2
94#dsa 1024 bits   0.0018s   0.0023s    545.0    442.2
95#dsa 2048 bits   0.0061s   0.0075s    163.5    132.8
96#
97#        Performance increase of ~60%
98#
99#	If you have comments or suggestions to improve code send
100#	me a note at schari@us.ibm.com
101#
102
103$flavour = shift;
104
105if ($flavour =~ /32/) {
106	$BITS=	32;
107	$BNSZ=	$BITS/8;
108	$ISA=	"\"ppc\"";
109
110	$LD=	"lwz";		# load
111	$LDU=	"lwzu";		# load and update
112	$ST=	"stw";		# store
113	$STU=	"stwu";		# store and update
114	$UMULL=	"mullw";	# unsigned multiply low
115	$UMULH=	"mulhwu";	# unsigned multiply high
116	$UDIV=	"divwu";	# unsigned divide
117	$UCMPI=	"cmplwi";	# unsigned compare with immediate
118	$UCMP=	"cmplw";	# unsigned compare
119	$CNTLZ=	"cntlzw";	# count leading zeros
120	$SHL=	"slw";		# shift left
121	$SHR=	"srw";		# unsigned shift right
122	$SHRI=	"srwi";		# unsigned shift right by immediate
123	$SHLI=	"slwi";		# shift left by immediate
124	$CLRU=	"clrlwi";	# clear upper bits
125	$INSR=	"insrwi";	# insert right
126	$ROTL=	"rotlwi";	# rotate left by immediate
127	$TR=	"tw";		# conditional trap
128} elsif ($flavour =~ /64/) {
129	$BITS=	64;
130	$BNSZ=	$BITS/8;
131	$ISA=	"\"ppc64\"";
132
133	# same as above, but 64-bit mnemonics...
134	$LD=	"ld";		# load
135	$LDU=	"ldu";		# load and update
136	$ST=	"std";		# store
137	$STU=	"stdu";		# store and update
138	$UMULL=	"mulld";	# unsigned multiply low
139	$UMULH=	"mulhdu";	# unsigned multiply high
140	$UDIV=	"divdu";	# unsigned divide
141	$UCMPI=	"cmpldi";	# unsigned compare with immediate
142	$UCMP=	"cmpld";	# unsigned compare
143	$CNTLZ=	"cntlzd";	# count leading zeros
144	$SHL=	"sld";		# shift left
145	$SHR=	"srd";		# unsigned shift right
146	$SHRI=	"srdi";		# unsigned shift right by immediate
147	$SHLI=	"sldi";		# shift left by immediate
148	$CLRU=	"clrldi";	# clear upper bits
149	$INSR=	"insrdi";	# insert right
150	$ROTL=	"rotldi";	# rotate left by immediate
151	$TR=	"td";		# conditional trap
152} else { die "nonsense $flavour"; }
153
154$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
155( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
156( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
157die "can't locate ppc-xlate.pl";
158
159open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
160
161$data=<<EOF;
162#--------------------------------------------------------------------
163#
164#
165#
166#
167#	File:		ppc32.s
168#
169#	Created by:	Suresh Chari
170#			IBM Thomas J. Watson Research Library
171#			Hawthorne, NY
172#
173#
174#	Description:	Optimized assembly routines for OpenSSL crypto
175#			on the 32 bitPowerPC platform.
176#
177#
178#	Version History
179#
180#	2. Fixed bn_add,bn_sub and bn_div_words, added comments,
181#	   cleaned up code. Also made a single version which can
182#	   be used for both the AIX and Linux compilers. See NOTE
183#	   below.
184#				12/05/03		Suresh Chari
185#			(with lots of help from)        Andy Polyakov
186##
187#	1. Initial version	10/20/02		Suresh Chari
188#
189#
190#	The following file works for the xlc,cc
191#	and gcc compilers.
192#
193#	NOTE:	To get the file to link correctly with the gcc compiler
194#	        you have to change the names of the routines and remove
195#		the first .(dot) character. This should automatically
196#		be done in the build process.
197#
198#	Hand optimized assembly code for the following routines
199#
200#	bn_sqr_comba4
201#	bn_sqr_comba8
202#	bn_mul_comba4
203#	bn_mul_comba8
204#	bn_sub_words
205#	bn_add_words
206#	bn_div_words
207#	bn_sqr_words
208#	bn_mul_words
209#	bn_mul_add_words
210#
211#	NOTE:	It is possible to optimize this code more for
212#	specific PowerPC or Power architectures. On the Northstar
213#	architecture the optimizations in this file do
214#	 NOT provide much improvement.
215#
216#	If you have comments or suggestions to improve code send
217#	me a note at schari\@us.ibm.com
218#
219#--------------------------------------------------------------------------
220#
221#	Defines to be used in the assembly code.
222#
223#.set r0,0	# we use it as storage for value of 0
224#.set SP,1	# preserved
225#.set RTOC,2	# preserved
226#.set r3,3	# 1st argument/return value
227#.set r4,4	# 2nd argument/volatile register
228#.set r5,5	# 3rd argument/volatile register
229#.set r6,6	# ...
230#.set r7,7
231#.set r8,8
232#.set r9,9
233#.set r10,10
234#.set r11,11
235#.set r12,12
236#.set r13,13	# not used, nor any other "below" it...
237
238#	Declare function names to be global
239#	NOTE:	For gcc these names MUST be changed to remove
240#	        the first . i.e. for example change ".bn_sqr_comba4"
241#		to "bn_sqr_comba4". This should be automatically done
242#		in the build.
243
244	.globl	.bn_sqr_comba4
245	.globl	.bn_sqr_comba8
246	.globl	.bn_mul_comba4
247	.globl	.bn_mul_comba8
248	.globl	.bn_sub_words
249	.globl	.bn_add_words
250	.globl	.bn_div_words
251	.globl	.bn_sqr_words
252	.globl	.bn_mul_words
253	.globl	.bn_mul_add_words
254
255# .text section
256
257	.machine	"any"
258
259#
260#	NOTE:	The following label name should be changed to
261#		"bn_sqr_comba4" i.e. remove the first dot
262#		for the gcc compiler. This should be automatically
263#		done in the build
264#
265
266.align	4
267.bn_sqr_comba4:
268#
269# Optimized version of bn_sqr_comba4.
270#
271# void bn_sqr_comba4(BN_ULONG *r, BN_ULONG *a)
272# r3 contains r
273# r4 contains a
274#
275# Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows:
276#
277# r5,r6 are the two BN_ULONGs being multiplied.
278# r7,r8 are the results of the 32x32 giving 64 bit multiply.
279# r9,r10, r11 are the equivalents of c1,c2, c3.
280# Here's the assembly
281#
282#
283	xor		r0,r0,r0		# set r0 = 0. Used in the addze
284						# instructions below
285
286						#sqr_add_c(a,0,c1,c2,c3)
287	$LD		r5,`0*$BNSZ`(r4)
288	$UMULL		r9,r5,r5
289	$UMULH		r10,r5,r5		#in first iteration. No need
290						#to add since c1=c2=c3=0.
291						# Note c3(r11) is NOT set to 0
292						# but will be.
293
294	$ST		r9,`0*$BNSZ`(r3)	# r[0]=c1;
295						# sqr_add_c2(a,1,0,c2,c3,c1);
296	$LD		r6,`1*$BNSZ`(r4)
297	$UMULL		r7,r5,r6
298	$UMULH		r8,r5,r6
299
300	addc		r7,r7,r7		# compute (r7,r8)=2*(r7,r8)
301	adde		r8,r8,r8
302	addze		r9,r0			# catch carry if any.
303						# r9= r0(=0) and carry
304
305	addc		r10,r7,r10		# now add to temp result.
306	addze		r11,r8                  # r8 added to r11 which is 0
307	addze		r9,r9
308
309	$ST		r10,`1*$BNSZ`(r3)	#r[1]=c2;
310						#sqr_add_c(a,1,c3,c1,c2)
311	$UMULL		r7,r6,r6
312	$UMULH		r8,r6,r6
313	addc		r11,r7,r11
314	adde		r9,r8,r9
315	addze		r10,r0
316						#sqr_add_c2(a,2,0,c3,c1,c2)
317	$LD		r6,`2*$BNSZ`(r4)
318	$UMULL		r7,r5,r6
319	$UMULH		r8,r5,r6
320
321	addc		r7,r7,r7
322	adde		r8,r8,r8
323	addze		r10,r10
324
325	addc		r11,r7,r11
326	adde		r9,r8,r9
327	addze		r10,r10
328	$ST		r11,`2*$BNSZ`(r3)	#r[2]=c3
329						#sqr_add_c2(a,3,0,c1,c2,c3);
330	$LD		r6,`3*$BNSZ`(r4)
331	$UMULL		r7,r5,r6
332	$UMULH		r8,r5,r6
333	addc		r7,r7,r7
334	adde		r8,r8,r8
335	addze		r11,r0
336
337	addc		r9,r7,r9
338	adde		r10,r8,r10
339	addze		r11,r11
340						#sqr_add_c2(a,2,1,c1,c2,c3);
341	$LD		r5,`1*$BNSZ`(r4)
342	$LD		r6,`2*$BNSZ`(r4)
343	$UMULL		r7,r5,r6
344	$UMULH		r8,r5,r6
345
346	addc		r7,r7,r7
347	adde		r8,r8,r8
348	addze		r11,r11
349	addc		r9,r7,r9
350	adde		r10,r8,r10
351	addze		r11,r11
352	$ST		r9,`3*$BNSZ`(r3)	#r[3]=c1
353						#sqr_add_c(a,2,c2,c3,c1);
354	$UMULL		r7,r6,r6
355	$UMULH		r8,r6,r6
356	addc		r10,r7,r10
357	adde		r11,r8,r11
358	addze		r9,r0
359						#sqr_add_c2(a,3,1,c2,c3,c1);
360	$LD		r6,`3*$BNSZ`(r4)
361	$UMULL		r7,r5,r6
362	$UMULH		r8,r5,r6
363	addc		r7,r7,r7
364	adde		r8,r8,r8
365	addze		r9,r9
366
367	addc		r10,r7,r10
368	adde		r11,r8,r11
369	addze		r9,r9
370	$ST		r10,`4*$BNSZ`(r3)	#r[4]=c2
371						#sqr_add_c2(a,3,2,c3,c1,c2);
372	$LD		r5,`2*$BNSZ`(r4)
373	$UMULL		r7,r5,r6
374	$UMULH		r8,r5,r6
375	addc		r7,r7,r7
376	adde		r8,r8,r8
377	addze		r10,r0
378
379	addc		r11,r7,r11
380	adde		r9,r8,r9
381	addze		r10,r10
382	$ST		r11,`5*$BNSZ`(r3)	#r[5] = c3
383						#sqr_add_c(a,3,c1,c2,c3);
384	$UMULL		r7,r6,r6
385	$UMULH		r8,r6,r6
386	addc		r9,r7,r9
387	adde		r10,r8,r10
388
389	$ST		r9,`6*$BNSZ`(r3)	#r[6]=c1
390	$ST		r10,`7*$BNSZ`(r3)	#r[7]=c2
391	blr
392	.long	0
393	.byte	0,12,0x14,0,0,0,2,0
394	.long	0
395
396#
397#	NOTE:	The following label name should be changed to
398#		"bn_sqr_comba8" i.e. remove the first dot
399#		for the gcc compiler. This should be automatically
400#		done in the build
401#
402
403.align	4
404.bn_sqr_comba8:
405#
406# This is an optimized version of the bn_sqr_comba8 routine.
407# Tightly uses the adde instruction
408#
409#
410# void bn_sqr_comba8(BN_ULONG *r, BN_ULONG *a)
411# r3 contains r
412# r4 contains a
413#
414# Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows:
415#
416# r5,r6 are the two BN_ULONGs being multiplied.
417# r7,r8 are the results of the 32x32 giving 64 bit multiply.
418# r9,r10, r11 are the equivalents of c1,c2, c3.
419#
420# Possible optimization of loading all 8 longs of a into registers
421# doesnt provide any speedup
422#
423
424	xor		r0,r0,r0		#set r0 = 0.Used in addze
425						#instructions below.
426
427						#sqr_add_c(a,0,c1,c2,c3);
428	$LD		r5,`0*$BNSZ`(r4)
429	$UMULL		r9,r5,r5		#1st iteration:	no carries.
430	$UMULH		r10,r5,r5
431	$ST		r9,`0*$BNSZ`(r3)	# r[0]=c1;
432						#sqr_add_c2(a,1,0,c2,c3,c1);
433	$LD		r6,`1*$BNSZ`(r4)
434	$UMULL		r7,r5,r6
435	$UMULH		r8,r5,r6
436
437	addc		r10,r7,r10		#add the two register number
438	adde		r11,r8,r0 		# (r8,r7) to the three register
439	addze		r9,r0			# number (r9,r11,r10).NOTE:r0=0
440
441	addc		r10,r7,r10		#add the two register number
442	adde		r11,r8,r11 		# (r8,r7) to the three register
443	addze		r9,r9			# number (r9,r11,r10).
444
445	$ST		r10,`1*$BNSZ`(r3)	# r[1]=c2
446
447						#sqr_add_c(a,1,c3,c1,c2);
448	$UMULL		r7,r6,r6
449	$UMULH		r8,r6,r6
450	addc		r11,r7,r11
451	adde		r9,r8,r9
452	addze		r10,r0
453						#sqr_add_c2(a,2,0,c3,c1,c2);
454	$LD		r6,`2*$BNSZ`(r4)
455	$UMULL		r7,r5,r6
456	$UMULH		r8,r5,r6
457
458	addc		r11,r7,r11
459	adde		r9,r8,r9
460	addze		r10,r10
461
462	addc		r11,r7,r11
463	adde		r9,r8,r9
464	addze		r10,r10
465
466	$ST		r11,`2*$BNSZ`(r3)	#r[2]=c3
467						#sqr_add_c2(a,3,0,c1,c2,c3);
468	$LD		r6,`3*$BNSZ`(r4)	#r6 = a[3]. r5 is already a[0].
469	$UMULL		r7,r5,r6
470	$UMULH		r8,r5,r6
471
472	addc		r9,r7,r9
473	adde		r10,r8,r10
474	addze		r11,r0
475
476	addc		r9,r7,r9
477	adde		r10,r8,r10
478	addze		r11,r11
479						#sqr_add_c2(a,2,1,c1,c2,c3);
480	$LD		r5,`1*$BNSZ`(r4)
481	$LD		r6,`2*$BNSZ`(r4)
482	$UMULL		r7,r5,r6
483	$UMULH		r8,r5,r6
484
485	addc		r9,r7,r9
486	adde		r10,r8,r10
487	addze		r11,r11
488
489	addc		r9,r7,r9
490	adde		r10,r8,r10
491	addze		r11,r11
492
493	$ST		r9,`3*$BNSZ`(r3)	#r[3]=c1;
494						#sqr_add_c(a,2,c2,c3,c1);
495	$UMULL		r7,r6,r6
496	$UMULH		r8,r6,r6
497
498	addc		r10,r7,r10
499	adde		r11,r8,r11
500	addze		r9,r0
501						#sqr_add_c2(a,3,1,c2,c3,c1);
502	$LD		r6,`3*$BNSZ`(r4)
503	$UMULL		r7,r5,r6
504	$UMULH		r8,r5,r6
505
506	addc		r10,r7,r10
507	adde		r11,r8,r11
508	addze		r9,r9
509
510	addc		r10,r7,r10
511	adde		r11,r8,r11
512	addze		r9,r9
513						#sqr_add_c2(a,4,0,c2,c3,c1);
514	$LD		r5,`0*$BNSZ`(r4)
515	$LD		r6,`4*$BNSZ`(r4)
516	$UMULL		r7,r5,r6
517	$UMULH		r8,r5,r6
518
519	addc		r10,r7,r10
520	adde		r11,r8,r11
521	addze		r9,r9
522
523	addc		r10,r7,r10
524	adde		r11,r8,r11
525	addze		r9,r9
526	$ST		r10,`4*$BNSZ`(r3)	#r[4]=c2;
527						#sqr_add_c2(a,5,0,c3,c1,c2);
528	$LD		r6,`5*$BNSZ`(r4)
529	$UMULL		r7,r5,r6
530	$UMULH		r8,r5,r6
531
532	addc		r11,r7,r11
533	adde		r9,r8,r9
534	addze		r10,r0
535
536	addc		r11,r7,r11
537	adde		r9,r8,r9
538	addze		r10,r10
539						#sqr_add_c2(a,4,1,c3,c1,c2);
540	$LD		r5,`1*$BNSZ`(r4)
541	$LD		r6,`4*$BNSZ`(r4)
542	$UMULL		r7,r5,r6
543	$UMULH		r8,r5,r6
544
545	addc		r11,r7,r11
546	adde		r9,r8,r9
547	addze		r10,r10
548
549	addc		r11,r7,r11
550	adde		r9,r8,r9
551	addze		r10,r10
552						#sqr_add_c2(a,3,2,c3,c1,c2);
553	$LD		r5,`2*$BNSZ`(r4)
554	$LD		r6,`3*$BNSZ`(r4)
555	$UMULL		r7,r5,r6
556	$UMULH		r8,r5,r6
557
558	addc		r11,r7,r11
559	adde		r9,r8,r9
560	addze		r10,r10
561
562	addc		r11,r7,r11
563	adde		r9,r8,r9
564	addze		r10,r10
565	$ST		r11,`5*$BNSZ`(r3)	#r[5]=c3;
566						#sqr_add_c(a,3,c1,c2,c3);
567	$UMULL		r7,r6,r6
568	$UMULH		r8,r6,r6
569	addc		r9,r7,r9
570	adde		r10,r8,r10
571	addze		r11,r0
572						#sqr_add_c2(a,4,2,c1,c2,c3);
573	$LD		r6,`4*$BNSZ`(r4)
574	$UMULL		r7,r5,r6
575	$UMULH		r8,r5,r6
576
577	addc		r9,r7,r9
578	adde		r10,r8,r10
579	addze		r11,r11
580
581	addc		r9,r7,r9
582	adde		r10,r8,r10
583	addze		r11,r11
584						#sqr_add_c2(a,5,1,c1,c2,c3);
585	$LD		r5,`1*$BNSZ`(r4)
586	$LD		r6,`5*$BNSZ`(r4)
587	$UMULL		r7,r5,r6
588	$UMULH		r8,r5,r6
589
590	addc		r9,r7,r9
591	adde		r10,r8,r10
592	addze		r11,r11
593
594	addc		r9,r7,r9
595	adde		r10,r8,r10
596	addze		r11,r11
597						#sqr_add_c2(a,6,0,c1,c2,c3);
598	$LD		r5,`0*$BNSZ`(r4)
599	$LD		r6,`6*$BNSZ`(r4)
600	$UMULL		r7,r5,r6
601	$UMULH		r8,r5,r6
602	addc		r9,r7,r9
603	adde		r10,r8,r10
604	addze		r11,r11
605	addc		r9,r7,r9
606	adde		r10,r8,r10
607	addze		r11,r11
608	$ST		r9,`6*$BNSZ`(r3)	#r[6]=c1;
609						#sqr_add_c2(a,7,0,c2,c3,c1);
610	$LD		r6,`7*$BNSZ`(r4)
611	$UMULL		r7,r5,r6
612	$UMULH		r8,r5,r6
613
614	addc		r10,r7,r10
615	adde		r11,r8,r11
616	addze		r9,r0
617	addc		r10,r7,r10
618	adde		r11,r8,r11
619	addze		r9,r9
620						#sqr_add_c2(a,6,1,c2,c3,c1);
621	$LD		r5,`1*$BNSZ`(r4)
622	$LD		r6,`6*$BNSZ`(r4)
623	$UMULL		r7,r5,r6
624	$UMULH		r8,r5,r6
625
626	addc		r10,r7,r10
627	adde		r11,r8,r11
628	addze		r9,r9
629	addc		r10,r7,r10
630	adde		r11,r8,r11
631	addze		r9,r9
632						#sqr_add_c2(a,5,2,c2,c3,c1);
633	$LD		r5,`2*$BNSZ`(r4)
634	$LD		r6,`5*$BNSZ`(r4)
635	$UMULL		r7,r5,r6
636	$UMULH		r8,r5,r6
637	addc		r10,r7,r10
638	adde		r11,r8,r11
639	addze		r9,r9
640	addc		r10,r7,r10
641	adde		r11,r8,r11
642	addze		r9,r9
643						#sqr_add_c2(a,4,3,c2,c3,c1);
644	$LD		r5,`3*$BNSZ`(r4)
645	$LD		r6,`4*$BNSZ`(r4)
646	$UMULL		r7,r5,r6
647	$UMULH		r8,r5,r6
648
649	addc		r10,r7,r10
650	adde		r11,r8,r11
651	addze		r9,r9
652	addc		r10,r7,r10
653	adde		r11,r8,r11
654	addze		r9,r9
655	$ST		r10,`7*$BNSZ`(r3)	#r[7]=c2;
656						#sqr_add_c(a,4,c3,c1,c2);
657	$UMULL		r7,r6,r6
658	$UMULH		r8,r6,r6
659	addc		r11,r7,r11
660	adde		r9,r8,r9
661	addze		r10,r0
662						#sqr_add_c2(a,5,3,c3,c1,c2);
663	$LD		r6,`5*$BNSZ`(r4)
664	$UMULL		r7,r5,r6
665	$UMULH		r8,r5,r6
666	addc		r11,r7,r11
667	adde		r9,r8,r9
668	addze		r10,r10
669	addc		r11,r7,r11
670	adde		r9,r8,r9
671	addze		r10,r10
672						#sqr_add_c2(a,6,2,c3,c1,c2);
673	$LD		r5,`2*$BNSZ`(r4)
674	$LD		r6,`6*$BNSZ`(r4)
675	$UMULL		r7,r5,r6
676	$UMULH		r8,r5,r6
677	addc		r11,r7,r11
678	adde		r9,r8,r9
679	addze		r10,r10
680
681	addc		r11,r7,r11
682	adde		r9,r8,r9
683	addze		r10,r10
684						#sqr_add_c2(a,7,1,c3,c1,c2);
685	$LD		r5,`1*$BNSZ`(r4)
686	$LD		r6,`7*$BNSZ`(r4)
687	$UMULL		r7,r5,r6
688	$UMULH		r8,r5,r6
689	addc		r11,r7,r11
690	adde		r9,r8,r9
691	addze		r10,r10
692	addc		r11,r7,r11
693	adde		r9,r8,r9
694	addze		r10,r10
695	$ST		r11,`8*$BNSZ`(r3)	#r[8]=c3;
696						#sqr_add_c2(a,7,2,c1,c2,c3);
697	$LD		r5,`2*$BNSZ`(r4)
698	$UMULL		r7,r5,r6
699	$UMULH		r8,r5,r6
700
701	addc		r9,r7,r9
702	adde		r10,r8,r10
703	addze		r11,r0
704	addc		r9,r7,r9
705	adde		r10,r8,r10
706	addze		r11,r11
707						#sqr_add_c2(a,6,3,c1,c2,c3);
708	$LD		r5,`3*$BNSZ`(r4)
709	$LD		r6,`6*$BNSZ`(r4)
710	$UMULL		r7,r5,r6
711	$UMULH		r8,r5,r6
712	addc		r9,r7,r9
713	adde		r10,r8,r10
714	addze		r11,r11
715	addc		r9,r7,r9
716	adde		r10,r8,r10
717	addze		r11,r11
718						#sqr_add_c2(a,5,4,c1,c2,c3);
719	$LD		r5,`4*$BNSZ`(r4)
720	$LD		r6,`5*$BNSZ`(r4)
721	$UMULL		r7,r5,r6
722	$UMULH		r8,r5,r6
723	addc		r9,r7,r9
724	adde		r10,r8,r10
725	addze		r11,r11
726	addc		r9,r7,r9
727	adde		r10,r8,r10
728	addze		r11,r11
729	$ST		r9,`9*$BNSZ`(r3)	#r[9]=c1;
730						#sqr_add_c(a,5,c2,c3,c1);
731	$UMULL		r7,r6,r6
732	$UMULH		r8,r6,r6
733	addc		r10,r7,r10
734	adde		r11,r8,r11
735	addze		r9,r0
736						#sqr_add_c2(a,6,4,c2,c3,c1);
737	$LD		r6,`6*$BNSZ`(r4)
738	$UMULL		r7,r5,r6
739	$UMULH		r8,r5,r6
740	addc		r10,r7,r10
741	adde		r11,r8,r11
742	addze		r9,r9
743	addc		r10,r7,r10
744	adde		r11,r8,r11
745	addze		r9,r9
746						#sqr_add_c2(a,7,3,c2,c3,c1);
747	$LD		r5,`3*$BNSZ`(r4)
748	$LD		r6,`7*$BNSZ`(r4)
749	$UMULL		r7,r5,r6
750	$UMULH		r8,r5,r6
751	addc		r10,r7,r10
752	adde		r11,r8,r11
753	addze		r9,r9
754	addc		r10,r7,r10
755	adde		r11,r8,r11
756	addze		r9,r9
757	$ST		r10,`10*$BNSZ`(r3)	#r[10]=c2;
758						#sqr_add_c2(a,7,4,c3,c1,c2);
759	$LD		r5,`4*$BNSZ`(r4)
760	$UMULL		r7,r5,r6
761	$UMULH		r8,r5,r6
762	addc		r11,r7,r11
763	adde		r9,r8,r9
764	addze		r10,r0
765	addc		r11,r7,r11
766	adde		r9,r8,r9
767	addze		r10,r10
768						#sqr_add_c2(a,6,5,c3,c1,c2);
769	$LD		r5,`5*$BNSZ`(r4)
770	$LD		r6,`6*$BNSZ`(r4)
771	$UMULL		r7,r5,r6
772	$UMULH		r8,r5,r6
773	addc		r11,r7,r11
774	adde		r9,r8,r9
775	addze		r10,r10
776	addc		r11,r7,r11
777	adde		r9,r8,r9
778	addze		r10,r10
779	$ST		r11,`11*$BNSZ`(r3)	#r[11]=c3;
780						#sqr_add_c(a,6,c1,c2,c3);
781	$UMULL		r7,r6,r6
782	$UMULH		r8,r6,r6
783	addc		r9,r7,r9
784	adde		r10,r8,r10
785	addze		r11,r0
786						#sqr_add_c2(a,7,5,c1,c2,c3)
787	$LD		r6,`7*$BNSZ`(r4)
788	$UMULL		r7,r5,r6
789	$UMULH		r8,r5,r6
790	addc		r9,r7,r9
791	adde		r10,r8,r10
792	addze		r11,r11
793	addc		r9,r7,r9
794	adde		r10,r8,r10
795	addze		r11,r11
796	$ST		r9,`12*$BNSZ`(r3)	#r[12]=c1;
797
798						#sqr_add_c2(a,7,6,c2,c3,c1)
799	$LD		r5,`6*$BNSZ`(r4)
800	$UMULL		r7,r5,r6
801	$UMULH		r8,r5,r6
802	addc		r10,r7,r10
803	adde		r11,r8,r11
804	addze		r9,r0
805	addc		r10,r7,r10
806	adde		r11,r8,r11
807	addze		r9,r9
808	$ST		r10,`13*$BNSZ`(r3)	#r[13]=c2;
809						#sqr_add_c(a,7,c3,c1,c2);
810	$UMULL		r7,r6,r6
811	$UMULH		r8,r6,r6
812	addc		r11,r7,r11
813	adde		r9,r8,r9
814	$ST		r11,`14*$BNSZ`(r3)	#r[14]=c3;
815	$ST		r9, `15*$BNSZ`(r3)	#r[15]=c1;
816
817
818	blr
819	.long	0
820	.byte	0,12,0x14,0,0,0,2,0
821	.long	0
822
823#
824#	NOTE:	The following label name should be changed to
825#		"bn_mul_comba4" i.e. remove the first dot
826#		for the gcc compiler. This should be automatically
827#		done in the build
828#
829
830.align	4
831.bn_mul_comba4:
832#
833# This is an optimized version of the bn_mul_comba4 routine.
834#
835# void bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
836# r3 contains r
837# r4 contains a
838# r5 contains b
839# r6, r7 are the 2 BN_ULONGs being multiplied.
840# r8, r9 are the results of the 32x32 giving 64 multiply.
841# r10, r11, r12 are the equivalents of c1, c2, and c3.
842#
843	xor	r0,r0,r0		#r0=0. Used in addze below.
844					#mul_add_c(a[0],b[0],c1,c2,c3);
845	$LD	r6,`0*$BNSZ`(r4)
846	$LD	r7,`0*$BNSZ`(r5)
847	$UMULL	r10,r6,r7
848	$UMULH	r11,r6,r7
849	$ST	r10,`0*$BNSZ`(r3)	#r[0]=c1
850					#mul_add_c(a[0],b[1],c2,c3,c1);
851	$LD	r7,`1*$BNSZ`(r5)
852	$UMULL	r8,r6,r7
853	$UMULH	r9,r6,r7
854	addc	r11,r8,r11
855	adde	r12,r9,r0
856	addze	r10,r0
857					#mul_add_c(a[1],b[0],c2,c3,c1);
858	$LD	r6, `1*$BNSZ`(r4)
859	$LD	r7, `0*$BNSZ`(r5)
860	$UMULL	r8,r6,r7
861	$UMULH	r9,r6,r7
862	addc	r11,r8,r11
863	adde	r12,r9,r12
864	addze	r10,r10
865	$ST	r11,`1*$BNSZ`(r3)	#r[1]=c2
866					#mul_add_c(a[2],b[0],c3,c1,c2);
867	$LD	r6,`2*$BNSZ`(r4)
868	$UMULL	r8,r6,r7
869	$UMULH	r9,r6,r7
870	addc	r12,r8,r12
871	adde	r10,r9,r10
872	addze	r11,r0
873					#mul_add_c(a[1],b[1],c3,c1,c2);
874	$LD	r6,`1*$BNSZ`(r4)
875	$LD	r7,`1*$BNSZ`(r5)
876	$UMULL	r8,r6,r7
877	$UMULH	r9,r6,r7
878	addc	r12,r8,r12
879	adde	r10,r9,r10
880	addze	r11,r11
881					#mul_add_c(a[0],b[2],c3,c1,c2);
882	$LD	r6,`0*$BNSZ`(r4)
883	$LD	r7,`2*$BNSZ`(r5)
884	$UMULL	r8,r6,r7
885	$UMULH	r9,r6,r7
886	addc	r12,r8,r12
887	adde	r10,r9,r10
888	addze	r11,r11
889	$ST	r12,`2*$BNSZ`(r3)	#r[2]=c3
890					#mul_add_c(a[0],b[3],c1,c2,c3);
891	$LD	r7,`3*$BNSZ`(r5)
892	$UMULL	r8,r6,r7
893	$UMULH	r9,r6,r7
894	addc	r10,r8,r10
895	adde	r11,r9,r11
896	addze	r12,r0
897					#mul_add_c(a[1],b[2],c1,c2,c3);
898	$LD	r6,`1*$BNSZ`(r4)
899	$LD	r7,`2*$BNSZ`(r5)
900	$UMULL	r8,r6,r7
901	$UMULH	r9,r6,r7
902	addc	r10,r8,r10
903	adde	r11,r9,r11
904	addze	r12,r12
905					#mul_add_c(a[2],b[1],c1,c2,c3);
906	$LD	r6,`2*$BNSZ`(r4)
907	$LD	r7,`1*$BNSZ`(r5)
908	$UMULL	r8,r6,r7
909	$UMULH	r9,r6,r7
910	addc	r10,r8,r10
911	adde	r11,r9,r11
912	addze	r12,r12
913					#mul_add_c(a[3],b[0],c1,c2,c3);
914	$LD	r6,`3*$BNSZ`(r4)
915	$LD	r7,`0*$BNSZ`(r5)
916	$UMULL	r8,r6,r7
917	$UMULH	r9,r6,r7
918	addc	r10,r8,r10
919	adde	r11,r9,r11
920	addze	r12,r12
921	$ST	r10,`3*$BNSZ`(r3)	#r[3]=c1
922					#mul_add_c(a[3],b[1],c2,c3,c1);
923	$LD	r7,`1*$BNSZ`(r5)
924	$UMULL	r8,r6,r7
925	$UMULH	r9,r6,r7
926	addc	r11,r8,r11
927	adde	r12,r9,r12
928	addze	r10,r0
929					#mul_add_c(a[2],b[2],c2,c3,c1);
930	$LD	r6,`2*$BNSZ`(r4)
931	$LD	r7,`2*$BNSZ`(r5)
932	$UMULL	r8,r6,r7
933	$UMULH	r9,r6,r7
934	addc	r11,r8,r11
935	adde	r12,r9,r12
936	addze	r10,r10
937					#mul_add_c(a[1],b[3],c2,c3,c1);
938	$LD	r6,`1*$BNSZ`(r4)
939	$LD	r7,`3*$BNSZ`(r5)
940	$UMULL	r8,r6,r7
941	$UMULH	r9,r6,r7
942	addc	r11,r8,r11
943	adde	r12,r9,r12
944	addze	r10,r10
945	$ST	r11,`4*$BNSZ`(r3)	#r[4]=c2
946					#mul_add_c(a[2],b[3],c3,c1,c2);
947	$LD	r6,`2*$BNSZ`(r4)
948	$UMULL	r8,r6,r7
949	$UMULH	r9,r6,r7
950	addc	r12,r8,r12
951	adde	r10,r9,r10
952	addze	r11,r0
953					#mul_add_c(a[3],b[2],c3,c1,c2);
954	$LD	r6,`3*$BNSZ`(r4)
955	$LD	r7,`2*$BNSZ`(r5)
956	$UMULL	r8,r6,r7
957	$UMULH	r9,r6,r7
958	addc	r12,r8,r12
959	adde	r10,r9,r10
960	addze	r11,r11
961	$ST	r12,`5*$BNSZ`(r3)	#r[5]=c3
962					#mul_add_c(a[3],b[3],c1,c2,c3);
963	$LD	r7,`3*$BNSZ`(r5)
964	$UMULL	r8,r6,r7
965	$UMULH	r9,r6,r7
966	addc	r10,r8,r10
967	adde	r11,r9,r11
968
969	$ST	r10,`6*$BNSZ`(r3)	#r[6]=c1
970	$ST	r11,`7*$BNSZ`(r3)	#r[7]=c2
971	blr
972	.long	0
973	.byte	0,12,0x14,0,0,0,3,0
974	.long	0
975
976#
977#	NOTE:	The following label name should be changed to
978#		"bn_mul_comba8" i.e. remove the first dot
979#		for the gcc compiler. This should be automatically
980#		done in the build
981#
982
983.align	4
984.bn_mul_comba8:
985#
986# Optimized version of the bn_mul_comba8 routine.
987#
988# void bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
989# r3 contains r
990# r4 contains a
991# r5 contains b
992# r6, r7 are the 2 BN_ULONGs being multiplied.
993# r8, r9 are the results of the 32x32 giving 64 multiply.
994# r10, r11, r12 are the equivalents of c1, c2, and c3.
995#
996	xor	r0,r0,r0		#r0=0. Used in addze below.
997
998					#mul_add_c(a[0],b[0],c1,c2,c3);
999	$LD	r6,`0*$BNSZ`(r4)	#a[0]
1000	$LD	r7,`0*$BNSZ`(r5)	#b[0]
1001	$UMULL	r10,r6,r7
1002	$UMULH	r11,r6,r7
1003	$ST	r10,`0*$BNSZ`(r3)	#r[0]=c1;
1004					#mul_add_c(a[0],b[1],c2,c3,c1);
1005	$LD	r7,`1*$BNSZ`(r5)
1006	$UMULL	r8,r6,r7
1007	$UMULH	r9,r6,r7
1008	addc	r11,r11,r8
1009	addze	r12,r9			# since we didnt set r12 to zero before.
1010	addze	r10,r0
1011					#mul_add_c(a[1],b[0],c2,c3,c1);
1012	$LD	r6,`1*$BNSZ`(r4)
1013	$LD	r7,`0*$BNSZ`(r5)
1014	$UMULL	r8,r6,r7
1015	$UMULH	r9,r6,r7
1016	addc	r11,r11,r8
1017	adde	r12,r12,r9
1018	addze	r10,r10
1019	$ST	r11,`1*$BNSZ`(r3)	#r[1]=c2;
1020					#mul_add_c(a[2],b[0],c3,c1,c2);
1021	$LD	r6,`2*$BNSZ`(r4)
1022	$UMULL	r8,r6,r7
1023	$UMULH	r9,r6,r7
1024	addc	r12,r12,r8
1025	adde	r10,r10,r9
1026	addze	r11,r0
1027					#mul_add_c(a[1],b[1],c3,c1,c2);
1028	$LD	r6,`1*$BNSZ`(r4)
1029	$LD	r7,`1*$BNSZ`(r5)
1030	$UMULL	r8,r6,r7
1031	$UMULH	r9,r6,r7
1032	addc	r12,r12,r8
1033	adde	r10,r10,r9
1034	addze	r11,r11
1035					#mul_add_c(a[0],b[2],c3,c1,c2);
1036	$LD	r6,`0*$BNSZ`(r4)
1037	$LD	r7,`2*$BNSZ`(r5)
1038	$UMULL	r8,r6,r7
1039	$UMULH	r9,r6,r7
1040	addc	r12,r12,r8
1041	adde	r10,r10,r9
1042	addze	r11,r11
1043	$ST	r12,`2*$BNSZ`(r3)	#r[2]=c3;
1044					#mul_add_c(a[0],b[3],c1,c2,c3);
1045	$LD	r7,`3*$BNSZ`(r5)
1046	$UMULL	r8,r6,r7
1047	$UMULH	r9,r6,r7
1048	addc	r10,r10,r8
1049	adde	r11,r11,r9
1050	addze	r12,r0
1051					#mul_add_c(a[1],b[2],c1,c2,c3);
1052	$LD	r6,`1*$BNSZ`(r4)
1053	$LD	r7,`2*$BNSZ`(r5)
1054	$UMULL	r8,r6,r7
1055	$UMULH	r9,r6,r7
1056	addc	r10,r10,r8
1057	adde	r11,r11,r9
1058	addze	r12,r12
1059
1060					#mul_add_c(a[2],b[1],c1,c2,c3);
1061	$LD	r6,`2*$BNSZ`(r4)
1062	$LD	r7,`1*$BNSZ`(r5)
1063	$UMULL	r8,r6,r7
1064	$UMULH	r9,r6,r7
1065	addc	r10,r10,r8
1066	adde	r11,r11,r9
1067	addze	r12,r12
1068					#mul_add_c(a[3],b[0],c1,c2,c3);
1069	$LD	r6,`3*$BNSZ`(r4)
1070	$LD	r7,`0*$BNSZ`(r5)
1071	$UMULL	r8,r6,r7
1072	$UMULH	r9,r6,r7
1073	addc	r10,r10,r8
1074	adde	r11,r11,r9
1075	addze	r12,r12
1076	$ST	r10,`3*$BNSZ`(r3)	#r[3]=c1;
1077					#mul_add_c(a[4],b[0],c2,c3,c1);
1078	$LD	r6,`4*$BNSZ`(r4)
1079	$UMULL	r8,r6,r7
1080	$UMULH	r9,r6,r7
1081	addc	r11,r11,r8
1082	adde	r12,r12,r9
1083	addze	r10,r0
1084					#mul_add_c(a[3],b[1],c2,c3,c1);
1085	$LD	r6,`3*$BNSZ`(r4)
1086	$LD	r7,`1*$BNSZ`(r5)
1087	$UMULL	r8,r6,r7
1088	$UMULH	r9,r6,r7
1089	addc	r11,r11,r8
1090	adde	r12,r12,r9
1091	addze	r10,r10
1092					#mul_add_c(a[2],b[2],c2,c3,c1);
1093	$LD	r6,`2*$BNSZ`(r4)
1094	$LD	r7,`2*$BNSZ`(r5)
1095	$UMULL	r8,r6,r7
1096	$UMULH	r9,r6,r7
1097	addc	r11,r11,r8
1098	adde	r12,r12,r9
1099	addze	r10,r10
1100					#mul_add_c(a[1],b[3],c2,c3,c1);
1101	$LD	r6,`1*$BNSZ`(r4)
1102	$LD	r7,`3*$BNSZ`(r5)
1103	$UMULL	r8,r6,r7
1104	$UMULH	r9,r6,r7
1105	addc	r11,r11,r8
1106	adde	r12,r12,r9
1107	addze	r10,r10
1108					#mul_add_c(a[0],b[4],c2,c3,c1);
1109	$LD	r6,`0*$BNSZ`(r4)
1110	$LD	r7,`4*$BNSZ`(r5)
1111	$UMULL	r8,r6,r7
1112	$UMULH	r9,r6,r7
1113	addc	r11,r11,r8
1114	adde	r12,r12,r9
1115	addze	r10,r10
1116	$ST	r11,`4*$BNSZ`(r3)	#r[4]=c2;
1117					#mul_add_c(a[0],b[5],c3,c1,c2);
1118	$LD	r7,`5*$BNSZ`(r5)
1119	$UMULL	r8,r6,r7
1120	$UMULH	r9,r6,r7
1121	addc	r12,r12,r8
1122	adde	r10,r10,r9
1123	addze	r11,r0
1124					#mul_add_c(a[1],b[4],c3,c1,c2);
1125	$LD	r6,`1*$BNSZ`(r4)
1126	$LD	r7,`4*$BNSZ`(r5)
1127	$UMULL	r8,r6,r7
1128	$UMULH	r9,r6,r7
1129	addc	r12,r12,r8
1130	adde	r10,r10,r9
1131	addze	r11,r11
1132					#mul_add_c(a[2],b[3],c3,c1,c2);
1133	$LD	r6,`2*$BNSZ`(r4)
1134	$LD	r7,`3*$BNSZ`(r5)
1135	$UMULL	r8,r6,r7
1136	$UMULH	r9,r6,r7
1137	addc	r12,r12,r8
1138	adde	r10,r10,r9
1139	addze	r11,r11
1140					#mul_add_c(a[3],b[2],c3,c1,c2);
1141	$LD	r6,`3*$BNSZ`(r4)
1142	$LD	r7,`2*$BNSZ`(r5)
1143	$UMULL	r8,r6,r7
1144	$UMULH	r9,r6,r7
1145	addc	r12,r12,r8
1146	adde	r10,r10,r9
1147	addze	r11,r11
1148					#mul_add_c(a[4],b[1],c3,c1,c2);
1149	$LD	r6,`4*$BNSZ`(r4)
1150	$LD	r7,`1*$BNSZ`(r5)
1151	$UMULL	r8,r6,r7
1152	$UMULH	r9,r6,r7
1153	addc	r12,r12,r8
1154	adde	r10,r10,r9
1155	addze	r11,r11
1156					#mul_add_c(a[5],b[0],c3,c1,c2);
1157	$LD	r6,`5*$BNSZ`(r4)
1158	$LD	r7,`0*$BNSZ`(r5)
1159	$UMULL	r8,r6,r7
1160	$UMULH	r9,r6,r7
1161	addc	r12,r12,r8
1162	adde	r10,r10,r9
1163	addze	r11,r11
1164	$ST	r12,`5*$BNSZ`(r3)	#r[5]=c3;
1165					#mul_add_c(a[6],b[0],c1,c2,c3);
1166	$LD	r6,`6*$BNSZ`(r4)
1167	$UMULL	r8,r6,r7
1168	$UMULH	r9,r6,r7
1169	addc	r10,r10,r8
1170	adde	r11,r11,r9
1171	addze	r12,r0
1172					#mul_add_c(a[5],b[1],c1,c2,c3);
1173	$LD	r6,`5*$BNSZ`(r4)
1174	$LD	r7,`1*$BNSZ`(r5)
1175	$UMULL	r8,r6,r7
1176	$UMULH	r9,r6,r7
1177	addc	r10,r10,r8
1178	adde	r11,r11,r9
1179	addze	r12,r12
1180					#mul_add_c(a[4],b[2],c1,c2,c3);
1181	$LD	r6,`4*$BNSZ`(r4)
1182	$LD	r7,`2*$BNSZ`(r5)
1183	$UMULL	r8,r6,r7
1184	$UMULH	r9,r6,r7
1185	addc	r10,r10,r8
1186	adde	r11,r11,r9
1187	addze	r12,r12
1188					#mul_add_c(a[3],b[3],c1,c2,c3);
1189	$LD	r6,`3*$BNSZ`(r4)
1190	$LD	r7,`3*$BNSZ`(r5)
1191	$UMULL	r8,r6,r7
1192	$UMULH	r9,r6,r7
1193	addc	r10,r10,r8
1194	adde	r11,r11,r9
1195	addze	r12,r12
1196					#mul_add_c(a[2],b[4],c1,c2,c3);
1197	$LD	r6,`2*$BNSZ`(r4)
1198	$LD	r7,`4*$BNSZ`(r5)
1199	$UMULL	r8,r6,r7
1200	$UMULH	r9,r6,r7
1201	addc	r10,r10,r8
1202	adde	r11,r11,r9
1203	addze	r12,r12
1204					#mul_add_c(a[1],b[5],c1,c2,c3);
1205	$LD	r6,`1*$BNSZ`(r4)
1206	$LD	r7,`5*$BNSZ`(r5)
1207	$UMULL	r8,r6,r7
1208	$UMULH	r9,r6,r7
1209	addc	r10,r10,r8
1210	adde	r11,r11,r9
1211	addze	r12,r12
1212					#mul_add_c(a[0],b[6],c1,c2,c3);
1213	$LD	r6,`0*$BNSZ`(r4)
1214	$LD	r7,`6*$BNSZ`(r5)
1215	$UMULL	r8,r6,r7
1216	$UMULH	r9,r6,r7
1217	addc	r10,r10,r8
1218	adde	r11,r11,r9
1219	addze	r12,r12
1220	$ST	r10,`6*$BNSZ`(r3)	#r[6]=c1;
1221					#mul_add_c(a[0],b[7],c2,c3,c1);
1222	$LD	r7,`7*$BNSZ`(r5)
1223	$UMULL	r8,r6,r7
1224	$UMULH	r9,r6,r7
1225	addc	r11,r11,r8
1226	adde	r12,r12,r9
1227	addze	r10,r0
1228					#mul_add_c(a[1],b[6],c2,c3,c1);
1229	$LD	r6,`1*$BNSZ`(r4)
1230	$LD	r7,`6*$BNSZ`(r5)
1231	$UMULL	r8,r6,r7
1232	$UMULH	r9,r6,r7
1233	addc	r11,r11,r8
1234	adde	r12,r12,r9
1235	addze	r10,r10
1236					#mul_add_c(a[2],b[5],c2,c3,c1);
1237	$LD	r6,`2*$BNSZ`(r4)
1238	$LD	r7,`5*$BNSZ`(r5)
1239	$UMULL	r8,r6,r7
1240	$UMULH	r9,r6,r7
1241	addc	r11,r11,r8
1242	adde	r12,r12,r9
1243	addze	r10,r10
1244					#mul_add_c(a[3],b[4],c2,c3,c1);
1245	$LD	r6,`3*$BNSZ`(r4)
1246	$LD	r7,`4*$BNSZ`(r5)
1247	$UMULL	r8,r6,r7
1248	$UMULH	r9,r6,r7
1249	addc	r11,r11,r8
1250	adde	r12,r12,r9
1251	addze	r10,r10
1252					#mul_add_c(a[4],b[3],c2,c3,c1);
1253	$LD	r6,`4*$BNSZ`(r4)
1254	$LD	r7,`3*$BNSZ`(r5)
1255	$UMULL	r8,r6,r7
1256	$UMULH	r9,r6,r7
1257	addc	r11,r11,r8
1258	adde	r12,r12,r9
1259	addze	r10,r10
1260					#mul_add_c(a[5],b[2],c2,c3,c1);
1261	$LD	r6,`5*$BNSZ`(r4)
1262	$LD	r7,`2*$BNSZ`(r5)
1263	$UMULL	r8,r6,r7
1264	$UMULH	r9,r6,r7
1265	addc	r11,r11,r8
1266	adde	r12,r12,r9
1267	addze	r10,r10
1268					#mul_add_c(a[6],b[1],c2,c3,c1);
1269	$LD	r6,`6*$BNSZ`(r4)
1270	$LD	r7,`1*$BNSZ`(r5)
1271	$UMULL	r8,r6,r7
1272	$UMULH	r9,r6,r7
1273	addc	r11,r11,r8
1274	adde	r12,r12,r9
1275	addze	r10,r10
1276					#mul_add_c(a[7],b[0],c2,c3,c1);
1277	$LD	r6,`7*$BNSZ`(r4)
1278	$LD	r7,`0*$BNSZ`(r5)
1279	$UMULL	r8,r6,r7
1280	$UMULH	r9,r6,r7
1281	addc	r11,r11,r8
1282	adde	r12,r12,r9
1283	addze	r10,r10
1284	$ST	r11,`7*$BNSZ`(r3)	#r[7]=c2;
1285					#mul_add_c(a[7],b[1],c3,c1,c2);
1286	$LD	r7,`1*$BNSZ`(r5)
1287	$UMULL	r8,r6,r7
1288	$UMULH	r9,r6,r7
1289	addc	r12,r12,r8
1290	adde	r10,r10,r9
1291	addze	r11,r0
1292					#mul_add_c(a[6],b[2],c3,c1,c2);
1293	$LD	r6,`6*$BNSZ`(r4)
1294	$LD	r7,`2*$BNSZ`(r5)
1295	$UMULL	r8,r6,r7
1296	$UMULH	r9,r6,r7
1297	addc	r12,r12,r8
1298	adde	r10,r10,r9
1299	addze	r11,r11
1300					#mul_add_c(a[5],b[3],c3,c1,c2);
1301	$LD	r6,`5*$BNSZ`(r4)
1302	$LD	r7,`3*$BNSZ`(r5)
1303	$UMULL	r8,r6,r7
1304	$UMULH	r9,r6,r7
1305	addc	r12,r12,r8
1306	adde	r10,r10,r9
1307	addze	r11,r11
1308					#mul_add_c(a[4],b[4],c3,c1,c2);
1309	$LD	r6,`4*$BNSZ`(r4)
1310	$LD	r7,`4*$BNSZ`(r5)
1311	$UMULL	r8,r6,r7
1312	$UMULH	r9,r6,r7
1313	addc	r12,r12,r8
1314	adde	r10,r10,r9
1315	addze	r11,r11
1316					#mul_add_c(a[3],b[5],c3,c1,c2);
1317	$LD	r6,`3*$BNSZ`(r4)
1318	$LD	r7,`5*$BNSZ`(r5)
1319	$UMULL	r8,r6,r7
1320	$UMULH	r9,r6,r7
1321	addc	r12,r12,r8
1322	adde	r10,r10,r9
1323	addze	r11,r11
1324					#mul_add_c(a[2],b[6],c3,c1,c2);
1325	$LD	r6,`2*$BNSZ`(r4)
1326	$LD	r7,`6*$BNSZ`(r5)
1327	$UMULL	r8,r6,r7
1328	$UMULH	r9,r6,r7
1329	addc	r12,r12,r8
1330	adde	r10,r10,r9
1331	addze	r11,r11
1332					#mul_add_c(a[1],b[7],c3,c1,c2);
1333	$LD	r6,`1*$BNSZ`(r4)
1334	$LD	r7,`7*$BNSZ`(r5)
1335	$UMULL	r8,r6,r7
1336	$UMULH	r9,r6,r7
1337	addc	r12,r12,r8
1338	adde	r10,r10,r9
1339	addze	r11,r11
1340	$ST	r12,`8*$BNSZ`(r3)	#r[8]=c3;
1341					#mul_add_c(a[2],b[7],c1,c2,c3);
1342	$LD	r6,`2*$BNSZ`(r4)
1343	$UMULL	r8,r6,r7
1344	$UMULH	r9,r6,r7
1345	addc	r10,r10,r8
1346	adde	r11,r11,r9
1347	addze	r12,r0
1348					#mul_add_c(a[3],b[6],c1,c2,c3);
1349	$LD	r6,`3*$BNSZ`(r4)
1350	$LD	r7,`6*$BNSZ`(r5)
1351	$UMULL	r8,r6,r7
1352	$UMULH	r9,r6,r7
1353	addc	r10,r10,r8
1354	adde	r11,r11,r9
1355	addze	r12,r12
1356					#mul_add_c(a[4],b[5],c1,c2,c3);
1357	$LD	r6,`4*$BNSZ`(r4)
1358	$LD	r7,`5*$BNSZ`(r5)
1359	$UMULL	r8,r6,r7
1360	$UMULH	r9,r6,r7
1361	addc	r10,r10,r8
1362	adde	r11,r11,r9
1363	addze	r12,r12
1364					#mul_add_c(a[5],b[4],c1,c2,c3);
1365	$LD	r6,`5*$BNSZ`(r4)
1366	$LD	r7,`4*$BNSZ`(r5)
1367	$UMULL	r8,r6,r7
1368	$UMULH	r9,r6,r7
1369	addc	r10,r10,r8
1370	adde	r11,r11,r9
1371	addze	r12,r12
1372					#mul_add_c(a[6],b[3],c1,c2,c3);
1373	$LD	r6,`6*$BNSZ`(r4)
1374	$LD	r7,`3*$BNSZ`(r5)
1375	$UMULL	r8,r6,r7
1376	$UMULH	r9,r6,r7
1377	addc	r10,r10,r8
1378	adde	r11,r11,r9
1379	addze	r12,r12
1380					#mul_add_c(a[7],b[2],c1,c2,c3);
1381	$LD	r6,`7*$BNSZ`(r4)
1382	$LD	r7,`2*$BNSZ`(r5)
1383	$UMULL	r8,r6,r7
1384	$UMULH	r9,r6,r7
1385	addc	r10,r10,r8
1386	adde	r11,r11,r9
1387	addze	r12,r12
1388	$ST	r10,`9*$BNSZ`(r3)	#r[9]=c1;
1389					#mul_add_c(a[7],b[3],c2,c3,c1);
1390	$LD	r7,`3*$BNSZ`(r5)
1391	$UMULL	r8,r6,r7
1392	$UMULH	r9,r6,r7
1393	addc	r11,r11,r8
1394	adde	r12,r12,r9
1395	addze	r10,r0
1396					#mul_add_c(a[6],b[4],c2,c3,c1);
1397	$LD	r6,`6*$BNSZ`(r4)
1398	$LD	r7,`4*$BNSZ`(r5)
1399	$UMULL	r8,r6,r7
1400	$UMULH	r9,r6,r7
1401	addc	r11,r11,r8
1402	adde	r12,r12,r9
1403	addze	r10,r10
1404					#mul_add_c(a[5],b[5],c2,c3,c1);
1405	$LD	r6,`5*$BNSZ`(r4)
1406	$LD	r7,`5*$BNSZ`(r5)
1407	$UMULL	r8,r6,r7
1408	$UMULH	r9,r6,r7
1409	addc	r11,r11,r8
1410	adde	r12,r12,r9
1411	addze	r10,r10
1412					#mul_add_c(a[4],b[6],c2,c3,c1);
1413	$LD	r6,`4*$BNSZ`(r4)
1414	$LD	r7,`6*$BNSZ`(r5)
1415	$UMULL	r8,r6,r7
1416	$UMULH	r9,r6,r7
1417	addc	r11,r11,r8
1418	adde	r12,r12,r9
1419	addze	r10,r10
1420					#mul_add_c(a[3],b[7],c2,c3,c1);
1421	$LD	r6,`3*$BNSZ`(r4)
1422	$LD	r7,`7*$BNSZ`(r5)
1423	$UMULL	r8,r6,r7
1424	$UMULH	r9,r6,r7
1425	addc	r11,r11,r8
1426	adde	r12,r12,r9
1427	addze	r10,r10
1428	$ST	r11,`10*$BNSZ`(r3)	#r[10]=c2;
1429					#mul_add_c(a[4],b[7],c3,c1,c2);
1430	$LD	r6,`4*$BNSZ`(r4)
1431	$UMULL	r8,r6,r7
1432	$UMULH	r9,r6,r7
1433	addc	r12,r12,r8
1434	adde	r10,r10,r9
1435	addze	r11,r0
1436					#mul_add_c(a[5],b[6],c3,c1,c2);
1437	$LD	r6,`5*$BNSZ`(r4)
1438	$LD	r7,`6*$BNSZ`(r5)
1439	$UMULL	r8,r6,r7
1440	$UMULH	r9,r6,r7
1441	addc	r12,r12,r8
1442	adde	r10,r10,r9
1443	addze	r11,r11
1444					#mul_add_c(a[6],b[5],c3,c1,c2);
1445	$LD	r6,`6*$BNSZ`(r4)
1446	$LD	r7,`5*$BNSZ`(r5)
1447	$UMULL	r8,r6,r7
1448	$UMULH	r9,r6,r7
1449	addc	r12,r12,r8
1450	adde	r10,r10,r9
1451	addze	r11,r11
1452					#mul_add_c(a[7],b[4],c3,c1,c2);
1453	$LD	r6,`7*$BNSZ`(r4)
1454	$LD	r7,`4*$BNSZ`(r5)
1455	$UMULL	r8,r6,r7
1456	$UMULH	r9,r6,r7
1457	addc	r12,r12,r8
1458	adde	r10,r10,r9
1459	addze	r11,r11
1460	$ST	r12,`11*$BNSZ`(r3)	#r[11]=c3;
1461					#mul_add_c(a[7],b[5],c1,c2,c3);
1462	$LD	r7,`5*$BNSZ`(r5)
1463	$UMULL	r8,r6,r7
1464	$UMULH	r9,r6,r7
1465	addc	r10,r10,r8
1466	adde	r11,r11,r9
1467	addze	r12,r0
1468					#mul_add_c(a[6],b[6],c1,c2,c3);
1469	$LD	r6,`6*$BNSZ`(r4)
1470	$LD	r7,`6*$BNSZ`(r5)
1471	$UMULL	r8,r6,r7
1472	$UMULH	r9,r6,r7
1473	addc	r10,r10,r8
1474	adde	r11,r11,r9
1475	addze	r12,r12
1476					#mul_add_c(a[5],b[7],c1,c2,c3);
1477	$LD	r6,`5*$BNSZ`(r4)
1478	$LD	r7,`7*$BNSZ`(r5)
1479	$UMULL	r8,r6,r7
1480	$UMULH	r9,r6,r7
1481	addc	r10,r10,r8
1482	adde	r11,r11,r9
1483	addze	r12,r12
1484	$ST	r10,`12*$BNSZ`(r3)	#r[12]=c1;
1485					#mul_add_c(a[6],b[7],c2,c3,c1);
1486	$LD	r6,`6*$BNSZ`(r4)
1487	$UMULL	r8,r6,r7
1488	$UMULH	r9,r6,r7
1489	addc	r11,r11,r8
1490	adde	r12,r12,r9
1491	addze	r10,r0
1492					#mul_add_c(a[7],b[6],c2,c3,c1);
1493	$LD	r6,`7*$BNSZ`(r4)
1494	$LD	r7,`6*$BNSZ`(r5)
1495	$UMULL	r8,r6,r7
1496	$UMULH	r9,r6,r7
1497	addc	r11,r11,r8
1498	adde	r12,r12,r9
1499	addze	r10,r10
1500	$ST	r11,`13*$BNSZ`(r3)	#r[13]=c2;
1501					#mul_add_c(a[7],b[7],c3,c1,c2);
1502	$LD	r7,`7*$BNSZ`(r5)
1503	$UMULL	r8,r6,r7
1504	$UMULH	r9,r6,r7
1505	addc	r12,r12,r8
1506	adde	r10,r10,r9
1507	$ST	r12,`14*$BNSZ`(r3)	#r[14]=c3;
1508	$ST	r10,`15*$BNSZ`(r3)	#r[15]=c1;
1509	blr
1510	.long	0
1511	.byte	0,12,0x14,0,0,0,3,0
1512	.long	0
1513
1514#
1515#	NOTE:	The following label name should be changed to
1516#		"bn_sub_words" i.e. remove the first dot
1517#		for the gcc compiler. This should be automatically
1518#		done in the build
1519#
1520#
1521.align	4
1522.bn_sub_words:
1523#
1524#	Handcoded version of bn_sub_words
1525#
1526#BN_ULONG bn_sub_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n)
1527#
1528#	r3 = r
1529#	r4 = a
1530#	r5 = b
1531#	r6 = n
1532#
1533#       Note:	No loop unrolling done since this is not a performance
1534#               critical loop.
1535
1536	xor	r0,r0,r0	#set r0 = 0
1537#
1538#	check for r6 = 0 AND set carry bit.
1539#
1540	subfc.	r7,r0,r6        # If r6 is 0 then result is 0.
1541				# if r6 > 0 then result !=0
1542				# In either case carry bit is set.
1543	beq	Lppcasm_sub_adios
1544	addi	r4,r4,-$BNSZ
1545	addi	r3,r3,-$BNSZ
1546	addi	r5,r5,-$BNSZ
1547	mtctr	r6
1548Lppcasm_sub_mainloop:
1549	$LDU	r7,$BNSZ(r4)
1550	$LDU	r8,$BNSZ(r5)
1551	subfe	r6,r8,r7	# r6 = r7+carry bit + onescomplement(r8)
1552				# if carry = 1 this is r7-r8. Else it
1553				# is r7-r8 -1 as we need.
1554	$STU	r6,$BNSZ(r3)
1555	bdnz-	Lppcasm_sub_mainloop
1556Lppcasm_sub_adios:
1557	subfze	r3,r0		# if carry bit is set then r3 = 0 else -1
1558	andi.	r3,r3,1         # keep only last bit.
1559	blr
1560	.long	0
1561	.byte	0,12,0x14,0,0,0,4,0
1562	.long	0
1563
1564#
1565#	NOTE:	The following label name should be changed to
1566#		"bn_add_words" i.e. remove the first dot
1567#		for the gcc compiler. This should be automatically
1568#		done in the build
1569#
1570
1571.align	4
1572.bn_add_words:
1573#
1574#	Handcoded version of bn_add_words
1575#
1576#BN_ULONG bn_add_words(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b, int n)
1577#
1578#	r3 = r
1579#	r4 = a
1580#	r5 = b
1581#	r6 = n
1582#
1583#       Note:	No loop unrolling done since this is not a performance
1584#               critical loop.
1585
1586	xor	r0,r0,r0
1587#
1588#	check for r6 = 0. Is this needed?
1589#
1590	addic.	r6,r6,0		#test r6 and clear carry bit.
1591	beq	Lppcasm_add_adios
1592	addi	r4,r4,-$BNSZ
1593	addi	r3,r3,-$BNSZ
1594	addi	r5,r5,-$BNSZ
1595	mtctr	r6
1596Lppcasm_add_mainloop:
1597	$LDU	r7,$BNSZ(r4)
1598	$LDU	r8,$BNSZ(r5)
1599	adde	r8,r7,r8
1600	$STU	r8,$BNSZ(r3)
1601	bdnz-	Lppcasm_add_mainloop
1602Lppcasm_add_adios:
1603	addze	r3,r0			#return carry bit.
1604	blr
1605	.long	0
1606	.byte	0,12,0x14,0,0,0,4,0
1607	.long	0
1608
1609#
1610#	NOTE:	The following label name should be changed to
1611#		"bn_div_words" i.e. remove the first dot
1612#		for the gcc compiler. This should be automatically
1613#		done in the build
1614#
1615
1616.align	4
1617.bn_div_words:
1618#
1619#	This is a cleaned up version of code generated by
1620#	the AIX compiler. The only optimization is to use
1621#	the PPC instruction to count leading zeros instead
1622#	of call to num_bits_word. Since this was compiled
1623#	only at level -O2 we can possibly squeeze it more?
1624#
1625#	r3 = h
1626#	r4 = l
1627#	r5 = d
1628
1629	$UCMPI	0,r5,0			# compare r5 and 0
1630	bne	Lppcasm_div1		# proceed if d!=0
1631	li	r3,-1			# d=0 return -1
1632	blr
1633Lppcasm_div1:
1634	xor	r0,r0,r0		#r0=0
1635	li	r8,$BITS
1636	$CNTLZ.	r7,r5			#r7 = num leading 0s in d.
1637	beq	Lppcasm_div2		#proceed if no leading zeros
1638	subf	r8,r7,r8		#r8 = BN_num_bits_word(d)
1639	$SHR.	r9,r3,r8		#are there any bits above r8'th?
1640	$TR	16,r9,r0		#if there're, signal to dump core...
1641Lppcasm_div2:
1642	$UCMP	0,r3,r5			#h>=d?
1643	blt	Lppcasm_div3		#goto Lppcasm_div3 if not
1644	subf	r3,r5,r3		#h-=d ;
1645Lppcasm_div3:				#r7 = BN_BITS2-i. so r7=i
1646	cmpi	0,0,r7,0		# is (i == 0)?
1647	beq	Lppcasm_div4
1648	$SHL	r3,r3,r7		# h = (h<< i)
1649	$SHR	r8,r4,r8		# r8 = (l >> BN_BITS2 -i)
1650	$SHL	r5,r5,r7		# d<<=i
1651	or	r3,r3,r8		# h = (h<<i)|(l>>(BN_BITS2-i))
1652	$SHL	r4,r4,r7		# l <<=i
1653Lppcasm_div4:
1654	$SHRI	r9,r5,`$BITS/2`		# r9 = dh
1655					# dl will be computed when needed
1656					# as it saves registers.
1657	li	r6,2			#r6=2
1658	mtctr	r6			#counter will be in count.
1659Lppcasm_divouterloop:
1660	$SHRI	r8,r3,`$BITS/2`		#r8 = (h>>BN_BITS4)
1661	$SHRI	r11,r4,`$BITS/2`	#r11= (l&BN_MASK2h)>>BN_BITS4
1662					# compute here for innerloop.
1663	$UCMP	0,r8,r9			# is (h>>BN_BITS4)==dh
1664	bne	Lppcasm_div5		# goto Lppcasm_div5 if not
1665
1666	li	r8,-1
1667	$CLRU	r8,r8,`$BITS/2`		#q = BN_MASK2l
1668	b	Lppcasm_div6
1669Lppcasm_div5:
1670	$UDIV	r8,r3,r9		#q = h/dh
1671Lppcasm_div6:
1672	$UMULL	r12,r9,r8		#th = q*dh
1673	$CLRU	r10,r5,`$BITS/2`	#r10=dl
1674	$UMULL	r6,r8,r10		#tl = q*dl
1675
1676Lppcasm_divinnerloop:
1677	subf	r10,r12,r3		#t = h -th
1678	$SHRI	r7,r10,`$BITS/2`	#r7= (t &BN_MASK2H), sort of...
1679	addic.	r7,r7,0			#test if r7 == 0. used below.
1680					# now want to compute
1681					# r7 = (t<<BN_BITS4)|((l&BN_MASK2h)>>BN_BITS4)
1682					# the following 2 instructions do that
1683	$SHLI	r7,r10,`$BITS/2`	# r7 = (t<<BN_BITS4)
1684	or	r7,r7,r11		# r7|=((l&BN_MASK2h)>>BN_BITS4)
1685	$UCMP	cr1,r6,r7		# compare (tl <= r7)
1686	bne	Lppcasm_divinnerexit
1687	ble	cr1,Lppcasm_divinnerexit
1688	addi	r8,r8,-1		#q--
1689	subf	r12,r9,r12		#th -=dh
1690	$CLRU	r10,r5,`$BITS/2`	#r10=dl. t is no longer needed in loop.
1691	subf	r6,r10,r6		#tl -=dl
1692	b	Lppcasm_divinnerloop
1693Lppcasm_divinnerexit:
1694	$SHRI	r10,r6,`$BITS/2`	#t=(tl>>BN_BITS4)
1695	$SHLI	r11,r6,`$BITS/2`	#tl=(tl<<BN_BITS4)&BN_MASK2h;
1696	$UCMP	cr1,r4,r11		# compare l and tl
1697	add	r12,r12,r10		# th+=t
1698	bge	cr1,Lppcasm_div7	# if (l>=tl) goto Lppcasm_div7
1699	addi	r12,r12,1		# th++
1700Lppcasm_div7:
1701	subf	r11,r11,r4		#r11=l-tl
1702	$UCMP	cr1,r3,r12		#compare h and th
1703	bge	cr1,Lppcasm_div8	#if (h>=th) goto Lppcasm_div8
1704	addi	r8,r8,-1		# q--
1705	add	r3,r5,r3		# h+=d
1706Lppcasm_div8:
1707	subf	r12,r12,r3		#r12 = h-th
1708	$SHLI	r4,r11,`$BITS/2`	#l=(l&BN_MASK2l)<<BN_BITS4
1709					# want to compute
1710					# h = ((h<<BN_BITS4)|(l>>BN_BITS4))&BN_MASK2
1711					# the following 2 instructions will do this.
1712	$INSR	r11,r12,`$BITS/2`,`$BITS/2`	# r11 is the value we want rotated $BITS/2.
1713	$ROTL	r3,r11,`$BITS/2`	# rotate by $BITS/2 and store in r3
1714	bdz	Lppcasm_div9		#if (count==0) break ;
1715	$SHLI	r0,r8,`$BITS/2`		#ret =q<<BN_BITS4
1716	b	Lppcasm_divouterloop
1717Lppcasm_div9:
1718	or	r3,r8,r0
1719	blr
1720	.long	0
1721	.byte	0,12,0x14,0,0,0,3,0
1722	.long	0
1723
1724#
1725#	NOTE:	The following label name should be changed to
1726#		"bn_sqr_words" i.e. remove the first dot
1727#		for the gcc compiler. This should be automatically
1728#		done in the build
1729#
1730.align	4
1731.bn_sqr_words:
1732#
1733#	Optimized version of bn_sqr_words
1734#
1735#	void bn_sqr_words(BN_ULONG *r, BN_ULONG *a, int n)
1736#
1737#	r3 = r
1738#	r4 = a
1739#	r5 = n
1740#
1741#	r6 = a[i].
1742#	r7,r8 = product.
1743#
1744#	No unrolling done here. Not performance critical.
1745
1746	addic.	r5,r5,0			#test r5.
1747	beq	Lppcasm_sqr_adios
1748	addi	r4,r4,-$BNSZ
1749	addi	r3,r3,-$BNSZ
1750	mtctr	r5
1751Lppcasm_sqr_mainloop:
1752					#sqr(r[0],r[1],a[0]);
1753	$LDU	r6,$BNSZ(r4)
1754	$UMULL	r7,r6,r6
1755	$UMULH  r8,r6,r6
1756	$STU	r7,$BNSZ(r3)
1757	$STU	r8,$BNSZ(r3)
1758	bdnz-	Lppcasm_sqr_mainloop
1759Lppcasm_sqr_adios:
1760	blr
1761	.long	0
1762	.byte	0,12,0x14,0,0,0,3,0
1763	.long	0
1764
1765#
1766#	NOTE:	The following label name should be changed to
1767#		"bn_mul_words" i.e. remove the first dot
1768#		for the gcc compiler. This should be automatically
1769#		done in the build
1770#
1771
1772.align	4
1773.bn_mul_words:
1774#
1775# BN_ULONG bn_mul_words(BN_ULONG *rp, BN_ULONG *ap, int num, BN_ULONG w)
1776#
1777# r3 = rp
1778# r4 = ap
1779# r5 = num
1780# r6 = w
1781	xor	r0,r0,r0
1782	xor	r12,r12,r12		# used for carry
1783	rlwinm.	r7,r5,30,2,31		# num >> 2
1784	beq	Lppcasm_mw_REM
1785	mtctr	r7
1786Lppcasm_mw_LOOP:
1787					#mul(rp[0],ap[0],w,c1);
1788	$LD	r8,`0*$BNSZ`(r4)
1789	$UMULL	r9,r6,r8
1790	$UMULH  r10,r6,r8
1791	addc	r9,r9,r12
1792	#addze	r10,r10			#carry is NOT ignored.
1793					#will be taken care of
1794					#in second spin below
1795					#using adde.
1796	$ST	r9,`0*$BNSZ`(r3)
1797					#mul(rp[1],ap[1],w,c1);
1798	$LD	r8,`1*$BNSZ`(r4)
1799	$UMULL	r11,r6,r8
1800	$UMULH  r12,r6,r8
1801	adde	r11,r11,r10
1802	#addze	r12,r12
1803	$ST	r11,`1*$BNSZ`(r3)
1804					#mul(rp[2],ap[2],w,c1);
1805	$LD	r8,`2*$BNSZ`(r4)
1806	$UMULL	r9,r6,r8
1807	$UMULH  r10,r6,r8
1808	adde	r9,r9,r12
1809	#addze	r10,r10
1810	$ST	r9,`2*$BNSZ`(r3)
1811					#mul_add(rp[3],ap[3],w,c1);
1812	$LD	r8,`3*$BNSZ`(r4)
1813	$UMULL	r11,r6,r8
1814	$UMULH  r12,r6,r8
1815	adde	r11,r11,r10
1816	addze	r12,r12			#this spin we collect carry into
1817					#r12
1818	$ST	r11,`3*$BNSZ`(r3)
1819
1820	addi	r3,r3,`4*$BNSZ`
1821	addi	r4,r4,`4*$BNSZ`
1822	bdnz-	Lppcasm_mw_LOOP
1823
1824Lppcasm_mw_REM:
1825	andi.	r5,r5,0x3
1826	beq	Lppcasm_mw_OVER
1827					#mul(rp[0],ap[0],w,c1);
1828	$LD	r8,`0*$BNSZ`(r4)
1829	$UMULL	r9,r6,r8
1830	$UMULH  r10,r6,r8
1831	addc	r9,r9,r12
1832	addze	r10,r10
1833	$ST	r9,`0*$BNSZ`(r3)
1834	addi	r12,r10,0
1835
1836	addi	r5,r5,-1
1837	cmpli	0,0,r5,0
1838	beq	Lppcasm_mw_OVER
1839
1840
1841					#mul(rp[1],ap[1],w,c1);
1842	$LD	r8,`1*$BNSZ`(r4)
1843	$UMULL	r9,r6,r8
1844	$UMULH  r10,r6,r8
1845	addc	r9,r9,r12
1846	addze	r10,r10
1847	$ST	r9,`1*$BNSZ`(r3)
1848	addi	r12,r10,0
1849
1850	addi	r5,r5,-1
1851	cmpli	0,0,r5,0
1852	beq	Lppcasm_mw_OVER
1853
1854					#mul_add(rp[2],ap[2],w,c1);
1855	$LD	r8,`2*$BNSZ`(r4)
1856	$UMULL	r9,r6,r8
1857	$UMULH  r10,r6,r8
1858	addc	r9,r9,r12
1859	addze	r10,r10
1860	$ST	r9,`2*$BNSZ`(r3)
1861	addi	r12,r10,0
1862
1863Lppcasm_mw_OVER:
1864	addi	r3,r12,0
1865	blr
1866	.long	0
1867	.byte	0,12,0x14,0,0,0,4,0
1868	.long	0
1869
1870#
1871#	NOTE:	The following label name should be changed to
1872#		"bn_mul_add_words" i.e. remove the first dot
1873#		for the gcc compiler. This should be automatically
1874#		done in the build
1875#
1876
1877.align	4
1878.bn_mul_add_words:
1879#
1880# BN_ULONG bn_mul_add_words(BN_ULONG *rp, BN_ULONG *ap, int num, BN_ULONG w)
1881#
1882# r3 = rp
1883# r4 = ap
1884# r5 = num
1885# r6 = w
1886#
1887# empirical evidence suggests that unrolled version performs best!!
1888#
1889	xor	r0,r0,r0		#r0 = 0
1890	xor	r12,r12,r12  		#r12 = 0 . used for carry
1891	rlwinm.	r7,r5,30,2,31		# num >> 2
1892	beq	Lppcasm_maw_leftover	# if (num < 4) go LPPCASM_maw_leftover
1893	mtctr	r7
1894Lppcasm_maw_mainloop:
1895					#mul_add(rp[0],ap[0],w,c1);
1896	$LD	r8,`0*$BNSZ`(r4)
1897	$LD	r11,`0*$BNSZ`(r3)
1898	$UMULL	r9,r6,r8
1899	$UMULH  r10,r6,r8
1900	addc	r9,r9,r12		#r12 is carry.
1901	addze	r10,r10
1902	addc	r9,r9,r11
1903	#addze	r10,r10
1904					#the above instruction addze
1905					#is NOT needed. Carry will NOT
1906					#be ignored. It's not affected
1907					#by multiply and will be collected
1908					#in the next spin
1909	$ST	r9,`0*$BNSZ`(r3)
1910
1911					#mul_add(rp[1],ap[1],w,c1);
1912	$LD	r8,`1*$BNSZ`(r4)
1913	$LD	r9,`1*$BNSZ`(r3)
1914	$UMULL	r11,r6,r8
1915	$UMULH  r12,r6,r8
1916	adde	r11,r11,r10		#r10 is carry.
1917	addze	r12,r12
1918	addc	r11,r11,r9
1919	#addze	r12,r12
1920	$ST	r11,`1*$BNSZ`(r3)
1921
1922					#mul_add(rp[2],ap[2],w,c1);
1923	$LD	r8,`2*$BNSZ`(r4)
1924	$UMULL	r9,r6,r8
1925	$LD	r11,`2*$BNSZ`(r3)
1926	$UMULH  r10,r6,r8
1927	adde	r9,r9,r12
1928	addze	r10,r10
1929	addc	r9,r9,r11
1930	#addze	r10,r10
1931	$ST	r9,`2*$BNSZ`(r3)
1932
1933					#mul_add(rp[3],ap[3],w,c1);
1934	$LD	r8,`3*$BNSZ`(r4)
1935	$UMULL	r11,r6,r8
1936	$LD	r9,`3*$BNSZ`(r3)
1937	$UMULH  r12,r6,r8
1938	adde	r11,r11,r10
1939	addze	r12,r12
1940	addc	r11,r11,r9
1941	addze	r12,r12
1942	$ST	r11,`3*$BNSZ`(r3)
1943	addi	r3,r3,`4*$BNSZ`
1944	addi	r4,r4,`4*$BNSZ`
1945	bdnz-	Lppcasm_maw_mainloop
1946
1947Lppcasm_maw_leftover:
1948	andi.	r5,r5,0x3
1949	beq	Lppcasm_maw_adios
1950	addi	r3,r3,-$BNSZ
1951	addi	r4,r4,-$BNSZ
1952					#mul_add(rp[0],ap[0],w,c1);
1953	mtctr	r5
1954	$LDU	r8,$BNSZ(r4)
1955	$UMULL	r9,r6,r8
1956	$UMULH  r10,r6,r8
1957	$LDU	r11,$BNSZ(r3)
1958	addc	r9,r9,r11
1959	addze	r10,r10
1960	addc	r9,r9,r12
1961	addze	r12,r10
1962	$ST	r9,0(r3)
1963
1964	bdz	Lppcasm_maw_adios
1965					#mul_add(rp[1],ap[1],w,c1);
1966	$LDU	r8,$BNSZ(r4)
1967	$UMULL	r9,r6,r8
1968	$UMULH  r10,r6,r8
1969	$LDU	r11,$BNSZ(r3)
1970	addc	r9,r9,r11
1971	addze	r10,r10
1972	addc	r9,r9,r12
1973	addze	r12,r10
1974	$ST	r9,0(r3)
1975
1976	bdz	Lppcasm_maw_adios
1977					#mul_add(rp[2],ap[2],w,c1);
1978	$LDU	r8,$BNSZ(r4)
1979	$UMULL	r9,r6,r8
1980	$UMULH  r10,r6,r8
1981	$LDU	r11,$BNSZ(r3)
1982	addc	r9,r9,r11
1983	addze	r10,r10
1984	addc	r9,r9,r12
1985	addze	r12,r10
1986	$ST	r9,0(r3)
1987
1988Lppcasm_maw_adios:
1989	addi	r3,r12,0
1990	blr
1991	.long	0
1992	.byte	0,12,0x14,0,0,0,4,0
1993	.long	0
1994	.align	4
1995EOF
1996$data =~ s/\`([^\`]*)\`/eval $1/gem;
1997print $data;
1998close STDOUT;
1999