summaryrefslogtreecommitdiff
path: root/src/lib/libcrypto/bn
diff options
context:
space:
mode:
authorjsing <>2016-09-04 14:31:29 +0000
committerjsing <>2016-09-04 14:31:29 +0000
commite38c58272a121e2bc9a785ec4001bbc802d68f66 (patch)
tree492fd2a4355d8592de425463d194374bdc85aa0a /src/lib/libcrypto/bn
parenta9cbed3be03a99c87e2b07b16b511e65a90bf800 (diff)
downloadopenbsd-e38c58272a121e2bc9a785ec4001bbc802d68f66.tar.gz
openbsd-e38c58272a121e2bc9a785ec4001bbc802d68f66.tar.bz2
openbsd-e38c58272a121e2bc9a785ec4001bbc802d68f66.zip
Less S390.
ok deraadt@
Diffstat (limited to 'src/lib/libcrypto/bn')
-rw-r--r--src/lib/libcrypto/bn/asm/s390x-gf2m.pl221
-rw-r--r--src/lib/libcrypto/bn/asm/s390x-mont.pl277
-rwxr-xr-xsrc/lib/libcrypto/bn/asm/s390x.S678
3 files changed, 0 insertions, 1176 deletions
diff --git a/src/lib/libcrypto/bn/asm/s390x-gf2m.pl b/src/lib/libcrypto/bn/asm/s390x-gf2m.pl
deleted file mode 100644
index cd9f13eca2..0000000000
--- a/src/lib/libcrypto/bn/asm/s390x-gf2m.pl
+++ /dev/null
@@ -1,221 +0,0 @@
1#!/usr/bin/env perl
2#
3# ====================================================================
4# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8# ====================================================================
9#
10# May 2011
11#
12# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
13# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
14# the time being... gcc 4.3 appeared to generate poor code, therefore
15# the effort. And indeed, the module delivers 55%-90%(*) improvement
16# on haviest ECDSA verify and ECDH benchmarks for 163- and 571-bit
17# key lengths on z990, 30%-55%(*) - on z10, and 70%-110%(*) - on z196.
18# This is for 64-bit build. In 32-bit "highgprs" case improvement is
19# even higher, for example on z990 it was measured 80%-150%. ECDSA
20# sign is modest 9%-12% faster. Keep in mind that these coefficients
21# are not ones for bn_GF2m_mul_2x2 itself, as not all CPU time is
22# burnt in it...
23#
24# (*) gcc 4.1 was observed to deliver better results than gcc 4.3,
25# so that improvement coefficients can vary from one specific
26# setup to another.
27
28$flavour = shift;
29
30if ($flavour =~ /3[12]/) {
31 $SIZE_T=4;
32 $g="";
33} else {
34 $SIZE_T=8;
35 $g="g";
36}
37
38while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
39open STDOUT,">$output";
40
41$stdframe=16*$SIZE_T+4*8;
42
43$rp="%r2";
44$a1="%r3";
45$a0="%r4";
46$b1="%r5";
47$b0="%r6";
48
49$ra="%r14";
50$sp="%r15";
51
52@T=("%r0","%r1");
53@i=("%r12","%r13");
54
55($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(6..11));
56($lo,$hi,$b)=map("%r$_",(3..5)); $a=$lo; $mask=$a8;
57
58$code.=<<___;
59.text
60
61.type _mul_1x1,\@function
62.align 16
63_mul_1x1:
64 lgr $a1,$a
65 sllg $a2,$a,1
66 sllg $a4,$a,2
67 sllg $a8,$a,3
68
69 srag $lo,$a1,63 # broadcast 63rd bit
70 nihh $a1,0x1fff
71 srag @i[0],$a2,63 # broadcast 62nd bit
72 nihh $a2,0x3fff
73 srag @i[1],$a4,63 # broadcast 61st bit
74 nihh $a4,0x7fff
75 ngr $lo,$b
76 ngr @i[0],$b
77 ngr @i[1],$b
78
79 lghi @T[0],0
80 lgr $a12,$a1
81 stg @T[0],`$stdframe+0*8`($sp) # tab[0]=0
82 xgr $a12,$a2
83 stg $a1,`$stdframe+1*8`($sp) # tab[1]=a1
84 lgr $a48,$a4
85 stg $a2,`$stdframe+2*8`($sp) # tab[2]=a2
86 xgr $a48,$a8
87 stg $a12,`$stdframe+3*8`($sp) # tab[3]=a1^a2
88 xgr $a1,$a4
89
90 stg $a4,`$stdframe+4*8`($sp) # tab[4]=a4
91 xgr $a2,$a4
92 stg $a1,`$stdframe+5*8`($sp) # tab[5]=a1^a4
93 xgr $a12,$a4
94 stg $a2,`$stdframe+6*8`($sp) # tab[6]=a2^a4
95 xgr $a1,$a48
96 stg $a12,`$stdframe+7*8`($sp) # tab[7]=a1^a2^a4
97 xgr $a2,$a48
98
99 stg $a8,`$stdframe+8*8`($sp) # tab[8]=a8
100 xgr $a12,$a48
101 stg $a1,`$stdframe+9*8`($sp) # tab[9]=a1^a8
102 xgr $a1,$a4
103 stg $a2,`$stdframe+10*8`($sp) # tab[10]=a2^a8
104 xgr $a2,$a4
105 stg $a12,`$stdframe+11*8`($sp) # tab[11]=a1^a2^a8
106
107 xgr $a12,$a4
108 stg $a48,`$stdframe+12*8`($sp) # tab[12]=a4^a8
109 srlg $hi,$lo,1
110 stg $a1,`$stdframe+13*8`($sp) # tab[13]=a1^a4^a8
111 sllg $lo,$lo,63
112 stg $a2,`$stdframe+14*8`($sp) # tab[14]=a2^a4^a8
113 srlg @T[0],@i[0],2
114 stg $a12,`$stdframe+15*8`($sp) # tab[15]=a1^a2^a4^a8
115
116 lghi $mask,`0xf<<3`
117 sllg $a1,@i[0],62
118 sllg @i[0],$b,3
119 srlg @T[1],@i[1],3
120 ngr @i[0],$mask
121 sllg $a2,@i[1],61
122 srlg @i[1],$b,4-3
123 xgr $hi,@T[0]
124 ngr @i[1],$mask
125 xgr $lo,$a1
126 xgr $hi,@T[1]
127 xgr $lo,$a2
128
129 xg $lo,$stdframe(@i[0],$sp)
130 srlg @i[0],$b,8-3
131 ngr @i[0],$mask
132___
133for($n=1;$n<14;$n++) {
134$code.=<<___;
135 lg @T[1],$stdframe(@i[1],$sp)
136 srlg @i[1],$b,`($n+2)*4`-3
137 sllg @T[0],@T[1],`$n*4`
138 ngr @i[1],$mask
139 srlg @T[1],@T[1],`64-$n*4`
140 xgr $lo,@T[0]
141 xgr $hi,@T[1]
142___
143 push(@i,shift(@i)); push(@T,shift(@T));
144}
145$code.=<<___;
146 lg @T[1],$stdframe(@i[1],$sp)
147 sllg @T[0],@T[1],`$n*4`
148 srlg @T[1],@T[1],`64-$n*4`
149 xgr $lo,@T[0]
150 xgr $hi,@T[1]
151
152 lg @T[0],$stdframe(@i[0],$sp)
153 sllg @T[1],@T[0],`($n+1)*4`
154 srlg @T[0],@T[0],`64-($n+1)*4`
155 xgr $lo,@T[1]
156 xgr $hi,@T[0]
157
158 br $ra
159.size _mul_1x1,.-_mul_1x1
160
161.globl bn_GF2m_mul_2x2
162.type bn_GF2m_mul_2x2,\@function
163.align 16
164bn_GF2m_mul_2x2:
165 stm${g} %r3,%r15,3*$SIZE_T($sp)
166
167 lghi %r1,-$stdframe-128
168 la %r0,0($sp)
169 la $sp,0(%r1,$sp) # alloca
170 st${g} %r0,0($sp) # back chain
171___
172if ($SIZE_T==8) {
173my @r=map("%r$_",(6..9));
174$code.=<<___;
175 bras $ra,_mul_1x1 # a1·b1
176 stmg $lo,$hi,16($rp)
177
178 lg $a,`$stdframe+128+4*$SIZE_T`($sp)
179 lg $b,`$stdframe+128+6*$SIZE_T`($sp)
180 bras $ra,_mul_1x1 # a0·b0
181 stmg $lo,$hi,0($rp)
182
183 lg $a,`$stdframe+128+3*$SIZE_T`($sp)
184 lg $b,`$stdframe+128+5*$SIZE_T`($sp)
185 xg $a,`$stdframe+128+4*$SIZE_T`($sp)
186 xg $b,`$stdframe+128+6*$SIZE_T`($sp)
187 bras $ra,_mul_1x1 # (a0+a1)·(b0+b1)
188 lmg @r[0],@r[3],0($rp)
189
190 xgr $lo,$hi
191 xgr $hi,@r[1]
192 xgr $lo,@r[0]
193 xgr $hi,@r[2]
194 xgr $lo,@r[3]
195 xgr $hi,@r[3]
196 xgr $lo,$hi
197 stg $hi,16($rp)
198 stg $lo,8($rp)
199___
200} else {
201$code.=<<___;
202 sllg %r3,%r3,32
203 sllg %r5,%r5,32
204 or %r3,%r4
205 or %r5,%r6
206 bras $ra,_mul_1x1
207 rllg $lo,$lo,32
208 rllg $hi,$hi,32
209 stmg $lo,$hi,0($rp)
210___
211}
212$code.=<<___;
213 lm${g} %r6,%r15,`$stdframe+128+6*$SIZE_T`($sp)
214 br $ra
215.size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
216.string "GF(2^m) Multiplication for s390x, CRYPTOGAMS by <appro\@openssl.org>"
217___
218
219$code =~ s/\`([^\`]*)\`/eval($1)/gem;
220print $code;
221close STDOUT;
diff --git a/src/lib/libcrypto/bn/asm/s390x-mont.pl b/src/lib/libcrypto/bn/asm/s390x-mont.pl
deleted file mode 100644
index 9fd64e81ee..0000000000
--- a/src/lib/libcrypto/bn/asm/s390x-mont.pl
+++ /dev/null
@@ -1,277 +0,0 @@
1#!/usr/bin/env perl
2
3# ====================================================================
4# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5# project. The module is, however, dual licensed under OpenSSL and
6# CRYPTOGAMS licenses depending on where you obtain it. For further
7# details see http://www.openssl.org/~appro/cryptogams/.
8# ====================================================================
9
10# April 2007.
11#
12# Performance improvement over vanilla C code varies from 85% to 45%
13# depending on key length and benchmark. Unfortunately in this context
14# these are not very impressive results [for code that utilizes "wide"
15# 64x64=128-bit multiplication, which is not commonly available to C
16# programmers], at least hand-coded bn_asm.c replacement is known to
17# provide 30-40% better results for longest keys. Well, on a second
18# thought it's not very surprising, because z-CPUs are single-issue
19# and _strictly_ in-order execution, while bn_mul_mont is more or less
20# dependent on CPU ability to pipe-line instructions and have several
21# of them "in-flight" at the same time. I mean while other methods,
22# for example Karatsuba, aim to minimize amount of multiplications at
23# the cost of other operations increase, bn_mul_mont aim to neatly
24# "overlap" multiplications and the other operations [and on most
25# platforms even minimize the amount of the other operations, in
26# particular references to memory]. But it's possible to improve this
27# module performance by implementing dedicated squaring code-path and
28# possibly by unrolling loops...
29
30# January 2009.
31#
32# Reschedule to minimize/avoid Address Generation Interlock hazard,
33# make inner loops counter-based.
34
35# November 2010.
36#
37# Adapt for -m31 build. If kernel supports what's called "highgprs"
38# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
39# instructions and achieve "64-bit" performance even in 31-bit legacy
40# application context. The feature is not specific to any particular
41# processor, as long as it's "z-CPU". Latter implies that the code
42# remains z/Architecture specific. Compatibility with 32-bit BN_ULONG
43# is achieved by swapping words after 64-bit loads, follow _dswap-s.
44# On z990 it was measured to perform 2.6-2.2 times better than
45# compiler-generated code, less for longer keys...
46
47$flavour = shift;
48
49if ($flavour =~ /3[12]/) {
50 $SIZE_T=4;
51 $g="";
52} else {
53 $SIZE_T=8;
54 $g="g";
55}
56
57while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
58open STDOUT,">$output";
59
60$stdframe=16*$SIZE_T+4*8;
61
62$mn0="%r0";
63$num="%r1";
64
65# int bn_mul_mont(
66$rp="%r2"; # BN_ULONG *rp,
67$ap="%r3"; # const BN_ULONG *ap,
68$bp="%r4"; # const BN_ULONG *bp,
69$np="%r5"; # const BN_ULONG *np,
70$n0="%r6"; # const BN_ULONG *n0,
71#$num="160(%r15)" # int num);
72
73$bi="%r2"; # zaps rp
74$j="%r7";
75
76$ahi="%r8";
77$alo="%r9";
78$nhi="%r10";
79$nlo="%r11";
80$AHI="%r12";
81$NHI="%r13";
82$count="%r14";
83$sp="%r15";
84
85$code.=<<___;
86.text
87.globl bn_mul_mont
88.type bn_mul_mont,\@function
89bn_mul_mont:
90 lgf $num,`$stdframe+$SIZE_T-4`($sp) # pull $num
91 sla $num,`log($SIZE_T)/log(2)` # $num to enumerate bytes
92 la $bp,0($num,$bp)
93
94 st${g} %r2,2*$SIZE_T($sp)
95
96 cghi $num,16 #
97 lghi %r2,0 #
98 blr %r14 # if($num<16) return 0;
99___
100$code.=<<___ if ($flavour =~ /3[12]/);
101 tmll $num,4
102 bnzr %r14 # if ($num&1) return 0;
103___
104$code.=<<___ if ($flavour !~ /3[12]/);
105 cghi $num,96 #
106 bhr %r14 # if($num>96) return 0;
107___
108$code.=<<___;
109 stm${g} %r3,%r15,3*$SIZE_T($sp)
110
111 lghi $rp,-$stdframe-8 # leave room for carry bit
112 lcgr $j,$num # -$num
113 lgr %r0,$sp
114 la $rp,0($rp,$sp)
115 la $sp,0($j,$rp) # alloca
116 st${g} %r0,0($sp) # back chain
117
118 sra $num,3 # restore $num
119 la $bp,0($j,$bp) # restore $bp
120 ahi $num,-1 # adjust $num for inner loop
121 lg $n0,0($n0) # pull n0
122 _dswap $n0
123
124 lg $bi,0($bp)
125 _dswap $bi
126 lg $alo,0($ap)
127 _dswap $alo
128 mlgr $ahi,$bi # ap[0]*bp[0]
129 lgr $AHI,$ahi
130
131 lgr $mn0,$alo # "tp[0]"*n0
132 msgr $mn0,$n0
133
134 lg $nlo,0($np) #
135 _dswap $nlo
136 mlgr $nhi,$mn0 # np[0]*m1
137 algr $nlo,$alo # +="tp[0]"
138 lghi $NHI,0
139 alcgr $NHI,$nhi
140
141 la $j,8(%r0) # j=1
142 lr $count,$num
143
144.align 16
145.L1st:
146 lg $alo,0($j,$ap)
147 _dswap $alo
148 mlgr $ahi,$bi # ap[j]*bp[0]
149 algr $alo,$AHI
150 lghi $AHI,0
151 alcgr $AHI,$ahi
152
153 lg $nlo,0($j,$np)
154 _dswap $nlo
155 mlgr $nhi,$mn0 # np[j]*m1
156 algr $nlo,$NHI
157 lghi $NHI,0
158 alcgr $nhi,$NHI # +="tp[j]"
159 algr $nlo,$alo
160 alcgr $NHI,$nhi
161
162 stg $nlo,$stdframe-8($j,$sp) # tp[j-1]=
163 la $j,8($j) # j++
164 brct $count,.L1st
165
166 algr $NHI,$AHI
167 lghi $AHI,0
168 alcgr $AHI,$AHI # upmost overflow bit
169 stg $NHI,$stdframe-8($j,$sp)
170 stg $AHI,$stdframe($j,$sp)
171 la $bp,8($bp) # bp++
172
173.Louter:
174 lg $bi,0($bp) # bp[i]
175 _dswap $bi
176 lg $alo,0($ap)
177 _dswap $alo
178 mlgr $ahi,$bi # ap[0]*bp[i]
179 alg $alo,$stdframe($sp) # +=tp[0]
180 lghi $AHI,0
181 alcgr $AHI,$ahi
182
183 lgr $mn0,$alo
184 msgr $mn0,$n0 # tp[0]*n0
185
186 lg $nlo,0($np) # np[0]
187 _dswap $nlo
188 mlgr $nhi,$mn0 # np[0]*m1
189 algr $nlo,$alo # +="tp[0]"
190 lghi $NHI,0
191 alcgr $NHI,$nhi
192
193 la $j,8(%r0) # j=1
194 lr $count,$num
195
196.align 16
197.Linner:
198 lg $alo,0($j,$ap)
199 _dswap $alo
200 mlgr $ahi,$bi # ap[j]*bp[i]
201 algr $alo,$AHI
202 lghi $AHI,0
203 alcgr $ahi,$AHI
204 alg $alo,$stdframe($j,$sp)# +=tp[j]
205 alcgr $AHI,$ahi
206
207 lg $nlo,0($j,$np)
208 _dswap $nlo
209 mlgr $nhi,$mn0 # np[j]*m1
210 algr $nlo,$NHI
211 lghi $NHI,0
212 alcgr $nhi,$NHI
213 algr $nlo,$alo # +="tp[j]"
214 alcgr $NHI,$nhi
215
216 stg $nlo,$stdframe-8($j,$sp) # tp[j-1]=
217 la $j,8($j) # j++
218 brct $count,.Linner
219
220 algr $NHI,$AHI
221 lghi $AHI,0
222 alcgr $AHI,$AHI
223 alg $NHI,$stdframe($j,$sp)# accumulate previous upmost overflow bit
224 lghi $ahi,0
225 alcgr $AHI,$ahi # new upmost overflow bit
226 stg $NHI,$stdframe-8($j,$sp)
227 stg $AHI,$stdframe($j,$sp)
228
229 la $bp,8($bp) # bp++
230 cl${g} $bp,`$stdframe+8+4*$SIZE_T`($j,$sp) # compare to &bp[num]
231 jne .Louter
232
233 l${g} $rp,`$stdframe+8+2*$SIZE_T`($j,$sp) # reincarnate rp
234 la $ap,$stdframe($sp)
235 ahi $num,1 # restore $num, incidentally clears "borrow"
236
237 la $j,0(%r0)
238 lr $count,$num
239.Lsub: lg $alo,0($j,$ap)
240 lg $nlo,0($j,$np)
241 _dswap $nlo
242 slbgr $alo,$nlo
243 stg $alo,0($j,$rp)
244 la $j,8($j)
245 brct $count,.Lsub
246 lghi $ahi,0
247 slbgr $AHI,$ahi # handle upmost carry
248
249 ngr $ap,$AHI
250 lghi $np,-1
251 xgr $np,$AHI
252 ngr $np,$rp
253 ogr $ap,$np # ap=borrow?tp:rp
254
255 la $j,0(%r0)
256 lgr $count,$num
257.Lcopy: lg $alo,0($j,$ap) # copy or in-place refresh
258 _dswap $alo
259 stg $j,$stdframe($j,$sp) # zap tp
260 stg $alo,0($j,$rp)
261 la $j,8($j)
262 brct $count,.Lcopy
263
264 la %r1,`$stdframe+8+6*$SIZE_T`($j,$sp)
265 lm${g} %r6,%r15,0(%r1)
266 lghi %r2,1 # signal "processed"
267 br %r14
268.size bn_mul_mont,.-bn_mul_mont
269.string "Montgomery Multiplication for s390x, CRYPTOGAMS by <appro\@openssl.org>"
270___
271
272foreach (split("\n",$code)) {
273 s/\`([^\`]*)\`/eval $1/ge;
274 s/_dswap\s+(%r[0-9]+)/sprintf("rllg\t%s,%s,32",$1,$1) if($SIZE_T==4)/e;
275 print $_,"\n";
276}
277close STDOUT;
diff --git a/src/lib/libcrypto/bn/asm/s390x.S b/src/lib/libcrypto/bn/asm/s390x.S
deleted file mode 100755
index 43fcb79bc0..0000000000
--- a/src/lib/libcrypto/bn/asm/s390x.S
+++ /dev/null
@@ -1,678 +0,0 @@
1.ident "s390x.S, version 1.1"
2// ====================================================================
3// Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
4// project.
5//
6// Rights for redistribution and usage in source and binary forms are
7// granted according to the OpenSSL license. Warranty of any kind is
8// disclaimed.
9// ====================================================================
10
11.text
12
13#define zero %r0
14
15// BN_ULONG bn_mul_add_words(BN_ULONG *r2,BN_ULONG *r3,int r4,BN_ULONG r5);
16.globl bn_mul_add_words
17.type bn_mul_add_words,@function
18.align 4
19bn_mul_add_words:
20 lghi zero,0 // zero = 0
21 la %r1,0(%r2) // put rp aside
22 lghi %r2,0 // i=0;
23 ltgfr %r4,%r4
24 bler %r14 // if (len<=0) return 0;
25
26 stmg %r6,%r10,48(%r15)
27 lghi %r10,3
28 lghi %r8,0 // carry = 0
29 nr %r10,%r4 // len%4
30 sra %r4,2 // cnt=len/4
31 jz .Loop1_madd // carry is incidentally cleared if branch taken
32 algr zero,zero // clear carry
33
34.Loop4_madd:
35 lg %r7,0(%r2,%r3) // ap[i]
36 mlgr %r6,%r5 // *=w
37 alcgr %r7,%r8 // +=carry
38 alcgr %r6,zero
39 alg %r7,0(%r2,%r1) // +=rp[i]
40 stg %r7,0(%r2,%r1) // rp[i]=
41
42 lg %r9,8(%r2,%r3)
43 mlgr %r8,%r5
44 alcgr %r9,%r6
45 alcgr %r8,zero
46 alg %r9,8(%r2,%r1)
47 stg %r9,8(%r2,%r1)
48
49 lg %r7,16(%r2,%r3)
50 mlgr %r6,%r5
51 alcgr %r7,%r8
52 alcgr %r6,zero
53 alg %r7,16(%r2,%r1)
54 stg %r7,16(%r2,%r1)
55
56 lg %r9,24(%r2,%r3)
57 mlgr %r8,%r5
58 alcgr %r9,%r6
59 alcgr %r8,zero
60 alg %r9,24(%r2,%r1)
61 stg %r9,24(%r2,%r1)
62
63 la %r2,32(%r2) // i+=4
64 brct %r4,.Loop4_madd
65
66 la %r10,1(%r10) // see if len%4 is zero ...
67 brct %r10,.Loop1_madd // without touching condition code:-)
68
69.Lend_madd:
70 alcgr %r8,zero // collect carry bit
71 lgr %r2,%r8
72 lmg %r6,%r10,48(%r15)
73 br %r14
74
75.Loop1_madd:
76 lg %r7,0(%r2,%r3) // ap[i]
77 mlgr %r6,%r5 // *=w
78 alcgr %r7,%r8 // +=carry
79 alcgr %r6,zero
80 alg %r7,0(%r2,%r1) // +=rp[i]
81 stg %r7,0(%r2,%r1) // rp[i]=
82
83 lgr %r8,%r6
84 la %r2,8(%r2) // i++
85 brct %r10,.Loop1_madd
86
87 j .Lend_madd
88.size bn_mul_add_words,.-bn_mul_add_words
89
90// BN_ULONG bn_mul_words(BN_ULONG *r2,BN_ULONG *r3,int r4,BN_ULONG r5);
91.globl bn_mul_words
92.type bn_mul_words,@function
93.align 4
94bn_mul_words:
95 lghi zero,0 // zero = 0
96 la %r1,0(%r2) // put rp aside
97 lghi %r2,0 // i=0;
98 ltgfr %r4,%r4
99 bler %r14 // if (len<=0) return 0;
100
101 stmg %r6,%r10,48(%r15)
102 lghi %r10,3
103 lghi %r8,0 // carry = 0
104 nr %r10,%r4 // len%4
105 sra %r4,2 // cnt=len/4
106 jz .Loop1_mul // carry is incidentally cleared if branch taken
107 algr zero,zero // clear carry
108
109.Loop4_mul:
110 lg %r7,0(%r2,%r3) // ap[i]
111 mlgr %r6,%r5 // *=w
112 alcgr %r7,%r8 // +=carry
113 stg %r7,0(%r2,%r1) // rp[i]=
114
115 lg %r9,8(%r2,%r3)
116 mlgr %r8,%r5
117 alcgr %r9,%r6
118 stg %r9,8(%r2,%r1)
119
120 lg %r7,16(%r2,%r3)
121 mlgr %r6,%r5
122 alcgr %r7,%r8
123 stg %r7,16(%r2,%r1)
124
125 lg %r9,24(%r2,%r3)
126 mlgr %r8,%r5
127 alcgr %r9,%r6
128 stg %r9,24(%r2,%r1)
129
130 la %r2,32(%r2) // i+=4
131 brct %r4,.Loop4_mul
132
133 la %r10,1(%r10) // see if len%4 is zero ...
134 brct %r10,.Loop1_mul // without touching condition code:-)
135
136.Lend_mul:
137 alcgr %r8,zero // collect carry bit
138 lgr %r2,%r8
139 lmg %r6,%r10,48(%r15)
140 br %r14
141
142.Loop1_mul:
143 lg %r7,0(%r2,%r3) // ap[i]
144 mlgr %r6,%r5 // *=w
145 alcgr %r7,%r8 // +=carry
146 stg %r7,0(%r2,%r1) // rp[i]=
147
148 lgr %r8,%r6
149 la %r2,8(%r2) // i++
150 brct %r10,.Loop1_mul
151
152 j .Lend_mul
153.size bn_mul_words,.-bn_mul_words
154
155// void bn_sqr_words(BN_ULONG *r2,BN_ULONG *r2,int r4)
156.globl bn_sqr_words
157.type bn_sqr_words,@function
158.align 4
159bn_sqr_words:
160 ltgfr %r4,%r4
161 bler %r14
162
163 stmg %r6,%r7,48(%r15)
164 srag %r1,%r4,2 // cnt=len/4
165 jz .Loop1_sqr
166
167.Loop4_sqr:
168 lg %r7,0(%r3)
169 mlgr %r6,%r7
170 stg %r7,0(%r2)
171 stg %r6,8(%r2)
172
173 lg %r7,8(%r3)
174 mlgr %r6,%r7
175 stg %r7,16(%r2)
176 stg %r6,24(%r2)
177
178 lg %r7,16(%r3)
179 mlgr %r6,%r7
180 stg %r7,32(%r2)
181 stg %r6,40(%r2)
182
183 lg %r7,24(%r3)
184 mlgr %r6,%r7
185 stg %r7,48(%r2)
186 stg %r6,56(%r2)
187
188 la %r3,32(%r3)
189 la %r2,64(%r2)
190 brct %r1,.Loop4_sqr
191
192 lghi %r1,3
193 nr %r4,%r1 // cnt=len%4
194 jz .Lend_sqr
195
196.Loop1_sqr:
197 lg %r7,0(%r3)
198 mlgr %r6,%r7
199 stg %r7,0(%r2)
200 stg %r6,8(%r2)
201
202 la %r3,8(%r3)
203 la %r2,16(%r2)
204 brct %r4,.Loop1_sqr
205
206.Lend_sqr:
207 lmg %r6,%r7,48(%r15)
208 br %r14
209.size bn_sqr_words,.-bn_sqr_words
210
211// BN_ULONG bn_div_words(BN_ULONG h,BN_ULONG l,BN_ULONG d);
212.globl bn_div_words
213.type bn_div_words,@function
214.align 4
215bn_div_words:
216 dlgr %r2,%r4
217 lgr %r2,%r3
218 br %r14
219.size bn_div_words,.-bn_div_words
220
221// BN_ULONG bn_add_words(BN_ULONG *r2,BN_ULONG *r3,BN_ULONG *r4,int r5);
222.globl bn_add_words
223.type bn_add_words,@function
224.align 4
225bn_add_words:
226 la %r1,0(%r2) // put rp aside
227 lghi %r2,0 // i=0
228 ltgfr %r5,%r5
229 bler %r14 // if (len<=0) return 0;
230
231 stg %r6,48(%r15)
232 lghi %r6,3
233 nr %r6,%r5 // len%4
234 sra %r5,2 // len/4, use sra because it sets condition code
235 jz .Loop1_add // carry is incidentally cleared if branch taken
236 algr %r2,%r2 // clear carry
237
238.Loop4_add:
239 lg %r0,0(%r2,%r3)
240 alcg %r0,0(%r2,%r4)
241 stg %r0,0(%r2,%r1)
242 lg %r0,8(%r2,%r3)
243 alcg %r0,8(%r2,%r4)
244 stg %r0,8(%r2,%r1)
245 lg %r0,16(%r2,%r3)
246 alcg %r0,16(%r2,%r4)
247 stg %r0,16(%r2,%r1)
248 lg %r0,24(%r2,%r3)
249 alcg %r0,24(%r2,%r4)
250 stg %r0,24(%r2,%r1)
251
252 la %r2,32(%r2) // i+=4
253 brct %r5,.Loop4_add
254
255 la %r6,1(%r6) // see if len%4 is zero ...
256 brct %r6,.Loop1_add // without touching condition code:-)
257
258.Lexit_add:
259 lghi %r2,0
260 alcgr %r2,%r2
261 lg %r6,48(%r15)
262 br %r14
263
264.Loop1_add:
265 lg %r0,0(%r2,%r3)
266 alcg %r0,0(%r2,%r4)
267 stg %r0,0(%r2,%r1)
268
269 la %r2,8(%r2) // i++
270 brct %r6,.Loop1_add
271
272 j .Lexit_add
273.size bn_add_words,.-bn_add_words
274
275// BN_ULONG bn_sub_words(BN_ULONG *r2,BN_ULONG *r3,BN_ULONG *r4,int r5);
276.globl bn_sub_words
277.type bn_sub_words,@function
278.align 4
279bn_sub_words:
280 la %r1,0(%r2) // put rp aside
281 lghi %r2,0 // i=0
282 ltgfr %r5,%r5
283 bler %r14 // if (len<=0) return 0;
284
285 stg %r6,48(%r15)
286 lghi %r6,3
287 nr %r6,%r5 // len%4
288 sra %r5,2 // len/4, use sra because it sets condition code
289 jnz .Loop4_sub // borrow is incidentally cleared if branch taken
290 slgr %r2,%r2 // clear borrow
291
292.Loop1_sub:
293 lg %r0,0(%r2,%r3)
294 slbg %r0,0(%r2,%r4)
295 stg %r0,0(%r2,%r1)
296
297 la %r2,8(%r2) // i++
298 brct %r6,.Loop1_sub
299 j .Lexit_sub
300
301.Loop4_sub:
302 lg %r0,0(%r2,%r3)
303 slbg %r0,0(%r2,%r4)
304 stg %r0,0(%r2,%r1)
305 lg %r0,8(%r2,%r3)
306 slbg %r0,8(%r2,%r4)
307 stg %r0,8(%r2,%r1)
308 lg %r0,16(%r2,%r3)
309 slbg %r0,16(%r2,%r4)
310 stg %r0,16(%r2,%r1)
311 lg %r0,24(%r2,%r3)
312 slbg %r0,24(%r2,%r4)
313 stg %r0,24(%r2,%r1)
314
315 la %r2,32(%r2) // i+=4
316 brct %r5,.Loop4_sub
317
318 la %r6,1(%r6) // see if len%4 is zero ...
319 brct %r6,.Loop1_sub // without touching condition code:-)
320
321.Lexit_sub:
322 lghi %r2,0
323 slbgr %r2,%r2
324 lcgr %r2,%r2
325 lg %r6,48(%r15)
326 br %r14
327.size bn_sub_words,.-bn_sub_words
328
329#define c1 %r1
330#define c2 %r5
331#define c3 %r8
332
333#define mul_add_c(ai,bi,c1,c2,c3) \
334 lg %r7,ai*8(%r3); \
335 mlg %r6,bi*8(%r4); \
336 algr c1,%r7; \
337 alcgr c2,%r6; \
338 alcgr c3,zero
339
340// void bn_mul_comba8(BN_ULONG *r2,BN_ULONG *r3,BN_ULONG *r4);
341.globl bn_mul_comba8
342.type bn_mul_comba8,@function
343.align 4
344bn_mul_comba8:
345 stmg %r6,%r8,48(%r15)
346
347 lghi c1,0
348 lghi c2,0
349 lghi c3,0
350 lghi zero,0
351
352 mul_add_c(0,0,c1,c2,c3);
353 stg c1,0*8(%r2)
354 lghi c1,0
355
356 mul_add_c(0,1,c2,c3,c1);
357 mul_add_c(1,0,c2,c3,c1);
358 stg c2,1*8(%r2)
359 lghi c2,0
360
361 mul_add_c(2,0,c3,c1,c2);
362 mul_add_c(1,1,c3,c1,c2);
363 mul_add_c(0,2,c3,c1,c2);
364 stg c3,2*8(%r2)
365 lghi c3,0
366
367 mul_add_c(0,3,c1,c2,c3);
368 mul_add_c(1,2,c1,c2,c3);
369 mul_add_c(2,1,c1,c2,c3);
370 mul_add_c(3,0,c1,c2,c3);
371 stg c1,3*8(%r2)
372 lghi c1,0
373
374 mul_add_c(4,0,c2,c3,c1);
375 mul_add_c(3,1,c2,c3,c1);
376 mul_add_c(2,2,c2,c3,c1);
377 mul_add_c(1,3,c2,c3,c1);
378 mul_add_c(0,4,c2,c3,c1);
379 stg c2,4*8(%r2)
380 lghi c2,0
381
382 mul_add_c(0,5,c3,c1,c2);
383 mul_add_c(1,4,c3,c1,c2);
384 mul_add_c(2,3,c3,c1,c2);
385 mul_add_c(3,2,c3,c1,c2);
386 mul_add_c(4,1,c3,c1,c2);
387 mul_add_c(5,0,c3,c1,c2);
388 stg c3,5*8(%r2)
389 lghi c3,0
390
391 mul_add_c(6,0,c1,c2,c3);
392 mul_add_c(5,1,c1,c2,c3);
393 mul_add_c(4,2,c1,c2,c3);
394 mul_add_c(3,3,c1,c2,c3);
395 mul_add_c(2,4,c1,c2,c3);
396 mul_add_c(1,5,c1,c2,c3);
397 mul_add_c(0,6,c1,c2,c3);
398 stg c1,6*8(%r2)
399 lghi c1,0
400
401 mul_add_c(0,7,c2,c3,c1);
402 mul_add_c(1,6,c2,c3,c1);
403 mul_add_c(2,5,c2,c3,c1);
404 mul_add_c(3,4,c2,c3,c1);
405 mul_add_c(4,3,c2,c3,c1);
406 mul_add_c(5,2,c2,c3,c1);
407 mul_add_c(6,1,c2,c3,c1);
408 mul_add_c(7,0,c2,c3,c1);
409 stg c2,7*8(%r2)
410 lghi c2,0
411
412 mul_add_c(7,1,c3,c1,c2);
413 mul_add_c(6,2,c3,c1,c2);
414 mul_add_c(5,3,c3,c1,c2);
415 mul_add_c(4,4,c3,c1,c2);
416 mul_add_c(3,5,c3,c1,c2);
417 mul_add_c(2,6,c3,c1,c2);
418 mul_add_c(1,7,c3,c1,c2);
419 stg c3,8*8(%r2)
420 lghi c3,0
421
422 mul_add_c(2,7,c1,c2,c3);
423 mul_add_c(3,6,c1,c2,c3);
424 mul_add_c(4,5,c1,c2,c3);
425 mul_add_c(5,4,c1,c2,c3);
426 mul_add_c(6,3,c1,c2,c3);
427 mul_add_c(7,2,c1,c2,c3);
428 stg c1,9*8(%r2)
429 lghi c1,0
430
431 mul_add_c(7,3,c2,c3,c1);
432 mul_add_c(6,4,c2,c3,c1);
433 mul_add_c(5,5,c2,c3,c1);
434 mul_add_c(4,6,c2,c3,c1);
435 mul_add_c(3,7,c2,c3,c1);
436 stg c2,10*8(%r2)
437 lghi c2,0
438
439 mul_add_c(4,7,c3,c1,c2);
440 mul_add_c(5,6,c3,c1,c2);
441 mul_add_c(6,5,c3,c1,c2);
442 mul_add_c(7,4,c3,c1,c2);
443 stg c3,11*8(%r2)
444 lghi c3,0
445
446 mul_add_c(7,5,c1,c2,c3);
447 mul_add_c(6,6,c1,c2,c3);
448 mul_add_c(5,7,c1,c2,c3);
449 stg c1,12*8(%r2)
450 lghi c1,0
451
452
453 mul_add_c(6,7,c2,c3,c1);
454 mul_add_c(7,6,c2,c3,c1);
455 stg c2,13*8(%r2)
456 lghi c2,0
457
458 mul_add_c(7,7,c3,c1,c2);
459 stg c3,14*8(%r2)
460 stg c1,15*8(%r2)
461
462 lmg %r6,%r8,48(%r15)
463 br %r14
464.size bn_mul_comba8,.-bn_mul_comba8
465
466// void bn_mul_comba4(BN_ULONG *r2,BN_ULONG *r3,BN_ULONG *r4);
467.globl bn_mul_comba4
468.type bn_mul_comba4,@function
469.align 4
470bn_mul_comba4:
471 stmg %r6,%r8,48(%r15)
472
473 lghi c1,0
474 lghi c2,0
475 lghi c3,0
476 lghi zero,0
477
478 mul_add_c(0,0,c1,c2,c3);
479 stg c1,0*8(%r3)
480 lghi c1,0
481
482 mul_add_c(0,1,c2,c3,c1);
483 mul_add_c(1,0,c2,c3,c1);
484 stg c2,1*8(%r2)
485 lghi c2,0
486
487 mul_add_c(2,0,c3,c1,c2);
488 mul_add_c(1,1,c3,c1,c2);
489 mul_add_c(0,2,c3,c1,c2);
490 stg c3,2*8(%r2)
491 lghi c3,0
492
493 mul_add_c(0,3,c1,c2,c3);
494 mul_add_c(1,2,c1,c2,c3);
495 mul_add_c(2,1,c1,c2,c3);
496 mul_add_c(3,0,c1,c2,c3);
497 stg c1,3*8(%r2)
498 lghi c1,0
499
500 mul_add_c(3,1,c2,c3,c1);
501 mul_add_c(2,2,c2,c3,c1);
502 mul_add_c(1,3,c2,c3,c1);
503 stg c2,4*8(%r2)
504 lghi c2,0
505
506 mul_add_c(2,3,c3,c1,c2);
507 mul_add_c(3,2,c3,c1,c2);
508 stg c3,5*8(%r2)
509 lghi c3,0
510
511 mul_add_c(3,3,c1,c2,c3);
512 stg c1,6*8(%r2)
513 stg c2,7*8(%r2)
514
515 stmg %r6,%r8,48(%r15)
516 br %r14
517.size bn_mul_comba4,.-bn_mul_comba4
518
519#define sqr_add_c(ai,c1,c2,c3) \
520 lg %r7,ai*8(%r3); \
521 mlgr %r6,%r7; \
522 algr c1,%r7; \
523 alcgr c2,%r6; \
524 alcgr c3,zero
525
526#define sqr_add_c2(ai,aj,c1,c2,c3) \
527 lg %r7,ai*8(%r3); \
528 mlg %r6,aj*8(%r3); \
529 algr c1,%r7; \
530 alcgr c2,%r6; \
531 alcgr c3,zero; \
532 algr c1,%r7; \
533 alcgr c2,%r6; \
534 alcgr c3,zero
535
536// void bn_sqr_comba8(BN_ULONG *r2,BN_ULONG *r3);
537.globl bn_sqr_comba8
538.type bn_sqr_comba8,@function
539.align 4
540bn_sqr_comba8:
541 stmg %r6,%r8,48(%r15)
542
543 lghi c1,0
544 lghi c2,0
545 lghi c3,0
546 lghi zero,0
547
548 sqr_add_c(0,c1,c2,c3);
549 stg c1,0*8(%r2)
550 lghi c1,0
551
552 sqr_add_c2(1,0,c2,c3,c1);
553 stg c2,1*8(%r2)
554 lghi c2,0
555
556 sqr_add_c(1,c3,c1,c2);
557 sqr_add_c2(2,0,c3,c1,c2);
558 stg c3,2*8(%r2)
559 lghi c3,0
560
561 sqr_add_c2(3,0,c1,c2,c3);
562 sqr_add_c2(2,1,c1,c2,c3);
563 stg c1,3*8(%r2)
564 lghi c1,0
565
566 sqr_add_c(2,c2,c3,c1);
567 sqr_add_c2(3,1,c2,c3,c1);
568 sqr_add_c2(4,0,c2,c3,c1);
569 stg c2,4*8(%r2)
570 lghi c2,0
571
572 sqr_add_c2(5,0,c3,c1,c2);
573 sqr_add_c2(4,1,c3,c1,c2);
574 sqr_add_c2(3,2,c3,c1,c2);
575 stg c3,5*8(%r2)
576 lghi c3,0
577
578 sqr_add_c(3,c1,c2,c3);
579 sqr_add_c2(4,2,c1,c2,c3);
580 sqr_add_c2(5,1,c1,c2,c3);
581 sqr_add_c2(6,0,c1,c2,c3);
582 stg c1,6*8(%r2)
583 lghi c1,0
584
585 sqr_add_c2(7,0,c2,c3,c1);
586 sqr_add_c2(6,1,c2,c3,c1);
587 sqr_add_c2(5,2,c2,c3,c1);
588 sqr_add_c2(4,3,c2,c3,c1);
589 stg c2,7*8(%r2)
590 lghi c2,0
591
592 sqr_add_c(4,c3,c1,c2);
593 sqr_add_c2(5,3,c3,c1,c2);
594 sqr_add_c2(6,2,c3,c1,c2);
595 sqr_add_c2(7,1,c3,c1,c2);
596 stg c3,8*8(%r2)
597 lghi c3,0
598
599 sqr_add_c2(7,2,c1,c2,c3);
600 sqr_add_c2(6,3,c1,c2,c3);
601 sqr_add_c2(5,4,c1,c2,c3);
602 stg c1,9*8(%r2)
603 lghi c1,0
604
605 sqr_add_c(5,c2,c3,c1);
606 sqr_add_c2(6,4,c2,c3,c1);
607 sqr_add_c2(7,3,c2,c3,c1);
608 stg c2,10*8(%r2)
609 lghi c2,0
610
611 sqr_add_c2(7,4,c3,c1,c2);
612 sqr_add_c2(6,5,c3,c1,c2);
613 stg c3,11*8(%r2)
614 lghi c3,0
615
616 sqr_add_c(6,c1,c2,c3);
617 sqr_add_c2(7,5,c1,c2,c3);
618 stg c1,12*8(%r2)
619 lghi c1,0
620
621 sqr_add_c2(7,6,c2,c3,c1);
622 stg c2,13*8(%r2)
623 lghi c2,0
624
625 sqr_add_c(7,c3,c1,c2);
626 stg c3,14*8(%r2)
627 stg c1,15*8(%r2)
628
629 lmg %r6,%r8,48(%r15)
630 br %r14
631.size bn_sqr_comba8,.-bn_sqr_comba8
632
633// void bn_sqr_comba4(BN_ULONG *r2,BN_ULONG *r3);
634.globl bn_sqr_comba4
635.type bn_sqr_comba4,@function
636.align 4
637bn_sqr_comba4:
638 stmg %r6,%r8,48(%r15)
639
640 lghi c1,0
641 lghi c2,0
642 lghi c3,0
643 lghi zero,0
644
645 sqr_add_c(0,c1,c2,c3);
646 stg c1,0*8(%r2)
647 lghi c1,0
648
649 sqr_add_c2(1,0,c2,c3,c1);
650 stg c2,1*8(%r2)
651 lghi c2,0
652
653 sqr_add_c(1,c3,c1,c2);
654 sqr_add_c2(2,0,c3,c1,c2);
655 stg c3,2*8(%r2)
656 lghi c3,0
657
658 sqr_add_c2(3,0,c1,c2,c3);
659 sqr_add_c2(2,1,c1,c2,c3);
660 stg c1,3*8(%r2)
661 lghi c1,0
662
663 sqr_add_c(2,c2,c3,c1);
664 sqr_add_c2(3,1,c2,c3,c1);
665 stg c2,4*8(%r2)
666 lghi c2,0
667
668 sqr_add_c2(3,2,c3,c1,c2);
669 stg c3,5*8(%r2)
670 lghi c3,0
671
672 sqr_add_c(3,c1,c2,c3);
673 stg c1,6*8(%r2)
674 stg c2,7*8(%r2)
675
676 lmg %r6,%r8,48(%r15)
677 br %r14
678.size bn_sqr_comba4,.-bn_sqr_comba4
generated by cgit v1.2.3-55-g6feb (git 2.39.1) at 2025-08-13 11:09:58 +0000