forked from miracl/MIRACL
-
Notifications
You must be signed in to change notification settings - Fork 0
/
sparc.txt
176 lines (127 loc) · 5.31 KB
/
sparc.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
NOTE: On Sparcs with hardware support for quad-precision long doubles,
it may be optimal to build a MIRACL library using a "double" underlying
type rather than use the approach described here. See double.txt
*************************************
These comments apply to the standard 32-bit SPARC (Version 8) processor with
hardware 32-bit multiplication. For 64-bit SPARC (Version 9) see below.
*************************************
If developing for the SPARC, or indeed any other new processor, you should
first build a C-only library.
For the SPARC, this mirdef.h header would be appropriate for an integer-
only build of the library.
--------------------------------------
/*
* MIRACL compiler/hardware definitions - mirdef.h
*/
#define MIRACL 32
#define MR_BIG_ENDIAN
#define mr_utype int
#define MR_IBITS 32
#define MR_LBITS 32
#define mr_dltype long long
#define mr_unsign32 unsigned int
#define mr_unsign64 unsigned long long
#define MAXBASE ((mr_small)1<<(MIRACL-1))
#define MR_NOASM
---------------------------------------------
Assuming that the mirdef.h, miracl.h and mr*.c files are all in the same
directory, then a suitable batch file for building a MIRACL library might
look like this:-
-------------------------------
gcc -I. -c -O2 mrcore.c
gcc -I. -c -O2 mrarth0.c
gcc -I. -c -O2 mrarth1.c
gcc -I. -c -O2 mrarth2.c
gcc -I. -c -O2 mralloc.c
gcc -I. -c -O2 mrsmall.c
gcc -I. -c -O2 mrio1.c
gcc -I. -c -O2 mrio2.c
gcc -I. -c -O2 mrgcd.c
gcc -I. -c -O2 mrjack.c
gcc -I. -c -O2 mrxgcd.c
gcc -I. -c -O2 mrarth3.c
gcc -I. -c -O2 mrbits.c
gcc -I. -c -O2 mrrand.c
gcc -I. -c -O2 mrprime.c
gcc -I. -c -O2 mrcrt.c
gcc -I. -c -O2 mrscrt.c
gcc -I. -c -O2 mrmonty.c
gcc -I. -c -O2 mrpower.c
gcc -I. -c -O2 mrsroot.c
gcc -I. -c -O2 mrcurve.c
gcc -I. -c -O2 mrfast.c
gcc -I. -c -O2 mrzzn2.c
gcc -I. -c -O2 mrzzn2b.c
gcc -I. -c -O2 mrzzn3.c
gcc -I. -c -O2 mrecn2.c
gcc -I. -c -O2 mrshs.c
gcc -I. -c -O2 mrshs256.c
gcc -I. -c -O2 mrshs512.c
gcc -I. -c -O2 mraes.c
gcc -I. -c -O2 mrlucas.c
gcc -I. -c -O2 mrstrong.c
gcc -I. -c -O2 mrbrick.c
gcc -I. -c -O2 mrebrick.c
gcc -I. -c -O2 mrec2m.c
gcc -I. -c -O2 mrgf2m.c
ar rc miracl.a mrcore.o mrarth0.o mrarth1.o mrarth2.o mralloc.o mrsmall.o
ar r miracl.a mrio1.o mrio2.o mrjack.o mrgcd.o mrxgcd.o mrarth3.o mrsroot.o
ar r miracl.a mrrand.o mrprime.o mrcrt.o mrscrt.o mrmonty.o mrcurve.o
ar r miracl.a mrfast.o mrshs.o mraes.o mrlucas.o mrstrong.o mrbrick.o
ar r miracl.a mrebrick.o mrgf2m.o mrec2m.o mrpower.o mrzzn2b.o
ar r miracl.a mrshs256.o mrshs512.o mrbits.o mrzzn2.o mrzzn3.o mrecn2.o
del mr*.o
gcc -I.-O2 pk-demo.c miracl.a -o pk-demo
--------------------------------------------
This may be fast enough for you. If its not you can use the assembly language
macros provided in sparc32.mcs for greater speed. See kcmcomba.txt.
For faster RSA and DH implementations replace the MR_NOASM definition with
MR_KCM n (where n is usually 4, 8 or 16 - experiment. n*MIRACL must divide the
modulus size in bits exactly, which it will for standard moduli of 1024 bit
for example). Compile and run the utility mex.c
c:\miracl>mex n sparc32 mrkcm
(Yes its the same n). Rebuild the MIRACL library, but this time include the
modules mrkcm.c and mrmuldv.c (you can find the latter as mrmuldv.ccc This
standard C version will do, although the SPARC asm versions from mrmuldv.any
are faster. These would need to be assembled rather than compiled)
For fast GF(p) elliptic curves, replace MR_NOASM with MR_COMBA n. This time
32*n is exactly the size of p in bits (assuming 32-bit processor).
c:\miracl>mex n sparc32 mrcomba
Rebuild the MIRACL library, but this time include the modules mrcomba.c and
mrmuldv.c.
Still not fast enough? If the prime p is of a "special" form, define in
mirdef.h MR_SPECIAL. Edit mrcomba.tpl to insert "special" code for modular
reduction - its quite easy and you will find examples there already. Run
mex as before, and rebuild MIRACL again.
For processors other than the SPARC, the basic procedure is the same. A C-only
build is always possible. To go faster you will need to create a .mcs file
for your processor, and then you can proceed as above.
An alternative is to do a C-only build and then go in and optimise the
generated assembly language. The time-critical routines are usually
multiply() and redc() which can be found in mrarth2.c and mrmonty.c
This will probably not be as fast as the highly optimised approach outlined
above.
----------------------------------------------------------
64-bit SPARC (Version 9). Alas not a "real" 64-bit processor in the sense that
there is no 64x64=128 bit multiply instruction.
The standard C header files mirdef.h should in this case look like
--------------------------------------
/*
* MIRACL compiler/hardware definitions - mirdef.h
*/
#define MIRACL 32
#define MR_BIG_ENDIAN
#define mr_utype int
#define MR_IBITS 32
#define MR_LBITS 32
#define mr_dltype long
#define mr_unsign32 unsigned int
#define mr_unsign64 unsigned long
#define MAXBASE ((mr_small)1<<(MIRACL-1))
#define MR_NOASM
---------------------------------------------
Compile as above, but include compiler flag -m64. Also you may need to change
-O2 to -O1 - when I tried it -O2 optimization was broken.
For faster assembly language implementation proceed as above, but this time
use macros from sparc64.mcs
____________________________________________