sparc.txt

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

____________________________________________