Note that the UNROLL option makes the 'inner' des loop unroll all 16 rounds
instead of the default 4.
RISC1 and RISC2 are 2 alternatives for the inner loop and
PTR means to use pointers arithmetic instead of arrays.

FreeBSD - Pentium Pro 200mhz - gcc 2.7.2.2 - assembler		577,000 4620k/s
IRIX 6.2 - R10000 195mhz - cc (-O3 -n32) - UNROLL RISC2 PTR	496,000 3968k/s
solaris 2.5.1 usparc 167mhz?? - SC4.0 - UNROLL RISC1 PTR [1]	459,400 3672k/s
FreeBSD - Pentium Pro 200mhz - gcc 2.7.2.2 - UNROLL RISC1	433,000 3468k/s
solaris 2.5.1 usparc 167mhz?? - gcc 2.7.2 - UNROLL 		380,000 3041k/s
linux - pentium 100mhz - gcc 2.7.0 - assembler			281,000 2250k/s
NT 4.0 - pentium 100mhz - VC 4.2 - assembler			281,000 2250k/s
AIX 4.1? - PPC604 100mhz - cc - UNROLL 				275,000 2200k/s
IRIX 5.3 - R4400 200mhz - gcc 2.6.3 - UNROLL RISC2 PTR		235,300 1882k/s
IRIX 5.3 - R4400 200mhz - cc - UNROLL RISC2 PTR			233,700 1869k/s
NT 4.0 - pentium 100mhz - VC 4.2 - UNROLL RISC1 PTR		191,000 1528k/s
DEC Alpha 165mhz??  - cc - RISC2 PTR [2]			181,000 1448k/s
linux - pentium 100mhz - gcc 2.7.0 - UNROLL RISC1 PTR		158,500 1268k/s
HPUX 10 - 9000/887 - cc - UNROLL [3]	 			148,000	1190k/s
solaris 2.5.1 - sparc 10 50mhz - gcc 2.7.2 - UNROLL		123,600  989k/s
IRIX 5.3 - R4000 100mhz - cc - UNROLL RISC2 PTR			101,000  808k/s
DGUX - 88100 50mhz(?) - gcc 2.6.3 - UNROLL			 81,000  648k/s
solaris 2.4 486 50mhz - gcc 2.6.3 - assembler			 65,000  522k/s
HPUX 10 - 9000/887 - k&r cc (default compiler) - UNROLL PTR	 76,000	 608k/s
solaris 2.4 486 50mhz - gcc 2.6.3 - UNROLL RISC2		 43,500  344k/s
AIX - old slow one :-) - cc -					 39,000  312k/s

Notes.
[1] For the ultra sparc, SunC 4.0 
    cc -xtarget=ultra -xarch=v8plus -Xa -xO5, running 'des_opts'
    gives a speed of 344,000 des/s while 'speed' gives 459,000 des/s.
    I'll record the higher since it is coming from the library but it
    is all rather weird.
[2] Similar to the ultra sparc ([1]), 181,000 for 'des_opts' vs 175,000.
[3] I was unable to get access to this machine when it was not heavily loaded.
    As such, my timing program was never able to get more that %30 of the CPU.
    This would cause the program to give much lower speed numbers because
    it would be 'fighting' to stay in the cache with the other CPU burning
    processes.