DES

Data Encryption Standard

My thesis work involved the optimization of DES in assembler for the Intel Pentium processor. This resulted in speed records for both the Pentium and other PC processors, as shown in the table. The Pentium III version is even faster (~9%) than May, Penna and Clark's bitslice implementation on the same processor.

Processor	Key setup (enc&dec)	ECB		CBC
Processor	Key setup (enc&dec)	Encrypt	Decrypt	Encrypt	Decrypt
Pentium	576	315	315	316	318
Pentium III	1124	319
Pentium IV	636	445
Athlon	496	240

Notes

This code was specifically optimized for the original Pentium. Only the encryption round function was then modified to improve performance on Pentium III and Athlon.

In March 2005 I made a DES implementation in C, and adapted it to the Linux kernel. It is available under the GNU General Public License. The table below provides some performance data. For practical reasons it's limited to the cases where the optimization target for gcc and the actual processor are the same.

Processor	Clock speed (MHz)	Approximate speed relative to old kernel implementation
Processor	Clock speed (MHz)	des_setkey	des_encrypt	des3_ede_setkey	des3_ede_encrypt
Pentium	120	6.8	0.82	7.2	0.86
Pentium 3	1266	5.6	1.19	5.8	1.34
Pentium M	1300	5.7	1.15	6.0	1.31
Pentium 4	2266	5.8	1.24	6.0	1.40
Pentium 4E	3000	5.4	1.27	5.5	1.48
StrongARM 1110	206	4.3	1.03	4.4	1.14
Athlon XP	2000	7.8	1.44	8.1	1.61
Athlon 64	2000	7.8	1.34	8.3	1.49