aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorDenys Vlasenko <vda.linux@googlemail.com>2007-11-06 09:15:19 -0500
committerHerbert Xu <herbert@gondor.apana.org.au>2008-01-10 16:16:08 -0500
commitd3e7480572bf882dee5baa2891bccbfa3db0b1a1 (patch)
tree40281dcc543e1b76b835b4f63a6feadcc9720166
parent1ce73e8d6d95ceb860184c34fa1a91a82e51cbb3 (diff)
[CRYPTO] camellia: De-unrolling
Move huge unrolled pieces of code (3 screenfuls) at the end of 128/256 key setup routines into common camellia_setup_tail(), convert it to loop there. Loop is still unrolled six times, so performance hit is very small, code size win is big. Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com> Acked-by: Noriaki TAKAMIYA <takamiya@po.ntts.co.jp> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
-rw-r--r--crypto/camellia.c107
1 files changed, 23 insertions, 84 deletions
diff --git a/crypto/camellia.c b/crypto/camellia.c
index 16529dddee72..2e129ab1a6a2 100644
--- a/crypto/camellia.c
+++ b/crypto/camellia.c
@@ -424,6 +424,27 @@ static const u32 camellia_sp4404[256] = {
424#define SUBKEY_L(INDEX) (subkey[(INDEX)*2]) 424#define SUBKEY_L(INDEX) (subkey[(INDEX)*2])
425#define SUBKEY_R(INDEX) (subkey[(INDEX)*2 + 1]) 425#define SUBKEY_R(INDEX) (subkey[(INDEX)*2 + 1])
426 426
427static void camellia_setup_tail(u32 *subkey, int max)
428{
429 u32 dw;
430 int i = 2;
431 do {
432 dw = SUBKEY_L(i + 0) ^ SUBKEY_R(i + 0); dw = ROL8(dw);/* round 1 */
433 SUBKEY_R(i + 0) = SUBKEY_L(i + 0) ^ dw; SUBKEY_L(i + 0) = dw;
434 dw = SUBKEY_L(i + 1) ^ SUBKEY_R(i + 1); dw = ROL8(dw);/* round 2 */
435 SUBKEY_R(i + 1) = SUBKEY_L(i + 1) ^ dw; SUBKEY_L(i + 1) = dw;
436 dw = SUBKEY_L(i + 2) ^ SUBKEY_R(i + 2); dw = ROL8(dw);/* round 3 */
437 SUBKEY_R(i + 2) = SUBKEY_L(i + 2) ^ dw; SUBKEY_L(i + 2) = dw;
438 dw = SUBKEY_L(i + 3) ^ SUBKEY_R(i + 3); dw = ROL8(dw);/* round 4 */
439 SUBKEY_R(i + 3) = SUBKEY_L(i + 3) ^ dw; SUBKEY_L(i + 3) = dw;
440 dw = SUBKEY_L(i + 4) ^ SUBKEY_R(i + 4); dw = ROL8(dw);/* round 5 */
441 SUBKEY_R(i + 4) = SUBKEY_L(i + 4) ^ dw; SUBKEY_L(i + 4) = dw;
442 dw = SUBKEY_L(i + 5) ^ SUBKEY_R(i + 5); dw = ROL8(dw);/* round 6 */
443 SUBKEY_R(i + 5) = SUBKEY_L(i + 5) ^ dw; SUBKEY_L(i + 5) = dw;
444 i += 8;
445 } while (i < max);
446}
447
427static void camellia_setup128(const unsigned char *key, u32 *subkey) 448static void camellia_setup128(const unsigned char *key, u32 *subkey)
428{ 449{
429 u32 kll, klr, krl, krr; 450 u32 kll, klr, krl, krr;
@@ -650,42 +671,7 @@ static void camellia_setup128(const unsigned char *key, u32 *subkey)
650 SUBKEY_R(24) = subR[24] ^ subR[23]; 671 SUBKEY_R(24) = subR[24] ^ subR[23];
651 672
652 /* apply the inverse of the last half of P-function */ 673 /* apply the inverse of the last half of P-function */
653 dw = SUBKEY_L(2) ^ SUBKEY_R(2); dw = ROL8(dw);/* round 1 */ 674 camellia_setup_tail(subkey, 24);
654 SUBKEY_R(2) = SUBKEY_L(2) ^ dw; SUBKEY_L(2) = dw;
655 dw = SUBKEY_L(3) ^ SUBKEY_R(3); dw = ROL8(dw);/* round 2 */
656 SUBKEY_R(3) = SUBKEY_L(3) ^ dw; SUBKEY_L(3) = dw;
657 dw = SUBKEY_L(4) ^ SUBKEY_R(4); dw = ROL8(dw);/* round 3 */
658 SUBKEY_R(4) = SUBKEY_L(4) ^ dw; SUBKEY_L(4) = dw;
659 dw = SUBKEY_L(5) ^ SUBKEY_R(5); dw = ROL8(dw);/* round 4 */
660 SUBKEY_R(5) = SUBKEY_L(5) ^ dw; SUBKEY_L(5) = dw;
661 dw = SUBKEY_L(6) ^ SUBKEY_R(6); dw = ROL8(dw);/* round 5 */
662 SUBKEY_R(6) = SUBKEY_L(6) ^ dw; SUBKEY_L(6) = dw;
663 dw = SUBKEY_L(7) ^ SUBKEY_R(7); dw = ROL8(dw);/* round 6 */
664 SUBKEY_R(7) = SUBKEY_L(7) ^ dw; SUBKEY_L(7) = dw;
665 dw = SUBKEY_L(10) ^ SUBKEY_R(10); dw = ROL8(dw);/* round 7 */
666 SUBKEY_R(10) = SUBKEY_L(10) ^ dw; SUBKEY_L(10) = dw;
667 dw = SUBKEY_L(11) ^ SUBKEY_R(11); dw = ROL8(dw);/* round 8 */
668 SUBKEY_R(11) = SUBKEY_L(11) ^ dw; SUBKEY_L(11) = dw;
669 dw = SUBKEY_L(12) ^ SUBKEY_R(12); dw = ROL8(dw);/* round 9 */
670 SUBKEY_R(12) = SUBKEY_L(12) ^ dw; SUBKEY_L(12) = dw;
671 dw = SUBKEY_L(13) ^ SUBKEY_R(13); dw = ROL8(dw);/* round 10 */
672 SUBKEY_R(13) = SUBKEY_L(13) ^ dw; SUBKEY_L(13) = dw;
673 dw = SUBKEY_L(14) ^ SUBKEY_R(14); dw = ROL8(dw);/* round 11 */
674 SUBKEY_R(14) = SUBKEY_L(14) ^ dw; SUBKEY_L(14) = dw;
675 dw = SUBKEY_L(15) ^ SUBKEY_R(15); dw = ROL8(dw);/* round 12 */
676 SUBKEY_R(15) = SUBKEY_L(15) ^ dw; SUBKEY_L(15) = dw;
677 dw = SUBKEY_L(18) ^ SUBKEY_R(18); dw = ROL8(dw);/* round 13 */
678 SUBKEY_R(18) = SUBKEY_L(18) ^ dw; SUBKEY_L(18) = dw;
679 dw = SUBKEY_L(19) ^ SUBKEY_R(19); dw = ROL8(dw);/* round 14 */
680 SUBKEY_R(19) = SUBKEY_L(19) ^ dw; SUBKEY_L(19) = dw;
681 dw = SUBKEY_L(20) ^ SUBKEY_R(20); dw = ROL8(dw);/* round 15 */
682 SUBKEY_R(20) = SUBKEY_L(20) ^ dw; SUBKEY_L(20) = dw;
683 dw = SUBKEY_L(21) ^ SUBKEY_R(21); dw = ROL8(dw);/* round 16 */
684 SUBKEY_R(21) = SUBKEY_L(21) ^ dw; SUBKEY_L(21) = dw;
685 dw = SUBKEY_L(22) ^ SUBKEY_R(22); dw = ROL8(dw);/* round 17 */
686 SUBKEY_R(22) = SUBKEY_L(22) ^ dw; SUBKEY_L(22) = dw;
687 dw = SUBKEY_L(23) ^ SUBKEY_R(23); dw = ROL8(dw);/* round 18 */
688 SUBKEY_R(23) = SUBKEY_L(23) ^ dw; SUBKEY_L(23) = dw;
689} 675}
690 676
691static void camellia_setup256(const unsigned char *key, u32 *subkey) 677static void camellia_setup256(const unsigned char *key, u32 *subkey)
@@ -995,54 +981,7 @@ static void camellia_setup256(const unsigned char *key, u32 *subkey)
995 SUBKEY_R(32) = subR[32] ^ subR[31]; 981 SUBKEY_R(32) = subR[32] ^ subR[31];
996 982
997 /* apply the inverse of the last half of P-function */ 983 /* apply the inverse of the last half of P-function */
998 dw = SUBKEY_L(2) ^ SUBKEY_R(2); dw = ROL8(dw);/* round 1 */ 984 camellia_setup_tail(subkey, 32);
999 SUBKEY_R(2) = SUBKEY_L(2) ^ dw; SUBKEY_L(2) = dw;
1000 dw = SUBKEY_L(3) ^ SUBKEY_R(3); dw = ROL8(dw);/* round 2 */
1001 SUBKEY_R(3) = SUBKEY_L(3) ^ dw; SUBKEY_L(3) = dw;
1002 dw = SUBKEY_L(4) ^ SUBKEY_R(4); dw = ROL8(dw);/* round 3 */
1003 SUBKEY_R(4) = SUBKEY_L(4) ^ dw; SUBKEY_L(4) = dw;
1004 dw = SUBKEY_L(5) ^ SUBKEY_R(5); dw = ROL8(dw);/* round 4 */
1005 SUBKEY_R(5) = SUBKEY_L(5) ^ dw; SUBKEY_L(5) = dw;
1006 dw = SUBKEY_L(6) ^ SUBKEY_R(6); dw = ROL8(dw);/* round 5 */
1007 SUBKEY_R(6) = SUBKEY_L(6) ^ dw; SUBKEY_L(6) = dw;
1008 dw = SUBKEY_L(7) ^ SUBKEY_R(7); dw = ROL8(dw);/* round 6 */
1009 SUBKEY_R(7) = SUBKEY_L(7) ^ dw; SUBKEY_L(7) = dw;
1010 dw = SUBKEY_L(10) ^ SUBKEY_R(10); dw = ROL8(dw);/* round 7 */
1011 SUBKEY_R(10) = SUBKEY_L(10) ^ dw; SUBKEY_L(10) = dw;
1012 dw = SUBKEY_L(11) ^ SUBKEY_R(11); dw = ROL8(dw);/* round 8 */
1013 SUBKEY_R(11) = SUBKEY_L(11) ^ dw; SUBKEY_L(11) = dw;
1014 dw = SUBKEY_L(12) ^ SUBKEY_R(12); dw = ROL8(dw);/* round 9 */
1015 SUBKEY_R(12) = SUBKEY_L(12) ^ dw; SUBKEY_L(12) = dw;
1016 dw = SUBKEY_L(13) ^ SUBKEY_R(13); dw = ROL8(dw);/* round 10 */
1017 SUBKEY_R(13) = SUBKEY_L(13) ^ dw; SUBKEY_L(13) = dw;
1018 dw = SUBKEY_L(14) ^ SUBKEY_R(14); dw = ROL8(dw);/* round 11 */
1019 SUBKEY_R(14) = SUBKEY_L(14) ^ dw; SUBKEY_L(14) = dw;
1020 dw = SUBKEY_L(15) ^ SUBKEY_R(15); dw = ROL8(dw);/* round 12 */
1021 SUBKEY_R(15) = SUBKEY_L(15) ^ dw; SUBKEY_L(15) = dw;
1022 dw = SUBKEY_L(18) ^ SUBKEY_R(18); dw = ROL8(dw);/* round 13 */
1023 SUBKEY_R(18) = SUBKEY_L(18) ^ dw; SUBKEY_L(18) = dw;
1024 dw = SUBKEY_L(19) ^ SUBKEY_R(19); dw = ROL8(dw);/* round 14 */
1025 SUBKEY_R(19) = SUBKEY_L(19) ^ dw; SUBKEY_L(19) = dw;
1026 dw = SUBKEY_L(20) ^ SUBKEY_R(20); dw = ROL8(dw);/* round 15 */
1027 SUBKEY_R(20) = SUBKEY_L(20) ^ dw; SUBKEY_L(20) = dw;
1028 dw = SUBKEY_L(21) ^ SUBKEY_R(21); dw = ROL8(dw);/* round 16 */
1029 SUBKEY_R(21) = SUBKEY_L(21) ^ dw; SUBKEY_L(21) = dw;
1030 dw = SUBKEY_L(22) ^ SUBKEY_R(22); dw = ROL8(dw);/* round 17 */
1031 SUBKEY_R(22) = SUBKEY_L(22) ^ dw; SUBKEY_L(22) = dw;
1032 dw = SUBKEY_L(23) ^ SUBKEY_R(23); dw = ROL8(dw);/* round 18 */
1033 SUBKEY_R(23) = SUBKEY_L(23) ^ dw; SUBKEY_L(23) = dw;
1034 dw = SUBKEY_L(26) ^ SUBKEY_R(26); dw = ROL8(dw);/* round 19 */
1035 SUBKEY_R(26) = SUBKEY_L(26) ^ dw; SUBKEY_L(26) = dw;
1036 dw = SUBKEY_L(27) ^ SUBKEY_R(27); dw = ROL8(dw);/* round 20 */
1037 SUBKEY_R(27) = SUBKEY_L(27) ^ dw; SUBKEY_L(27) = dw;
1038 dw = SUBKEY_L(28) ^ SUBKEY_R(28); dw = ROL8(dw);/* round 21 */
1039 SUBKEY_R(28) = SUBKEY_L(28) ^ dw; SUBKEY_L(28) = dw;
1040 dw = SUBKEY_L(29) ^ SUBKEY_R(29); dw = ROL8(dw);/* round 22 */
1041 SUBKEY_R(29) = SUBKEY_L(29) ^ dw; SUBKEY_L(29) = dw;
1042 dw = SUBKEY_L(30) ^ SUBKEY_R(30); dw = ROL8(dw);/* round 23 */
1043 SUBKEY_R(30) = SUBKEY_L(30) ^ dw; SUBKEY_L(30) = dw;
1044 dw = SUBKEY_L(31) ^ SUBKEY_R(31); dw = ROL8(dw);/* round 24 */
1045 SUBKEY_R(31) = SUBKEY_L(31) ^ dw; SUBKEY_L(31) = dw;
1046} 985}
1047 986
1048static void camellia_setup192(const unsigned char *key, u32 *subkey) 987static void camellia_setup192(const unsigned char *key, u32 *subkey)