emo/busybox
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

sha3: code shrink

Browse Source 
function                                             old     new   delta
KeccakF                                             1064    1053     -11

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
This commit is contained in:
Denys Vlasenko
2013-01-15 15:22:30 +01:00
parent 07a54e21dd
commit a55df27936
1 changed files with 18 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
libbb/hash_md5_sha.c
View File

@@ -988,24 +988,29 @@ static void KeccakF(uint64_t *state)
for (round = 0; round < cKeccakNumberOfRounds; ++round) { for (round = 0; round < cKeccakNumberOfRounds; ++round) {
/* Theta */ /* Theta */
{ {
uint64_t BC[5]; uint64_t BC[10];
for (x = 0; x < 5; ++x) { for (x = 0; x < 5; ++x) {
BC[x] = state[x] ^ state[5 + x] ^ state[10 + x] ^ BC[x + 5] = BC[x] = state[x]
state[15 + x] ^ state[20 + x]; ^ state[x + 5] ^ state[x + 10]
^ state[x + 15] ^ state[x + 20];
} }
/* Using 2x5 vector above eliminates the need to use
* [Mod5[x+N]] index trick below to calculate (x+N) % 5,
* and the code is a bit _smaller_.
*/
for (x = 0; x < 5; ++x) { for (x = 0; x < 5; ++x) {
uint64_t temp = BC[KeccakF_Mod5[x + 4]] ^ uint64_t temp = BC[x + 4] ^ rotl64(BC[x + 1], 1);
rotl64(BC[KeccakF_Mod5[x + 1]], 1);
if (SHA3_SMALL && !ARCH_IS_64BIT) { if (SHA3_SMALL && !ARCH_IS_64BIT) {
for (y = 0; y <= 20; y += 5) for (y = 0; y <= 20; y += 5)
state[y + x] ^= temp; state[x + y] ^= temp;
} else { } else {
/* on 64-bit arch, this is actually smaller too */ /* On 64-bit, this is also smaller,
state[0 + x] ^= temp; * not only faster, than loop */
state[5 + x] ^= temp; state[x] ^= temp;
state[10 + x] ^= temp; state[x + 5] ^= temp;
state[15 + x] ^= temp; state[x + 10] ^= temp;
state[20 + x] ^= temp; state[x + 15] ^= temp;
state[x + 20] ^= temp;
} }
} }
} }
@@ -1019,7 +1024,7 @@ static void KeccakF(uint64_t *state)
t1 = t0; t1 = t0;
} }
} else { } else {
/* Especially large benefit for 32-bit arch: /* Especially large benefit for 32-bit arch (75% faster):
* 64-bit rotations by non-constant usually are SLOW on those. * 64-bit rotations by non-constant usually are SLOW on those.
* We resort to unrolling here. * We resort to unrolling here.
* This optimizes out KeccakF_PiLane[] and KeccakF_RotationConstants[], * This optimizes out KeccakF_PiLane[] and KeccakF_RotationConstants[],