view aes.c @ 728:1fa3820b1a84 libavutil

ARM asm for AV_RN*() ARMv6 and later support unaligned loads and stores for single word/halfword but not double/multiple. GCC is ignorant of this and will always use bytewise accesses for unaligned data. Casting to an int32_t pointer is dangerous since a load/store double or multiple instruction might be used (this happens with some code in FFmpeg). Implementing the AV_[RW]* macros with inline asm using only supported instructions gives fast and safe unaligned accesses. ARM RVCT does the right thing with generic code. This gives an overall speedup of up to 10%.

/*
 * copyright (c) 2007 Michael Niedermayer <[e-mail]>
 *
 * some optimization ideas from aes128.c by Reimar Doeffinger
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "common.h"
#include "aes.h"

typedef struct AVAES{
    // Note: round_key[16] is accessed in the init code, but this only
    // overwrites state, which does not matter (see also r7471).
    uint8_t round_key[15][4][4];
    uint8_t state[2][4][4];
    int rounds;
}AVAES;

const int av_aes_size= sizeof(AVAES);

static const uint8_t rcon[10] = {
  0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1b, 0x36
};

static uint8_t     sbox[256];
static uint8_t inv_sbox[256];
#if CONFIG_SMALL
static uint32_t enc_multbl[1][256];
static uint32_t dec_multbl[1][256];
#else
static uint32_t enc_multbl[4][256];
static uint32_t dec_multbl[4][256];
#endif

static inline void addkey(uint64_t dst[2], const uint64_t src[2], const uint64_t round_key[2]){
    dst[0] = src[0] ^ round_key[0];
    dst[1] = src[1] ^ round_key[1];
}

static void subshift(uint8_t s0[2][16], int s, const uint8_t *box){
    uint8_t (*s1)[16]= s0[0] - s;
    uint8_t (*s3)[16]= s0[0] + s;
    s0[0][0]=box[s0[1][ 0]]; s0[0][ 4]=box[s0[1][ 4]]; s0[0][ 8]=box[s0[1][ 8]]; s0[0][12]=box[s0[1][12]];
    s1[0][3]=box[s1[1][ 7]]; s1[0][ 7]=box[s1[1][11]]; s1[0][11]=box[s1[1][15]]; s1[0][15]=box[s1[1][ 3]];
    s0[0][2]=box[s0[1][10]]; s0[0][10]=box[s0[1][ 2]]; s0[0][ 6]=box[s0[1][14]]; s0[0][14]=box[s0[1][ 6]];
    s3[0][1]=box[s3[1][13]]; s3[0][13]=box[s3[1][ 9]]; s3[0][ 9]=box[s3[1][ 5]]; s3[0][ 5]=box[s3[1][ 1]];
}

static inline int mix_core(uint32_t multbl[4][256], int a, int b, int c, int d){
#if CONFIG_SMALL
#define ROT(x,s) ((x<<s)|(x>>(32-s)))
    return multbl[0][a] ^ ROT(multbl[0][b], 8) ^ ROT(multbl[0][c], 16) ^ ROT(multbl[0][d], 24);
#else
    return multbl[0][a] ^ multbl[1][b] ^ multbl[2][c] ^ multbl[3][d];
#endif
}

static inline void mix(uint8_t state[2][4][4], uint32_t multbl[4][256], int s1, int s3){
    ((uint32_t *)(state))[0] = mix_core(multbl, state[1][0][0], state[1][s1  ][1], state[1][2][2], state[1][s3  ][3]);
    ((uint32_t *)(state))[1] = mix_core(multbl, state[1][1][0], state[1][s3-1][1], state[1][3][2], state[1][s1-1][3]);
    ((uint32_t *)(state))[2] = mix_core(multbl, state[1][2][0], state[1][s3  ][1], state[1][0][2], state[1][s1  ][3]);
    ((uint32_t *)(state))[3] = mix_core(multbl, state[1][3][0], state[1][s1-1][1], state[1][1][2], state[1][s3-1][3]);
}

static inline void crypt(AVAES *a, int s, const uint8_t *sbox, const uint32_t *multbl){
    int r;

    for(r=a->rounds-1; r>0; r--){
        mix(a->state, multbl, 3-s, 1+s);
        addkey(a->state[1], a->state[0], a->round_key[r]);
    }
    subshift(a->state[0][0], s, sbox);
}

void av_aes_crypt(AVAES *a, uint8_t *dst, const uint8_t *src, int count, uint8_t *iv, int decrypt){
    while(count--){
        addkey(a->state[1], src, a->round_key[a->rounds]);
        if(decrypt) {
            crypt(a, 0, inv_sbox, dec_multbl);
            if(iv){
                addkey(a->state[0], a->state[0], iv);
                memcpy(iv, src, 16);
            }
            addkey(dst, a->state[0], a->round_key[0]);
        }else{
            if(iv) addkey(a->state[1], a->state[1], iv);
            crypt(a, 2,     sbox, enc_multbl);
            addkey(dst, a->state[0], a->round_key[0]);
            if(iv) memcpy(iv, dst, 16);
        }
        src+=16;
        dst+=16;
    }
}

static void init_multbl2(uint8_t tbl[1024], const int c[4], const uint8_t *log8, const uint8_t *alog8, const uint8_t *sbox){
    int i, j;
    for(i=0; i<1024; i++){
        int x= sbox[i>>2];
        if(x) tbl[i]= alog8[ log8[x] + log8[c[i&3]] ];
    }
#if !CONFIG_SMALL
    for(j=256; j<1024; j++)
        for(i=0; i<4; i++)
            tbl[4*j+i]= tbl[4*j + ((i-1)&3) - 1024];
#endif
}

// this is based on the reference AES code by Paulo Barreto and Vincent Rijmen
int av_aes_init(AVAES *a, const uint8_t *key, int key_bits, int decrypt) {
    int i, j, t, rconpointer = 0;
    uint8_t tk[8][4];
    int KC= key_bits>>5;
    int rounds= KC + 6;
    uint8_t  log8[256];
    uint8_t alog8[512];

    if(!enc_multbl[0][sizeof(enc_multbl)/sizeof(enc_multbl[0][0])-1]){
        j=1;
        for(i=0; i<255; i++){
            alog8[i]=
            alog8[i+255]= j;
            log8[j]= i;
            j^= j+j;
            if(j>255) j^= 0x11B;
        }
        for(i=0; i<256; i++){
            j= i ? alog8[255-log8[i]] : 0;
            j ^= (j<<1) ^ (j<<2) ^ (j<<3) ^ (j<<4);
            j = (j ^ (j>>8) ^ 99) & 255;
            inv_sbox[j]= i;
            sbox    [i]= j;
        }
        init_multbl2(dec_multbl[0], (const int[4]){0xe, 0x9, 0xd, 0xb}, log8, alog8, inv_sbox);
        init_multbl2(enc_multbl[0], (const int[4]){0x2, 0x1, 0x1, 0x3}, log8, alog8, sbox);
    }

    if(key_bits!=128 && key_bits!=192 && key_bits!=256)
        return -1;

    a->rounds= rounds;

    memcpy(tk, key, KC*4);

    for(t= 0; t < (rounds+1)*16;) {
        memcpy(a->round_key[0][0]+t, tk, KC*4);
        t+= KC*4;

        for(i = 0; i < 4; i++)
            tk[0][i] ^= sbox[tk[KC-1][(i+1)&3]];
        tk[0][0] ^= rcon[rconpointer++];

        for(j = 1; j < KC; j++){
            if(KC != 8 || j != KC>>1)
                for(i = 0; i < 4; i++) tk[j][i] ^=      tk[j-1][i];
            else
                for(i = 0; i < 4; i++) tk[j][i] ^= sbox[tk[j-1][i]];
        }
    }

    if(decrypt){
        for(i=1; i<rounds; i++){
            uint8_t tmp[3][16];
            memcpy(tmp[2], a->round_key[i][0], 16);
            subshift(tmp[1], 0, sbox);
            mix(tmp, dec_multbl, 1, 3);
            memcpy(a->round_key[i][0], tmp[0], 16);
        }
    }else{
        for(i=0; i<(rounds+1)>>1; i++){
            for(j=0; j<16; j++)
                FFSWAP(int, a->round_key[i][0][j], a->round_key[rounds-i][0][j]);
        }
    }

    return 0;
}

#ifdef TEST
#include "lfg.h"
#include "log.h"

int main(void){
    int i,j;
    AVAES ae, ad, b;
    uint8_t rkey[2][16]= {
        {0},
        {0x10, 0xa5, 0x88, 0x69, 0xd7, 0x4b, 0xe5, 0xa3, 0x74, 0xcf, 0x86, 0x7c, 0xfb, 0x47, 0x38, 0x59}};
    uint8_t pt[16], rpt[2][16]= {
        {0x6a, 0x84, 0x86, 0x7c, 0xd7, 0x7e, 0x12, 0xad, 0x07, 0xea, 0x1b, 0xe8, 0x95, 0xc5, 0x3f, 0xa3},
        {0}};
    uint8_t rct[2][16]= {
        {0x73, 0x22, 0x81, 0xc0, 0xa0, 0xaa, 0xb8, 0xf7, 0xa5, 0x4a, 0x0c, 0x67, 0xa0, 0xc4, 0x5e, 0xcf},
        {0x6d, 0x25, 0x1e, 0x69, 0x44, 0xb0, 0x51, 0xe0, 0x4e, 0xaa, 0x6f, 0xb4, 0xdb, 0xf7, 0x84, 0x65}};
    uint8_t temp[16];
    AVLFG prng;

    av_aes_init(&ae, "PI=3.141592654..", 128, 0);
    av_aes_init(&ad, "PI=3.141592654..", 128, 1);
    av_log_set_level(AV_LOG_DEBUG);
    av_lfg_init(&prng, 1);

    for(i=0; i<2; i++){
        av_aes_init(&b, rkey[i], 128, 1);
        av_aes_crypt(&b, temp, rct[i], 1, NULL, 1);
        for(j=0; j<16; j++)
            if(rpt[i][j] != temp[j])
                av_log(NULL, AV_LOG_ERROR, "%d %02X %02X\n", j, rpt[i][j], temp[j]);
    }

    for(i=0; i<10000; i++){
        for(j=0; j<16; j++){
            pt[j] = av_lfg_get(&prng);
        }
{START_TIMER
        av_aes_crypt(&ae, temp, pt, 1, NULL, 0);
        if(!(i&(i-1)))
            av_log(NULL, AV_LOG_ERROR, "%02X %02X %02X %02X\n", temp[0], temp[5], temp[10], temp[15]);
        av_aes_crypt(&ad, temp, temp, 1, NULL, 1);
STOP_TIMER("aes")}
        for(j=0; j<16; j++){
            if(pt[j] != temp[j]){
                av_log(NULL, AV_LOG_ERROR, "%d %d %02X %02X\n", i,j, pt[j], temp[j]);
            }
        }
    }
    return 0;
}
#endif