OLD | NEW |
(Empty) | |
| 1 /* This Source Code Form is subject to the terms of the Mozilla Public |
| 2 * License, v. 2.0. If a copy of the MPL was not distributed with this |
| 3 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ |
| 4 |
| 5 /* This implementation of poly1305 is by Andrew Moon |
| 6 * (https://github.com/floodyberry/poly1305-donna) and released as public |
| 7 * domain. It implements SIMD vectorization based on the algorithm described in |
| 8 * http://cr.yp.to/papers.html#neoncrypto. Unrolled to 2 powers, i.e. 64 byte |
| 9 * block size. */ |
| 10 |
| 11 #include <emmintrin.h> |
| 12 #include <stdint.h> |
| 13 |
| 14 #include "poly1305.h" |
| 15 |
| 16 #define ALIGN(x) __attribute__((aligned(x))) |
| 17 #define INLINE inline |
| 18 #define U8TO64_LE(m) (*(uint64_t*)(m)) |
| 19 #define U8TO32_LE(m) (*(uint32_t*)(m)) |
| 20 #define U64TO8_LE(m,v) (*(uint64_t*)(m)) = v |
| 21 |
| 22 typedef __m128i xmmi; |
| 23 typedef unsigned __int128 uint128_t; |
| 24 |
| 25 static const uint32_t ALIGN(16) poly1305_x64_sse2_message_mask[4] = {(1 << 26) -
1, 0, (1 << 26) - 1, 0}; |
| 26 static const uint32_t ALIGN(16) poly1305_x64_sse2_5[4] = {5, 0, 5, 0}; |
| 27 static const uint32_t ALIGN(16) poly1305_x64_sse2_1shl128[4] = {(1 << 24), 0, (1
<< 24), 0}; |
| 28 |
| 29 static uint128_t INLINE |
| 30 add128(uint128_t a, uint128_t b) { |
| 31 return a + b; |
| 32 } |
| 33 |
| 34 static uint128_t INLINE |
| 35 add128_64(uint128_t a, uint64_t b) { |
| 36 return a + b; |
| 37 } |
| 38 |
| 39 static uint128_t INLINE |
| 40 mul64x64_128(uint64_t a, uint64_t b) { |
| 41 return (uint128_t)a * b; |
| 42 } |
| 43 |
| 44 static uint64_t INLINE |
| 45 lo128(uint128_t a) { |
| 46 return (uint64_t)a; |
| 47 } |
| 48 |
| 49 static uint64_t INLINE |
| 50 shr128(uint128_t v, const int shift) { |
| 51 return (uint64_t)(v >> shift); |
| 52 } |
| 53 |
| 54 static uint64_t INLINE |
| 55 shr128_pair(uint64_t hi, uint64_t lo, const int shift) { |
| 56 return (uint64_t)((((uint128_t)hi << 64) | lo) >> shift); |
| 57 } |
| 58 |
| 59 typedef struct poly1305_power_t { |
| 60 union { |
| 61 xmmi v; |
| 62 uint64_t u[2]; |
| 63 uint32_t d[4]; |
| 64 } R20,R21,R22,R23,R24,S21,S22,S23,S24; |
| 65 } poly1305_power; |
| 66 |
| 67 typedef struct poly1305_state_internal_t { |
| 68 poly1305_power P[2]; /* 288 bytes, top 32 bit halves unused = 144 by
tes of free storage */ |
| 69 union { |
| 70 xmmi H[5]; /* 80 bytes */ |
| 71 uint64_t HH[10]; |
| 72 }; |
| 73 /* uint64_t r0,r1,r2; [24 bytes] */ |
| 74 /* uint64_t pad0,pad1; [16 bytes] */ |
| 75 uint64_t started; /* 8 bytes */ |
| 76 uint64_t leftover; /* 8 bytes */ |
| 77 uint8_t buffer[64]; /* 64 bytes */ |
| 78 } poly1305_state_internal; /* 448 bytes total + 63 bytes for alignment = 511 b
ytes raw */ |
| 79 |
| 80 static poly1305_state_internal INLINE |
| 81 *poly1305_aligned_state(poly1305_state *state) { |
| 82 return (poly1305_state_internal *)(((uint64_t)state + 63) & ~63); |
| 83 } |
| 84 |
| 85 /* copy 0-63 bytes */ |
| 86 static void INLINE |
| 87 poly1305_block_copy(uint8_t *dst, const uint8_t *src, size_t bytes) { |
| 88 size_t offset = src - dst; |
| 89 if (bytes & 32) { |
| 90 _mm_storeu_si128((xmmi *)(dst + 0), _mm_loadu_si128((xmmi *)(dst
+ offset + 0))); |
| 91 _mm_storeu_si128((xmmi *)(dst + 16), _mm_loadu_si128((xmmi *)(ds
t + offset + 16))); |
| 92 dst += 32; |
| 93 } |
| 94 if (bytes & 16) { _mm_storeu_si128((xmmi *)dst, _mm_loadu_si128((xmmi *)
(dst + offset))); dst += 16; } |
| 95 if (bytes & 8) { *(uint64_t *)dst = *(uint64_t *)(dst + offset); dst +=
8; } |
| 96 if (bytes & 4) { *(uint32_t *)dst = *(uint32_t *)(dst + offset); dst +=
4; } |
| 97 if (bytes & 2) { *(uint16_t *)dst = *(uint16_t *)(dst + offset); dst +=
2; } |
| 98 if (bytes & 1) { *( uint8_t *)dst = *( uint8_t *)(dst + offset);
} |
| 99 } |
| 100 |
| 101 /* zero 0-15 bytes */ |
| 102 static void INLINE |
| 103 poly1305_block_zero(uint8_t *dst, size_t bytes) { |
| 104 if (bytes & 8) { *(uint64_t *)dst = 0; dst += 8; } |
| 105 if (bytes & 4) { *(uint32_t *)dst = 0; dst += 4; } |
| 106 if (bytes & 2) { *(uint16_t *)dst = 0; dst += 2; } |
| 107 if (bytes & 1) { *( uint8_t *)dst = 0; } |
| 108 } |
| 109 |
| 110 static size_t INLINE |
| 111 poly1305_min(size_t a, size_t b) { |
| 112 return (a < b) ? a : b; |
| 113 } |
| 114 |
| 115 void |
| 116 Poly1305Init(poly1305_state *state, const unsigned char key[32]) { |
| 117 poly1305_state_internal *st = poly1305_aligned_state(state); |
| 118 poly1305_power *p; |
| 119 uint64_t r0,r1,r2; |
| 120 uint64_t t0,t1; |
| 121 |
| 122 /* clamp key */ |
| 123 t0 = U8TO64_LE(key + 0); |
| 124 t1 = U8TO64_LE(key + 8); |
| 125 r0 = t0 & 0xffc0fffffff; t0 >>= 44; t0 |= t1 << 20; |
| 126 r1 = t0 & 0xfffffc0ffff; t1 >>= 24; |
| 127 r2 = t1 & 0x00ffffffc0f; |
| 128 |
| 129 /* store r in un-used space of st->P[1] */ |
| 130 p = &st->P[1]; |
| 131 p->R20.d[1] = (uint32_t)(r0 ); |
| 132 p->R20.d[3] = (uint32_t)(r0 >> 32); |
| 133 p->R21.d[1] = (uint32_t)(r1 ); |
| 134 p->R21.d[3] = (uint32_t)(r1 >> 32); |
| 135 p->R22.d[1] = (uint32_t)(r2 ); |
| 136 p->R22.d[3] = (uint32_t)(r2 >> 32); |
| 137 |
| 138 /* store pad */ |
| 139 p->R23.d[1] = U8TO32_LE(key + 16); |
| 140 p->R23.d[3] = U8TO32_LE(key + 20); |
| 141 p->R24.d[1] = U8TO32_LE(key + 24); |
| 142 p->R24.d[3] = U8TO32_LE(key + 28); |
| 143 |
| 144 /* H = 0 */ |
| 145 st->H[0] = _mm_setzero_si128(); |
| 146 st->H[1] = _mm_setzero_si128(); |
| 147 st->H[2] = _mm_setzero_si128(); |
| 148 st->H[3] = _mm_setzero_si128(); |
| 149 st->H[4] = _mm_setzero_si128(); |
| 150 |
| 151 st->started = 0; |
| 152 st->leftover = 0; |
| 153 } |
| 154 |
| 155 static void |
| 156 poly1305_first_block(poly1305_state_internal *st, const uint8_t *m) { |
| 157 const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask
); |
| 158 const xmmi FIVE = _mm_load_si128((xmmi*)poly1305_x64_sse2_5); |
| 159 const xmmi HIBIT = _mm_load_si128((xmmi*)poly1305_x64_sse2_1shl128); |
| 160 xmmi T5,T6; |
| 161 poly1305_power *p; |
| 162 uint128_t d[3]; |
| 163 uint64_t r0,r1,r2; |
| 164 uint64_t r20,r21,r22,s22; |
| 165 uint64_t pad0,pad1; |
| 166 uint64_t c; |
| 167 uint64_t i; |
| 168 |
| 169 /* pull out stored info */ |
| 170 p = &st->P[1]; |
| 171 |
| 172 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1]; |
| 173 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1]; |
| 174 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1]; |
| 175 pad0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1]; |
| 176 pad1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1]; |
| 177 |
| 178 /* compute powers r^2,r^4 */ |
| 179 r20 = r0; |
| 180 r21 = r1; |
| 181 r22 = r2; |
| 182 for (i = 0; i < 2; i++) { |
| 183 s22 = r22 * (5 << 2); |
| 184 |
| 185 d[0] = add128(mul64x64_128(r20, r20), mul64x64_128(r21 * 2, s22)
); |
| 186 d[1] = add128(mul64x64_128(r22, s22), mul64x64_128(r20 * 2, r21)
); |
| 187 d[2] = add128(mul64x64_128(r21, r21), mul64x64_128(r22 * 2, r20)
); |
| 188 |
| 189 r20 = lo128(d[0]) & 0xfffffffffff; c
= shr128(d[0], 44); |
| 190 d[1] = add128_64(d[1], c); r21 = lo128(d[1]) & 0xfffffffffff; c
= shr128(d[1], 44); |
| 191 d[2] = add128_64(d[2], c); r22 = lo128(d[2]) & 0x3ffffffffff; c
= shr128(d[2], 42); |
| 192 r20 += c * 5; c = (r20 >> 44); r20 = r20 & 0xfffffffffff; |
| 193 r21 += c; |
| 194 |
| 195 p->R20.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)( r20
) & 0x3ffffff), _MM_SHUFFLE(1,0,1,0)); |
| 196 p->R21.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r20 >
> 26) | (r21 << 18)) & 0x3ffffff), _MM_SHUFFLE(1,0,1,0)); |
| 197 p->R22.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r21 >
> 8) ) & 0x3ffffff), _MM_SHUFFLE(1,0,1,0)); |
| 198 p->R23.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r21 >
> 34) | (r22 << 10)) & 0x3ffffff), _MM_SHUFFLE(1,0,1,0)); |
| 199 p->R24.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r22 >
> 16) ) ), _MM_SHUFFLE(1,0,1,0)); |
| 200 p->S21.v = _mm_mul_epu32(p->R21.v, FIVE); |
| 201 p->S22.v = _mm_mul_epu32(p->R22.v, FIVE); |
| 202 p->S23.v = _mm_mul_epu32(p->R23.v, FIVE); |
| 203 p->S24.v = _mm_mul_epu32(p->R24.v, FIVE); |
| 204 p--; |
| 205 } |
| 206 |
| 207 /* put saved info back */ |
| 208 p = &st->P[1]; |
| 209 p->R20.d[1] = (uint32_t)(r0 ); |
| 210 p->R20.d[3] = (uint32_t)(r0 >> 32); |
| 211 p->R21.d[1] = (uint32_t)(r1 ); |
| 212 p->R21.d[3] = (uint32_t)(r1 >> 32); |
| 213 p->R22.d[1] = (uint32_t)(r2 ); |
| 214 p->R22.d[3] = (uint32_t)(r2 >> 32); |
| 215 p->R23.d[1] = (uint32_t)(pad0 ); |
| 216 p->R23.d[3] = (uint32_t)(pad0 >> 32); |
| 217 p->R24.d[1] = (uint32_t)(pad1 ); |
| 218 p->R24.d[3] = (uint32_t)(pad1 >> 32); |
| 219 |
| 220 /* H = [Mx,My] */ |
| 221 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)), _mm_loadl_epi6
4((xmmi *)(m + 16))); |
| 222 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)), _mm_loadl_epi6
4((xmmi *)(m + 24))); |
| 223 st->H[0] = _mm_and_si128(MMASK, T5); |
| 224 st->H[1] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 225 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12)); |
| 226 st->H[2] = _mm_and_si128(MMASK, T5); |
| 227 st->H[3] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 228 st->H[4] = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT); |
| 229 } |
| 230 |
| 231 static void |
| 232 poly1305_blocks(poly1305_state_internal *st, const uint8_t *m, size_t bytes) { |
| 233 const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask
); |
| 234 const xmmi FIVE = _mm_load_si128((xmmi*)poly1305_x64_sse2_5); |
| 235 const xmmi HIBIT = _mm_load_si128((xmmi*)poly1305_x64_sse2_1shl128); |
| 236 |
| 237 poly1305_power *p; |
| 238 xmmi H0,H1,H2,H3,H4; |
| 239 xmmi T0,T1,T2,T3,T4,T5,T6; |
| 240 xmmi M0,M1,M2,M3,M4; |
| 241 xmmi C1,C2; |
| 242 |
| 243 H0 = st->H[0]; |
| 244 H1 = st->H[1]; |
| 245 H2 = st->H[2]; |
| 246 H3 = st->H[3]; |
| 247 H4 = st->H[4]; |
| 248 |
| 249 while (bytes >= 64) { |
| 250 /* H *= [r^4,r^4] */ |
| 251 p = &st->P[0]; |
| 252 T0 = _mm_mul_epu32(H0, p->R20.v); |
| 253 T1 = _mm_mul_epu32(H0, p->R21.v); |
| 254 T2 = _mm_mul_epu32(H0, p->R22.v); |
| 255 T3 = _mm_mul_epu32(H0, p->R23.v); |
| 256 T4 = _mm_mul_epu32(H0, p->R24.v); |
| 257 T5 = _mm_mul_epu32(H1, p->S24.v); T6 = _mm_mul_epu32(H1, p->R20.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 258 T5 = _mm_mul_epu32(H2, p->S23.v); T6 = _mm_mul_epu32(H2, p->S24.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 259 T5 = _mm_mul_epu32(H3, p->S22.v); T6 = _mm_mul_epu32(H3, p->S23.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 260 T5 = _mm_mul_epu32(H4, p->S21.v); T6 = _mm_mul_epu32(H4, p->S22.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 261 T5 = _mm_mul_epu32(H1, p->R21.v); T6 = _mm_mul_epu32(H1, p->R22.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 262 T5 = _mm_mul_epu32(H2, p->R20.v); T6 = _mm_mul_epu32(H2, p->R21.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 263 T5 = _mm_mul_epu32(H3, p->S24.v); T6 = _mm_mul_epu32(H3, p->R20.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 264 T5 = _mm_mul_epu32(H4, p->S23.v); T6 = _mm_mul_epu32(H4, p->S24.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 265 T5 = _mm_mul_epu32(H1, p->R23.v);
T4 = _mm_add_epi64(T4, T5); |
| 266 T5 = _mm_mul_epu32(H2, p->R22.v);
T4 = _mm_add_epi64(T4, T5); |
| 267 T5 = _mm_mul_epu32(H3, p->R21.v);
T4 = _mm_add_epi64(T4, T5); |
| 268 T5 = _mm_mul_epu32(H4, p->R20.v);
T4 = _mm_add_epi64(T4, T5); |
| 269 |
| 270 /* H += [Mx,My]*[r^2,r^2] */ |
| 271 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)), _mm_lo
adl_epi64((xmmi *)(m + 16))); |
| 272 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)), _mm_lo
adl_epi64((xmmi *)(m + 24))); |
| 273 M0 = _mm_and_si128(MMASK, T5); |
| 274 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 275 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12)
); |
| 276 M2 = _mm_and_si128(MMASK, T5); |
| 277 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 278 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT); |
| 279 |
| 280 p = &st->P[1]; |
| 281 T5 = _mm_mul_epu32(M0, p->R20.v); T6 = _mm_mul_epu32(M0, p->R21.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 282 T5 = _mm_mul_epu32(M1, p->S24.v); T6 = _mm_mul_epu32(M1, p->R20.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 283 T5 = _mm_mul_epu32(M2, p->S23.v); T6 = _mm_mul_epu32(M2, p->S24.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 284 T5 = _mm_mul_epu32(M3, p->S22.v); T6 = _mm_mul_epu32(M3, p->S23.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 285 T5 = _mm_mul_epu32(M4, p->S21.v); T6 = _mm_mul_epu32(M4, p->S22.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 286 T5 = _mm_mul_epu32(M0, p->R22.v); T6 = _mm_mul_epu32(M0, p->R23.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 287 T5 = _mm_mul_epu32(M1, p->R21.v); T6 = _mm_mul_epu32(M1, p->R22.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 288 T5 = _mm_mul_epu32(M2, p->R20.v); T6 = _mm_mul_epu32(M2, p->R21.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 289 T5 = _mm_mul_epu32(M3, p->S24.v); T6 = _mm_mul_epu32(M3, p->R20.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 290 T5 = _mm_mul_epu32(M4, p->S23.v); T6 = _mm_mul_epu32(M4, p->S24.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 291 T5 = _mm_mul_epu32(M0, p->R24.v);
T4 = _mm_add_epi64(T4, T5); |
| 292 T5 = _mm_mul_epu32(M1, p->R23.v);
T4 = _mm_add_epi64(T4, T5); |
| 293 T5 = _mm_mul_epu32(M2, p->R22.v);
T4 = _mm_add_epi64(T4, T5); |
| 294 T5 = _mm_mul_epu32(M3, p->R21.v);
T4 = _mm_add_epi64(T4, T5); |
| 295 T5 = _mm_mul_epu32(M4, p->R20.v);
T4 = _mm_add_epi64(T4, T5); |
| 296 |
| 297 /* H += [Mx,My] */ |
| 298 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 32)), _mm_l
oadl_epi64((xmmi *)(m + 48))); |
| 299 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 40)), _mm_l
oadl_epi64((xmmi *)(m + 56))); |
| 300 M0 = _mm_and_si128(MMASK, T5); |
| 301 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 302 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12)
); |
| 303 M2 = _mm_and_si128(MMASK, T5); |
| 304 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 305 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT); |
| 306 |
| 307 T0 = _mm_add_epi64(T0, M0); |
| 308 T1 = _mm_add_epi64(T1, M1); |
| 309 T2 = _mm_add_epi64(T2, M2); |
| 310 T3 = _mm_add_epi64(T3, M3); |
| 311 T4 = _mm_add_epi64(T4, M4); |
| 312 |
| 313 /* reduce */ |
| 314 C1 = _mm_srli_epi64(T0, 26); C2 = _mm_srli_epi64(T3, 26); T0 = _
mm_and_si128(T0, MMASK); T3 = _mm_and_si128(T3, MMASK); T1 = _mm_add_epi64(T1, C
1); T4 = _mm_add_epi64(T4, C2); |
| 315 C1 = _mm_srli_epi64(T1, 26); C2 = _mm_srli_epi64(T4, 26); T1 = _
mm_and_si128(T1, MMASK); T4 = _mm_and_si128(T4, MMASK); T2 = _mm_add_epi64(T2, C
1); T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE)); |
| 316 C1 = _mm_srli_epi64(T2, 26); C2 = _mm_srli_epi64(T0, 26); T2 = _
mm_and_si128(T2, MMASK); T0 = _mm_and_si128(T0, MMASK); T3 = _mm_add_epi64(T3, C
1); T1 = _mm_add_epi64(T1, C2); |
| 317 C1 = _mm_srli_epi64(T3, 26); T3 = _
mm_and_si128(T3, MMASK); T4 = _mm_add_epi64(T4, C
1); |
| 318 |
| 319 /* H = (H*[r^4,r^4] + [Mx,My]*[r^2,r^2] + [Mx,My]) */ |
| 320 H0 = T0; |
| 321 H1 = T1; |
| 322 H2 = T2; |
| 323 H3 = T3; |
| 324 H4 = T4; |
| 325 |
| 326 m += 64; |
| 327 bytes -= 64; |
| 328 } |
| 329 |
| 330 st->H[0] = H0; |
| 331 st->H[1] = H1; |
| 332 st->H[2] = H2; |
| 333 st->H[3] = H3; |
| 334 st->H[4] = H4; |
| 335 } |
| 336 |
| 337 static size_t |
| 338 poly1305_combine(poly1305_state_internal *st, const uint8_t *m, size_t bytes) { |
| 339 const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask
); |
| 340 const xmmi HIBIT = _mm_load_si128((xmmi*)poly1305_x64_sse2_1shl128); |
| 341 const xmmi FIVE = _mm_load_si128((xmmi*)poly1305_x64_sse2_5); |
| 342 |
| 343 poly1305_power *p; |
| 344 xmmi H0,H1,H2,H3,H4; |
| 345 xmmi M0,M1,M2,M3,M4; |
| 346 xmmi T0,T1,T2,T3,T4,T5,T6; |
| 347 xmmi C1,C2; |
| 348 |
| 349 uint64_t r0,r1,r2; |
| 350 uint64_t t0,t1,t2,t3,t4; |
| 351 uint64_t c; |
| 352 size_t consumed = 0; |
| 353 |
| 354 H0 = st->H[0]; |
| 355 H1 = st->H[1]; |
| 356 H2 = st->H[2]; |
| 357 H3 = st->H[3]; |
| 358 H4 = st->H[4]; |
| 359 |
| 360 /* p = [r^2,r^2] */ |
| 361 p = &st->P[1]; |
| 362 |
| 363 if (bytes >= 32) { |
| 364 /* H *= [r^2,r^2] */ |
| 365 T0 = _mm_mul_epu32(H0, p->R20.v); |
| 366 T1 = _mm_mul_epu32(H0, p->R21.v); |
| 367 T2 = _mm_mul_epu32(H0, p->R22.v); |
| 368 T3 = _mm_mul_epu32(H0, p->R23.v); |
| 369 T4 = _mm_mul_epu32(H0, p->R24.v); |
| 370 T5 = _mm_mul_epu32(H1, p->S24.v); T6 = _mm_mul_epu32(H1, p->R20.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 371 T5 = _mm_mul_epu32(H2, p->S23.v); T6 = _mm_mul_epu32(H2, p->S24.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 372 T5 = _mm_mul_epu32(H3, p->S22.v); T6 = _mm_mul_epu32(H3, p->S23.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 373 T5 = _mm_mul_epu32(H4, p->S21.v); T6 = _mm_mul_epu32(H4, p->S22.
v); T0 = _mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 374 T5 = _mm_mul_epu32(H1, p->R21.v); T6 = _mm_mul_epu32(H1, p->R22.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 375 T5 = _mm_mul_epu32(H2, p->R20.v); T6 = _mm_mul_epu32(H2, p->R21.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 376 T5 = _mm_mul_epu32(H3, p->S24.v); T6 = _mm_mul_epu32(H3, p->R20.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 377 T5 = _mm_mul_epu32(H4, p->S23.v); T6 = _mm_mul_epu32(H4, p->S24.
v); T2 = _mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 378 T5 = _mm_mul_epu32(H1, p->R23.v);
T4 = _mm_add_epi64(T4, T5); |
| 379 T5 = _mm_mul_epu32(H2, p->R22.v);
T4 = _mm_add_epi64(T4, T5); |
| 380 T5 = _mm_mul_epu32(H3, p->R21.v);
T4 = _mm_add_epi64(T4, T5); |
| 381 T5 = _mm_mul_epu32(H4, p->R20.v);
T4 = _mm_add_epi64(T4, T5); |
| 382 |
| 383 /* H += [Mx,My] */ |
| 384 T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)), _mm_lo
adl_epi64((xmmi *)(m + 16))); |
| 385 T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)), _mm_lo
adl_epi64((xmmi *)(m + 24))); |
| 386 M0 = _mm_and_si128(MMASK, T5); |
| 387 M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 388 T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12)
); |
| 389 M2 = _mm_and_si128(MMASK, T5); |
| 390 M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26)); |
| 391 M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT); |
| 392 |
| 393 T0 = _mm_add_epi64(T0, M0); |
| 394 T1 = _mm_add_epi64(T1, M1); |
| 395 T2 = _mm_add_epi64(T2, M2); |
| 396 T3 = _mm_add_epi64(T3, M3); |
| 397 T4 = _mm_add_epi64(T4, M4); |
| 398 |
| 399 /* reduce */ |
| 400 C1 = _mm_srli_epi64(T0, 26); C2 = _mm_srli_epi64(T3, 26); T0 = _
mm_and_si128(T0, MMASK); T3 = _mm_and_si128(T3, MMASK); T1 = _mm_add_epi64(T1, C
1); T4 = _mm_add_epi64(T4, C2); |
| 401 C1 = _mm_srli_epi64(T1, 26); C2 = _mm_srli_epi64(T4, 26); T1 = _
mm_and_si128(T1, MMASK); T4 = _mm_and_si128(T4, MMASK); T2 = _mm_add_epi64(T2, C
1); T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE)); |
| 402 C1 = _mm_srli_epi64(T2, 26); C2 = _mm_srli_epi64(T0, 26); T2 = _
mm_and_si128(T2, MMASK); T0 = _mm_and_si128(T0, MMASK); T3 = _mm_add_epi64(T3, C
1); T1 = _mm_add_epi64(T1, C2); |
| 403 C1 = _mm_srli_epi64(T3, 26); T3 = _
mm_and_si128(T3, MMASK); T4 = _mm_add_epi64(T4, C
1); |
| 404 |
| 405 /* H = (H*[r^2,r^2] + [Mx,My]) */ |
| 406 H0 = T0; |
| 407 H1 = T1; |
| 408 H2 = T2; |
| 409 H3 = T3; |
| 410 H4 = T4; |
| 411 |
| 412 consumed = 32; |
| 413 } |
| 414 |
| 415 /* finalize, H *= [r^2,r] */ |
| 416 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1]; |
| 417 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1]; |
| 418 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1]; |
| 419 |
| 420 p->R20.d[2] = (uint32_t)( r0 ) & 0x3ffffff; |
| 421 p->R21.d[2] = (uint32_t)((r0 >> 26) | (r1 << 18)) & 0x3ffffff; |
| 422 p->R22.d[2] = (uint32_t)((r1 >> 8) ) & 0x3ffffff; |
| 423 p->R23.d[2] = (uint32_t)((r1 >> 34) | (r2 << 10)) & 0x3ffffff; |
| 424 p->R24.d[2] = (uint32_t)((r2 >> 16) ) ; |
| 425 p->S21.d[2] = p->R21.d[2] * 5; |
| 426 p->S22.d[2] = p->R22.d[2] * 5; |
| 427 p->S23.d[2] = p->R23.d[2] * 5; |
| 428 p->S24.d[2] = p->R24.d[2] * 5; |
| 429 |
| 430 /* H *= [r^2,r] */ |
| 431 T0 = _mm_mul_epu32(H0, p->R20.v); |
| 432 T1 = _mm_mul_epu32(H0, p->R21.v); |
| 433 T2 = _mm_mul_epu32(H0, p->R22.v); |
| 434 T3 = _mm_mul_epu32(H0, p->R23.v); |
| 435 T4 = _mm_mul_epu32(H0, p->R24.v); |
| 436 T5 = _mm_mul_epu32(H1, p->S24.v); T6 = _mm_mul_epu32(H1, p->R20.v); T0 =
_mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 437 T5 = _mm_mul_epu32(H2, p->S23.v); T6 = _mm_mul_epu32(H2, p->S24.v); T0 =
_mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 438 T5 = _mm_mul_epu32(H3, p->S22.v); T6 = _mm_mul_epu32(H3, p->S23.v); T0 =
_mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 439 T5 = _mm_mul_epu32(H4, p->S21.v); T6 = _mm_mul_epu32(H4, p->S22.v); T0 =
_mm_add_epi64(T0, T5); T1 = _mm_add_epi64(T1, T6); |
| 440 T5 = _mm_mul_epu32(H1, p->R21.v); T6 = _mm_mul_epu32(H1, p->R22.v); T2 =
_mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 441 T5 = _mm_mul_epu32(H2, p->R20.v); T6 = _mm_mul_epu32(H2, p->R21.v); T2 =
_mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 442 T5 = _mm_mul_epu32(H3, p->S24.v); T6 = _mm_mul_epu32(H3, p->R20.v); T2 =
_mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 443 T5 = _mm_mul_epu32(H4, p->S23.v); T6 = _mm_mul_epu32(H4, p->S24.v); T2 =
_mm_add_epi64(T2, T5); T3 = _mm_add_epi64(T3, T6); |
| 444 T5 = _mm_mul_epu32(H1, p->R23.v); T4 =
_mm_add_epi64(T4, T5); |
| 445 T5 = _mm_mul_epu32(H2, p->R22.v); T4 =
_mm_add_epi64(T4, T5); |
| 446 T5 = _mm_mul_epu32(H3, p->R21.v); T4 =
_mm_add_epi64(T4, T5); |
| 447 T5 = _mm_mul_epu32(H4, p->R20.v); T4 =
_mm_add_epi64(T4, T5); |
| 448 |
| 449 C1 = _mm_srli_epi64(T0, 26); C2 = _mm_srli_epi64(T3, 26); T0 = _mm_and_s
i128(T0, MMASK); T3 = _mm_and_si128(T3, MMASK); T1 = _mm_add_epi64(T1, C1); T4 =
_mm_add_epi64(T4, C2); |
| 450 C1 = _mm_srli_epi64(T1, 26); C2 = _mm_srli_epi64(T4, 26); T1 = _mm_and_s
i128(T1, MMASK); T4 = _mm_and_si128(T4, MMASK); T2 = _mm_add_epi64(T2, C1); T0 =
_mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE)); |
| 451 C1 = _mm_srli_epi64(T2, 26); C2 = _mm_srli_epi64(T0, 26); T2 = _mm_and_s
i128(T2, MMASK); T0 = _mm_and_si128(T0, MMASK); T3 = _mm_add_epi64(T3, C1); T1 =
_mm_add_epi64(T1, C2); |
| 452 C1 = _mm_srli_epi64(T3, 26); T3 = _mm_and_s
i128(T3, MMASK); T4 = _mm_add_epi64(T4, C1); |
| 453 |
| 454 /* H = H[0]+H[1] */ |
| 455 H0 = _mm_add_epi64(T0, _mm_srli_si128(T0, 8)); |
| 456 H1 = _mm_add_epi64(T1, _mm_srli_si128(T1, 8)); |
| 457 H2 = _mm_add_epi64(T2, _mm_srli_si128(T2, 8)); |
| 458 H3 = _mm_add_epi64(T3, _mm_srli_si128(T3, 8)); |
| 459 H4 = _mm_add_epi64(T4, _mm_srli_si128(T4, 8)); |
| 460 |
| 461 t0 = _mm_cvtsi128_si32(H0) ; c = (t0 >> 26); t0 &= 0x3ffffff; |
| 462 t1 = _mm_cvtsi128_si32(H1) + c; c = (t1 >> 26); t1 &= 0x3ffffff; |
| 463 t2 = _mm_cvtsi128_si32(H2) + c; c = (t2 >> 26); t2 &= 0x3ffffff; |
| 464 t3 = _mm_cvtsi128_si32(H3) + c; c = (t3 >> 26); t3 &= 0x3ffffff; |
| 465 t4 = _mm_cvtsi128_si32(H4) + c; c = (t4 >> 26); t4 &= 0x3ffffff; |
| 466 t0 = t0 + (c * 5); c = (t0 >> 26); t0 &= 0x3ffffff; |
| 467 t1 = t1 + c; |
| 468 |
| 469 st->HH[0] = ((t0 ) | (t1 << 26) ) & 0xfffffffffffull; |
| 470 st->HH[1] = ((t1 >> 18) | (t2 << 8) | (t3 << 34)) & 0xfffffffffffull; |
| 471 st->HH[2] = ((t3 >> 10) | (t4 << 16) ) & 0x3ffffffffffull; |
| 472 |
| 473 return consumed; |
| 474 } |
| 475 |
| 476 void |
| 477 Poly1305Update(poly1305_state *state, const unsigned char *m, size_t bytes) { |
| 478 poly1305_state_internal *st = poly1305_aligned_state(state); |
| 479 size_t want; |
| 480 |
| 481 /* need at least 32 initial bytes to start the accelerated branch */ |
| 482 if (!st->started) { |
| 483 if ((st->leftover == 0) && (bytes > 32)) { |
| 484 poly1305_first_block(st, m); |
| 485 m += 32; |
| 486 bytes -= 32; |
| 487 } else { |
| 488 want = poly1305_min(32 - st->leftover, bytes); |
| 489 poly1305_block_copy(st->buffer + st->leftover, m, want); |
| 490 bytes -= want; |
| 491 m += want; |
| 492 st->leftover += want; |
| 493 if ((st->leftover < 32) || (bytes == 0)) |
| 494 return; |
| 495 poly1305_first_block(st, st->buffer); |
| 496 st->leftover = 0; |
| 497 } |
| 498 st->started = 1; |
| 499 } |
| 500 |
| 501 /* handle leftover */ |
| 502 if (st->leftover) { |
| 503 want = poly1305_min(64 - st->leftover, bytes); |
| 504 poly1305_block_copy(st->buffer + st->leftover, m, want); |
| 505 bytes -= want; |
| 506 m += want; |
| 507 st->leftover += want; |
| 508 if (st->leftover < 64) |
| 509 return; |
| 510 poly1305_blocks(st, st->buffer, 64); |
| 511 st->leftover = 0; |
| 512 } |
| 513 |
| 514 /* process 64 byte blocks */ |
| 515 if (bytes >= 64) { |
| 516 want = (bytes & ~63); |
| 517 poly1305_blocks(st, m, want); |
| 518 m += want; |
| 519 bytes -= want; |
| 520 } |
| 521 |
| 522 if (bytes) { |
| 523 poly1305_block_copy(st->buffer + st->leftover, m, bytes); |
| 524 st->leftover += bytes; |
| 525 } |
| 526 } |
| 527 |
| 528 void |
| 529 Poly1305Finish(poly1305_state *state, unsigned char mac[16]) { |
| 530 poly1305_state_internal *st = poly1305_aligned_state(state); |
| 531 size_t leftover = st->leftover; |
| 532 uint8_t *m = st->buffer; |
| 533 uint128_t d[3]; |
| 534 uint64_t h0,h1,h2; |
| 535 uint64_t t0,t1; |
| 536 uint64_t g0,g1,g2,c,nc; |
| 537 uint64_t r0,r1,r2,s1,s2; |
| 538 poly1305_power *p; |
| 539 |
| 540 if (st->started) { |
| 541 size_t consumed = poly1305_combine(st, m, leftover); |
| 542 leftover -= consumed; |
| 543 m += consumed; |
| 544 } |
| 545 |
| 546 /* st->HH will either be 0 or have the combined result */ |
| 547 h0 = st->HH[0]; |
| 548 h1 = st->HH[1]; |
| 549 h2 = st->HH[2]; |
| 550 |
| 551 p = &st->P[1]; |
| 552 r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1]; |
| 553 r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1]; |
| 554 r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1]; |
| 555 s1 = r1 * (5 << 2); |
| 556 s2 = r2 * (5 << 2); |
| 557 |
| 558 if (leftover < 16) |
| 559 goto poly1305_donna_atmost15bytes; |
| 560 |
| 561 poly1305_donna_atleast16bytes: |
| 562 t0 = U8TO64_LE(m + 0); |
| 563 t1 = U8TO64_LE(m + 8); |
| 564 h0 += t0 & 0xfffffffffff; |
| 565 t0 = shr128_pair(t1, t0, 44); |
| 566 h1 += t0 & 0xfffffffffff; |
| 567 h2 += (t1 >> 24) | ((uint64_t)1 << 40); |
| 568 |
| 569 poly1305_donna_mul: |
| 570 d[0] = add128(add128(mul64x64_128(h0, r0), mul64x64_128(h1, s2)), mul64x
64_128(h2, s1)); |
| 571 d[1] = add128(add128(mul64x64_128(h0, r1), mul64x64_128(h1, r0)), mul64x
64_128(h2, s2)); |
| 572 d[2] = add128(add128(mul64x64_128(h0, r2), mul64x64_128(h1, r1)), mul64x
64_128(h2, r0)); |
| 573 h0 = lo128(d[0]) & 0xfffffffffff; c = shr128(
d[0], 44); |
| 574 d[1] = add128_64(d[1], c); h1 = lo128(d[1]) & 0xfffffffffff; c = shr128(
d[1], 44); |
| 575 d[2] = add128_64(d[2], c); h2 = lo128(d[2]) & 0x3ffffffffff; c = shr128(
d[2], 42); |
| 576 h0 += c * 5; |
| 577 |
| 578 m += 16; |
| 579 leftover -= 16; |
| 580 if (leftover >= 16) goto poly1305_donna_atleast16bytes; |
| 581 |
| 582 /* final bytes */ |
| 583 poly1305_donna_atmost15bytes: |
| 584 if (!leftover) goto poly1305_donna_finish; |
| 585 |
| 586 m[leftover++] = 1; |
| 587 poly1305_block_zero(m + leftover, 16 - leftover); |
| 588 leftover = 16; |
| 589 |
| 590 t0 = U8TO64_LE(m+0); |
| 591 t1 = U8TO64_LE(m+8); |
| 592 h0 += t0 & 0xfffffffffff; t0 = shr128_pair(t1, t0, 44); |
| 593 h1 += t0 & 0xfffffffffff; |
| 594 h2 += (t1 >> 24); |
| 595 |
| 596 goto poly1305_donna_mul; |
| 597 |
| 598 poly1305_donna_finish: |
| 599 c = (h0 >> 44); h0 &= 0xfffffffffff; |
| 600 h1 += c; c = (h1 >> 44); h1 &= 0xfffffffffff; |
| 601 h2 += c; c = (h2 >> 42); h2 &= 0x3ffffffffff; |
| 602 h0 += c * 5; |
| 603 |
| 604 g0 = h0 + 5; c = (g0 >> 44); g0 &= 0xfffffffffff; |
| 605 g1 = h1 + c; c = (g1 >> 44); g1 &= 0xfffffffffff; |
| 606 g2 = h2 + c - ((uint64_t)1 << 42); |
| 607 |
| 608 c = (g2 >> 63) - 1; |
| 609 nc = ~c; |
| 610 h0 = (h0 & nc) | (g0 & c); |
| 611 h1 = (h1 & nc) | (g1 & c); |
| 612 h2 = (h2 & nc) | (g2 & c); |
| 613 |
| 614 /* pad */ |
| 615 t0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1]; |
| 616 t1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1]; |
| 617 h0 += (t0 & 0xfffffffffff) ; c = (h0 >> 44); h0 &= 0xfffffffffff; t0
= shr128_pair(t1, t0, 44); |
| 618 h1 += (t0 & 0xfffffffffff) + c; c = (h1 >> 44); h1 &= 0xfffffffffff; t1
= (t1 >> 24); |
| 619 h2 += (t1 ) + c; |
| 620 |
| 621 U64TO8_LE(mac + 0, ((h0 ) | (h1 << 44))); |
| 622 U64TO8_LE(mac + 8, ((h1 >> 20) | (h2 << 24))); |
| 623 } |
OLD | NEW |