|
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892 |
- /* Copyright (c) 2014, Google Inc.
- *
- * Permission to use, copy, modify, and/or distribute this software for any
- * purpose with or without fee is hereby granted, provided that the above
- * copyright notice and this permission notice appear in all copies.
- *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
- * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
- * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
- * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
-
- /* This implementation of poly1305 is by Andrew Moon
- * (https://github.com/floodyberry/poly1305-donna) and released as public
- * domain. It implements SIMD vectorization based on the algorithm described in
- * http://cr.yp.to/papers.html#neoncrypto. Unrolled to 2 powers, i.e. 64 byte
- * block size */
-
- #include <openssl/poly1305.h>
-
-
- #if !defined(OPENSSL_WINDOWS) && defined(OPENSSL_X86_64)
-
- #include <emmintrin.h>
-
- #define ALIGN(x) __attribute__((aligned(x)))
- /* inline is not a keyword in C89. */
- #define INLINE
- #define U8TO64_LE(m) (*(uint64_t *)(m))
- #define U8TO32_LE(m) (*(uint32_t *)(m))
- #define U64TO8_LE(m, v) (*(uint64_t *)(m)) = v
-
- typedef __m128i xmmi;
- typedef unsigned __int128 uint128_t;
-
- static const uint32_t ALIGN(16) poly1305_x64_sse2_message_mask[4] = {
- (1 << 26) - 1, 0, (1 << 26) - 1, 0};
- static const uint32_t ALIGN(16) poly1305_x64_sse2_5[4] = {5, 0, 5, 0};
- static const uint32_t ALIGN(16) poly1305_x64_sse2_1shl128[4] = {(1 << 24), 0,
- (1 << 24), 0};
-
- static uint128_t INLINE add128(uint128_t a, uint128_t b) { return a + b; }
-
- static uint128_t INLINE add128_64(uint128_t a, uint64_t b) { return a + b; }
-
- static uint128_t INLINE mul64x64_128(uint64_t a, uint64_t b) {
- return (uint128_t)a * b;
- }
-
- static uint64_t INLINE lo128(uint128_t a) { return (uint64_t)a; }
-
- static uint64_t INLINE shr128(uint128_t v, const int shift) {
- return (uint64_t)(v >> shift);
- }
-
- static uint64_t INLINE shr128_pair(uint64_t hi, uint64_t lo, const int shift) {
- return (uint64_t)((((uint128_t)hi << 64) | lo) >> shift);
- }
-
- typedef struct poly1305_power_t {
- union {
- xmmi v;
- uint64_t u[2];
- uint32_t d[4];
- } R20, R21, R22, R23, R24, S21, S22, S23, S24;
- } poly1305_power;
-
- typedef struct poly1305_state_internal_t {
- poly1305_power P[2]; /* 288 bytes, top 32 bit halves unused = 144
- bytes of free storage */
- union {
- xmmi H[5]; /* 80 bytes */
- uint64_t HH[10];
- };
- /* uint64_t r0,r1,r2; [24 bytes] */
- /* uint64_t pad0,pad1; [16 bytes] */
- uint64_t started; /* 8 bytes */
- uint64_t leftover; /* 8 bytes */
- uint8_t buffer[64]; /* 64 bytes */
- } poly1305_state_internal; /* 448 bytes total + 63 bytes for
- alignment = 511 bytes raw */
-
- static poly1305_state_internal INLINE *poly1305_aligned_state(
- poly1305_state *state) {
- return (poly1305_state_internal *)(((uint64_t)state + 63) & ~63);
- }
-
- /* copy 0-63 bytes */
- static void INLINE
- poly1305_block_copy(uint8_t *dst, const uint8_t *src, size_t bytes) {
- size_t offset = src - dst;
- if (bytes & 32) {
- _mm_storeu_si128((xmmi *)(dst + 0),
- _mm_loadu_si128((xmmi *)(dst + offset + 0)));
- _mm_storeu_si128((xmmi *)(dst + 16),
- _mm_loadu_si128((xmmi *)(dst + offset + 16)));
- dst += 32;
- }
- if (bytes & 16) {
- _mm_storeu_si128((xmmi *)dst, _mm_loadu_si128((xmmi *)(dst + offset)));
- dst += 16;
- }
- if (bytes & 8) {
- *(uint64_t *)dst = *(uint64_t *)(dst + offset);
- dst += 8;
- }
- if (bytes & 4) {
- *(uint32_t *)dst = *(uint32_t *)(dst + offset);
- dst += 4;
- }
- if (bytes & 2) {
- *(uint16_t *)dst = *(uint16_t *)(dst + offset);
- dst += 2;
- }
- if (bytes & 1) {
- *(uint8_t *)dst = *(uint8_t *)(dst + offset);
- }
- }
-
- /* zero 0-15 bytes */
- static void INLINE poly1305_block_zero(uint8_t *dst, size_t bytes) {
- if (bytes & 8) {
- *(uint64_t *)dst = 0;
- dst += 8;
- }
- if (bytes & 4) {
- *(uint32_t *)dst = 0;
- dst += 4;
- }
- if (bytes & 2) {
- *(uint16_t *)dst = 0;
- dst += 2;
- }
- if (bytes & 1) {
- *(uint8_t *)dst = 0;
- }
- }
-
- static size_t INLINE poly1305_min(size_t a, size_t b) {
- return (a < b) ? a : b;
- }
-
- void CRYPTO_poly1305_init(poly1305_state *state, const uint8_t key[32]) {
- poly1305_state_internal *st = poly1305_aligned_state(state);
- poly1305_power *p;
- uint64_t r0, r1, r2;
- uint64_t t0, t1;
-
- /* clamp key */
- t0 = U8TO64_LE(key + 0);
- t1 = U8TO64_LE(key + 8);
- r0 = t0 & 0xffc0fffffff;
- t0 >>= 44;
- t0 |= t1 << 20;
- r1 = t0 & 0xfffffc0ffff;
- t1 >>= 24;
- r2 = t1 & 0x00ffffffc0f;
-
- /* store r in un-used space of st->P[1] */
- p = &st->P[1];
- p->R20.d[1] = (uint32_t)(r0);
- p->R20.d[3] = (uint32_t)(r0 >> 32);
- p->R21.d[1] = (uint32_t)(r1);
- p->R21.d[3] = (uint32_t)(r1 >> 32);
- p->R22.d[1] = (uint32_t)(r2);
- p->R22.d[3] = (uint32_t)(r2 >> 32);
-
- /* store pad */
- p->R23.d[1] = U8TO32_LE(key + 16);
- p->R23.d[3] = U8TO32_LE(key + 20);
- p->R24.d[1] = U8TO32_LE(key + 24);
- p->R24.d[3] = U8TO32_LE(key + 28);
-
- /* H = 0 */
- st->H[0] = _mm_setzero_si128();
- st->H[1] = _mm_setzero_si128();
- st->H[2] = _mm_setzero_si128();
- st->H[3] = _mm_setzero_si128();
- st->H[4] = _mm_setzero_si128();
-
- st->started = 0;
- st->leftover = 0;
- }
-
- static void poly1305_first_block(poly1305_state_internal *st,
- const uint8_t *m) {
- const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask);
- const xmmi FIVE = _mm_load_si128((xmmi *)poly1305_x64_sse2_5);
- const xmmi HIBIT = _mm_load_si128((xmmi *)poly1305_x64_sse2_1shl128);
- xmmi T5, T6;
- poly1305_power *p;
- uint128_t d[3];
- uint64_t r0, r1, r2;
- uint64_t r20, r21, r22, s22;
- uint64_t pad0, pad1;
- uint64_t c;
- uint64_t i;
-
- /* pull out stored info */
- p = &st->P[1];
-
- r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
- r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
- r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
- pad0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1];
- pad1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1];
-
- /* compute powers r^2,r^4 */
- r20 = r0;
- r21 = r1;
- r22 = r2;
- for (i = 0; i < 2; i++) {
- s22 = r22 * (5 << 2);
-
- d[0] = add128(mul64x64_128(r20, r20), mul64x64_128(r21 * 2, s22));
- d[1] = add128(mul64x64_128(r22, s22), mul64x64_128(r20 * 2, r21));
- d[2] = add128(mul64x64_128(r21, r21), mul64x64_128(r22 * 2, r20));
-
- r20 = lo128(d[0]) & 0xfffffffffff;
- c = shr128(d[0], 44);
- d[1] = add128_64(d[1], c);
- r21 = lo128(d[1]) & 0xfffffffffff;
- c = shr128(d[1], 44);
- d[2] = add128_64(d[2], c);
- r22 = lo128(d[2]) & 0x3ffffffffff;
- c = shr128(d[2], 42);
- r20 += c * 5;
- c = (r20 >> 44);
- r20 = r20 & 0xfffffffffff;
- r21 += c;
-
- p->R20.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)(r20)&0x3ffffff),
- _MM_SHUFFLE(1, 0, 1, 0));
- p->R21.v = _mm_shuffle_epi32(
- _mm_cvtsi32_si128((uint32_t)((r20 >> 26) | (r21 << 18)) & 0x3ffffff),
- _MM_SHUFFLE(1, 0, 1, 0));
- p->R22.v =
- _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r21 >> 8)) & 0x3ffffff),
- _MM_SHUFFLE(1, 0, 1, 0));
- p->R23.v = _mm_shuffle_epi32(
- _mm_cvtsi32_si128((uint32_t)((r21 >> 34) | (r22 << 10)) & 0x3ffffff),
- _MM_SHUFFLE(1, 0, 1, 0));
- p->R24.v = _mm_shuffle_epi32(_mm_cvtsi32_si128((uint32_t)((r22 >> 16))),
- _MM_SHUFFLE(1, 0, 1, 0));
- p->S21.v = _mm_mul_epu32(p->R21.v, FIVE);
- p->S22.v = _mm_mul_epu32(p->R22.v, FIVE);
- p->S23.v = _mm_mul_epu32(p->R23.v, FIVE);
- p->S24.v = _mm_mul_epu32(p->R24.v, FIVE);
- p--;
- }
-
- /* put saved info back */
- p = &st->P[1];
- p->R20.d[1] = (uint32_t)(r0);
- p->R20.d[3] = (uint32_t)(r0 >> 32);
- p->R21.d[1] = (uint32_t)(r1);
- p->R21.d[3] = (uint32_t)(r1 >> 32);
- p->R22.d[1] = (uint32_t)(r2);
- p->R22.d[3] = (uint32_t)(r2 >> 32);
- p->R23.d[1] = (uint32_t)(pad0);
- p->R23.d[3] = (uint32_t)(pad0 >> 32);
- p->R24.d[1] = (uint32_t)(pad1);
- p->R24.d[3] = (uint32_t)(pad1 >> 32);
-
- /* H = [Mx,My] */
- T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)),
- _mm_loadl_epi64((xmmi *)(m + 16)));
- T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)),
- _mm_loadl_epi64((xmmi *)(m + 24)));
- st->H[0] = _mm_and_si128(MMASK, T5);
- st->H[1] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
- st->H[2] = _mm_and_si128(MMASK, T5);
- st->H[3] = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- st->H[4] = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
- }
-
- static void poly1305_blocks(poly1305_state_internal *st, const uint8_t *m,
- size_t bytes) {
- const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask);
- const xmmi FIVE = _mm_load_si128((xmmi *)poly1305_x64_sse2_5);
- const xmmi HIBIT = _mm_load_si128((xmmi *)poly1305_x64_sse2_1shl128);
-
- poly1305_power *p;
- xmmi H0, H1, H2, H3, H4;
- xmmi T0, T1, T2, T3, T4, T5, T6;
- xmmi M0, M1, M2, M3, M4;
- xmmi C1, C2;
-
- H0 = st->H[0];
- H1 = st->H[1];
- H2 = st->H[2];
- H3 = st->H[3];
- H4 = st->H[4];
-
- while (bytes >= 64) {
- /* H *= [r^4,r^4] */
- p = &st->P[0];
- T0 = _mm_mul_epu32(H0, p->R20.v);
- T1 = _mm_mul_epu32(H0, p->R21.v);
- T2 = _mm_mul_epu32(H0, p->R22.v);
- T3 = _mm_mul_epu32(H0, p->R23.v);
- T4 = _mm_mul_epu32(H0, p->R24.v);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- T5 = _mm_mul_epu32(H1, p->R23.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H2, p->R22.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H3, p->R21.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H4, p->R20.v);
- T4 = _mm_add_epi64(T4, T5);
-
- /* H += [Mx,My]*[r^2,r^2] */
- T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)),
- _mm_loadl_epi64((xmmi *)(m + 16)));
- T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)),
- _mm_loadl_epi64((xmmi *)(m + 24)));
- M0 = _mm_and_si128(MMASK, T5);
- M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
- M2 = _mm_and_si128(MMASK, T5);
- M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
-
- p = &st->P[1];
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- T5 = _mm_mul_epu32(M0, p->R24.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(M1, p->R23.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(M2, p->R22.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(M3, p->R21.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(M4, p->R20.v);
- T4 = _mm_add_epi64(T4, T5);
-
- /* H += [Mx,My] */
- T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 32)),
- _mm_loadl_epi64((xmmi *)(m + 48)));
- T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 40)),
- _mm_loadl_epi64((xmmi *)(m + 56)));
- M0 = _mm_and_si128(MMASK, T5);
- M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
- M2 = _mm_and_si128(MMASK, T5);
- M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
-
- T0 = _mm_add_epi64(T0, M0);
- T1 = _mm_add_epi64(T1, M1);
- T2 = _mm_add_epi64(T2, M2);
- T3 = _mm_add_epi64(T3, M3);
- T4 = _mm_add_epi64(T4, M4);
-
- /* reduce */
- 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, C1);
- T4 = _mm_add_epi64(T4, C2);
- 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, C1);
- T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
- 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, C1);
- T1 = _mm_add_epi64(T1, C2);
- C1 = _mm_srli_epi64(T3, 26);
- T3 = _mm_and_si128(T3, MMASK);
- T4 = _mm_add_epi64(T4, C1);
-
- /* H = (H*[r^4,r^4] + [Mx,My]*[r^2,r^2] + [Mx,My]) */
- H0 = T0;
- H1 = T1;
- H2 = T2;
- H3 = T3;
- H4 = T4;
-
- m += 64;
- bytes -= 64;
- }
-
- st->H[0] = H0;
- st->H[1] = H1;
- st->H[2] = H2;
- st->H[3] = H3;
- st->H[4] = H4;
- }
-
- static size_t poly1305_combine(poly1305_state_internal *st, const uint8_t *m,
- size_t bytes) {
- const xmmi MMASK = _mm_load_si128((xmmi *)poly1305_x64_sse2_message_mask);
- const xmmi HIBIT = _mm_load_si128((xmmi *)poly1305_x64_sse2_1shl128);
- const xmmi FIVE = _mm_load_si128((xmmi *)poly1305_x64_sse2_5);
-
- poly1305_power *p;
- xmmi H0, H1, H2, H3, H4;
- xmmi M0, M1, M2, M3, M4;
- xmmi T0, T1, T2, T3, T4, T5, T6;
- xmmi C1, C2;
-
- uint64_t r0, r1, r2;
- uint64_t t0, t1, t2, t3, t4;
- uint64_t c;
- size_t consumed = 0;
-
- H0 = st->H[0];
- H1 = st->H[1];
- H2 = st->H[2];
- H3 = st->H[3];
- H4 = st->H[4];
-
- /* p = [r^2,r^2] */
- p = &st->P[1];
-
- if (bytes >= 32) {
- /* H *= [r^2,r^2] */
- T0 = _mm_mul_epu32(H0, p->R20.v);
- T1 = _mm_mul_epu32(H0, p->R21.v);
- T2 = _mm_mul_epu32(H0, p->R22.v);
- T3 = _mm_mul_epu32(H0, p->R23.v);
- T4 = _mm_mul_epu32(H0, p->R24.v);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- T5 = _mm_mul_epu32(H1, p->R23.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H2, p->R22.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H3, p->R21.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H4, p->R20.v);
- T4 = _mm_add_epi64(T4, T5);
-
- /* H += [Mx,My] */
- T5 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 0)),
- _mm_loadl_epi64((xmmi *)(m + 16)));
- T6 = _mm_unpacklo_epi64(_mm_loadl_epi64((xmmi *)(m + 8)),
- _mm_loadl_epi64((xmmi *)(m + 24)));
- M0 = _mm_and_si128(MMASK, T5);
- M1 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- T5 = _mm_or_si128(_mm_srli_epi64(T5, 52), _mm_slli_epi64(T6, 12));
- M2 = _mm_and_si128(MMASK, T5);
- M3 = _mm_and_si128(MMASK, _mm_srli_epi64(T5, 26));
- M4 = _mm_or_si128(_mm_srli_epi64(T6, 40), HIBIT);
-
- T0 = _mm_add_epi64(T0, M0);
- T1 = _mm_add_epi64(T1, M1);
- T2 = _mm_add_epi64(T2, M2);
- T3 = _mm_add_epi64(T3, M3);
- T4 = _mm_add_epi64(T4, M4);
-
- /* reduce */
- 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, C1);
- T4 = _mm_add_epi64(T4, C2);
- 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, C1);
- T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
- 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, C1);
- T1 = _mm_add_epi64(T1, C2);
- C1 = _mm_srli_epi64(T3, 26);
- T3 = _mm_and_si128(T3, MMASK);
- T4 = _mm_add_epi64(T4, C1);
-
- /* H = (H*[r^2,r^2] + [Mx,My]) */
- H0 = T0;
- H1 = T1;
- H2 = T2;
- H3 = T3;
- H4 = T4;
-
- consumed = 32;
- }
-
- /* finalize, H *= [r^2,r] */
- r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
- r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
- r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
-
- p->R20.d[2] = (uint32_t)(r0)&0x3ffffff;
- p->R21.d[2] = (uint32_t)((r0 >> 26) | (r1 << 18)) & 0x3ffffff;
- p->R22.d[2] = (uint32_t)((r1 >> 8)) & 0x3ffffff;
- p->R23.d[2] = (uint32_t)((r1 >> 34) | (r2 << 10)) & 0x3ffffff;
- p->R24.d[2] = (uint32_t)((r2 >> 16));
- p->S21.d[2] = p->R21.d[2] * 5;
- p->S22.d[2] = p->R22.d[2] * 5;
- p->S23.d[2] = p->R23.d[2] * 5;
- p->S24.d[2] = p->R24.d[2] * 5;
-
- /* H *= [r^2,r] */
- T0 = _mm_mul_epu32(H0, p->R20.v);
- T1 = _mm_mul_epu32(H0, p->R21.v);
- T2 = _mm_mul_epu32(H0, p->R22.v);
- T3 = _mm_mul_epu32(H0, p->R23.v);
- T4 = _mm_mul_epu32(H0, p->R24.v);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- 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);
- T5 = _mm_mul_epu32(H1, p->R23.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H2, p->R22.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H3, p->R21.v);
- T4 = _mm_add_epi64(T4, T5);
- T5 = _mm_mul_epu32(H4, p->R20.v);
- T4 = _mm_add_epi64(T4, T5);
-
- 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, C1);
- T4 = _mm_add_epi64(T4, C2);
- 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, C1);
- T0 = _mm_add_epi64(T0, _mm_mul_epu32(C2, FIVE));
- 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, C1);
- T1 = _mm_add_epi64(T1, C2);
- C1 = _mm_srli_epi64(T3, 26);
- T3 = _mm_and_si128(T3, MMASK);
- T4 = _mm_add_epi64(T4, C1);
-
- /* H = H[0]+H[1] */
- H0 = _mm_add_epi64(T0, _mm_srli_si128(T0, 8));
- H1 = _mm_add_epi64(T1, _mm_srli_si128(T1, 8));
- H2 = _mm_add_epi64(T2, _mm_srli_si128(T2, 8));
- H3 = _mm_add_epi64(T3, _mm_srli_si128(T3, 8));
- H4 = _mm_add_epi64(T4, _mm_srli_si128(T4, 8));
-
- t0 = _mm_cvtsi128_si32(H0);
- c = (t0 >> 26);
- t0 &= 0x3ffffff;
- t1 = _mm_cvtsi128_si32(H1) + c;
- c = (t1 >> 26);
- t1 &= 0x3ffffff;
- t2 = _mm_cvtsi128_si32(H2) + c;
- c = (t2 >> 26);
- t2 &= 0x3ffffff;
- t3 = _mm_cvtsi128_si32(H3) + c;
- c = (t3 >> 26);
- t3 &= 0x3ffffff;
- t4 = _mm_cvtsi128_si32(H4) + c;
- c = (t4 >> 26);
- t4 &= 0x3ffffff;
- t0 = t0 + (c * 5);
- c = (t0 >> 26);
- t0 &= 0x3ffffff;
- t1 = t1 + c;
-
- st->HH[0] = ((t0) | (t1 << 26)) & 0xfffffffffffull;
- st->HH[1] = ((t1 >> 18) | (t2 << 8) | (t3 << 34)) & 0xfffffffffffull;
- st->HH[2] = ((t3 >> 10) | (t4 << 16)) & 0x3ffffffffffull;
-
- return consumed;
- }
-
- void CRYPTO_poly1305_update(poly1305_state *state, const uint8_t *m,
- size_t bytes) {
- poly1305_state_internal *st = poly1305_aligned_state(state);
- size_t want;
-
- /* need at least 32 initial bytes to start the accelerated branch */
- if (!st->started) {
- if ((st->leftover == 0) && (bytes > 32)) {
- poly1305_first_block(st, m);
- m += 32;
- bytes -= 32;
- } else {
- want = poly1305_min(32 - st->leftover, bytes);
- poly1305_block_copy(st->buffer + st->leftover, m, want);
- bytes -= want;
- m += want;
- st->leftover += want;
- if ((st->leftover < 32) || (bytes == 0)) {
- return;
- }
- poly1305_first_block(st, st->buffer);
- st->leftover = 0;
- }
- st->started = 1;
- }
-
- /* handle leftover */
- if (st->leftover) {
- want = poly1305_min(64 - st->leftover, bytes);
- poly1305_block_copy(st->buffer + st->leftover, m, want);
- bytes -= want;
- m += want;
- st->leftover += want;
- if (st->leftover < 64) {
- return;
- }
- poly1305_blocks(st, st->buffer, 64);
- st->leftover = 0;
- }
-
- /* process 64 byte blocks */
- if (bytes >= 64) {
- want = (bytes & ~63);
- poly1305_blocks(st, m, want);
- m += want;
- bytes -= want;
- }
-
- if (bytes) {
- poly1305_block_copy(st->buffer + st->leftover, m, bytes);
- st->leftover += bytes;
- }
- }
-
- void CRYPTO_poly1305_finish(poly1305_state *state, uint8_t mac[16]) {
- poly1305_state_internal *st = poly1305_aligned_state(state);
- size_t leftover = st->leftover;
- uint8_t *m = st->buffer;
- uint128_t d[3];
- uint64_t h0, h1, h2;
- uint64_t t0, t1;
- uint64_t g0, g1, g2, c, nc;
- uint64_t r0, r1, r2, s1, s2;
- poly1305_power *p;
-
- if (st->started) {
- size_t consumed = poly1305_combine(st, m, leftover);
- leftover -= consumed;
- m += consumed;
- }
-
- /* st->HH will either be 0 or have the combined result */
- h0 = st->HH[0];
- h1 = st->HH[1];
- h2 = st->HH[2];
-
- p = &st->P[1];
- r0 = ((uint64_t)p->R20.d[3] << 32) | (uint64_t)p->R20.d[1];
- r1 = ((uint64_t)p->R21.d[3] << 32) | (uint64_t)p->R21.d[1];
- r2 = ((uint64_t)p->R22.d[3] << 32) | (uint64_t)p->R22.d[1];
- s1 = r1 * (5 << 2);
- s2 = r2 * (5 << 2);
-
- if (leftover < 16) {
- goto poly1305_donna_atmost15bytes;
- }
-
- poly1305_donna_atleast16bytes:
- t0 = U8TO64_LE(m + 0);
- t1 = U8TO64_LE(m + 8);
- h0 += t0 & 0xfffffffffff;
- t0 = shr128_pair(t1, t0, 44);
- h1 += t0 & 0xfffffffffff;
- h2 += (t1 >> 24) | ((uint64_t)1 << 40);
-
- poly1305_donna_mul:
- d[0] = add128(add128(mul64x64_128(h0, r0), mul64x64_128(h1, s2)),
- mul64x64_128(h2, s1));
- d[1] = add128(add128(mul64x64_128(h0, r1), mul64x64_128(h1, r0)),
- mul64x64_128(h2, s2));
- d[2] = add128(add128(mul64x64_128(h0, r2), mul64x64_128(h1, r1)),
- mul64x64_128(h2, r0));
- h0 = lo128(d[0]) & 0xfffffffffff;
- c = shr128(d[0], 44);
- d[1] = add128_64(d[1], c);
- h1 = lo128(d[1]) & 0xfffffffffff;
- c = shr128(d[1], 44);
- d[2] = add128_64(d[2], c);
- h2 = lo128(d[2]) & 0x3ffffffffff;
- c = shr128(d[2], 42);
- h0 += c * 5;
-
- m += 16;
- leftover -= 16;
- if (leftover >= 16) {
- goto poly1305_donna_atleast16bytes;
- }
-
- /* final bytes */
- poly1305_donna_atmost15bytes:
- if (!leftover) {
- goto poly1305_donna_finish;
- }
-
- m[leftover++] = 1;
- poly1305_block_zero(m + leftover, 16 - leftover);
- leftover = 16;
-
- t0 = U8TO64_LE(m + 0);
- t1 = U8TO64_LE(m + 8);
- h0 += t0 & 0xfffffffffff;
- t0 = shr128_pair(t1, t0, 44);
- h1 += t0 & 0xfffffffffff;
- h2 += (t1 >> 24);
-
- goto poly1305_donna_mul;
-
- poly1305_donna_finish:
- c = (h0 >> 44);
- h0 &= 0xfffffffffff;
- h1 += c;
- c = (h1 >> 44);
- h1 &= 0xfffffffffff;
- h2 += c;
- c = (h2 >> 42);
- h2 &= 0x3ffffffffff;
- h0 += c * 5;
-
- g0 = h0 + 5;
- c = (g0 >> 44);
- g0 &= 0xfffffffffff;
- g1 = h1 + c;
- c = (g1 >> 44);
- g1 &= 0xfffffffffff;
- g2 = h2 + c - ((uint64_t)1 << 42);
-
- c = (g2 >> 63) - 1;
- nc = ~c;
- h0 = (h0 & nc) | (g0 & c);
- h1 = (h1 & nc) | (g1 & c);
- h2 = (h2 & nc) | (g2 & c);
-
- /* pad */
- t0 = ((uint64_t)p->R23.d[3] << 32) | (uint64_t)p->R23.d[1];
- t1 = ((uint64_t)p->R24.d[3] << 32) | (uint64_t)p->R24.d[1];
- h0 += (t0 & 0xfffffffffff);
- c = (h0 >> 44);
- h0 &= 0xfffffffffff;
- t0 = shr128_pair(t1, t0, 44);
- h1 += (t0 & 0xfffffffffff) + c;
- c = (h1 >> 44);
- h1 &= 0xfffffffffff;
- t1 = (t1 >> 24);
- h2 += (t1)+c;
-
- U64TO8_LE(mac + 0, ((h0) | (h1 << 44)));
- U64TO8_LE(mac + 8, ((h1 >> 20) | (h2 << 24)));
- }
-
- #endif /* !OPENSSL_WINDOWS && OPENSSL_X86_64 */
|