Link chacha20_vec

src/libsodium/Makefile.am

@@ -285,7 +285,9 @@ libssse3_la_LDFLAGS = $(libsodium_la_LDFLAGS)
 libssse3_la_CPPFLAGS = $(libsodium_la_CPPFLAGS) \
 	@CFLAGS_SSE2@ @CFLAGS_SSSE3@
 libssse3_la_SOURCES = \
-	crypto_generichash/blake2/ref/blake2b-compress-ssse3.c
+	crypto_generichash/blake2/ref/blake2b-compress-ssse3.c \
+	crypto_stream/chacha20/vec/stream_chacha20_vec.h \
+	crypto_stream/chacha20/vec/stream_chacha20_vec.c
 
 libsse41_la_LDFLAGS = $(libsodium_la_LDFLAGS)
 libsse41_la_CPPFLAGS = $(libsodium_la_CPPFLAGS) \

src/libsodium/crypto_stream/chacha20/vec/stream_chacha20_vec.c

@@ -0,0 +1,351 @@
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "export.h"
+#include "utils.h"
+#include "crypto_stream_chacha20.h"
+#include "stream_chacha20_vec.h"
+#include "../stream_chacha20.h"
+
+#if (defined(HAVE_EMMINTRIN_H) && defined(HAVE_TMMINTRIN_H)) || \
+    (defined(_MSC_VER) && (defined(_M_X64) || defined(_M_AMD64)))
+
+#pragma GCC target("sse2")
+#pragma GCC target("ssse3")
+
+#define CHACHA_RNDS 20
+
+typedef unsigned vec __attribute__((vector_size(16)));
+
+#include <emmintrin.h>
+#include <tmmintrin.h>
+
+# if __clang__
+#  define VBPI 4
+# else
+#  define VBPI 3
+# endif
+# define ONE (vec) _mm_set_epi32(0, 0, 0, 1)
+# define LOAD(m) (vec) _mm_loadu_si128((__m128i *)(m))
+# define LOAD_ALIGNED(m) (vec) _mm_load_si128((__m128i *)(m))
+# define STORE(m, r) _mm_storeu_si128((__m128i *)(m), (__m128i)(r))
+# define ROTV1(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(0, 3, 2, 1))
+# define ROTV2(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(1, 0, 3, 2))
+# define ROTV3(x) (vec) _mm_shuffle_epi32((__m128i)x, _MM_SHUFFLE(2, 1, 0, 3))
+# define ROTW7(x) \
+    (vec)(_mm_slli_epi32((__m128i)x, 7) ^ _mm_srli_epi32((__m128i)x, 25))
+# define ROTW12(x) \
+    (vec)(_mm_slli_epi32((__m128i)x, 12) ^ _mm_srli_epi32((__m128i)x, 20))
+# define ROTW8(x) \
+    (vec)(_mm_slli_epi32((__m128i)x, 8) ^ _mm_srli_epi32((__m128i)x, 24))
+#define ROTW16(x) \
+    (vec)(_mm_slli_epi32((__m128i)x, 16) ^ _mm_srli_epi32((__m128i)x, 16))
+
+#ifndef REVV_BE
+# define REVV_BE(x) (x)
+#endif
+
+#ifndef REVW_BE
+# define REVW_BE(x) (x)
+#endif
+
+#define BPI (VBPI + 0) /* Blocks computed per loop iteration   */
+
+#define DQROUND_VECTORS(a, b, c, d) \
+    a += b;                         \
+    d ^= a;                         \
+    d = ROTW16(d);                  \
+    c += d;                         \
+    b ^= c;                         \
+    b = ROTW12(b);                  \
+    a += b;                         \
+    d ^= a;                         \
+    d = ROTW8(d);                   \
+    c += d;                         \
+    b ^= c;                         \
+    b = ROTW7(b);                   \
+    b = ROTV1(b);                   \
+    c = ROTV2(c);                   \
+    d = ROTV3(d);                   \
+    a += b;                         \
+    d ^= a;                         \
+    d = ROTW16(d);                  \
+    c += d;                         \
+    b ^= c;                         \
+    b = ROTW12(b);                  \
+    a += b;                         \
+    d ^= a;                         \
+    d = ROTW8(d);                   \
+    c += d;                         \
+    b ^= c;                         \
+    b = ROTW7(b);                   \
+    b = ROTV3(b);                   \
+    c = ROTV2(c);                   \
+    d = ROTV1(d);
+
+#define QROUND_WORDS(a, b, c, d) \
+    a = a + b;                   \
+    d ^= a;                      \
+    d = d << 16 | d >> 16;       \
+    c = c + d;                   \
+    b ^= c;                      \
+    b = b << 12 | b >> 20;       \
+    a = a + b;                   \
+    d ^= a;                      \
+    d = d << 8 | d >> 24;        \
+    c = c + d;                   \
+    b ^= c;                      \
+    b = b << 7 | b >> 25;
+
+#define WRITE_XOR(in, op, d, v0, v1, v2, v3)           \
+    STORE(op + d + 0, LOAD(in + d + 0) ^ REVV_BE(v0)); \
+    STORE(op + d + 4, LOAD(in + d + 4) ^ REVV_BE(v1)); \
+    STORE(op + d + 8, LOAD(in + d + 8) ^ REVV_BE(v2)); \
+    STORE(op + d + 12, LOAD(in + d + 12) ^ REVV_BE(v3));
+
+struct chacha_ctx {
+    vec s1;
+    vec s2;
+    vec s3;
+};
+
+typedef struct chacha_ctx chacha_ctx;
+
+static void
+chacha_ivsetup(chacha_ctx *ctx, const uint8_t *iv, uint64_t ic)
+{
+    ctx->s3 = (vec) {
+            ic,
+            ic >> 32,
+            ((uint32_t *) iv)[0],
+            ((uint32_t *) iv)[1]
+    };
+}
+
+static void
+chacha_ietf_ivsetup(chacha_ctx *ctx, const uint8_t *iv, uint32_t ic)
+{
+    ctx->s3 = (vec) {
+            ic,
+            ((uint32_t *) iv)[0],
+            ((uint32_t *) iv)[1],
+            ((uint32_t *) iv)[2]
+    };
+}
+
+static void
+chacha_keysetup(chacha_ctx *ctx, const uint8_t *k)
+{
+    unsigned int *kp;
+
+    kp = (unsigned int *) k;
+    ctx->s1 = LOAD(&((vec *)kp)[0]);
+    ctx->s2 = LOAD(&((vec *)kp)[1]);
+}
+
+static void
+chacha_encrypt_bytes(chacha_ctx *ctx, const uint8_t *in, uint8_t *out,
+                     unsigned long long inlen)
+{
+    unsigned long long iters;
+    unsigned int i;
+    unsigned int *op = (unsigned *)out;
+    unsigned int *ip = (unsigned *)in;
+    unsigned int *kp;
+    vec s0, s1, s2, s3;
+    CRYPTO_ALIGN(16) unsigned chacha_const[]
+        = { 0x61707865, 0x3320646E, 0x79622D32, 0x6B206574 };
+
+    s0 = LOAD_ALIGNED(chacha_const);
+    s1 = ctx->s1;
+    s2 = ctx->s2;
+    s3 = ctx->s3;
+
+    for (iters = 0; iters < inlen / (BPI * 64); iters++) {
+#if VBPI > 2
+        vec v8, v9, v10, v11;
+#endif
+#if VBPI > 3
+        vec v12, v13, v14, v15;
+#endif
+        vec v0, v1, v2, v3, v4, v5, v6, v7;
+        v4 = v0 = s0;
+        v5 = v1 = s1;
+        v6 = v2 = s2;
+        v3 = s3;
+        v7 = v3 + ONE;
+#if VBPI > 2
+        v8 = v4;
+        v9 = v5;
+        v10 = v6;
+        v11 = v7 + ONE;
+#endif
+#if VBPI > 3
+        v12 = v8;
+        v13 = v9;
+        v14 = v10;
+        v15 = v11 + ONE;
+#endif
+        for (i = CHACHA_RNDS / 2; i; i--) {
+            DQROUND_VECTORS(v0, v1, v2, v3)
+            DQROUND_VECTORS(v4, v5, v6, v7)
+#if VBPI > 2
+            DQROUND_VECTORS(v8, v9, v10, v11)
+#endif
+#if VBPI > 3
+            DQROUND_VECTORS(v12, v13, v14, v15)
+#endif
+        }
+
+        WRITE_XOR(ip, op, 0, v0 + s0, v1 + s1, v2 + s2, v3 + s3)
+        s3 += ONE;
+        WRITE_XOR(ip, op, 16, v4 + s0, v5 + s1, v6 + s2, v7 + s3)
+        s3 += ONE;
+#if VBPI > 2
+        WRITE_XOR(ip, op, 32, v8 + s0, v9 + s1, v10 + s2, v11 + s3)
+        s3 += ONE;
+#endif
+#if VBPI > 3
+        WRITE_XOR(ip, op, 48, v12 + s0, v13 + s1, v14 + s2, v15 + s3)
+        s3 += ONE;
+#endif
+        ip += VBPI * 16;
+        op += VBPI * 16;
+    }
+
+    for (iters = inlen % (BPI * 64) / 64; iters != 0; iters--) {
+        vec v0 = s0, v1 = s1, v2 = s2, v3 = s3;
+        for (i = CHACHA_RNDS / 2; i; i--) {
+            DQROUND_VECTORS(v0, v1, v2, v3);
+        }
+        WRITE_XOR(ip, op, 0, v0 + s0, v1 + s1, v2 + s2, v3 + s3)
+        s3 += ONE;
+        ip += 16;
+        op += 16;
+    }
+
+    inlen = inlen % 64;
+    if (inlen) {
+        CRYPTO_ALIGN(16) vec buf[4];
+        vec v0, v1, v2, v3;
+        v0 = s0;
+        v1 = s1;
+        v2 = s2;
+        v3 = s3;
+        for (i = CHACHA_RNDS / 2; i; i--) {
+            DQROUND_VECTORS(v0, v1, v2, v3);
+        }
+
+        if (inlen >= 16) {
+            STORE(op + 0, LOAD(ip + 0) ^ REVV_BE(v0 + s0));
+            if (inlen >= 32) {
+                STORE(op + 4, LOAD(ip + 4) ^ REVV_BE(v1 + s1));
+                if (inlen >= 48) {
+                    STORE(op + 8, LOAD(ip + 8) ^ REVV_BE(v2 + s2));
+                    buf[3] = REVV_BE(v3 + s3);
+                } else {
+                    buf[2] = REVV_BE(v2 + s2);
+                }
+            } else {
+                buf[1] = REVV_BE(v1 + s1);
+            }
+        } else {
+            buf[0] = REVV_BE(v0 + s0);
+        }
+        for (i = inlen & ~15; i < inlen; i++) {
+            ((char *)op)[i] = ((char *)ip)[i] ^ ((char *)buf)[i];
+        }
+    }
+}
+
+static int
+stream_vec(unsigned char *c, unsigned long long clen,
+           const unsigned char *n, const unsigned char *k)
+{
+    struct chacha_ctx ctx;
+
+    if (!clen) {
+        return 0;
+    }
+    (void) sizeof(int[crypto_stream_chacha20_KEYBYTES == 256 / 8 ? 1 : -1]);
+    chacha_keysetup(&ctx, k);
+    chacha_ivsetup(&ctx, n, 0ULL);
+    memset(c, 0, clen);
+    chacha_encrypt_bytes(&ctx, c, c, clen);
+    sodium_memzero(&ctx, sizeof ctx);
+
+    return 0;
+}
+
+static int
+stream_ietf_vec(unsigned char *c, unsigned long long clen,
+                const unsigned char *n, const unsigned char *k)
+{
+    struct chacha_ctx ctx;
+
+    if (!clen) {
+        return 0;
+    }
+    if (clen > 64ULL * (1ULL << 32) - 64ULL) {
+        abort();
+    }
+    (void) sizeof(int[crypto_stream_chacha20_KEYBYTES == 256 / 8 ? 1 : -1]);
+    chacha_keysetup(&ctx, k);
+    chacha_ietf_ivsetup(&ctx, n, 0ULL);
+    memset(c, 0, clen);
+    chacha_encrypt_bytes(&ctx, c, c, clen);
+    sodium_memzero(&ctx, sizeof ctx);
+
+    return 0;
+}
+
+static int
+stream_vec_xor_ic(unsigned char *c, const unsigned char *m,
+                  unsigned long long mlen,
+                  const unsigned char *n, uint64_t ic,
+                  const unsigned char *k)
+{
+    struct chacha_ctx ctx;
+
+    if (!mlen) {
+        return 0;
+    }
+    chacha_keysetup(&ctx, k);
+    chacha_ivsetup(&ctx, n, ic);
+    chacha_encrypt_bytes(&ctx, m, c, mlen);
+
+    sodium_memzero(&ctx, sizeof ctx);
+
+    return 0;
+}
+
+static int
+stream_ietf_vec_xor_ic(unsigned char *c, const unsigned char *m,
+                       unsigned long long mlen,
+                       const unsigned char *n, uint32_t ic,
+                       const unsigned char *k)
+{
+    struct chacha_ctx ctx;
+
+    if (!mlen) {
+        return 0;
+    }
+    chacha_keysetup(&ctx, k);
+    chacha_ietf_ivsetup(&ctx, n, ic);
+    chacha_encrypt_bytes(&ctx, m, c, mlen);
+    sodium_memzero(&ctx, sizeof ctx);
+
+    return 0;
+}
+
+struct crypto_stream_chacha20_implementation
+crypto_stream_chacha20_vec_implementation = {
+    SODIUM_C99(.stream =) stream_vec,
+    SODIUM_C99(.stream_ietf =) stream_ietf_vec,
+    SODIUM_C99(.stream_xor_ic =) stream_vec_xor_ic,
+    SODIUM_C99(.stream_ietf_xor_ic =) stream_ietf_vec_xor_ic
+};
+
+#endif

src/libsodium/crypto_stream/chacha20/vec/stream_chacha20_vec.h

@@ -0,0 +1,27 @@
+
+#include <stdint.h>
+
+#include "crypto_stream_chacha20.h"
+
+extern struct crypto_stream_chacha20_implementation
+    crypto_stream_chacha20_vec_implementation;
+
+int
+crypto_stream_chacha20_vec(unsigned char *c, unsigned long long clen,
+                           const unsigned char *n, const unsigned char *k);
+
+int
+crypto_stream_chacha20_vec_xor_ic(unsigned char *c, const unsigned char *m,
+                                  unsigned long long mlen,
+                                  const unsigned char *n, uint64_t ic,
+                                  const unsigned char *k);
+
+int
+crypto_stream_chacha20_ietf_vec(unsigned char *c, unsigned long long clen,
+                                const unsigned char *n, const unsigned char *k);
+
+int
+crypto_stream_chacha20_ietf_vec_xor_ic(unsigned char *c, const unsigned char *m,
+                                       unsigned long long mlen,
+                                       const unsigned char *n, uint32_t ic,
+                                       const unsigned char *k);