Add NEON accelerated functions
Reference implementation:
ScryptTest.TestVectors (2588 ms)
NEON acceleration:
ScryptTest.TestVectors (1169 ms)
Change-Id: I23974b2f0f64435696e75acfbdee1115a699abae
diff --git a/NOTICE b/NOTICE
index e88e916..b0b9311 100644
--- a/NOTICE
+++ b/NOTICE
@@ -26,3 +26,11 @@
* This file was originally written by Colin Percival as part of the Tarsnap
* online backup system.
*/
+
+/*
+ * version 20110505
+ * D. J. Bernstein
+ * Public domain.
+ *
+ * Based on crypto_core/salsa208/armneon/core.c from SUPERCOP 20130419
+ */
diff --git a/Scrypt-config.mk b/Scrypt-config.mk
index b32f9bd..30db285 100644
--- a/Scrypt-config.mk
+++ b/Scrypt-config.mk
@@ -37,9 +37,11 @@
arm_c_flags :=
-arm_src_files :=
+arm_src_files := \
+ lib/crypto/crypto_scrypt-neon.c \
-arm_exclude_files :=
+arm_exclude_files := \
+ lib/crypto/crypto_scrypt-ref.c \
x86_c_flags :=
diff --git a/import_scrypt.sh b/import_scrypt.sh
index 2270206..684379d 100755
--- a/import_scrypt.sh
+++ b/import_scrypt.sh
@@ -351,8 +351,6 @@
generate_build_config_mk ../build-config.mk
- # Extract license from source file
- sed '/^\/\*-/,/^ \*\//!d' lib/crypto/crypto_scrypt.h > ../NOTICE
touch ../MODULE_LICENSE_BSD_LIKE
cd ..
diff --git a/lib/crypto/crypto_scrypt-neon-salsa208.h b/lib/crypto/crypto_scrypt-neon-salsa208.h
new file mode 100644
index 0000000..a3b1019
--- /dev/null
+++ b/lib/crypto/crypto_scrypt-neon-salsa208.h
@@ -0,0 +1,120 @@
+/*
+ * version 20110505
+ * D. J. Bernstein
+ * Public domain.
+ *
+ * Based on crypto_core/salsa208/armneon/core.c from SUPERCOP 20130419
+ */
+
+#define ROUNDS 8
+static void
+salsa20_8_intrinsic(void * input)
+{
+ int i;
+
+ const uint32x4_t abab = {-1,0,-1,0};
+
+ /*
+ * This is modified since we only have one argument. Usually you'd rearrange
+ * the constant, key, and input bytes, but we just have one linear array to
+ * rearrange which is a bit easier.
+ */
+
+ /*
+ * Change the input to be diagonals as if it's a 4x4 matrix of 32-bit values.
+ */
+ uint32x4_t x0x5x10x15;
+ uint32x4_t x12x1x6x11;
+ uint32x4_t x8x13x2x7;
+ uint32x4_t x4x9x14x3;
+
+ uint32x4_t x0x1x10x11;
+ uint32x4_t x12x13x6x7;
+ uint32x4_t x8x9x2x3;
+ uint32x4_t x4x5x14x15;
+
+ uint32x4_t x0x1x2x3;
+ uint32x4_t x4x5x6x7;
+ uint32x4_t x8x9x10x11;
+ uint32x4_t x12x13x14x15;
+
+ x0x1x2x3 = vld1q_u8((uint8_t *) input);
+ x4x5x6x7 = vld1q_u8(16 + (uint8_t *) input);
+ x8x9x10x11 = vld1q_u8(32 + (uint8_t *) input);
+ x12x13x14x15 = vld1q_u8(48 + (uint8_t *) input);
+
+ x0x1x10x11 = vcombine_u32(vget_low_u32(x0x1x2x3), vget_high_u32(x8x9x10x11));
+ x4x5x14x15 = vcombine_u32(vget_low_u32(x4x5x6x7), vget_high_u32(x12x13x14x15));
+ x8x9x2x3 = vcombine_u32(vget_low_u32(x8x9x10x11), vget_high_u32(x0x1x2x3));
+ x12x13x6x7 = vcombine_u32(vget_low_u32(x12x13x14x15), vget_high_u32(x4x5x6x7));
+
+ x0x5x10x15 = vbslq_u32(abab,x0x1x10x11,x4x5x14x15);
+ x8x13x2x7 = vbslq_u32(abab,x8x9x2x3,x12x13x6x7);
+ x4x9x14x3 = vbslq_u32(abab,x4x5x14x15,x8x9x2x3);
+ x12x1x6x11 = vbslq_u32(abab,x12x13x6x7,x0x1x10x11);
+
+ uint32x4_t start0 = x0x5x10x15;
+ uint32x4_t start1 = x12x1x6x11;
+ uint32x4_t start3 = x4x9x14x3;
+ uint32x4_t start2 = x8x13x2x7;
+
+ /* From here on this should be the same as the SUPERCOP version. */
+
+ uint32x4_t diag0 = start0;
+ uint32x4_t diag1 = start1;
+ uint32x4_t diag2 = start2;
+ uint32x4_t diag3 = start3;
+
+ uint32x4_t a0;
+ uint32x4_t a1;
+ uint32x4_t a2;
+ uint32x4_t a3;
+
+ for (i = ROUNDS;i > 0;i -= 2) {
+ a0 = diag1 + diag0;
+ diag3 ^= vsriq_n_u32(vshlq_n_u32(a0,7),a0,25);
+ a1 = diag0 + diag3;
+ diag2 ^= vsriq_n_u32(vshlq_n_u32(a1,9),a1,23);
+ a2 = diag3 + diag2;
+ diag1 ^= vsriq_n_u32(vshlq_n_u32(a2,13),a2,19);
+ a3 = diag2 + diag1;
+ diag0 ^= vsriq_n_u32(vshlq_n_u32(a3,18),a3,14);
+
+ diag3 = vextq_u32(diag3,diag3,3);
+ diag2 = vextq_u32(diag2,diag2,2);
+ diag1 = vextq_u32(diag1,diag1,1);
+
+ a0 = diag3 + diag0;
+ diag1 ^= vsriq_n_u32(vshlq_n_u32(a0,7),a0,25);
+ a1 = diag0 + diag1;
+ diag2 ^= vsriq_n_u32(vshlq_n_u32(a1,9),a1,23);
+ a2 = diag1 + diag2;
+ diag3 ^= vsriq_n_u32(vshlq_n_u32(a2,13),a2,19);
+ a3 = diag2 + diag3;
+ diag0 ^= vsriq_n_u32(vshlq_n_u32(a3,18),a3,14);
+
+ diag1 = vextq_u32(diag1,diag1,3);
+ diag2 = vextq_u32(diag2,diag2,2);
+ diag3 = vextq_u32(diag3,diag3,1);
+ }
+
+ x0x5x10x15 = diag0 + start0;
+ x12x1x6x11 = diag1 + start1;
+ x8x13x2x7 = diag2 + start2;
+ x4x9x14x3 = diag3 + start3;
+
+ x0x1x10x11 = vbslq_u32(abab,x0x5x10x15,x12x1x6x11);
+ x12x13x6x7 = vbslq_u32(abab,x12x1x6x11,x8x13x2x7);
+ x8x9x2x3 = vbslq_u32(abab,x8x13x2x7,x4x9x14x3);
+ x4x5x14x15 = vbslq_u32(abab,x4x9x14x3,x0x5x10x15);
+
+ x0x1x2x3 = vcombine_u32(vget_low_u32(x0x1x10x11),vget_high_u32(x8x9x2x3));
+ x4x5x6x7 = vcombine_u32(vget_low_u32(x4x5x14x15),vget_high_u32(x12x13x6x7));
+ x8x9x10x11 = vcombine_u32(vget_low_u32(x8x9x2x3),vget_high_u32(x0x1x10x11));
+ x12x13x14x15 = vcombine_u32(vget_low_u32(x12x13x6x7),vget_high_u32(x4x5x14x15));
+
+ vst1q_u8((uint8_t *) input,(uint8x16_t) x0x1x2x3);
+ vst1q_u8(16 + (uint8_t *) input,(uint8x16_t) x4x5x6x7);
+ vst1q_u8(32 + (uint8_t *) input,(uint8x16_t) x8x9x10x11);
+ vst1q_u8(48 + (uint8_t *) input,(uint8x16_t) x12x13x14x15);
+}
diff --git a/lib/crypto/crypto_scrypt-neon.c b/lib/crypto/crypto_scrypt-neon.c
new file mode 100644
index 0000000..a3bf052
--- /dev/null
+++ b/lib/crypto/crypto_scrypt-neon.c
@@ -0,0 +1,305 @@
+/*-
+ * Copyright 2009 Colin Percival
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in the
+ * documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * This file was originally written by Colin Percival as part of the Tarsnap
+ * online backup system.
+ */
+#include "scrypt_platform.h"
+
+#include <machine/cpu-features.h>
+#include <arm_neon.h>
+
+#include <errno.h>
+#include <stdint.h>
+#include <limits.h>
+#include <stdlib.h>
+#include <string.h>
+
+#ifdef USE_OPENSSL_PBKDF2
+#include <openssl/evp.h>
+#else
+#include "sha256.h"
+#endif
+#include "sysendian.h"
+
+#include "crypto_scrypt.h"
+
+#include "crypto_scrypt-neon-salsa208.h"
+
+static void blkcpy(void *, void *, size_t);
+static void blkxor(void *, void *, size_t);
+void crypto_core_salsa208_armneon2(void *);
+static void blockmix_salsa8(uint8x16_t *, uint8x16_t *, uint8x16_t *, size_t);
+static uint64_t integerify(void *, size_t);
+static void smix(uint8_t *, size_t, uint64_t, void *, void *);
+
+static void
+blkcpy(void * dest, void * src, size_t len)
+{
+ uint8x16_t * D = dest;
+ uint8x16_t * S = src;
+ size_t L = len / 16;
+ size_t i;
+
+ for (i = 0; i < L; i++)
+ D[i] = S[i];
+}
+
+static void
+blkxor(void * dest, void * src, size_t len)
+{
+ uint8x16_t * D = dest;
+ uint8x16_t * S = src;
+ size_t L = len / 16;
+ size_t i;
+
+ for (i = 0; i < L; i++)
+ D[i] = veorq_u8(D[i], S[i]);
+}
+
+/**
+ * blockmix_salsa8(B, Y, r):
+ * Compute B = BlockMix_{salsa20/8, r}(B). The input B must be 128r bytes in
+ * length; the temporary space Y must also be the same size.
+ */
+static void
+blockmix_salsa8(uint8x16_t * Bin, uint8x16_t * Bout, uint8x16_t * X, size_t r)
+{
+ size_t i;
+
+ /* 1: X <-- B_{2r - 1} */
+ blkcpy(X, &Bin[8 * r - 4], 64);
+
+ /* 2: for i = 0 to 2r - 1 do */
+ for (i = 0; i < r; i++) {
+ /* 3: X <-- H(X \xor B_i) */
+ blkxor(X, &Bin[i * 8], 64);
+ salsa20_8_intrinsic((void *) X);
+
+ /* 4: Y_i <-- X */
+ /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
+ blkcpy(&Bout[i * 4], X, 64);
+
+ /* 3: X <-- H(X \xor B_i) */
+ blkxor(X, &Bin[i * 8 + 4], 64);
+ salsa20_8_intrinsic((void *) X);
+
+ /* 4: Y_i <-- X */
+ /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
+ blkcpy(&Bout[(r + i) * 4], X, 64);
+ }
+}
+
+/**
+ * integerify(B, r):
+ * Return the result of parsing B_{2r-1} as a little-endian integer.
+ */
+static uint64_t
+integerify(void * B, size_t r)
+{
+ uint8_t * X = (void*)((uintptr_t)(B) + (2 * r - 1) * 64);
+
+ return (le64dec(X));
+}
+
+/**
+ * smix(B, r, N, V, XY):
+ * Compute B = SMix_r(B, N). The input B must be 128r bytes in length; the
+ * temporary storage V must be 128rN bytes in length; the temporary storage
+ * XY must be 256r bytes in length. The value N must be a power of 2.
+ */
+static void
+smix(uint8_t * B, size_t r, uint64_t N, void * V, void * XY)
+{
+ uint8x16_t * X = XY;
+ uint8x16_t * Y = (void *)((uintptr_t)(XY) + 128 * r);
+ uint8x16_t * Z = (void *)((uintptr_t)(XY) + 256 * r);
+ uint32_t * X32 = (void *)X;
+ uint64_t i, j;
+ size_t k;
+
+ /* 1: X <-- B */
+ blkcpy(X, B, 128 * r);
+
+ /* 2: for i = 0 to N - 1 do */
+ for (i = 0; i < N; i += 2) {
+ /* 3: V_i <-- X */
+ blkcpy((void *)((uintptr_t)(V) + i * 128 * r), X, 128 * r);
+
+ /* 4: X <-- H(X) */
+ blockmix_salsa8(X, Y, Z, r);
+
+ /* 3: V_i <-- X */
+ blkcpy((void *)((uintptr_t)(V) + (i + 1) * 128 * r),
+ Y, 128 * r);
+
+ /* 4: X <-- H(X) */
+ blockmix_salsa8(Y, X, Z, r);
+ }
+
+ /* 6: for i = 0 to N - 1 do */
+ for (i = 0; i < N; i += 2) {
+ /* 7: j <-- Integerify(X) mod N */
+ j = integerify(X, r) & (N - 1);
+
+ /* 8: X <-- H(X \xor V_j) */
+ blkxor(X, (void *)((uintptr_t)(V) + j * 128 * r), 128 * r);
+ blockmix_salsa8(X, Y, Z, r);
+
+ /* 7: j <-- Integerify(X) mod N */
+ j = integerify(Y, r) & (N - 1);
+
+ /* 8: X <-- H(X \xor V_j) */
+ blkxor(Y, (void *)((uintptr_t)(V) + j * 128 * r), 128 * r);
+ blockmix_salsa8(Y, X, Z, r);
+ }
+
+ /* 10: B' <-- X */
+ blkcpy(B, X, 128 * r);
+}
+
+/**
+ * crypto_scrypt(passwd, passwdlen, salt, saltlen, N, r, p, buf, buflen):
+ * Compute scrypt(passwd[0 .. passwdlen - 1], salt[0 .. saltlen - 1], N, r,
+ * p, buflen) and write the result into buf. The parameters r, p, and buflen
+ * must satisfy r * p < 2^30 and buflen <= (2^32 - 1) * 32. The parameter N
+ * must be a power of 2.
+ *
+ * Return 0 on success; or -1 on error.
+ */
+int
+crypto_scrypt(const uint8_t * passwd, size_t passwdlen,
+ const uint8_t * salt, size_t saltlen, uint64_t N, uint32_t r, uint32_t p,
+ uint8_t * buf, size_t buflen)
+{
+ void * B0, * V0, * XY0;
+ uint8_t * B;
+ uint32_t * V;
+ uint32_t * XY;
+ uint32_t i;
+
+ /* Sanity-check parameters. */
+#if SIZE_MAX > UINT32_MAX
+ if (buflen > (((uint64_t)(1) << 32) - 1) * 32) {
+ errno = EFBIG;
+ goto err0;
+ }
+#endif
+ if ((uint64_t)(r) * (uint64_t)(p) >= (1 << 30)) {
+ errno = EFBIG;
+ goto err0;
+ }
+ if (((N & (N - 1)) != 0) || (N == 0)) {
+ errno = EINVAL;
+ goto err0;
+ }
+ if ((r > SIZE_MAX / 128 / p) ||
+#if SIZE_MAX / 256 <= UINT32_MAX
+ (r > SIZE_MAX / 256) ||
+#endif
+ (N > SIZE_MAX / 128 / r)) {
+ errno = ENOMEM;
+ goto err0;
+ }
+
+ /* Allocate memory. */
+#ifdef HAVE_POSIX_MEMALIGN
+ if ((errno = posix_memalign(&B0, 64, 128 * r * p)) != 0)
+ goto err0;
+ B = (uint8_t *)(B0);
+ if ((errno = posix_memalign(&XY0, 64, 256 * r + 64)) != 0)
+ goto err1;
+ XY = (uint32_t *)(XY0);
+#ifndef MAP_ANON
+ if ((errno = posix_memalign(&V0, 64, 128 * r * N)) != 0)
+ goto err2;
+ V = (uint32_t *)(V0);
+#endif
+#else
+ if ((B0 = malloc(128 * r * p + 63)) == NULL)
+ goto err0;
+ B = (uint8_t *)(((uintptr_t)(B0) + 63) & ~ (uintptr_t)(63));
+ if ((XY0 = malloc(256 * r + 64 + 63)) == NULL)
+ goto err1;
+ XY = (uint32_t *)(((uintptr_t)(XY0) + 63) & ~ (uintptr_t)(63));
+#ifndef MAP_ANON
+ if ((V0 = malloc(128 * r * N + 63)) == NULL)
+ goto err2;
+ V = (uint32_t *)(((uintptr_t)(V0) + 63) & ~ (uintptr_t)(63));
+#endif
+#endif
+#ifdef MAP_ANON
+ if ((V0 = mmap(NULL, 128 * r * N, PROT_READ | PROT_WRITE,
+#ifdef MAP_NOCORE
+ MAP_ANON | MAP_PRIVATE | MAP_NOCORE,
+#else
+ MAP_ANON | MAP_PRIVATE,
+#endif
+ -1, 0)) == MAP_FAILED)
+ goto err2;
+ V = (uint32_t *)(V0);
+#endif
+
+ /* 1: (B_0 ... B_{p-1}) <-- PBKDF2(P, S, 1, p * MFLen) */
+#ifdef USE_OPENSSL_PBKDF2
+ PKCS5_PBKDF2_HMAC((const char *)passwd, passwdlen, salt, saltlen, 1, EVP_sha256(), p * 128 * r, B);
+#else
+ PBKDF2_SHA256(passwd, passwdlen, salt, saltlen, 1, B, p * 128 * r);
+#endif
+
+ /* 2: for i = 0 to p - 1 do */
+ for (i = 0; i < p; i++) {
+ /* 3: B_i <-- MF(B_i, N) */
+ smix(&B[i * 128 * r], r, N, V, XY);
+ }
+
+ /* 5: DK <-- PBKDF2(P, B, 1, dkLen) */
+#ifdef USE_OPENSSL_PBKDF2
+ PKCS5_PBKDF2_HMAC((const char *)passwd, passwdlen, B, p * 128 * r, 1, EVP_sha256(), buflen, buf);
+#else
+ PBKDF2_SHA256(passwd, passwdlen, B, p * 128 * r, 1, buf, buflen);
+#endif
+
+ /* Free memory. */
+#ifdef MAP_ANON
+ if (munmap(V0, 128 * r * N))
+ goto err2;
+#else
+ free(V0);
+#endif
+ free(XY0);
+ free(B0);
+
+ /* Success! */
+ return (0);
+
+err2:
+ free(XY0);
+err1:
+ free(B0);
+err0:
+ /* Failure! */
+ return (-1);
+}
diff --git a/patches/README b/patches/README
index f606938..353ddbb 100644
--- a/patches/README
+++ b/patches/README
@@ -5,3 +5,7 @@
use_openssl_pbkdf2.patch:
Uses the PBKDF2 function from OpenSSL (it uses accelerated SHA256)
+
+arm-neon.patch:
+
+Adds NEON acceleration for the Salsa20/8 mixing function.
diff --git a/patches/arm-neon.patch b/patches/arm-neon.patch
new file mode 100644
index 0000000..7197f99
--- /dev/null
+++ b/patches/arm-neon.patch
@@ -0,0 +1,437 @@
+diff --git a/lib/crypto/crypto_scrypt-neon-salsa208.h b/lib/crypto/crypto_scrypt-neon-salsa208.h
+new file mode 100644
+index 0000000..a3b1019
+--- /dev/null
++++ b/lib/crypto/crypto_scrypt-neon-salsa208.h
+@@ -0,0 +1,120 @@
++/*
++ * version 20110505
++ * D. J. Bernstein
++ * Public domain.
++ *
++ * Based on crypto_core/salsa208/armneon/core.c from SUPERCOP 20130419
++ */
++
++#define ROUNDS 8
++static void
++salsa20_8_intrinsic(void * input)
++{
++ int i;
++
++ const uint32x4_t abab = {-1,0,-1,0};
++
++ /*
++ * This is modified since we only have one argument. Usually you'd rearrange
++ * the constant, key, and input bytes, but we just have one linear array to
++ * rearrange which is a bit easier.
++ */
++
++ /*
++ * Change the input to be diagonals as if it's a 4x4 matrix of 32-bit values.
++ */
++ uint32x4_t x0x5x10x15;
++ uint32x4_t x12x1x6x11;
++ uint32x4_t x8x13x2x7;
++ uint32x4_t x4x9x14x3;
++
++ uint32x4_t x0x1x10x11;
++ uint32x4_t x12x13x6x7;
++ uint32x4_t x8x9x2x3;
++ uint32x4_t x4x5x14x15;
++
++ uint32x4_t x0x1x2x3;
++ uint32x4_t x4x5x6x7;
++ uint32x4_t x8x9x10x11;
++ uint32x4_t x12x13x14x15;
++
++ x0x1x2x3 = vld1q_u8((uint8_t *) input);
++ x4x5x6x7 = vld1q_u8(16 + (uint8_t *) input);
++ x8x9x10x11 = vld1q_u8(32 + (uint8_t *) input);
++ x12x13x14x15 = vld1q_u8(48 + (uint8_t *) input);
++
++ x0x1x10x11 = vcombine_u32(vget_low_u32(x0x1x2x3), vget_high_u32(x8x9x10x11));
++ x4x5x14x15 = vcombine_u32(vget_low_u32(x4x5x6x7), vget_high_u32(x12x13x14x15));
++ x8x9x2x3 = vcombine_u32(vget_low_u32(x8x9x10x11), vget_high_u32(x0x1x2x3));
++ x12x13x6x7 = vcombine_u32(vget_low_u32(x12x13x14x15), vget_high_u32(x4x5x6x7));
++
++ x0x5x10x15 = vbslq_u32(abab,x0x1x10x11,x4x5x14x15);
++ x8x13x2x7 = vbslq_u32(abab,x8x9x2x3,x12x13x6x7);
++ x4x9x14x3 = vbslq_u32(abab,x4x5x14x15,x8x9x2x3);
++ x12x1x6x11 = vbslq_u32(abab,x12x13x6x7,x0x1x10x11);
++
++ uint32x4_t start0 = x0x5x10x15;
++ uint32x4_t start1 = x12x1x6x11;
++ uint32x4_t start3 = x4x9x14x3;
++ uint32x4_t start2 = x8x13x2x7;
++
++ /* From here on this should be the same as the SUPERCOP version. */
++
++ uint32x4_t diag0 = start0;
++ uint32x4_t diag1 = start1;
++ uint32x4_t diag2 = start2;
++ uint32x4_t diag3 = start3;
++
++ uint32x4_t a0;
++ uint32x4_t a1;
++ uint32x4_t a2;
++ uint32x4_t a3;
++
++ for (i = ROUNDS;i > 0;i -= 2) {
++ a0 = diag1 + diag0;
++ diag3 ^= vsriq_n_u32(vshlq_n_u32(a0,7),a0,25);
++ a1 = diag0 + diag3;
++ diag2 ^= vsriq_n_u32(vshlq_n_u32(a1,9),a1,23);
++ a2 = diag3 + diag2;
++ diag1 ^= vsriq_n_u32(vshlq_n_u32(a2,13),a2,19);
++ a3 = diag2 + diag1;
++ diag0 ^= vsriq_n_u32(vshlq_n_u32(a3,18),a3,14);
++
++ diag3 = vextq_u32(diag3,diag3,3);
++ diag2 = vextq_u32(diag2,diag2,2);
++ diag1 = vextq_u32(diag1,diag1,1);
++
++ a0 = diag3 + diag0;
++ diag1 ^= vsriq_n_u32(vshlq_n_u32(a0,7),a0,25);
++ a1 = diag0 + diag1;
++ diag2 ^= vsriq_n_u32(vshlq_n_u32(a1,9),a1,23);
++ a2 = diag1 + diag2;
++ diag3 ^= vsriq_n_u32(vshlq_n_u32(a2,13),a2,19);
++ a3 = diag2 + diag3;
++ diag0 ^= vsriq_n_u32(vshlq_n_u32(a3,18),a3,14);
++
++ diag1 = vextq_u32(diag1,diag1,3);
++ diag2 = vextq_u32(diag2,diag2,2);
++ diag3 = vextq_u32(diag3,diag3,1);
++ }
++
++ x0x5x10x15 = diag0 + start0;
++ x12x1x6x11 = diag1 + start1;
++ x8x13x2x7 = diag2 + start2;
++ x4x9x14x3 = diag3 + start3;
++
++ x0x1x10x11 = vbslq_u32(abab,x0x5x10x15,x12x1x6x11);
++ x12x13x6x7 = vbslq_u32(abab,x12x1x6x11,x8x13x2x7);
++ x8x9x2x3 = vbslq_u32(abab,x8x13x2x7,x4x9x14x3);
++ x4x5x14x15 = vbslq_u32(abab,x4x9x14x3,x0x5x10x15);
++
++ x0x1x2x3 = vcombine_u32(vget_low_u32(x0x1x10x11),vget_high_u32(x8x9x2x3));
++ x4x5x6x7 = vcombine_u32(vget_low_u32(x4x5x14x15),vget_high_u32(x12x13x6x7));
++ x8x9x10x11 = vcombine_u32(vget_low_u32(x8x9x2x3),vget_high_u32(x0x1x10x11));
++ x12x13x14x15 = vcombine_u32(vget_low_u32(x12x13x6x7),vget_high_u32(x4x5x14x15));
++
++ vst1q_u8((uint8_t *) input,(uint8x16_t) x0x1x2x3);
++ vst1q_u8(16 + (uint8_t *) input,(uint8x16_t) x4x5x6x7);
++ vst1q_u8(32 + (uint8_t *) input,(uint8x16_t) x8x9x10x11);
++ vst1q_u8(48 + (uint8_t *) input,(uint8x16_t) x12x13x14x15);
++}
+diff --git a/lib/crypto/crypto_scrypt-neon.c b/lib/crypto/crypto_scrypt-neon.c
+new file mode 100644
+index 0000000..a3bf052
+--- /dev/null
++++ b/lib/crypto/crypto_scrypt-neon.c
+@@ -0,0 +1,305 @@
++/*-
++ * Copyright 2009 Colin Percival
++ * All rights reserved.
++ *
++ * Redistribution and use in source and binary forms, with or without
++ * modification, are permitted provided that the following conditions
++ * are met:
++ * 1. Redistributions of source code must retain the above copyright
++ * notice, this list of conditions and the following disclaimer.
++ * 2. Redistributions in binary form must reproduce the above copyright
++ * notice, this list of conditions and the following disclaimer in the
++ * documentation and/or other materials provided with the distribution.
++ *
++ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
++ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
++ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
++ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
++ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
++ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
++ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
++ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
++ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
++ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
++ * SUCH DAMAGE.
++ *
++ * This file was originally written by Colin Percival as part of the Tarsnap
++ * online backup system.
++ */
++#include "scrypt_platform.h"
++
++#include <machine/cpu-features.h>
++#include <arm_neon.h>
++
++#include <errno.h>
++#include <stdint.h>
++#include <limits.h>
++#include <stdlib.h>
++#include <string.h>
++
++#ifdef USE_OPENSSL_PBKDF2
++#include <openssl/evp.h>
++#else
++#include "sha256.h"
++#endif
++#include "sysendian.h"
++
++#include "crypto_scrypt.h"
++
++#include "crypto_scrypt-neon-salsa208.h"
++
++static void blkcpy(void *, void *, size_t);
++static void blkxor(void *, void *, size_t);
++void crypto_core_salsa208_armneon2(void *);
++static void blockmix_salsa8(uint8x16_t *, uint8x16_t *, uint8x16_t *, size_t);
++static uint64_t integerify(void *, size_t);
++static void smix(uint8_t *, size_t, uint64_t, void *, void *);
++
++static void
++blkcpy(void * dest, void * src, size_t len)
++{
++ uint8x16_t * D = dest;
++ uint8x16_t * S = src;
++ size_t L = len / 16;
++ size_t i;
++
++ for (i = 0; i < L; i++)
++ D[i] = S[i];
++}
++
++static void
++blkxor(void * dest, void * src, size_t len)
++{
++ uint8x16_t * D = dest;
++ uint8x16_t * S = src;
++ size_t L = len / 16;
++ size_t i;
++
++ for (i = 0; i < L; i++)
++ D[i] = veorq_u8(D[i], S[i]);
++}
++
++/**
++ * blockmix_salsa8(B, Y, r):
++ * Compute B = BlockMix_{salsa20/8, r}(B). The input B must be 128r bytes in
++ * length; the temporary space Y must also be the same size.
++ */
++static void
++blockmix_salsa8(uint8x16_t * Bin, uint8x16_t * Bout, uint8x16_t * X, size_t r)
++{
++ size_t i;
++
++ /* 1: X <-- B_{2r - 1} */
++ blkcpy(X, &Bin[8 * r - 4], 64);
++
++ /* 2: for i = 0 to 2r - 1 do */
++ for (i = 0; i < r; i++) {
++ /* 3: X <-- H(X \xor B_i) */
++ blkxor(X, &Bin[i * 8], 64);
++ salsa20_8_intrinsic((void *) X);
++
++ /* 4: Y_i <-- X */
++ /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
++ blkcpy(&Bout[i * 4], X, 64);
++
++ /* 3: X <-- H(X \xor B_i) */
++ blkxor(X, &Bin[i * 8 + 4], 64);
++ salsa20_8_intrinsic((void *) X);
++
++ /* 4: Y_i <-- X */
++ /* 6: B' <-- (Y_0, Y_2 ... Y_{2r-2}, Y_1, Y_3 ... Y_{2r-1}) */
++ blkcpy(&Bout[(r + i) * 4], X, 64);
++ }
++}
++
++/**
++ * integerify(B, r):
++ * Return the result of parsing B_{2r-1} as a little-endian integer.
++ */
++static uint64_t
++integerify(void * B, size_t r)
++{
++ uint8_t * X = (void*)((uintptr_t)(B) + (2 * r - 1) * 64);
++
++ return (le64dec(X));
++}
++
++/**
++ * smix(B, r, N, V, XY):
++ * Compute B = SMix_r(B, N). The input B must be 128r bytes in length; the
++ * temporary storage V must be 128rN bytes in length; the temporary storage
++ * XY must be 256r bytes in length. The value N must be a power of 2.
++ */
++static void
++smix(uint8_t * B, size_t r, uint64_t N, void * V, void * XY)
++{
++ uint8x16_t * X = XY;
++ uint8x16_t * Y = (void *)((uintptr_t)(XY) + 128 * r);
++ uint8x16_t * Z = (void *)((uintptr_t)(XY) + 256 * r);
++ uint32_t * X32 = (void *)X;
++ uint64_t i, j;
++ size_t k;
++
++ /* 1: X <-- B */
++ blkcpy(X, B, 128 * r);
++
++ /* 2: for i = 0 to N - 1 do */
++ for (i = 0; i < N; i += 2) {
++ /* 3: V_i <-- X */
++ blkcpy((void *)((uintptr_t)(V) + i * 128 * r), X, 128 * r);
++
++ /* 4: X <-- H(X) */
++ blockmix_salsa8(X, Y, Z, r);
++
++ /* 3: V_i <-- X */
++ blkcpy((void *)((uintptr_t)(V) + (i + 1) * 128 * r),
++ Y, 128 * r);
++
++ /* 4: X <-- H(X) */
++ blockmix_salsa8(Y, X, Z, r);
++ }
++
++ /* 6: for i = 0 to N - 1 do */
++ for (i = 0; i < N; i += 2) {
++ /* 7: j <-- Integerify(X) mod N */
++ j = integerify(X, r) & (N - 1);
++
++ /* 8: X <-- H(X \xor V_j) */
++ blkxor(X, (void *)((uintptr_t)(V) + j * 128 * r), 128 * r);
++ blockmix_salsa8(X, Y, Z, r);
++
++ /* 7: j <-- Integerify(X) mod N */
++ j = integerify(Y, r) & (N - 1);
++
++ /* 8: X <-- H(X \xor V_j) */
++ blkxor(Y, (void *)((uintptr_t)(V) + j * 128 * r), 128 * r);
++ blockmix_salsa8(Y, X, Z, r);
++ }
++
++ /* 10: B' <-- X */
++ blkcpy(B, X, 128 * r);
++}
++
++/**
++ * crypto_scrypt(passwd, passwdlen, salt, saltlen, N, r, p, buf, buflen):
++ * Compute scrypt(passwd[0 .. passwdlen - 1], salt[0 .. saltlen - 1], N, r,
++ * p, buflen) and write the result into buf. The parameters r, p, and buflen
++ * must satisfy r * p < 2^30 and buflen <= (2^32 - 1) * 32. The parameter N
++ * must be a power of 2.
++ *
++ * Return 0 on success; or -1 on error.
++ */
++int
++crypto_scrypt(const uint8_t * passwd, size_t passwdlen,
++ const uint8_t * salt, size_t saltlen, uint64_t N, uint32_t r, uint32_t p,
++ uint8_t * buf, size_t buflen)
++{
++ void * B0, * V0, * XY0;
++ uint8_t * B;
++ uint32_t * V;
++ uint32_t * XY;
++ uint32_t i;
++
++ /* Sanity-check parameters. */
++#if SIZE_MAX > UINT32_MAX
++ if (buflen > (((uint64_t)(1) << 32) - 1) * 32) {
++ errno = EFBIG;
++ goto err0;
++ }
++#endif
++ if ((uint64_t)(r) * (uint64_t)(p) >= (1 << 30)) {
++ errno = EFBIG;
++ goto err0;
++ }
++ if (((N & (N - 1)) != 0) || (N == 0)) {
++ errno = EINVAL;
++ goto err0;
++ }
++ if ((r > SIZE_MAX / 128 / p) ||
++#if SIZE_MAX / 256 <= UINT32_MAX
++ (r > SIZE_MAX / 256) ||
++#endif
++ (N > SIZE_MAX / 128 / r)) {
++ errno = ENOMEM;
++ goto err0;
++ }
++
++ /* Allocate memory. */
++#ifdef HAVE_POSIX_MEMALIGN
++ if ((errno = posix_memalign(&B0, 64, 128 * r * p)) != 0)
++ goto err0;
++ B = (uint8_t *)(B0);
++ if ((errno = posix_memalign(&XY0, 64, 256 * r + 64)) != 0)
++ goto err1;
++ XY = (uint32_t *)(XY0);
++#ifndef MAP_ANON
++ if ((errno = posix_memalign(&V0, 64, 128 * r * N)) != 0)
++ goto err2;
++ V = (uint32_t *)(V0);
++#endif
++#else
++ if ((B0 = malloc(128 * r * p + 63)) == NULL)
++ goto err0;
++ B = (uint8_t *)(((uintptr_t)(B0) + 63) & ~ (uintptr_t)(63));
++ if ((XY0 = malloc(256 * r + 64 + 63)) == NULL)
++ goto err1;
++ XY = (uint32_t *)(((uintptr_t)(XY0) + 63) & ~ (uintptr_t)(63));
++#ifndef MAP_ANON
++ if ((V0 = malloc(128 * r * N + 63)) == NULL)
++ goto err2;
++ V = (uint32_t *)(((uintptr_t)(V0) + 63) & ~ (uintptr_t)(63));
++#endif
++#endif
++#ifdef MAP_ANON
++ if ((V0 = mmap(NULL, 128 * r * N, PROT_READ | PROT_WRITE,
++#ifdef MAP_NOCORE
++ MAP_ANON | MAP_PRIVATE | MAP_NOCORE,
++#else
++ MAP_ANON | MAP_PRIVATE,
++#endif
++ -1, 0)) == MAP_FAILED)
++ goto err2;
++ V = (uint32_t *)(V0);
++#endif
++
++ /* 1: (B_0 ... B_{p-1}) <-- PBKDF2(P, S, 1, p * MFLen) */
++#ifdef USE_OPENSSL_PBKDF2
++ PKCS5_PBKDF2_HMAC((const char *)passwd, passwdlen, salt, saltlen, 1, EVP_sha256(), p * 128 * r, B);
++#else
++ PBKDF2_SHA256(passwd, passwdlen, salt, saltlen, 1, B, p * 128 * r);
++#endif
++
++ /* 2: for i = 0 to p - 1 do */
++ for (i = 0; i < p; i++) {
++ /* 3: B_i <-- MF(B_i, N) */
++ smix(&B[i * 128 * r], r, N, V, XY);
++ }
++
++ /* 5: DK <-- PBKDF2(P, B, 1, dkLen) */
++#ifdef USE_OPENSSL_PBKDF2
++ PKCS5_PBKDF2_HMAC((const char *)passwd, passwdlen, B, p * 128 * r, 1, EVP_sha256(), buflen, buf);
++#else
++ PBKDF2_SHA256(passwd, passwdlen, B, p * 128 * r, 1, buf, buflen);
++#endif
++
++ /* Free memory. */
++#ifdef MAP_ANON
++ if (munmap(V0, 128 * r * N))
++ goto err2;
++#else
++ free(V0);
++#endif
++ free(XY0);
++ free(B0);
++
++ /* Success! */
++ return (0);
++
++err2:
++ free(XY0);
++err1:
++ free(B0);
++err0:
++ /* Failure! */
++ return (-1);
++}
diff --git a/scrypt.config b/scrypt.config
index 3d3ec90..13fd4c9 100644
--- a/scrypt.config
+++ b/scrypt.config
@@ -44,9 +44,11 @@
"
SCRYPT_SOURCES_arm="\
+lib/crypto/crypto_scrypt-neon.c \
"
SCRYPT_SOURCES_EXCLUDES_arm="\
+lib/crypto/crypto_scrypt-ref.c \
"
SCRYPT_SOURCES_mips="\
@@ -73,8 +75,14 @@
SCRYPT_PATCHES="\
use_openssl_pbkdf2.patch \
+arm-neon.patch \
"
SCRYPT_PATCHES_use_openssl_pbkdf2_SOURCES="\
lib/crypto/crypto_scrypt-ref.c \
"
+
+SCRYPT_PATCHES_bionic_SOURCES="\
+lib/crypto/crypto_scrypt-neon.c \
+lib/crypto/crypto_scrypt-neon-salsa208.h \
+"