core/cortex-m0/curve25519: Integrate fast curve25519 implementation

- Move generic implementation to curve25519-generic.o
- Always use optimized version on cortex-m0.

- Rename .s files to .S, remove unnecessary lines in assembly files.
- Rename crypto_scalarmult_curve25519 to x25519_scalar_mult to match
  the signature provided by the generic implementation.
- Replace some handcoded memcpy with function calls
- Remove unnecessary "volatile" specifications in the code.

BRANCH=none
BUG=b:62813194
TEST=To test old implementation only:
        - Increase CONFIG_RO_SIZE to 60kb
        - Increase console stack size to 2048
     make BOARD=hammer PROJECT=x25519 TEST_BUILD=y
     ./util/flash_ec --board=hammer --image=build/hammer/x25519.bin
     EC console: runtest, taskinfo
     => Used to takes ~4'17" to run (X25519 duration 256347 us).
        1496/2048 stack size usage in CONSOLE task
     => Now takes ~1'25" to run (X25519 duration 84520 us)
        732/2048 stack size usage in CONSOLE task
TEST=In test/x25519.c, uncomment #define TEST_X25519_1M_ITERATIONS
     make BOARD=hammer PROJECT=x25519 TEST_BUILD=y
     ./util/flash_ec --board=hammer --image=build/hammer/x25519.bin
     EC console: runtest, wait ~23 hours, test passes.
TEST=- Define CONFIG_CURVE25519_CORTEXM0 (next patch)
     makes newsizes
     build/hammer/RW/ec.RW.flat shrank by 1888 bytes: (52208 to 50320)

Change-Id: Icce38d3c32f431a85ac0f951cf34456b490dc665
Reviewed-on: https://chromium-review.googlesource.com/540962
Commit-Ready: Nicolas Boichat <drinkcat@chromium.org>
Tested-by: Nicolas Boichat <drinkcat@chromium.org>
Reviewed-by: Nicolas Boichat <drinkcat@chromium.org>

10 files changed, 858 insertions, 1012 deletions

diff --git a/common/build.mk b/common/build.mk
index 9954798988..7907a87e30 100644
--- a/common/build.mk
+++ b/common/build.mk
@@ -45,6 +45,9 @@ common-$(CONFIG_COMMON_RUNTIME)+=hooks.o main.o system.o
 common-$(CONFIG_COMMON_TIMER)+=timer.o
 common-$(CONFIG_CRC8)+= crc8.o
 common-$(CONFIG_CURVE25519)+=curve25519.o
+ifneq ($(CORE),cortex-m0)
+common-$(CONFIG_CURVE25519)+=curve25519-generic.o
+endif
 common-$(CONFIG_DEVICE_EVENT)+=device_event.o
 common-$(CONFIG_DEVICE_STATE)+=device_state.o
 common-$(CONFIG_DPTF)+=dptf.o
diff --git a/common/curve25519-generic.c b/common/curve25519-generic.c
new file mode 100644
index 0000000000..4bc1114546
--- /dev/null
+++ b/common/curve25519-generic.c
@@ -0,0 +1,821 @@
+/* Copyright (c) 2015, 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 code is mostly taken from the ref10 version of Ed25519 in SUPERCOP
+ * 20141124 (http://bench.cr.yp.to/supercop.html). That code is released as
+ * public domain but this file has the ISC license just to keep licencing
+ * simple.
+ *
+ * The field functions are shared by Ed25519 and X25519 where possible. */
+
+#include "curve25519.h"
+#include "util.h"
+
+/*
+ * fe means field element. Here the field is \Z/(2^255-19). An element t,
+ * entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
+ * t[3]+2^102 t[4]+...+2^230 t[9]. Bounds on each t[i] vary depending on
+ * context.
+ */
+typedef int32_t fe[10];
+
+static const int64_t kBottom25Bits = INT64_C(0x1ffffff);
+static const int64_t kBottom26Bits = INT64_C(0x3ffffff);
+static const int64_t kTop39Bits = INT64_C(0xfffffffffe000000);
+static const int64_t kTop38Bits = INT64_C(0xfffffffffc000000);
+
+static uint64_t load_3(const uint8_t *in) {
+  uint64_t result;
+  result = (uint64_t)in[0];
+  result |= ((uint64_t)in[1]) << 8;
+  result |= ((uint64_t)in[2]) << 16;
+  return result;
+}
+
+static uint64_t load_4(const uint8_t *in) {
+  uint64_t result;
+  result = (uint64_t)in[0];
+  result |= ((uint64_t)in[1]) << 8;
+  result |= ((uint64_t)in[2]) << 16;
+  result |= ((uint64_t)in[3]) << 24;
+  return result;
+}
+
+static void fe_frombytes(fe h, const uint8_t *s) {
+  /* Ignores top bit of h. */
+  int64_t h0 = load_4(s);
+  int64_t h1 = load_3(s + 4) << 6;
+  int64_t h2 = load_3(s + 7) << 5;
+  int64_t h3 = load_3(s + 10) << 3;
+  int64_t h4 = load_3(s + 13) << 2;
+  int64_t h5 = load_4(s + 16);
+  int64_t h6 = load_3(s + 20) << 7;
+  int64_t h7 = load_3(s + 23) << 5;
+  int64_t h8 = load_3(s + 26) << 4;
+  int64_t h9 = (load_3(s + 29) & 8388607) << 2;
+  int64_t carry0;
+  int64_t carry1;
+  int64_t carry2;
+  int64_t carry3;
+  int64_t carry4;
+  int64_t carry5;
+  int64_t carry6;
+  int64_t carry7;
+  int64_t carry8;
+  int64_t carry9;
+
+  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
+  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
+  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
+  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
+  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
+  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
+
+  h[0] = h0;
+  h[1] = h1;
+  h[2] = h2;
+  h[3] = h3;
+  h[4] = h4;
+  h[5] = h5;
+  h[6] = h6;
+  h[7] = h7;
+  h[8] = h8;
+  h[9] = h9;
+}
+
+/* Preconditions:
+ *  |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+ *
+ * Write p=2^255-19; q=floor(h/p).
+ * Basic claim: q = floor(2^(-255)(h + 19 2^(-25)h9 + 2^(-1))).
+ *
+ * Proof:
+ *   Have |h|<=p so |q|<=1 so |19^2 2^(-255) q|<1/4.
+ *   Also have |h-2^230 h9|<2^231 so |19 2^(-255)(h-2^230 h9)|<1/4.
+ *
+ *   Write y=2^(-1)-19^2 2^(-255)q-19 2^(-255)(h-2^230 h9).
+ *   Then 0<y<1.
+ *
+ *   Write r=h-pq.
+ *   Have 0<=r<=p-1=2^255-20.
+ *   Thus 0<=r+19(2^-255)r<r+19(2^-255)2^255<=2^255-1.
+ *
+ *   Write x=r+19(2^-255)r+y.
+ *   Then 0<x<2^255 so floor(2^(-255)x) = 0 so floor(q+2^(-255)x) = q.
+ *
+ *   Have q+2^(-255)x = 2^(-255)(h + 19 2^(-25) h9 + 2^(-1))
+ *   so floor(2^(-255)(h + 19 2^(-25) h9 + 2^(-1))) = q. */
+static void fe_tobytes(uint8_t *s, const fe h) {
+  int32_t h0 = h[0];
+  int32_t h1 = h[1];
+  int32_t h2 = h[2];
+  int32_t h3 = h[3];
+  int32_t h4 = h[4];
+  int32_t h5 = h[5];
+  int32_t h6 = h[6];
+  int32_t h7 = h[7];
+  int32_t h8 = h[8];
+  int32_t h9 = h[9];
+  int32_t q;
+
+  q = (19 * h9 + (((int32_t) 1) << 24)) >> 25;
+  q = (h0 + q) >> 26;
+  q = (h1 + q) >> 25;
+  q = (h2 + q) >> 26;
+  q = (h3 + q) >> 25;
+  q = (h4 + q) >> 26;
+  q = (h5 + q) >> 25;
+  q = (h6 + q) >> 26;
+  q = (h7 + q) >> 25;
+  q = (h8 + q) >> 26;
+  q = (h9 + q) >> 25;
+
+  /* Goal: Output h-(2^255-19)q, which is between 0 and 2^255-20. */
+  h0 += 19 * q;
+  /* Goal: Output h-2^255 q, which is between 0 and 2^255-20. */
+
+  h1 += h0 >> 26; h0 &= kBottom26Bits;
+  h2 += h1 >> 25; h1 &= kBottom25Bits;
+  h3 += h2 >> 26; h2 &= kBottom26Bits;
+  h4 += h3 >> 25; h3 &= kBottom25Bits;
+  h5 += h4 >> 26; h4 &= kBottom26Bits;
+  h6 += h5 >> 25; h5 &= kBottom25Bits;
+  h7 += h6 >> 26; h6 &= kBottom26Bits;
+  h8 += h7 >> 25; h7 &= kBottom25Bits;
+  h9 += h8 >> 26; h8 &= kBottom26Bits;
+                  h9 &= kBottom25Bits;
+                  /* h10 = carry9 */
+
+  /* Goal: Output h0+...+2^255 h10-2^255 q, which is between 0 and 2^255-20.
+   * Have h0+...+2^230 h9 between 0 and 2^255-1;
+   * evidently 2^255 h10-2^255 q = 0.
+   * Goal: Output h0+...+2^230 h9.  */
+
+  s[0] = h0 >> 0;
+  s[1] = h0 >> 8;
+  s[2] = h0 >> 16;
+  s[3] = (h0 >> 24) | ((uint32_t)(h1) << 2);
+  s[4] = h1 >> 6;
+  s[5] = h1 >> 14;
+  s[6] = (h1 >> 22) | ((uint32_t)(h2) << 3);
+  s[7] = h2 >> 5;
+  s[8] = h2 >> 13;
+  s[9] = (h2 >> 21) | ((uint32_t)(h3) << 5);
+  s[10] = h3 >> 3;
+  s[11] = h3 >> 11;
+  s[12] = (h3 >> 19) | ((uint32_t)(h4) << 6);
+  s[13] = h4 >> 2;
+  s[14] = h4 >> 10;
+  s[15] = h4 >> 18;
+  s[16] = h5 >> 0;
+  s[17] = h5 >> 8;
+  s[18] = h5 >> 16;
+  s[19] = (h5 >> 24) | ((uint32_t)(h6) << 1);
+  s[20] = h6 >> 7;
+  s[21] = h6 >> 15;
+  s[22] = (h6 >> 23) | ((uint32_t)(h7) << 3);
+  s[23] = h7 >> 5;
+  s[24] = h7 >> 13;
+  s[25] = (h7 >> 21) | ((uint32_t)(h8) << 4);
+  s[26] = h8 >> 4;
+  s[27] = h8 >> 12;
+  s[28] = (h8 >> 20) | ((uint32_t)(h9) << 6);
+  s[29] = h9 >> 2;
+  s[30] = h9 >> 10;
+  s[31] = h9 >> 18;
+}
+
+/* h = f */
+static void fe_copy(fe h, const fe f) {
+  memmove(h, f, sizeof(int32_t) * 10);
+}
+
+/* h = 0 */
+static void fe_0(fe h) { memset(h, 0, sizeof(int32_t) * 10); }
+
+/* h = 1 */
+static void fe_1(fe h) {
+  memset(h, 0, sizeof(int32_t) * 10);
+  h[0] = 1;
+}
+
+/* h = f + g
+ * Can overlap h with f or g.
+ *
+ * Preconditions:
+ *    |f| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+ *    |g| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+ *
+ * Postconditions:
+ *    |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc. */
+static void fe_add(fe h, const fe f, const fe g) {
+  unsigned i;
+  for (i = 0; i < 10; i++) {
+    h[i] = f[i] + g[i];
+  }
+}
+
+/* h = f - g
+ * Can overlap h with f or g.
+ *
+ * Preconditions:
+ *    |f| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+ *    |g| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
+ *
+ * Postconditions:
+ *    |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc. */
+static void fe_sub(fe h, const fe f, const fe g) {
+  unsigned i;
+  for (i = 0; i < 10; i++) {
+    h[i] = f[i] - g[i];
+  }
+}
+
+/* h = f * g
+ * Can overlap h with f or g.
+ *
+ * Preconditions:
+ *    |f| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
+ *    |g| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
+ *
+ * Postconditions:
+ *    |h| bounded by 1.01*2^25,1.01*2^24,1.01*2^25,1.01*2^24,etc.
+ *
+ * Notes on implementation strategy:
+ *
+ * Using schoolbook multiplication.
+ * Karatsuba would save a little in some cost models.
+ *
+ * Most multiplications by 2 and 19 are 32-bit precomputations;
+ * cheaper than 64-bit postcomputations.
+ *
+ * There is one remaining multiplication by 19 in the carry chain;
+ * one *19 precomputation can be merged into this,
+ * but the resulting data flow is considerably less clean.
+ *
+ * There are 12 carries below.
+ * 10 of them are 2-way parallelizable and vectorizable.
+ * Can get away with 11 carries, but then data flow is much deeper.
+ *
+ * With tighter constraints on inputs can squeeze carries into int32. */
+static void fe_mul(fe h, const fe f, const fe g) {
+  int32_t f0 = f[0];
+  int32_t f1 = f[1];
+  int32_t f2 = f[2];
+  int32_t f3 = f[3];
+  int32_t f4 = f[4];
+  int32_t f5 = f[5];
+  int32_t f6 = f[6];
+  int32_t f7 = f[7];
+  int32_t f8 = f[8];
+  int32_t f9 = f[9];
+  int32_t g0 = g[0];
+  int32_t g1 = g[1];
+  int32_t g2 = g[2];
+  int32_t g3 = g[3];
+  int32_t g4 = g[4];
+  int32_t g5 = g[5];
+  int32_t g6 = g[6];
+  int32_t g7 = g[7];
+  int32_t g8 = g[8];
+  int32_t g9 = g[9];
+  int32_t g1_19 = 19 * g1; /* 1.959375*2^29 */
+  int32_t g2_19 = 19 * g2; /* 1.959375*2^30; still ok */
+  int32_t g3_19 = 19 * g3;
+  int32_t g4_19 = 19 * g4;
+  int32_t g5_19 = 19 * g5;
+  int32_t g6_19 = 19 * g6;
+  int32_t g7_19 = 19 * g7;
+  int32_t g8_19 = 19 * g8;
+  int32_t g9_19 = 19 * g9;
+  int32_t f1_2 = 2 * f1;
+  int32_t f3_2 = 2 * f3;
+  int32_t f5_2 = 2 * f5;
+  int32_t f7_2 = 2 * f7;
+  int32_t f9_2 = 2 * f9;
+  int64_t f0g0    = f0   * (int64_t) g0;
+  int64_t f0g1    = f0   * (int64_t) g1;
+  int64_t f0g2    = f0   * (int64_t) g2;
+  int64_t f0g3    = f0   * (int64_t) g3;
+  int64_t f0g4    = f0   * (int64_t) g4;
+  int64_t f0g5    = f0   * (int64_t) g5;
+  int64_t f0g6    = f0   * (int64_t) g6;
+  int64_t f0g7    = f0   * (int64_t) g7;
+  int64_t f0g8    = f0   * (int64_t) g8;
+  int64_t f0g9    = f0   * (int64_t) g9;
+  int64_t f1g0    = f1   * (int64_t) g0;
+  int64_t f1g1_2  = f1_2 * (int64_t) g1;
+  int64_t f1g2    = f1   * (int64_t) g2;
+  int64_t f1g3_2  = f1_2 * (int64_t) g3;
+  int64_t f1g4    = f1   * (int64_t) g4;
+  int64_t f1g5_2  = f1_2 * (int64_t) g5;
+  int64_t f1g6    = f1   * (int64_t) g6;
+  int64_t f1g7_2  = f1_2 * (int64_t) g7;
+  int64_t f1g8    = f1   * (int64_t) g8;
+  int64_t f1g9_38 = f1_2 * (int64_t) g9_19;
+  int64_t f2g0    = f2   * (int64_t) g0;
+  int64_t f2g1    = f2   * (int64_t) g1;
+  int64_t f2g2    = f2   * (int64_t) g2;
+  int64_t f2g3    = f2   * (int64_t) g3;
+  int64_t f2g4    = f2   * (int64_t) g4;
+  int64_t f2g5    = f2   * (int64_t) g5;
+  int64_t f2g6    = f2   * (int64_t) g6;
+  int64_t f2g7    = f2   * (int64_t) g7;
+  int64_t f2g8_19 = f2   * (int64_t) g8_19;
+  int64_t f2g9_19 = f2   * (int64_t) g9_19;
+  int64_t f3g0    = f3   * (int64_t) g0;
+  int64_t f3g1_2  = f3_2 * (int64_t) g1;
+  int64_t f3g2    = f3   * (int64_t) g2;
+  int64_t f3g3_2  = f3_2 * (int64_t) g3;
+  int64_t f3g4    = f3   * (int64_t) g4;
+  int64_t f3g5_2  = f3_2 * (int64_t) g5;
+  int64_t f3g6    = f3   * (int64_t) g6;
+  int64_t f3g7_38 = f3_2 * (int64_t) g7_19;
+  int64_t f3g8_19 = f3   * (int64_t) g8_19;
+  int64_t f3g9_38 = f3_2 * (int64_t) g9_19;
+  int64_t f4g0    = f4   * (int64_t) g0;
+  int64_t f4g1    = f4   * (int64_t) g1;
+  int64_t f4g2    = f4   * (int64_t) g2;
+  int64_t f4g3    = f4   * (int64_t) g3;
+  int64_t f4g4    = f4   * (int64_t) g4;
+  int64_t f4g5    = f4   * (int64_t) g5;
+  int64_t f4g6_19 = f4   * (int64_t) g6_19;
+  int64_t f4g7_19 = f4   * (int64_t) g7_19;
+  int64_t f4g8_19 = f4   * (int64_t) g8_19;
+  int64_t f4g9_19 = f4   * (int64_t) g9_19;
+  int64_t f5g0    = f5   * (int64_t) g0;
+  int64_t f5g1_2  = f5_2 * (int64_t) g1;
+  int64_t f5g2    = f5   * (int64_t) g2;
+  int64_t f5g3_2  = f5_2 * (int64_t) g3;
+  int64_t f5g4    = f5   * (int64_t) g4;
+  int64_t f5g5_38 = f5_2 * (int64_t) g5_19;
+  int64_t f5g6_19 = f5   * (int64_t) g6_19;
+  int64_t f5g7_38 = f5_2 * (int64_t) g7_19;
+  int64_t f5g8_19 = f5   * (int64_t) g8_19;
+  int64_t f5g9_38 = f5_2 * (int64_t) g9_19;
+  int64_t f6g0    = f6   * (int64_t) g0;
+  int64_t f6g1    = f6   * (int64_t) g1;
+  int64_t f6g2    = f6   * (int64_t) g2;
+  int64_t f6g3    = f6   * (int64_t) g3;
+  int64_t f6g4_19 = f6   * (int64_t) g4_19;
+  int64_t f6g5_19 = f6   * (int64_t) g5_19;
+  int64_t f6g6_19 = f6   * (int64_t) g6_19;
+  int64_t f6g7_19 = f6   * (int64_t) g7_19;
+  int64_t f6g8_19 = f6   * (int64_t) g8_19;
+  int64_t f6g9_19 = f6   * (int64_t) g9_19;
+  int64_t f7g0    = f7   * (int64_t) g0;
+  int64_t f7g1_2  = f7_2 * (int64_t) g1;
+  int64_t f7g2    = f7   * (int64_t) g2;
+  int64_t f7g3_38 = f7_2 * (int64_t) g3_19;
+  int64_t f7g4_19 = f7   * (int64_t) g4_19;
+  int64_t f7g5_38 = f7_2 * (int64_t) g5_19;
+  int64_t f7g6_19 = f7   * (int64_t) g6_19;
+  int64_t f7g7_38 = f7_2 * (int64_t) g7_19;
+  int64_t f7g8_19 = f7   * (int64_t) g8_19;
+  int64_t f7g9_38 = f7_2 * (int64_t) g9_19;
+  int64_t f8g0    = f8   * (int64_t) g0;
+  int64_t f8g1    = f8   * (int64_t) g1;
+  int64_t f8g2_19 = f8   * (int64_t) g2_19;
+  int64_t f8g3_19 = f8   * (int64_t) g3_19;
+  int64_t f8g4_19 = f8   * (int64_t) g4_19;
+  int64_t f8g5_19 = f8   * (int64_t) g5_19;
+  int64_t f8g6_19 = f8   * (int64_t) g6_19;
+  int64_t f8g7_19 = f8   * (int64_t) g7_19;
+  int64_t f8g8_19 = f8   * (int64_t) g8_19;
+  int64_t f8g9_19 = f8   * (int64_t) g9_19;
+  int64_t f9g0    = f9   * (int64_t) g0;
+  int64_t f9g1_38 = f9_2 * (int64_t) g1_19;
+  int64_t f9g2_19 = f9   * (int64_t) g2_19;
+  int64_t f9g3_38 = f9_2 * (int64_t) g3_19;
+  int64_t f9g4_19 = f9   * (int64_t) g4_19;
+  int64_t f9g5_38 = f9_2 * (int64_t) g5_19;
+  int64_t f9g6_19 = f9   * (int64_t) g6_19;
+  int64_t f9g7_38 = f9_2 * (int64_t) g7_19;
+  int64_t f9g8_19 = f9   * (int64_t) g8_19;
+  int64_t f9g9_38 = f9_2 * (int64_t) g9_19;
+  int64_t h0 = f0g0+f1g9_38+f2g8_19+f3g7_38+f4g6_19+f5g5_38+f6g4_19+f7g3_38+f8g2_19+f9g1_38;
+  int64_t h1 = f0g1+f1g0   +f2g9_19+f3g8_19+f4g7_19+f5g6_19+f6g5_19+f7g4_19+f8g3_19+f9g2_19;
+  int64_t h2 = f0g2+f1g1_2 +f2g0   +f3g9_38+f4g8_19+f5g7_38+f6g6_19+f7g5_38+f8g4_19+f9g3_38;
+  int64_t h3 = f0g3+f1g2   +f2g1   +f3g0   +f4g9_19+f5g8_19+f6g7_19+f7g6_19+f8g5_19+f9g4_19;
+  int64_t h4 = f0g4+f1g3_2 +f2g2   +f3g1_2 +f4g0   +f5g9_38+f6g8_19+f7g7_38+f8g6_19+f9g5_38;
+  int64_t h5 = f0g5+f1g4   +f2g3   +f3g2   +f4g1   +f5g0   +f6g9_19+f7g8_19+f8g7_19+f9g6_19;
+  int64_t h6 = f0g6+f1g5_2 +f2g4   +f3g3_2 +f4g2   +f5g1_2 +f6g0   +f7g9_38+f8g8_19+f9g7_38;
+  int64_t h7 = f0g7+f1g6   +f2g5   +f3g4   +f4g3   +f5g2   +f6g1   +f7g0   +f8g9_19+f9g8_19;
+  int64_t h8 = f0g8+f1g7_2 +f2g6   +f3g5_2 +f4g4   +f5g3_2 +f6g2   +f7g1_2 +f8g0   +f9g9_38;
+  int64_t h9 = f0g9+f1g8   +f2g7   +f3g6   +f4g5   +f5g4   +f6g3   +f7g2   +f8g1   +f9g0   ;
+  int64_t carry0;
+  int64_t carry1;
+  int64_t carry2;
+  int64_t carry3;
+  int64_t carry4;
+  int64_t carry5;
+  int64_t carry6;
+  int64_t carry7;
+  int64_t carry8;
+  int64_t carry9;
+
+  /* |h0| <= (1.65*1.65*2^52*(1+19+19+19+19)+1.65*1.65*2^50*(38+38+38+38+38))
+   *   i.e. |h0| <= 1.4*2^60; narrower ranges for h2, h4, h6, h8
+   * |h1| <= (1.65*1.65*2^51*(1+1+19+19+19+19+19+19+19+19))
+   *   i.e. |h1| <= 1.7*2^59; narrower ranges for h3, h5, h7, h9 */
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+  /* |h0| <= 2^25 */
+  /* |h4| <= 2^25 */
+  /* |h1| <= 1.71*2^59 */
+  /* |h5| <= 1.71*2^59 */
+
+  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
+  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
+  /* |h1| <= 2^24; from now on fits into int32 */
+  /* |h5| <= 2^24; from now on fits into int32 */
+  /* |h2| <= 1.41*2^60 */
+  /* |h6| <= 1.41*2^60 */
+
+  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
+  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
+  /* |h2| <= 2^25; from now on fits into int32 unchanged */
+  /* |h6| <= 2^25; from now on fits into int32 unchanged */
+  /* |h3| <= 1.71*2^59 */
+  /* |h7| <= 1.71*2^59 */
+
+  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
+  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
+  /* |h3| <= 2^24; from now on fits into int32 unchanged */
+  /* |h7| <= 2^24; from now on fits into int32 unchanged */
+  /* |h4| <= 1.72*2^34 */
+  /* |h8| <= 1.41*2^60 */
+
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
+  /* |h4| <= 2^25; from now on fits into int32 unchanged */
+  /* |h8| <= 2^25; from now on fits into int32 unchanged */
+  /* |h5| <= 1.01*2^24 */
+  /* |h9| <= 1.71*2^59 */
+
+  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
+  /* |h9| <= 2^24; from now on fits into int32 unchanged */
+  /* |h0| <= 1.1*2^39 */
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+  /* |h0| <= 2^25; from now on fits into int32 unchanged */
+  /* |h1| <= 1.01*2^24 */
+
+  h[0] = h0;
+  h[1] = h1;
+  h[2] = h2;
+  h[3] = h3;
+  h[4] = h4;
+  h[5] = h5;
+  h[6] = h6;
+  h[7] = h7;
+  h[8] = h8;
+  h[9] = h9;
+}
+
+/* h = f * f
+ * Can overlap h with f.
+ *
+ * Preconditions:
+ *    |f| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
+ *
+ * Postconditions:
+ *    |h| bounded by 1.01*2^25,1.01*2^24,1.01*2^25,1.01*2^24,etc.
+ *
+ * See fe_mul.c for discussion of implementation strategy. */
+static void fe_sq(fe h, const fe f) {
+  int32_t f0 = f[0];
+  int32_t f1 = f[1];
+  int32_t f2 = f[2];
+  int32_t f3 = f[3];
+  int32_t f4 = f[4];
+  int32_t f5 = f[5];
+  int32_t f6 = f[6];
+  int32_t f7 = f[7];
+  int32_t f8 = f[8];
+  int32_t f9 = f[9];
+  int32_t f0_2 = 2 * f0;
+  int32_t f1_2 = 2 * f1;
+  int32_t f2_2 = 2 * f2;
+  int32_t f3_2 = 2 * f3;
+  int32_t f4_2 = 2 * f4;
+  int32_t f5_2 = 2 * f5;
+  int32_t f6_2 = 2 * f6;
+  int32_t f7_2 = 2 * f7;
+  int32_t f5_38 = 38 * f5; /* 1.959375*2^30 */
+  int32_t f6_19 = 19 * f6; /* 1.959375*2^30 */
+  int32_t f7_38 = 38 * f7; /* 1.959375*2^30 */
+  int32_t f8_19 = 19 * f8; /* 1.959375*2^30 */
+  int32_t f9_38 = 38 * f9; /* 1.959375*2^30 */
+  int64_t f0f0    = f0   * (int64_t) f0;
+  int64_t f0f1_2  = f0_2 * (int64_t) f1;
+  int64_t f0f2_2  = f0_2 * (int64_t) f2;
+  int64_t f0f3_2  = f0_2 * (int64_t) f3;
+  int64_t f0f4_2  = f0_2 * (int64_t) f4;
+  int64_t f0f5_2  = f0_2 * (int64_t) f5;
+  int64_t f0f6_2  = f0_2 * (int64_t) f6;
+  int64_t f0f7_2  = f0_2 * (int64_t) f7;
+  int64_t f0f8_2  = f0_2 * (int64_t) f8;
+  int64_t f0f9_2  = f0_2 * (int64_t) f9;
+  int64_t f1f1_2  = f1_2 * (int64_t) f1;
+  int64_t f1f2_2  = f1_2 * (int64_t) f2;
+  int64_t f1f3_4  = f1_2 * (int64_t) f3_2;
+  int64_t f1f4_2  = f1_2 * (int64_t) f4;
+  int64_t f1f5_4  = f1_2 * (int64_t) f5_2;
+  int64_t f1f6_2  = f1_2 * (int64_t) f6;
+  int64_t f1f7_4  = f1_2 * (int64_t) f7_2;
+  int64_t f1f8_2  = f1_2 * (int64_t) f8;
+  int64_t f1f9_76 = f1_2 * (int64_t) f9_38;
+  int64_t f2f2    = f2   * (int64_t) f2;
+  int64_t f2f3_2  = f2_2 * (int64_t) f3;
+  int64_t f2f4_2  = f2_2 * (int64_t) f4;
+  int64_t f2f5_2  = f2_2 * (int64_t) f5;
+  int64_t f2f6_2  = f2_2 * (int64_t) f6;
+  int64_t f2f7_2  = f2_2 * (int64_t) f7;
+  int64_t f2f8_38 = f2_2 * (int64_t) f8_19;
+  int64_t f2f9_38 = f2   * (int64_t) f9_38;
+  int64_t f3f3_2  = f3_2 * (int64_t) f3;
+  int64_t f3f4_2  = f3_2 * (int64_t) f4;
+  int64_t f3f5_4  = f3_2 * (int64_t) f5_2;
+  int64_t f3f6_2  = f3_2 * (int64_t) f6;
+  int64_t f3f7_76 = f3_2 * (int64_t) f7_38;
+  int64_t f3f8_38 = f3_2 * (int64_t) f8_19;
+  int64_t f3f9_76 = f3_2 * (int64_t) f9_38;
+  int64_t f4f4    = f4   * (int64_t) f4;
+  int64_t f4f5_2  = f4_2 * (int64_t) f5;
+  int64_t f4f6_38 = f4_2 * (int64_t) f6_19;
+  int64_t f4f7_38 = f4   * (int64_t) f7_38;
+  int64_t f4f8_38 = f4_2 * (int64_t) f8_19;
+  int64_t f4f9_38 = f4   * (int64_t) f9_38;
+  int64_t f5f5_38 = f5   * (int64_t) f5_38;
+  int64_t f5f6_38 = f5_2 * (int64_t) f6_19;
+  int64_t f5f7_76 = f5_2 * (int64_t) f7_38;
+  int64_t f5f8_38 = f5_2 * (int64_t) f8_19;
+  int64_t f5f9_76 = f5_2 * (int64_t) f9_38;
+  int64_t f6f6_19 = f6   * (int64_t) f6_19;
+  int64_t f6f7_38 = f6   * (int64_t) f7_38;
+  int64_t f6f8_38 = f6_2 * (int64_t) f8_19;
+  int64_t f6f9_38 = f6   * (int64_t) f9_38;
+  int64_t f7f7_38 = f7   * (int64_t) f7_38;
+  int64_t f7f8_38 = f7_2 * (int64_t) f8_19;
+  int64_t f7f9_76 = f7_2 * (int64_t) f9_38;
+  int64_t f8f8_19 = f8   * (int64_t) f8_19;
+  int64_t f8f9_38 = f8   * (int64_t) f9_38;
+  int64_t f9f9_38 = f9   * (int64_t) f9_38;
+  int64_t h0 = f0f0  +f1f9_76+f2f8_38+f3f7_76+f4f6_38+f5f5_38;
+  int64_t h1 = f0f1_2+f2f9_38+f3f8_38+f4f7_38+f5f6_38;
+  int64_t h2 = f0f2_2+f1f1_2 +f3f9_76+f4f8_38+f5f7_76+f6f6_19;
+  int64_t h3 = f0f3_2+f1f2_2 +f4f9_38+f5f8_38+f6f7_38;
+  int64_t h4 = f0f4_2+f1f3_4 +f2f2   +f5f9_76+f6f8_38+f7f7_38;
+  int64_t h5 = f0f5_2+f1f4_2 +f2f3_2 +f6f9_38+f7f8_38;
+  int64_t h6 = f0f6_2+f1f5_4 +f2f4_2 +f3f3_2 +f7f9_76+f8f8_19;
+  int64_t h7 = f0f7_2+f1f6_2 +f2f5_2 +f3f4_2 +f8f9_38;
+  int64_t h8 = f0f8_2+f1f7_4 +f2f6_2 +f3f5_4 +f4f4   +f9f9_38;
+  int64_t h9 = f0f9_2+f1f8_2 +f2f7_2 +f3f6_2 +f4f5_2;
+  int64_t carry0;
+  int64_t carry1;
+  int64_t carry2;
+  int64_t carry3;
+  int64_t carry4;
+  int64_t carry5;
+  int64_t carry6;
+  int64_t carry7;
+  int64_t carry8;
+  int64_t carry9;
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+
+  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
+  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
+
+  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
+  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
+
+  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
+  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
+
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
+
+  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+
+  h[0] = h0;
+  h[1] = h1;
+  h[2] = h2;
+  h[3] = h3;
+  h[4] = h4;
+  h[5] = h5;
+  h[6] = h6;
+  h[7] = h7;
+  h[8] = h8;
+  h[9] = h9;
+}
+
+static void fe_invert(fe out, const fe z) {
+  fe t0;
+  fe t1;
+  fe t2;
+  fe t3;
+  int i;
+
+  fe_sq(t0, z);
+  fe_sq(t1, t0);
+  for (i = 1; i < 2; ++i) {
+    fe_sq(t1, t1);
+  }
+  fe_mul(t1, z, t1);
+  fe_mul(t0, t0, t1);
+  fe_sq(t2, t0);
+  fe_mul(t1, t1, t2);
+  fe_sq(t2, t1);
+  for (i = 1; i < 5; ++i) {
+    fe_sq(t2, t2);
+  }
+  fe_mul(t1, t2, t1);
+  fe_sq(t2, t1);
+  for (i = 1; i < 10; ++i) {
+    fe_sq(t2, t2);
+  }
+  fe_mul(t2, t2, t1);
+  fe_sq(t3, t2);
+  for (i = 1; i < 20; ++i) {
+    fe_sq(t3, t3);
+  }
+  fe_mul(t2, t3, t2);
+  fe_sq(t2, t2);
+  for (i = 1; i < 10; ++i) {
+    fe_sq(t2, t2);
+  }
+  fe_mul(t1, t2, t1);
+  fe_sq(t2, t1);
+  for (i = 1; i < 50; ++i) {
+    fe_sq(t2, t2);
+  }
+  fe_mul(t2, t2, t1);
+  fe_sq(t3, t2);
+  for (i = 1; i < 100; ++i) {
+    fe_sq(t3, t3);
+  }
+  fe_mul(t2, t3, t2);
+  fe_sq(t2, t2);
+  for (i = 1; i < 50; ++i) {
+    fe_sq(t2, t2);
+  }
+  fe_mul(t1, t2, t1);
+  fe_sq(t1, t1);
+  for (i = 1; i < 5; ++i) {
+    fe_sq(t1, t1);
+  }
+  fe_mul(out, t1, t0);
+}
+
+/* Replace (f,g) with (g,f) if b == 1;
+ * replace (f,g) with (f,g) if b == 0.
+ *
+ * Preconditions: b in {0,1}. */
+static void fe_cswap(fe f, fe g, unsigned int b) {
+  unsigned i;
+  b = 0-b;
+  for (i = 0; i < 10; i++) {
+    int32_t x = f[i] ^ g[i];
+    x &= b;
+    f[i] ^= x;
+    g[i] ^= x;
+  }
+}
+
+/* h = f * 121666
+ * Can overlap h with f.
+ *
+ * Preconditions:
+ *    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+ *
+ * Postconditions:
+ *    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc. */
+static void fe_mul121666(fe h, fe f) {
+  int32_t f0 = f[0];
+  int32_t f1 = f[1];
+  int32_t f2 = f[2];
+  int32_t f3 = f[3];
+  int32_t f4 = f[4];
+  int32_t f5 = f[5];
+  int32_t f6 = f[6];
+  int32_t f7 = f[7];
+  int32_t f8 = f[8];
+  int32_t f9 = f[9];
+  int64_t h0 = f0 * (int64_t) 121666;
+  int64_t h1 = f1 * (int64_t) 121666;
+  int64_t h2 = f2 * (int64_t) 121666;
+  int64_t h3 = f3 * (int64_t) 121666;
+  int64_t h4 = f4 * (int64_t) 121666;
+  int64_t h5 = f5 * (int64_t) 121666;
+  int64_t h6 = f6 * (int64_t) 121666;
+  int64_t h7 = f7 * (int64_t) 121666;
+  int64_t h8 = f8 * (int64_t) 121666;
+  int64_t h9 = f9 * (int64_t) 121666;
+  int64_t carry0;
+  int64_t carry1;
+  int64_t carry2;
+  int64_t carry3;
+  int64_t carry4;
+  int64_t carry5;
+  int64_t carry6;
+  int64_t carry7;
+  int64_t carry8;
+  int64_t carry9;
+
+  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
+  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
+  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
+  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
+  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
+
+  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
+  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
+  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
+  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
+  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
+
+  h[0] = h0;
+  h[1] = h1;
+  h[2] = h2;
+  h[3] = h3;
+  h[4] = h4;
+  h[5] = h5;
+  h[6] = h6;
+  h[7] = h7;
+  h[8] = h8;
+  h[9] = h9;
+}
+
+void x25519_scalar_mult(uint8_t out[32],
+			const uint8_t scalar[32],
+			const uint8_t point[32]) {
+  fe x1, x2, z2, x3, z3, tmp0, tmp1;
+  unsigned swap;
+  int pos;
+
+  uint8_t e[32];
+  memcpy(e, scalar, 32);
+  e[0] &= 248;
+  e[31] &= 127;
+  e[31] |= 64;
+  fe_frombytes(x1, point);
+  fe_1(x2);
+  fe_0(z2);
+  fe_copy(x3, x1);
+  fe_1(z3);
+
+  swap = 0;
+  for (pos = 254; pos >= 0; --pos) {
+    unsigned b = 1 & (e[pos / 8] >> (pos & 7));
+    swap ^= b;
+    fe_cswap(x2, x3, swap);
+    fe_cswap(z2, z3, swap);
+    swap = b;
+    fe_sub(tmp0, x3, z3);
+    fe_sub(tmp1, x2, z2);
+    fe_add(x2, x2, z2);
+    fe_add(z2, x3, z3);
+    fe_mul(z3, tmp0, x2);
+    fe_mul(z2, z2, tmp1);
+    fe_sq(tmp0, tmp1);
+    fe_sq(tmp1, x2);
+    fe_add(x3, z3, z2);
+    fe_sub(z2, z3, z2);
+    fe_mul(x2, tmp1, tmp0);
+    fe_sub(tmp1, tmp1, tmp0);
+    fe_sq(z2, z2);
+    fe_mul121666(z3, tmp1);
+    fe_sq(x3, x3);
+    fe_add(tmp0, tmp0, z3);
+    fe_mul(z3, x1, z2);
+    fe_mul(z2, tmp1, tmp0);
+  }
+  fe_cswap(x2, x3, swap);
+  fe_cswap(z2, z3, swap);
+
+  fe_invert(z2, z2);
+  fe_mul(x2, x2, z2);
+  fe_tobytes(out, x2);
+}
diff --git a/common/curve25519.c b/common/curve25519.c
index 5faa53dc11..5e3a8b81f6 100644
--- a/common/curve25519.c
+++ b/common/curve25519.c
@@ -19,815 +19,12 @@
  *
  * The field functions are shared by Ed25519 and X25519 where possible. */
 
+#include "common.h"
 #include "curve25519.h"
 #include "trng.h"
 #include "util.h"
 #define CRYPTO_memcmp safe_memcmp
 
-
-/*
- * fe means field element. Here the field is \Z/(2^255-19). An element t,
- * entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
- * t[3]+2^102 t[4]+...+2^230 t[9]. Bounds on each t[i] vary depending on
- * context.
- */
-typedef int32_t fe[10];
-
-static const int64_t kBottom25Bits = INT64_C(0x1ffffff);
-static const int64_t kBottom26Bits = INT64_C(0x3ffffff);
-static const int64_t kTop39Bits = INT64_C(0xfffffffffe000000);
-static const int64_t kTop38Bits = INT64_C(0xfffffffffc000000);
-
-static uint64_t load_3(const uint8_t *in) {
-  uint64_t result;
-  result = (uint64_t)in[0];
-  result |= ((uint64_t)in[1]) << 8;
-  result |= ((uint64_t)in[2]) << 16;
-  return result;
-}
-
-static uint64_t load_4(const uint8_t *in) {
-  uint64_t result;
-  result = (uint64_t)in[0];
-  result |= ((uint64_t)in[1]) << 8;
-  result |= ((uint64_t)in[2]) << 16;
-  result |= ((uint64_t)in[3]) << 24;
-  return result;
-}
-
-static void fe_frombytes(fe h, const uint8_t *s) {
-  /* Ignores top bit of h. */
-  int64_t h0 = load_4(s);
-  int64_t h1 = load_3(s + 4) << 6;
-  int64_t h2 = load_3(s + 7) << 5;
-  int64_t h3 = load_3(s + 10) << 3;
-  int64_t h4 = load_3(s + 13) << 2;
-  int64_t h5 = load_4(s + 16);
-  int64_t h6 = load_3(s + 20) << 7;
-  int64_t h7 = load_3(s + 23) << 5;
-  int64_t h8 = load_3(s + 26) << 4;
-  int64_t h9 = (load_3(s + 29) & 8388607) << 2;
-  int64_t carry0;
-  int64_t carry1;
-  int64_t carry2;
-  int64_t carry3;
-  int64_t carry4;
-  int64_t carry5;
-  int64_t carry6;
-  int64_t carry7;
-  int64_t carry8;
-  int64_t carry9;
-
-  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
-  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
-  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
-  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
-  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
-  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
-
-  h[0] = h0;
-  h[1] = h1;
-  h[2] = h2;
-  h[3] = h3;
-  h[4] = h4;
-  h[5] = h5;
-  h[6] = h6;
-  h[7] = h7;
-  h[8] = h8;
-  h[9] = h9;
-}
-
-/* Preconditions:
- *  |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
- *
- * Write p=2^255-19; q=floor(h/p).
- * Basic claim: q = floor(2^(-255)(h + 19 2^(-25)h9 + 2^(-1))).
- *
- * Proof:
- *   Have |h|<=p so |q|<=1 so |19^2 2^(-255) q|<1/4.
- *   Also have |h-2^230 h9|<2^231 so |19 2^(-255)(h-2^230 h9)|<1/4.
- *
- *   Write y=2^(-1)-19^2 2^(-255)q-19 2^(-255)(h-2^230 h9).
- *   Then 0<y<1.
- *
- *   Write r=h-pq.
- *   Have 0<=r<=p-1=2^255-20.
- *   Thus 0<=r+19(2^-255)r<r+19(2^-255)2^255<=2^255-1.
- *
- *   Write x=r+19(2^-255)r+y.
- *   Then 0<x<2^255 so floor(2^(-255)x) = 0 so floor(q+2^(-255)x) = q.
- *
- *   Have q+2^(-255)x = 2^(-255)(h + 19 2^(-25) h9 + 2^(-1))
- *   so floor(2^(-255)(h + 19 2^(-25) h9 + 2^(-1))) = q. */
-static void fe_tobytes(uint8_t *s, const fe h) {
-  int32_t h0 = h[0];
-  int32_t h1 = h[1];
-  int32_t h2 = h[2];
-  int32_t h3 = h[3];
-  int32_t h4 = h[4];
-  int32_t h5 = h[5];
-  int32_t h6 = h[6];
-  int32_t h7 = h[7];
-  int32_t h8 = h[8];
-  int32_t h9 = h[9];
-  int32_t q;
-
-  q = (19 * h9 + (((int32_t) 1) << 24)) >> 25;
-  q = (h0 + q) >> 26;
-  q = (h1 + q) >> 25;
-  q = (h2 + q) >> 26;
-  q = (h3 + q) >> 25;
-  q = (h4 + q) >> 26;
-  q = (h5 + q) >> 25;
-  q = (h6 + q) >> 26;
-  q = (h7 + q) >> 25;
-  q = (h8 + q) >> 26;
-  q = (h9 + q) >> 25;
-
-  /* Goal: Output h-(2^255-19)q, which is between 0 and 2^255-20. */
-  h0 += 19 * q;
-  /* Goal: Output h-2^255 q, which is between 0 and 2^255-20. */
-
-  h1 += h0 >> 26; h0 &= kBottom26Bits;
-  h2 += h1 >> 25; h1 &= kBottom25Bits;
-  h3 += h2 >> 26; h2 &= kBottom26Bits;
-  h4 += h3 >> 25; h3 &= kBottom25Bits;
-  h5 += h4 >> 26; h4 &= kBottom26Bits;
-  h6 += h5 >> 25; h5 &= kBottom25Bits;
-  h7 += h6 >> 26; h6 &= kBottom26Bits;
-  h8 += h7 >> 25; h7 &= kBottom25Bits;
-  h9 += h8 >> 26; h8 &= kBottom26Bits;
-                  h9 &= kBottom25Bits;
-                  /* h10 = carry9 */
-
-  /* Goal: Output h0+...+2^255 h10-2^255 q, which is between 0 and 2^255-20.
-   * Have h0+...+2^230 h9 between 0 and 2^255-1;
-   * evidently 2^255 h10-2^255 q = 0.
-   * Goal: Output h0+...+2^230 h9.  */
-
-  s[0] = h0 >> 0;
-  s[1] = h0 >> 8;
-  s[2] = h0 >> 16;
-  s[3] = (h0 >> 24) | ((uint32_t)(h1) << 2);
-  s[4] = h1 >> 6;
-  s[5] = h1 >> 14;
-  s[6] = (h1 >> 22) | ((uint32_t)(h2) << 3);
-  s[7] = h2 >> 5;
-  s[8] = h2 >> 13;
-  s[9] = (h2 >> 21) | ((uint32_t)(h3) << 5);
-  s[10] = h3 >> 3;
-  s[11] = h3 >> 11;
-  s[12] = (h3 >> 19) | ((uint32_t)(h4) << 6);
-  s[13] = h4 >> 2;
-  s[14] = h4 >> 10;
-  s[15] = h4 >> 18;
-  s[16] = h5 >> 0;
-  s[17] = h5 >> 8;
-  s[18] = h5 >> 16;
-  s[19] = (h5 >> 24) | ((uint32_t)(h6) << 1);
-  s[20] = h6 >> 7;
-  s[21] = h6 >> 15;
-  s[22] = (h6 >> 23) | ((uint32_t)(h7) << 3);
-  s[23] = h7 >> 5;
-  s[24] = h7 >> 13;
-  s[25] = (h7 >> 21) | ((uint32_t)(h8) << 4);
-  s[26] = h8 >> 4;
-  s[27] = h8 >> 12;
-  s[28] = (h8 >> 20) | ((uint32_t)(h9) << 6);
-  s[29] = h9 >> 2;
-  s[30] = h9 >> 10;
-  s[31] = h9 >> 18;
-}
-
-/* h = f */
-static void fe_copy(fe h, const fe f) {
-  memmove(h, f, sizeof(int32_t) * 10);
-}
-
-/* h = 0 */
-static void fe_0(fe h) { memset(h, 0, sizeof(int32_t) * 10); }
-
-/* h = 1 */
-static void fe_1(fe h) {
-  memset(h, 0, sizeof(int32_t) * 10);
-  h[0] = 1;
-}
-
-/* h = f + g
- * Can overlap h with f or g.
- *
- * Preconditions:
- *    |f| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
- *    |g| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
- *
- * Postconditions:
- *    |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc. */
-static void fe_add(fe h, const fe f, const fe g) {
-  unsigned i;
-  for (i = 0; i < 10; i++) {
-    h[i] = f[i] + g[i];
-  }
-}
-
-/* h = f - g
- * Can overlap h with f or g.
- *
- * Preconditions:
- *    |f| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
- *    |g| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
- *
- * Postconditions:
- *    |h| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc. */
-static void fe_sub(fe h, const fe f, const fe g) {
-  unsigned i;
-  for (i = 0; i < 10; i++) {
-    h[i] = f[i] - g[i];
-  }
-}
-
-/* h = f * g
- * Can overlap h with f or g.
- *
- * Preconditions:
- *    |f| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
- *    |g| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
- *
- * Postconditions:
- *    |h| bounded by 1.01*2^25,1.01*2^24,1.01*2^25,1.01*2^24,etc.
- *
- * Notes on implementation strategy:
- *
- * Using schoolbook multiplication.
- * Karatsuba would save a little in some cost models.
- *
- * Most multiplications by 2 and 19 are 32-bit precomputations;
- * cheaper than 64-bit postcomputations.
- *
- * There is one remaining multiplication by 19 in the carry chain;
- * one *19 precomputation can be merged into this,
- * but the resulting data flow is considerably less clean.
- *
- * There are 12 carries below.
- * 10 of them are 2-way parallelizable and vectorizable.
- * Can get away with 11 carries, but then data flow is much deeper.
- *
- * With tighter constraints on inputs can squeeze carries into int32. */
-static void fe_mul(fe h, const fe f, const fe g) {
-  int32_t f0 = f[0];
-  int32_t f1 = f[1];
-  int32_t f2 = f[2];
-  int32_t f3 = f[3];
-  int32_t f4 = f[4];
-  int32_t f5 = f[5];
-  int32_t f6 = f[6];
-  int32_t f7 = f[7];
-  int32_t f8 = f[8];
-  int32_t f9 = f[9];
-  int32_t g0 = g[0];
-  int32_t g1 = g[1];
-  int32_t g2 = g[2];
-  int32_t g3 = g[3];
-  int32_t g4 = g[4];
-  int32_t g5 = g[5];
-  int32_t g6 = g[6];
-  int32_t g7 = g[7];
-  int32_t g8 = g[8];
-  int32_t g9 = g[9];
-  int32_t g1_19 = 19 * g1; /* 1.959375*2^29 */
-  int32_t g2_19 = 19 * g2; /* 1.959375*2^30; still ok */
-  int32_t g3_19 = 19 * g3;
-  int32_t g4_19 = 19 * g4;
-  int32_t g5_19 = 19 * g5;
-  int32_t g6_19 = 19 * g6;
-  int32_t g7_19 = 19 * g7;
-  int32_t g8_19 = 19 * g8;
-  int32_t g9_19 = 19 * g9;
-  int32_t f1_2 = 2 * f1;
-  int32_t f3_2 = 2 * f3;
-  int32_t f5_2 = 2 * f5;
-  int32_t f7_2 = 2 * f7;
-  int32_t f9_2 = 2 * f9;
-  int64_t f0g0    = f0   * (int64_t) g0;
-  int64_t f0g1    = f0   * (int64_t) g1;
-  int64_t f0g2    = f0   * (int64_t) g2;
-  int64_t f0g3    = f0   * (int64_t) g3;
-  int64_t f0g4    = f0   * (int64_t) g4;
-  int64_t f0g5    = f0   * (int64_t) g5;
-  int64_t f0g6    = f0   * (int64_t) g6;
-  int64_t f0g7    = f0   * (int64_t) g7;
-  int64_t f0g8    = f0   * (int64_t) g8;
-  int64_t f0g9    = f0   * (int64_t) g9;
-  int64_t f1g0    = f1   * (int64_t) g0;
-  int64_t f1g1_2  = f1_2 * (int64_t) g1;
-  int64_t f1g2    = f1   * (int64_t) g2;
-  int64_t f1g3_2  = f1_2 * (int64_t) g3;
-  int64_t f1g4    = f1   * (int64_t) g4;
-  int64_t f1g5_2  = f1_2 * (int64_t) g5;
-  int64_t f1g6    = f1   * (int64_t) g6;
-  int64_t f1g7_2  = f1_2 * (int64_t) g7;
-  int64_t f1g8    = f1   * (int64_t) g8;
-  int64_t f1g9_38 = f1_2 * (int64_t) g9_19;
-  int64_t f2g0    = f2   * (int64_t) g0;
-  int64_t f2g1    = f2   * (int64_t) g1;
-  int64_t f2g2    = f2   * (int64_t) g2;
-  int64_t f2g3    = f2   * (int64_t) g3;
-  int64_t f2g4    = f2   * (int64_t) g4;
-  int64_t f2g5    = f2   * (int64_t) g5;
-  int64_t f2g6    = f2   * (int64_t) g6;
-  int64_t f2g7    = f2   * (int64_t) g7;
-  int64_t f2g8_19 = f2   * (int64_t) g8_19;
-  int64_t f2g9_19 = f2   * (int64_t) g9_19;
-  int64_t f3g0    = f3   * (int64_t) g0;
-  int64_t f3g1_2  = f3_2 * (int64_t) g1;
-  int64_t f3g2    = f3   * (int64_t) g2;
-  int64_t f3g3_2  = f3_2 * (int64_t) g3;
-  int64_t f3g4    = f3   * (int64_t) g4;
-  int64_t f3g5_2  = f3_2 * (int64_t) g5;
-  int64_t f3g6    = f3   * (int64_t) g6;
-  int64_t f3g7_38 = f3_2 * (int64_t) g7_19;
-  int64_t f3g8_19 = f3   * (int64_t) g8_19;
-  int64_t f3g9_38 = f3_2 * (int64_t) g9_19;
-  int64_t f4g0    = f4   * (int64_t) g0;
-  int64_t f4g1    = f4   * (int64_t) g1;
-  int64_t f4g2    = f4   * (int64_t) g2;
-  int64_t f4g3    = f4   * (int64_t) g3;
-  int64_t f4g4    = f4   * (int64_t) g4;
-  int64_t f4g5    = f4   * (int64_t) g5;
-  int64_t f4g6_19 = f4   * (int64_t) g6_19;
-  int64_t f4g7_19 = f4   * (int64_t) g7_19;
-  int64_t f4g8_19 = f4   * (int64_t) g8_19;
-  int64_t f4g9_19 = f4   * (int64_t) g9_19;
-  int64_t f5g0    = f5   * (int64_t) g0;
-  int64_t f5g1_2  = f5_2 * (int64_t) g1;
-  int64_t f5g2    = f5   * (int64_t) g2;
-  int64_t f5g3_2  = f5_2 * (int64_t) g3;
-  int64_t f5g4    = f5   * (int64_t) g4;
-  int64_t f5g5_38 = f5_2 * (int64_t) g5_19;
-  int64_t f5g6_19 = f5   * (int64_t) g6_19;
-  int64_t f5g7_38 = f5_2 * (int64_t) g7_19;
-  int64_t f5g8_19 = f5   * (int64_t) g8_19;
-  int64_t f5g9_38 = f5_2 * (int64_t) g9_19;
-  int64_t f6g0    = f6   * (int64_t) g0;
-  int64_t f6g1    = f6   * (int64_t) g1;
-  int64_t f6g2    = f6   * (int64_t) g2;
-  int64_t f6g3    = f6   * (int64_t) g3;
-  int64_t f6g4_19 = f6   * (int64_t) g4_19;
-  int64_t f6g5_19 = f6   * (int64_t) g5_19;
-  int64_t f6g6_19 = f6   * (int64_t) g6_19;
-  int64_t f6g7_19 = f6   * (int64_t) g7_19;
-  int64_t f6g8_19 = f6   * (int64_t) g8_19;
-  int64_t f6g9_19 = f6   * (int64_t) g9_19;
-  int64_t f7g0    = f7   * (int64_t) g0;
-  int64_t f7g1_2  = f7_2 * (int64_t) g1;
-  int64_t f7g2    = f7   * (int64_t) g2;
-  int64_t f7g3_38 = f7_2 * (int64_t) g3_19;
-  int64_t f7g4_19 = f7   * (int64_t) g4_19;
-  int64_t f7g5_38 = f7_2 * (int64_t) g5_19;
-  int64_t f7g6_19 = f7   * (int64_t) g6_19;
-  int64_t f7g7_38 = f7_2 * (int64_t) g7_19;
-  int64_t f7g8_19 = f7   * (int64_t) g8_19;
-  int64_t f7g9_38 = f7_2 * (int64_t) g9_19;
-  int64_t f8g0    = f8   * (int64_t) g0;
-  int64_t f8g1    = f8   * (int64_t) g1;
-  int64_t f8g2_19 = f8   * (int64_t) g2_19;
-  int64_t f8g3_19 = f8   * (int64_t) g3_19;
-  int64_t f8g4_19 = f8   * (int64_t) g4_19;
-  int64_t f8g5_19 = f8   * (int64_t) g5_19;
-  int64_t f8g6_19 = f8   * (int64_t) g6_19;
-  int64_t f8g7_19 = f8   * (int64_t) g7_19;
-  int64_t f8g8_19 = f8   * (int64_t) g8_19;
-  int64_t f8g9_19 = f8   * (int64_t) g9_19;
-  int64_t f9g0    = f9   * (int64_t) g0;
-  int64_t f9g1_38 = f9_2 * (int64_t) g1_19;
-  int64_t f9g2_19 = f9   * (int64_t) g2_19;
-  int64_t f9g3_38 = f9_2 * (int64_t) g3_19;
-  int64_t f9g4_19 = f9   * (int64_t) g4_19;
-  int64_t f9g5_38 = f9_2 * (int64_t) g5_19;
-  int64_t f9g6_19 = f9   * (int64_t) g6_19;
-  int64_t f9g7_38 = f9_2 * (int64_t) g7_19;
-  int64_t f9g8_19 = f9   * (int64_t) g8_19;
-  int64_t f9g9_38 = f9_2 * (int64_t) g9_19;
-  int64_t h0 = f0g0+f1g9_38+f2g8_19+f3g7_38+f4g6_19+f5g5_38+f6g4_19+f7g3_38+f8g2_19+f9g1_38;
-  int64_t h1 = f0g1+f1g0   +f2g9_19+f3g8_19+f4g7_19+f5g6_19+f6g5_19+f7g4_19+f8g3_19+f9g2_19;
-  int64_t h2 = f0g2+f1g1_2 +f2g0   +f3g9_38+f4g8_19+f5g7_38+f6g6_19+f7g5_38+f8g4_19+f9g3_38;
-  int64_t h3 = f0g3+f1g2   +f2g1   +f3g0   +f4g9_19+f5g8_19+f6g7_19+f7g6_19+f8g5_19+f9g4_19;
-  int64_t h4 = f0g4+f1g3_2 +f2g2   +f3g1_2 +f4g0   +f5g9_38+f6g8_19+f7g7_38+f8g6_19+f9g5_38;
-  int64_t h5 = f0g5+f1g4   +f2g3   +f3g2   +f4g1   +f5g0   +f6g9_19+f7g8_19+f8g7_19+f9g6_19;
-  int64_t h6 = f0g6+f1g5_2 +f2g4   +f3g3_2 +f4g2   +f5g1_2 +f6g0   +f7g9_38+f8g8_19+f9g7_38;
-  int64_t h7 = f0g7+f1g6   +f2g5   +f3g4   +f4g3   +f5g2   +f6g1   +f7g0   +f8g9_19+f9g8_19;
-  int64_t h8 = f0g8+f1g7_2 +f2g6   +f3g5_2 +f4g4   +f5g3_2 +f6g2   +f7g1_2 +f8g0   +f9g9_38;
-  int64_t h9 = f0g9+f1g8   +f2g7   +f3g6   +f4g5   +f5g4   +f6g3   +f7g2   +f8g1   +f9g0   ;
-  int64_t carry0;
-  int64_t carry1;
-  int64_t carry2;
-  int64_t carry3;
-  int64_t carry4;
-  int64_t carry5;
-  int64_t carry6;
-  int64_t carry7;
-  int64_t carry8;
-  int64_t carry9;
-
-  /* |h0| <= (1.65*1.65*2^52*(1+19+19+19+19)+1.65*1.65*2^50*(38+38+38+38+38))
-   *   i.e. |h0| <= 1.4*2^60; narrower ranges for h2, h4, h6, h8
-   * |h1| <= (1.65*1.65*2^51*(1+1+19+19+19+19+19+19+19+19))
-   *   i.e. |h1| <= 1.7*2^59; narrower ranges for h3, h5, h7, h9 */
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-  /* |h0| <= 2^25 */
-  /* |h4| <= 2^25 */
-  /* |h1| <= 1.71*2^59 */
-  /* |h5| <= 1.71*2^59 */
-
-  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
-  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
-  /* |h1| <= 2^24; from now on fits into int32 */
-  /* |h5| <= 2^24; from now on fits into int32 */
-  /* |h2| <= 1.41*2^60 */
-  /* |h6| <= 1.41*2^60 */
-
-  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
-  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
-  /* |h2| <= 2^25; from now on fits into int32 unchanged */
-  /* |h6| <= 2^25; from now on fits into int32 unchanged */
-  /* |h3| <= 1.71*2^59 */
-  /* |h7| <= 1.71*2^59 */
-
-  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
-  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
-  /* |h3| <= 2^24; from now on fits into int32 unchanged */
-  /* |h7| <= 2^24; from now on fits into int32 unchanged */
-  /* |h4| <= 1.72*2^34 */
-  /* |h8| <= 1.41*2^60 */
-
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
-  /* |h4| <= 2^25; from now on fits into int32 unchanged */
-  /* |h8| <= 2^25; from now on fits into int32 unchanged */
-  /* |h5| <= 1.01*2^24 */
-  /* |h9| <= 1.71*2^59 */
-
-  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
-  /* |h9| <= 2^24; from now on fits into int32 unchanged */
-  /* |h0| <= 1.1*2^39 */
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-  /* |h0| <= 2^25; from now on fits into int32 unchanged */
-  /* |h1| <= 1.01*2^24 */
-
-  h[0] = h0;
-  h[1] = h1;
-  h[2] = h2;
-  h[3] = h3;
-  h[4] = h4;
-  h[5] = h5;
-  h[6] = h6;
-  h[7] = h7;
-  h[8] = h8;
-  h[9] = h9;
-}
-
-/* h = f * f
- * Can overlap h with f.
- *
- * Preconditions:
- *    |f| bounded by 1.65*2^26,1.65*2^25,1.65*2^26,1.65*2^25,etc.
- *
- * Postconditions:
- *    |h| bounded by 1.01*2^25,1.01*2^24,1.01*2^25,1.01*2^24,etc.
- *
- * See fe_mul.c for discussion of implementation strategy. */
-static void fe_sq(fe h, const fe f) {
-  int32_t f0 = f[0];
-  int32_t f1 = f[1];
-  int32_t f2 = f[2];
-  int32_t f3 = f[3];
-  int32_t f4 = f[4];
-  int32_t f5 = f[5];
-  int32_t f6 = f[6];
-  int32_t f7 = f[7];
-  int32_t f8 = f[8];
-  int32_t f9 = f[9];
-  int32_t f0_2 = 2 * f0;
-  int32_t f1_2 = 2 * f1;
-  int32_t f2_2 = 2 * f2;
-  int32_t f3_2 = 2 * f3;
-  int32_t f4_2 = 2 * f4;
-  int32_t f5_2 = 2 * f5;
-  int32_t f6_2 = 2 * f6;
-  int32_t f7_2 = 2 * f7;
-  int32_t f5_38 = 38 * f5; /* 1.959375*2^30 */
-  int32_t f6_19 = 19 * f6; /* 1.959375*2^30 */
-  int32_t f7_38 = 38 * f7; /* 1.959375*2^30 */
-  int32_t f8_19 = 19 * f8; /* 1.959375*2^30 */
-  int32_t f9_38 = 38 * f9; /* 1.959375*2^30 */
-  int64_t f0f0    = f0   * (int64_t) f0;
-  int64_t f0f1_2  = f0_2 * (int64_t) f1;
-  int64_t f0f2_2  = f0_2 * (int64_t) f2;
-  int64_t f0f3_2  = f0_2 * (int64_t) f3;
-  int64_t f0f4_2  = f0_2 * (int64_t) f4;
-  int64_t f0f5_2  = f0_2 * (int64_t) f5;
-  int64_t f0f6_2  = f0_2 * (int64_t) f6;
-  int64_t f0f7_2  = f0_2 * (int64_t) f7;
-  int64_t f0f8_2  = f0_2 * (int64_t) f8;
-  int64_t f0f9_2  = f0_2 * (int64_t) f9;
-  int64_t f1f1_2  = f1_2 * (int64_t) f1;
-  int64_t f1f2_2  = f1_2 * (int64_t) f2;
-  int64_t f1f3_4  = f1_2 * (int64_t) f3_2;
-  int64_t f1f4_2  = f1_2 * (int64_t) f4;
-  int64_t f1f5_4  = f1_2 * (int64_t) f5_2;
-  int64_t f1f6_2  = f1_2 * (int64_t) f6;
-  int64_t f1f7_4  = f1_2 * (int64_t) f7_2;
-  int64_t f1f8_2  = f1_2 * (int64_t) f8;
-  int64_t f1f9_76 = f1_2 * (int64_t) f9_38;
-  int64_t f2f2    = f2   * (int64_t) f2;
-  int64_t f2f3_2  = f2_2 * (int64_t) f3;
-  int64_t f2f4_2  = f2_2 * (int64_t) f4;
-  int64_t f2f5_2  = f2_2 * (int64_t) f5;
-  int64_t f2f6_2  = f2_2 * (int64_t) f6;
-  int64_t f2f7_2  = f2_2 * (int64_t) f7;
-  int64_t f2f8_38 = f2_2 * (int64_t) f8_19;
-  int64_t f2f9_38 = f2   * (int64_t) f9_38;
-  int64_t f3f3_2  = f3_2 * (int64_t) f3;
-  int64_t f3f4_2  = f3_2 * (int64_t) f4;
-  int64_t f3f5_4  = f3_2 * (int64_t) f5_2;
-  int64_t f3f6_2  = f3_2 * (int64_t) f6;
-  int64_t f3f7_76 = f3_2 * (int64_t) f7_38;
-  int64_t f3f8_38 = f3_2 * (int64_t) f8_19;
-  int64_t f3f9_76 = f3_2 * (int64_t) f9_38;
-  int64_t f4f4    = f4   * (int64_t) f4;
-  int64_t f4f5_2  = f4_2 * (int64_t) f5;
-  int64_t f4f6_38 = f4_2 * (int64_t) f6_19;
-  int64_t f4f7_38 = f4   * (int64_t) f7_38;
-  int64_t f4f8_38 = f4_2 * (int64_t) f8_19;
-  int64_t f4f9_38 = f4   * (int64_t) f9_38;
-  int64_t f5f5_38 = f5   * (int64_t) f5_38;
-  int64_t f5f6_38 = f5_2 * (int64_t) f6_19;
-  int64_t f5f7_76 = f5_2 * (int64_t) f7_38;
-  int64_t f5f8_38 = f5_2 * (int64_t) f8_19;
-  int64_t f5f9_76 = f5_2 * (int64_t) f9_38;
-  int64_t f6f6_19 = f6   * (int64_t) f6_19;
-  int64_t f6f7_38 = f6   * (int64_t) f7_38;
-  int64_t f6f8_38 = f6_2 * (int64_t) f8_19;
-  int64_t f6f9_38 = f6   * (int64_t) f9_38;
-  int64_t f7f7_38 = f7   * (int64_t) f7_38;
-  int64_t f7f8_38 = f7_2 * (int64_t) f8_19;
-  int64_t f7f9_76 = f7_2 * (int64_t) f9_38;
-  int64_t f8f8_19 = f8   * (int64_t) f8_19;
-  int64_t f8f9_38 = f8   * (int64_t) f9_38;
-  int64_t f9f9_38 = f9   * (int64_t) f9_38;
-  int64_t h0 = f0f0  +f1f9_76+f2f8_38+f3f7_76+f4f6_38+f5f5_38;
-  int64_t h1 = f0f1_2+f2f9_38+f3f8_38+f4f7_38+f5f6_38;
-  int64_t h2 = f0f2_2+f1f1_2 +f3f9_76+f4f8_38+f5f7_76+f6f6_19;
-  int64_t h3 = f0f3_2+f1f2_2 +f4f9_38+f5f8_38+f6f7_38;
-  int64_t h4 = f0f4_2+f1f3_4 +f2f2   +f5f9_76+f6f8_38+f7f7_38;
-  int64_t h5 = f0f5_2+f1f4_2 +f2f3_2 +f6f9_38+f7f8_38;
-  int64_t h6 = f0f6_2+f1f5_4 +f2f4_2 +f3f3_2 +f7f9_76+f8f8_19;
-  int64_t h7 = f0f7_2+f1f6_2 +f2f5_2 +f3f4_2 +f8f9_38;
-  int64_t h8 = f0f8_2+f1f7_4 +f2f6_2 +f3f5_4 +f4f4   +f9f9_38;
-  int64_t h9 = f0f9_2+f1f8_2 +f2f7_2 +f3f6_2 +f4f5_2;
-  int64_t carry0;
-  int64_t carry1;
-  int64_t carry2;
-  int64_t carry3;
-  int64_t carry4;
-  int64_t carry5;
-  int64_t carry6;
-  int64_t carry7;
-  int64_t carry8;
-  int64_t carry9;
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-
-  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
-  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
-
-  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
-  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
-
-  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
-  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
-
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
-
-  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-
-  h[0] = h0;
-  h[1] = h1;
-  h[2] = h2;
-  h[3] = h3;
-  h[4] = h4;
-  h[5] = h5;
-  h[6] = h6;
-  h[7] = h7;
-  h[8] = h8;
-  h[9] = h9;
-}
-
-static void fe_invert(fe out, const fe z) {
-  fe t0;
-  fe t1;
-  fe t2;
-  fe t3;
-  int i;
-
-  fe_sq(t0, z);
-  fe_sq(t1, t0);
-  for (i = 1; i < 2; ++i) {
-    fe_sq(t1, t1);
-  }
-  fe_mul(t1, z, t1);
-  fe_mul(t0, t0, t1);
-  fe_sq(t2, t0);
-  fe_mul(t1, t1, t2);
-  fe_sq(t2, t1);
-  for (i = 1; i < 5; ++i) {
-    fe_sq(t2, t2);
-  }
-  fe_mul(t1, t2, t1);
-  fe_sq(t2, t1);
-  for (i = 1; i < 10; ++i) {
-    fe_sq(t2, t2);
-  }
-  fe_mul(t2, t2, t1);
-  fe_sq(t3, t2);
-  for (i = 1; i < 20; ++i) {
-    fe_sq(t3, t3);
-  }
-  fe_mul(t2, t3, t2);
-  fe_sq(t2, t2);
-  for (i = 1; i < 10; ++i) {
-    fe_sq(t2, t2);
-  }
-  fe_mul(t1, t2, t1);
-  fe_sq(t2, t1);
-  for (i = 1; i < 50; ++i) {
-    fe_sq(t2, t2);
-  }
-  fe_mul(t2, t2, t1);
-  fe_sq(t3, t2);
-  for (i = 1; i < 100; ++i) {
-    fe_sq(t3, t3);
-  }
-  fe_mul(t2, t3, t2);
-  fe_sq(t2, t2);
-  for (i = 1; i < 50; ++i) {
-    fe_sq(t2, t2);
-  }
-  fe_mul(t1, t2, t1);
-  fe_sq(t1, t1);
-  for (i = 1; i < 5; ++i) {
-    fe_sq(t1, t1);
-  }
-  fe_mul(out, t1, t0);
-}
-
-/* Replace (f,g) with (g,f) if b == 1;
- * replace (f,g) with (f,g) if b == 0.
- *
- * Preconditions: b in {0,1}. */
-static void fe_cswap(fe f, fe g, unsigned int b) {
-  unsigned i;
-  b = 0-b;
-  for (i = 0; i < 10; i++) {
-    int32_t x = f[i] ^ g[i];
-    x &= b;
-    f[i] ^= x;
-    g[i] ^= x;
-  }
-}
-
-/* h = f * 121666
- * Can overlap h with f.
- *
- * Preconditions:
- *    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
- *
- * Postconditions:
- *    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc. */
-static void fe_mul121666(fe h, fe f) {
-  int32_t f0 = f[0];
-  int32_t f1 = f[1];
-  int32_t f2 = f[2];
-  int32_t f3 = f[3];
-  int32_t f4 = f[4];
-  int32_t f5 = f[5];
-  int32_t f6 = f[6];
-  int32_t f7 = f[7];
-  int32_t f8 = f[8];
-  int32_t f9 = f[9];
-  int64_t h0 = f0 * (int64_t) 121666;
-  int64_t h1 = f1 * (int64_t) 121666;
-  int64_t h2 = f2 * (int64_t) 121666;
-  int64_t h3 = f3 * (int64_t) 121666;
-  int64_t h4 = f4 * (int64_t) 121666;
-  int64_t h5 = f5 * (int64_t) 121666;
-  int64_t h6 = f6 * (int64_t) 121666;
-  int64_t h7 = f7 * (int64_t) 121666;
-  int64_t h8 = f8 * (int64_t) 121666;
-  int64_t h9 = f9 * (int64_t) 121666;
-  int64_t carry0;
-  int64_t carry1;
-  int64_t carry2;
-  int64_t carry3;
-  int64_t carry4;
-  int64_t carry5;
-  int64_t carry6;
-  int64_t carry7;
-  int64_t carry8;
-  int64_t carry9;
-
-  carry9 = h9 + (1 << 24); h0 += (carry9 >> 25) * 19; h9 -= carry9 & kTop39Bits;
-  carry1 = h1 + (1 << 24); h2 += carry1 >> 25; h1 -= carry1 & kTop39Bits;
-  carry3 = h3 + (1 << 24); h4 += carry3 >> 25; h3 -= carry3 & kTop39Bits;
-  carry5 = h5 + (1 << 24); h6 += carry5 >> 25; h5 -= carry5 & kTop39Bits;
-  carry7 = h7 + (1 << 24); h8 += carry7 >> 25; h7 -= carry7 & kTop39Bits;
-
-  carry0 = h0 + (1 << 25); h1 += carry0 >> 26; h0 -= carry0 & kTop38Bits;
-  carry2 = h2 + (1 << 25); h3 += carry2 >> 26; h2 -= carry2 & kTop38Bits;
-  carry4 = h4 + (1 << 25); h5 += carry4 >> 26; h4 -= carry4 & kTop38Bits;
-  carry6 = h6 + (1 << 25); h7 += carry6 >> 26; h6 -= carry6 & kTop38Bits;
-  carry8 = h8 + (1 << 25); h9 += carry8 >> 26; h8 -= carry8 & kTop38Bits;
-
-  h[0] = h0;
-  h[1] = h1;
-  h[2] = h2;
-  h[3] = h3;
-  h[4] = h4;
-  h[5] = h5;
-  h[6] = h6;
-  h[7] = h7;
-  h[8] = h8;
-  h[9] = h9;
-}
-
-static void x25519_scalar_mult_generic(uint8_t out[32],
-                                       const uint8_t scalar[32],
-                                       const uint8_t point[32]) {
-  fe x1, x2, z2, x3, z3, tmp0, tmp1;
-  unsigned swap;
-  int pos;
-
-  uint8_t e[32];
-  memcpy(e, scalar, 32);
-  e[0] &= 248;
-  e[31] &= 127;
-  e[31] |= 64;
-  fe_frombytes(x1, point);
-  fe_1(x2);
-  fe_0(z2);
-  fe_copy(x3, x1);
-  fe_1(z3);
-
-  swap = 0;
-  for (pos = 254; pos >= 0; --pos) {
-    unsigned b = 1 & (e[pos / 8] >> (pos & 7));
-    swap ^= b;
-    fe_cswap(x2, x3, swap);
-    fe_cswap(z2, z3, swap);
-    swap = b;
-    fe_sub(tmp0, x3, z3);
-    fe_sub(tmp1, x2, z2);
-    fe_add(x2, x2, z2);
-    fe_add(z2, x3, z3);
-    fe_mul(z3, tmp0, x2);
-    fe_mul(z2, z2, tmp1);
-    fe_sq(tmp0, tmp1);
-    fe_sq(tmp1, x2);
-    fe_add(x3, z3, z2);
-    fe_sub(z2, z3, z2);
-    fe_mul(x2, tmp1, tmp0);
-    fe_sub(tmp1, tmp1, tmp0);
-    fe_sq(z2, z2);
-    fe_mul121666(z3, tmp1);
-    fe_sq(x3, x3);
-    fe_add(tmp0, tmp0, z3);
-    fe_mul(z3, x1, z2);
-    fe_mul(z2, tmp1, tmp0);
-  }
-  fe_cswap(x2, x3, swap);
-  fe_cswap(z2, z3, swap);
-
-  fe_invert(z2, z2);
-  fe_mul(x2, x2, z2);
-  fe_tobytes(out, x2);
-}
-
-static void x25519_scalar_mult(uint8_t out[32], const uint8_t scalar[32],
-                               const uint8_t point[32]) {
-  x25519_scalar_mult_generic(out, scalar, point);
-}
-
 #ifdef CONFIG_RNG
 void X25519_keypair(uint8_t out_public_value[32], uint8_t out_private_key[32]) {
   rand_bytes(out_private_key, 32);
diff --git a/core/cortex-m0/build.mk b/core/cortex-m0/build.mk
index 5f2bc32eb5..f8f33aa2a3 100644
--- a/core/cortex-m0/build.mk
+++ b/core/cortex-m0/build.mk
@@ -21,4 +21,13 @@ endif
 core-y=cpu.o init.o thumb_case.o div.o lmul.o ldivmod.o uldivmod.o
 core-$(CONFIG_COMMON_PANIC_OUTPUT)+=panic.o
 core-$(CONFIG_COMMON_RUNTIME)+=switch.o task.o
+
+dirs-y += core/$(CORE)/curve25519
+
+core-$(CONFIG_CURVE25519)+=curve25519/mpy121666.o
+core-$(CONFIG_CURVE25519)+=curve25519/reduce25519.o
+core-$(CONFIG_CURVE25519)+=curve25519/mul.o
+core-$(CONFIG_CURVE25519)+=curve25519/scalarmult.o
+core-$(CONFIG_CURVE25519)+=curve25519/sqr.o
+
 core-$(CONFIG_WATCHDOG)+=watchdog.o
diff --git a/core/cortex-m0/curve25519/cortex_m0_mpy121666.s b/core/cortex-m0/curve25519/mpy121666.S
index 8e74dd0265..d2a467459b 100644
--- a/core/cortex-m0/curve25519/cortex_m0_mpy121666.s
+++ b/core/cortex-m0/curve25519/mpy121666.S
@@ -11,21 +11,7 @@
 // Not yet tested on target hardware.
 
 
-	.cpu cortex-m0
-	.fpu softvfp
-	.eabi_attribute 20, 1
-	.eabi_attribute 21, 1
-	.eabi_attribute 23, 3
-	.eabi_attribute 24, 1
-	.eabi_attribute 25, 1
-	.eabi_attribute 26, 1
-	.eabi_attribute 30, 2
-	.eabi_attribute 34, 0
-	.eabi_attribute 18, 4
 	.code	16
-
-	.file	"cortex_m0_reduce25519.s"
-
 	.text
 	.align	2
 
@@ -36,7 +22,7 @@
 
 fe25519_mpyWith121666_asm:
     push {r4,r5,r6,r7,r14}
-    ldr r7,__label_for_immediate_56130
+    ldr r7,=56130
     ldr r2,[r1,#28]
     lsl r5,r2,#16
     lsr r6,r2,#16
@@ -191,9 +177,5 @@ fe25519_mpyWith121666_asm:
     str r6,[r0,#28]
     pop {r4,r5,r6,r7,r15}
 
-	.align	2
-__label_for_immediate_56130:
-	.word 56130
-
 	.size	fe25519_mpyWith121666_asm, .-fe25519_mpyWith121666_asm
 
diff --git a/core/cortex-m0/curve25519/mul.s b/core/cortex-m0/curve25519/mul.S
index 366713a7a3..366713a7a3 100644
--- a/core/cortex-m0/curve25519/mul.s
+++ b/core/cortex-m0/curve25519/mul.S
diff --git a/core/cortex-m0/curve25519/cortex_m0_reduce25519.s b/core/cortex-m0/curve25519/reduce25519.S
index 8984752a38..9a3c29a0f6 100644
--- a/core/cortex-m0/curve25519/cortex_m0_reduce25519.s
+++ b/core/cortex-m0/curve25519/reduce25519.S
@@ -8,21 +8,8 @@
 // Generated and tested with C++ functions in the test subdirectory and on the target.
 //
 
-	.cpu cortex-m0
-	.fpu softvfp
-	.eabi_attribute 20, 1
-	.eabi_attribute 21, 1
-	.eabi_attribute 23, 3
-	.eabi_attribute 24, 1
-	.eabi_attribute 25, 1
-	.eabi_attribute 26, 1
-	.eabi_attribute 30, 2
-	.eabi_attribute 34, 0
-	.eabi_attribute 18, 4
 	.code	16
 
-	.file	"cortex_m0_reduce25519.s"
-
 	.text
 	.align	2
 
diff --git a/core/cortex-m0/curve25519/scalarmult.c b/core/cortex-m0/curve25519/scalarmult.c
index d4e8c06c3e..f7370d1f43 100644
--- a/core/cortex-m0/curve25519/scalarmult.c
+++ b/core/cortex-m0/curve25519/scalarmult.c
@@ -57,12 +57,14 @@
     Creative Commons CC0 1.0 Universal public domain dedication
   ============================================================================*/
 
-#include <inttypes.h>
+#include "curve25519.h"
+#include "util.h"
 
 // comment out this line if implementing conditional swaps by data moves
 //#define DH_SWAP_BY_POINTERS
 
 // Define the symbol to 0 in order to only use ladder steps
+#define DH_REPLACE_LAST_THREE_LADDERSTEPS_WITH_DOUBLINGS 0
 //#define DH_REPLACE_LAST_THREE_LADDERSTEPS_WITH_DOUBLINGS 1
 
 typedef uint8_t  uint8;
@@ -142,26 +144,17 @@ fe25519_cpy(
     const fe25519* source
 )
 {
-    uint32 ctr;
-
-    for (ctr = 0; ctr < 8; ctr++)
-    {
-        dest->as_uint32[ctr] = source->as_uint32[ctr];
-    }
+    memcpy(dest, source, 32);
 }
 
 static void
 fe25519_unpack(
-    volatile fe25519*            out,
+    fe25519*            out,
     const unsigned char in[32]
 )
 {
-    uint8 ctr;
+    memcpy(out, in, 32);
 
-    for (ctr = 0; ctr < 32; ctr++)
-    {
-        out->as_uint8[ctr] = in[ctr];
-    }
     out->as_uint8[31] &= 0x7f; // make sure that the last bit is cleared.
 }
 
@@ -257,7 +250,7 @@ fe25519_square(
 
 static void
 fe25519_reduceCompletely(
-    volatile fe25519* inout
+    fe25519* inout
 )
 {
     uint32 numberOfTimesToSubstractPrime;
@@ -307,17 +300,12 @@ fe25519_reduceCompletely(
 static void
 fe25519_pack(
     unsigned char out[32],
-    volatile fe25519*      in
+    fe25519*      in
 )
 {
-    uint8 ctr;
-
     fe25519_reduceCompletely(in);
 
-    for (ctr = 0; ctr < 32; ctr++)
-    {
-        out[ctr] = in->as_uint8[ctr];
-    }
+    memcpy(out, in, 32);
 }
 
 // Note, that r and x are allowed to overlap!
@@ -511,7 +499,7 @@ typedef struct _ST_curve25519ladderstepWorkingState
     fe25519 xq;
     fe25519 zq;
 
-    volatile UN_256bitValue s;
+    UN_256bitValue s;
 
     int nextScalarBitToProcess;
     uint8 previousProcessedBit;
@@ -649,11 +637,11 @@ curve25519_doublePointP (ST_curve25519ladderstepWorkingState* pState)
 
 #endif // #ifdef DH_REPLACE_LAST_THREE_LADDERSTEPS_WITH_DOUBLINGS
 
-int
-crypto_scalarmult_curve25519(
-    unsigned char*       r,
-    const unsigned char* s,
-    const unsigned char* p
+void
+x25519_scalar_mult(
+    uint8_t r[32],
+    const uint8_t s[32],
+    const uint8_t p[32]
 )
 {
     ST_curve25519ladderstepWorkingState state;
@@ -741,21 +729,4 @@ crypto_scalarmult_curve25519(
 
     fe25519_pack (r, &state.xp);
 #endif
-
-    return 0;
-}
-
-int
-crypto_scalarmult_curve25519_base(
-    unsigned char*       q,
-    const unsigned char* n
-)
-{
-    static const uint8 base[32] =
-    {
-        9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
-    };
-
-    return crypto_scalarmult_curve25519(q, n, base);
 }
diff --git a/core/cortex-m0/curve25519/sqr.s b/core/cortex-m0/curve25519/sqr.S
index 9666a1643c..b62121adb7 100644
--- a/core/cortex-m0/curve25519/sqr.s
+++ b/core/cortex-m0/curve25519/sqr.S
@@ -17,17 +17,9 @@ square256_asm:
     push {r1,r4,r5,r6,r7,r14}
     .syntax unified
     mov r3,r8
-    .syntax divided
-    .syntax unified
     mov r4,r9
-    .syntax divided
-    .syntax unified
     mov r5,r10
-    .syntax divided
-    .syntax unified
     mov r6,r11
-    .syntax divided
-    .syntax unified
     mov r7,r12
     .syntax divided
     push {r3,r4,r5,r6,r7}
@@ -41,8 +33,6 @@ square256_asm:
  // clobbers all registers except for r14
     .syntax unified
     mov r0,r4
-    .syntax divided
-    .syntax unified
     mov r1,r5
     .syntax divided
     sub r0,r6
@@ -54,11 +44,7 @@ square256_asm:
     sbc r1,r2
     .syntax unified
     mov r8,r0
-    .syntax divided
-    .syntax unified
     mov r9,r1
-    .syntax divided
-    .syntax unified
     mov r10,r6
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -144,17 +130,9 @@ square256_asm:
  // Leaves r6 zero.
     .syntax unified
     mov r6,r10
-    .syntax divided
-    .syntax unified
     mov r10,r0
-    .syntax divided
-    .syntax unified
     mov r11,r1
-    .syntax divided
-    .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r1,r3
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -247,17 +225,9 @@ square256_asm:
     adc r5,r6
     .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r2,r8
-    .syntax divided
-    .syntax unified
     mov r8,r3
-    .syntax divided
-    .syntax unified
     mov r3,r9
-    .syntax divided
-    .syntax unified
     mov r9,r4
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -343,19 +313,13 @@ square256_asm:
  // Returns r4 as zero.
     .syntax unified
     mov r2,r12
-    .syntax divided
-    .syntax unified
     mov r3,r8
-    .syntax divided
-    .syntax unified
     mov r4,r9
     .syntax divided
     sub r2,r6
     sbc r3,r7
     .syntax unified
     mov r6,r4
-    .syntax divided
-    .syntax unified
     mov r7,r5
     .syntax divided
     sbc r4,r0
@@ -401,8 +365,6 @@ square256_asm:
  // clobbers all registers except for r14
     .syntax unified
     mov r0,r4
-    .syntax divided
-    .syntax unified
     mov r1,r5
     .syntax divided
     sub r0,r6
@@ -414,11 +376,7 @@ square256_asm:
     sbc r1,r2
     .syntax unified
     mov r8,r0
-    .syntax divided
-    .syntax unified
     mov r9,r1
-    .syntax divided
-    .syntax unified
     mov r10,r6
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -504,17 +462,9 @@ square256_asm:
  // Leaves r6 zero.
     .syntax unified
     mov r6,r10
-    .syntax divided
-    .syntax unified
     mov r10,r0
-    .syntax divided
-    .syntax unified
     mov r11,r1
-    .syntax divided
-    .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r1,r3
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -607,17 +557,9 @@ square256_asm:
     adc r5,r6
     .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r2,r8
-    .syntax divided
-    .syntax unified
     mov r8,r3
-    .syntax divided
-    .syntax unified
     mov r3,r9
-    .syntax divided
-    .syntax unified
     mov r9,r4
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -703,19 +645,13 @@ square256_asm:
  // Returns r4 as zero.
     .syntax unified
     mov r2,r12
-    .syntax divided
-    .syntax unified
     mov r3,r8
-    .syntax divided
-    .syntax unified
     mov r4,r9
     .syntax divided
     sub r2,r6
     sbc r3,r7
     .syntax unified
     mov r6,r4
-    .syntax divided
-    .syntax unified
     mov r7,r5
     .syntax divided
     sbc r4,r0
@@ -749,14 +685,8 @@ square256_asm:
  // Result in r0 ... r7
     .syntax unified
     mov r8,r4
-    .syntax divided
-    .syntax unified
     mov r9,r5
-    .syntax divided
-    .syntax unified
     mov r10,r6
-    .syntax divided
-    .syntax unified
     mov r11,r7
     .syntax divided
     pop {r4,r5,r6,r7}
@@ -766,17 +696,9 @@ square256_asm:
     adc r3,r7
     .syntax unified
     mov r4,r8
-    .syntax divided
-    .syntax unified
     mov r5,r9
-    .syntax divided
-    .syntax unified
     mov r6,r10
-    .syntax divided
-    .syntax unified
     mov r7,r11
-    .syntax divided
-    .syntax unified
     mov r8,r0
     .syntax divided
     mov r0,#0
@@ -809,8 +731,6 @@ square256_asm:
  // clobbers all registers except for r14
     .syntax unified
     mov r0,r4
-    .syntax divided
-    .syntax unified
     mov r1,r5
     .syntax divided
     sub r0,r6
@@ -822,11 +742,7 @@ square256_asm:
     sbc r1,r2
     .syntax unified
     mov r8,r0
-    .syntax divided
-    .syntax unified
     mov r9,r1
-    .syntax divided
-    .syntax unified
     mov r10,r6
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -912,17 +828,9 @@ square256_asm:
  // Leaves r6 zero.
     .syntax unified
     mov r6,r10
-    .syntax divided
-    .syntax unified
     mov r10,r0
-    .syntax divided
-    .syntax unified
     mov r11,r1
-    .syntax divided
-    .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r1,r3
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -1015,17 +923,9 @@ square256_asm:
     adc r5,r6
     .syntax unified
     mov r12,r2
-    .syntax divided
-    .syntax unified
     mov r2,r8
-    .syntax divided
-    .syntax unified
     mov r8,r3
-    .syntax divided
-    .syntax unified
     mov r3,r9
-    .syntax divided
-    .syntax unified
     mov r9,r4
     .syntax divided
  // START: sqr 64 Refined Karatsuba
@@ -1111,19 +1011,13 @@ square256_asm:
  // Returns r4 as zero.
     .syntax unified
     mov r2,r12
-    .syntax divided
-    .syntax unified
     mov r3,r8
-    .syntax divided
-    .syntax unified
     mov r4,r9
     .syntax divided
     sub r2,r6
     sbc r3,r7
     .syntax unified
     mov r6,r4
-    .syntax divided
-    .syntax unified
     mov r7,r5
     .syntax divided
     sbc r4,r0
@@ -1165,14 +1059,8 @@ square256_asm:
     mvn r7,r7
     .syntax unified
     mov r8,r4
-    .syntax divided
-    .syntax unified
     mov r9,r5
-    .syntax divided
-    .syntax unified
     mov r10,r6
-    .syntax divided
-    .syntax unified
     mov r11,r7
     .syntax divided
     mov r4,#143
@@ -1190,29 +1078,21 @@ square256_asm:
     stm r4!,{r0,r1,r2,r3}
     .syntax unified
     mov r4,r12
-    .syntax divided
-    .syntax unified
     mov r0,r8
     .syntax divided
     adc r0,r4
     .syntax unified
     mov r8,r0
-    .syntax divided
-    .syntax unified
     mov r1,r9
     .syntax divided
     adc r1,r5
     .syntax unified
     mov r9,r1
-    .syntax divided
-    .syntax unified
     mov r2,r10
     .syntax divided
     adc r2,r6
     .syntax unified
     mov r10,r2
-    .syntax divided
-    .syntax unified
     mov r3,r11
     .syntax divided
     adc r3,r7
@@ -1223,8 +1103,6 @@ square256_asm:
     adc r0,r0
     .syntax unified
     mov r12,r0
-    .syntax divided
-    .syntax unified
     mov r0,r14
     .syntax divided
     ldm r0,{r0,r1,r2,r3,r4,r5,r6,r7}
@@ -1237,8 +1115,6 @@ square256_asm:
     stm r4!,{r0,r1,r2,r3}
     .syntax unified
     mov r14,r4
-    .syntax divided
-    .syntax unified
     mov r0,r13
     .syntax divided
     ldm r0!,{r4,r5,r6,r7}
@@ -1278,17 +1154,9 @@ square256_asm:
     pop {r3,r4,r5,r6,r7}
     .syntax unified
     mov r8,r3
-    .syntax divided
-    .syntax unified
     mov r9,r4
-    .syntax divided
-    .syntax unified
     mov r10,r5
-    .syntax divided
-    .syntax unified
     mov r11,r6
-    .syntax divided
-    .syntax unified
     mov r12,r7
     .syntax divided
     pop {r0,r4,r5,r6,r7,r15}
diff --git a/include/curve25519.h b/include/curve25519.h
index 4b75373da1..8287c94466 100644
--- a/include/curve25519.h
+++ b/include/curve25519.h
@@ -59,4 +59,12 @@ int X25519(uint8_t out_shared_key[32], const uint8_t private_key[32],
 void X25519_public_from_private(uint8_t out_public_value[32],
 				const uint8_t private_key[32]);
 
+/*
+ * Low-level x25519 function, defined by either the generic or cortex-m0
+ * implementation. Must not be called directly.
+ */
+void x25519_scalar_mult(uint8_t out[32],
+			const uint8_t scalar[32],
+			const uint8_t point[32]);
+
 #endif /* __CROS_EC_CURVE25519_H */