x25519-termux 1.1.0

data/ext/x25519_precomputed/x25519_precomputed.c

@@ -0,0 +1,99 @@
+/*
+Ruby C extension providing bindings to the rfc7748_precomputed implementation of
+the X25519 Diffie-Hellman algorithm
+*/
+
+#include "ruby.h"
+#include "x25519_precomputed.h"
+
+static VALUE mX25519 = Qnil;
+static VALUE mX25519_Provider = Qnil;
+static VALUE mX25519_Provider_Precomputed = Qnil;
+
+static VALUE mX25519_Provider_Precomputed_scalarmult(VALUE self, VALUE scalar, VALUE montgomery_u);
+static VALUE mX25519_Provider_Precomputed_scalarmult_base(VALUE self, VALUE scalar);
+static VALUE mX25519_is_available(VALUE self);
+static VALUE mX25519_disabled(VALUE self);
+
+/* Initialize the x25519_precomputed C extension */
+void Init_x25519_precomputed()
+{
+    mX25519 = rb_define_module("X25519");
+    mX25519_Provider = rb_define_module_under(mX25519, "Provider");
+    mX25519_Provider_Precomputed = rb_define_module_under(mX25519_Provider, "Precomputed");
+
+    #ifdef DISABLE_PRECOMPUTED
+      rb_define_singleton_method(mX25519_Provider_Precomputed, "available?", mX25519_disabled, 0);
+    #else
+      rb_define_singleton_method(mX25519_Provider_Precomputed, "scalarmult", mX25519_Provider_Precomputed_scalarmult, 2);
+      rb_define_singleton_method(mX25519_Provider_Precomputed, "scalarmult_base", mX25519_Provider_Precomputed_scalarmult_base, 1);
+      rb_define_singleton_method(mX25519_Provider_Precomputed, "available?", mX25519_is_available, 0);
+    #endif
+}
+
+/* Variable-base scalar multiplication */
+static VALUE mX25519_Provider_Precomputed_scalarmult(VALUE self, VALUE scalar, VALUE montgomery_u)
+{
+    /* X25519_KEY ensures inputs are aligned at 32-bytes */
+    X25519_KEY raw_scalar, raw_montgomery_u, product;
+
+    StringValue(scalar);
+    if(RSTRING_LEN(scalar) != X25519_KEYSIZE_BYTES) {
+        rb_raise(
+            rb_eArgError,
+            "expected %d-byte scalar, got %ld",
+            X25519_KEYSIZE_BYTES,
+            RSTRING_LEN(scalar)
+        );
+    }
+
+    StringValue(montgomery_u);
+    if(RSTRING_LEN(montgomery_u) != X25519_KEYSIZE_BYTES) {
+        rb_raise(
+            rb_eArgError,
+            "expected %d-byte Montgomery-u coordinate, got %ld",
+            X25519_KEYSIZE_BYTES,
+            RSTRING_LEN(montgomery_u)
+        );
+    }
+
+    memcpy(raw_scalar, RSTRING_PTR(scalar), X25519_KEYSIZE_BYTES);
+    memcpy(raw_montgomery_u, RSTRING_PTR(montgomery_u), X25519_KEYSIZE_BYTES);
+    x25519_precomputed_scalarmult(product, raw_scalar, raw_montgomery_u);
+
+    return rb_str_new((const char *)product, X25519_KEYSIZE_BYTES);
+}
+
+/* Fixed-base scalar multiplication */
+static VALUE mX25519_Provider_Precomputed_scalarmult_base(VALUE self, VALUE scalar)
+{
+    /* X25519_KEY ensures inputs are aligned at 32-bytes */
+    X25519_KEY raw_scalar, product;
+
+    StringValue(scalar);
+    if(RSTRING_LEN(scalar) != X25519_KEYSIZE_BYTES) {
+        rb_raise(
+            rb_eArgError,
+            "expected %d-byte scalar, got %ld",
+            X25519_KEYSIZE_BYTES,
+            RSTRING_LEN(scalar)
+        );
+    }
+
+    memcpy(raw_scalar, RSTRING_PTR(scalar), X25519_KEYSIZE_BYTES);
+    x25519_precomputed_scalarmult_base(product, raw_scalar);
+
+    return rb_str_new((const char *)product, X25519_KEYSIZE_BYTES);
+}
+
+/* Is the x25519_precomputed backend supported on this CPU? */
+static VALUE mX25519_is_available(VALUE self)
+{
+    return check_4th_gen_intel_core_features() ? Qtrue : Qfalse;
+}
+
+/* Set availability to return false if extension is skipped */
+static VALUE mX25519_disabled(VALUE self)
+{
+    return Qfalse;
+}

data/ext/x25519_precomputed/x25519_precomputed.h

@@ -0,0 +1,58 @@
+/**
+ * Copyright (c) 2017, Armando Faz <armfazh@ic.unicamp.br>. All rights reserved.
+ * Institute of Computing.
+ * University of Campinas, Brazil.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * 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.
+ *  * Neither the name of University of Campinas nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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
+ * COPYRIGHT OWNER 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.
+ */
+
+#ifndef X25519_PRECOMPUTED_H
+#define X25519_PRECOMPUTED_H
+
+#include <stdint.h>
+
+#ifndef ALIGN_BYTES
+#define ALIGN_BYTES 32
+#endif
+
+#ifndef ALIGN
+#ifdef __INTEL_COMPILER
+#define ALIGN __declspec(align(ALIGN_BYTES))
+#else
+#define ALIGN __attribute__((aligned(ALIGN_BYTES)))
+#endif
+#endif
+
+#define X25519_KEYSIZE_BYTES 32
+typedef ALIGN uint8_t X25519_KEY[X25519_KEYSIZE_BYTES];
+
+void x25519_precomputed_scalarmult(uint8_t *shared, uint8_t *private_key, uint8_t *session_key);
+void x25519_precomputed_scalarmult_base(uint8_t *session_key, uint8_t *private_key);
+int check_4th_gen_intel_core_features();
+
+#endif /* X25519_PRECOMPUTED_H */

data/ext/x25519_precomputed/x25519_x64.c

@@ -0,0 +1,251 @@
+/**
+ * Copyright (c) 2017, Armando Faz <armfazh@ic.unicamp.br>. All rights reserved.
+ * Institute of Computing.
+ * University of Campinas, Brazil.
+ *
+ * Copyright (C) 2018 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * 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.
+ *  * Neither the name of University of Campinas nor the names of its
+ *    contributors may be used to endorse or promote products derived
+ *    from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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
+ * COPYRIGHT OWNER 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.
+ */
+
+#include <string.h>
+#include "fp25519_x64.h"
+#include "x25519_precomputed.h"
+#include "table_ladder_x25519.h"
+
+static inline void cswap_x64(uint64_t bit, uint64_t *const px,
+                             uint64_t *const py) {
+  int i = 0;
+  uint64_t mask = (uint64_t)0 - bit;
+  for (i = 0; i < NUM_WORDS_ELTFP25519_X64; i++) {
+    uint64_t t = mask & (px[i] ^ py[i]);
+    px[i] = px[i] ^ t;
+    py[i] = py[i] ^ t;
+  }
+}
+
+
+/** Original rfc7748_precomputed name: 'x25519_shared_secret_x64' */
+void x25519_precomputed_scalarmult(uint8_t *shared, uint8_t *private_key,
+                                   uint8_t *session_key) {
+  ALIGN uint64_t buffer[4 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint64_t coordinates[4 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint64_t workspace[6 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint8_t session[X25519_KEYSIZE_BYTES];
+  ALIGN uint8_t private[X25519_KEYSIZE_BYTES];
+
+  int i = 0, j = 0;
+  uint64_t prev = 0;
+  uint64_t *const X1 = (uint64_t *)session;
+  uint64_t *const key = (uint64_t *)private;
+  uint64_t *const Px = coordinates + 0;
+  uint64_t *const Pz = coordinates + 4;
+  uint64_t *const Qx = coordinates + 8;
+  uint64_t *const Qz = coordinates + 12;
+  uint64_t *const X2 = Qx;
+  uint64_t *const Z2 = Qz;
+  uint64_t *const X3 = Px;
+  uint64_t *const Z3 = Pz;
+  uint64_t *const X2Z2 = Qx;
+  uint64_t *const X3Z3 = Px;
+
+  uint64_t *const A = workspace + 0;
+  uint64_t *const B = workspace + 4;
+  uint64_t *const D = workspace + 8;
+  uint64_t *const C = workspace + 12;
+  uint64_t *const DA = workspace + 16;
+  uint64_t *const CB = workspace + 20;
+  uint64_t *const AB = A;
+  uint64_t *const DC = D;
+  uint64_t *const DACB = DA;
+  uint64_t *const buffer_1w = buffer;
+  uint64_t *const buffer_2w = buffer;
+
+  memcpy(private, private_key, sizeof(private));
+  memcpy(session, session_key, sizeof(session));
+
+  /* clampC function */
+ private
+  [0] = private[0] & (~(uint8_t)0x7);
+ private
+  [X25519_KEYSIZE_BYTES - 1] =
+      (uint8_t)64 | (private[X25519_KEYSIZE_BYTES - 1] & (uint8_t)0x7F);
+
+  /**
+  * As in the draft:
+  * When receiving such an array, implementations of curve25519
+  * MUST mask the most-significant bit in the final byte. This
+  * is done to preserve compatibility with point formats which
+  * reserve the sign bit for use in other protocols and to
+  * increase resistance to implementation fingerprinting
+  **/
+  session[X25519_KEYSIZE_BYTES - 1] &= (1 << (255 % 8)) - 1;
+
+  copy_EltFp25519_1w_x64(Px, X1);
+  setzero_EltFp25519_1w_x64(Pz);
+  setzero_EltFp25519_1w_x64(Qx);
+  setzero_EltFp25519_1w_x64(Qz);
+
+  Pz[0] = 1;
+  Qx[0] = 1;
+
+  /* main-loop */
+  prev = 0;
+  j = 62;
+  for (i = 3; i >= 0; i--) {
+    while (j >= 0) {
+      uint64_t bit = (key[i] >> j) & 0x1;
+      uint64_t swap = bit ^ prev;
+      prev = bit;
+
+      add_EltFp25519_1w_x64(A, X2, Z2);    /* A = (X2+Z2)                   */
+      sub_EltFp25519_1w_x64(B, X2, Z2);    /* B = (X2-Z2)                   */
+      add_EltFp25519_1w_x64(C, X3, Z3);    /* C = (X3+Z3)                   */
+      sub_EltFp25519_1w_x64(D, X3, Z3);    /* D = (X3-Z3)                   */
+      mul_EltFp25519_2w_x64(DACB, AB, DC); /* [DA|CB] = [A|B]*[D|C]         */
+
+      cswap_x64(swap, A, C);
+      cswap_x64(swap, B, D);
+
+      sqr_EltFp25519_2w_x64(AB);         /* [AA|BB] = [A^2|B^2]           */
+      add_EltFp25519_1w_x64(X3, DA, CB); /* X3 = (DA+CB)                  */
+      sub_EltFp25519_1w_x64(Z3, DA, CB); /* Z3 = (DA-CB)                  */
+      sqr_EltFp25519_2w_x64(X3Z3);       /* [X3|Z3] = [(DA+CB)|(DA+CB)]^2 */
+
+      copy_EltFp25519_1w_x64(X2, B);   /* X2 = B^2                      */
+      sub_EltFp25519_1w_x64(Z2, A, B); /* Z2 = E = AA-BB                */
+
+      mul_a24_EltFp25519_1w_x64(B, Z2);      /* B = a24*E                     */
+      add_EltFp25519_1w_x64(B, B, X2);       /* B = a24*E+B                   */
+      mul_EltFp25519_2w_x64(X2Z2, X2Z2, AB); /* [X2|Z2] = [B|E]*[A|a24*E+B]   */
+      mul_EltFp25519_1w_x64(Z3, Z3, X1);     /* Z3 = Z3*X1                    */
+      j--;
+    }
+    j = 63;
+  }
+
+  inv_EltFp25519_1w_x64(A, Qz);
+  mul_EltFp25519_1w_x64((uint64_t *)shared, Qx, A);
+  fred_EltFp25519_1w_x64((uint64_t *)shared);
+}
+
+/* Original rfc7748_precomputed name: 'x25519_keygen_precmp_x64' */
+void x25519_precomputed_scalarmult_base(uint8_t *session_key, uint8_t *private_key) {
+  ALIGN uint64_t buffer[4 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint64_t coordinates[4 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint64_t workspace[4 * NUM_WORDS_ELTFP25519_X64];
+  ALIGN uint8_t private[X25519_KEYSIZE_BYTES];
+
+  int i = 0, j = 0, k = 0;
+  uint64_t *const key = (uint64_t *)private;
+  uint64_t *const Ur1 = coordinates + 0;
+  uint64_t *const Zr1 = coordinates + 4;
+  uint64_t *const Ur2 = coordinates + 8;
+  uint64_t *const Zr2 = coordinates + 12;
+
+  uint64_t *const UZr1 = coordinates + 0;
+  uint64_t *const ZUr2 = coordinates + 8;
+
+  uint64_t *const A = workspace + 0;
+  uint64_t *const B = workspace + 4;
+  uint64_t *const C = workspace + 8;
+  uint64_t *const D = workspace + 12;
+
+  uint64_t *const AB = workspace + 0;
+  uint64_t *const CD = workspace + 8;
+
+  uint64_t *const buffer_1w = buffer;
+  uint64_t *const buffer_2w = buffer;
+  uint64_t *P = (uint64_t *)Table_Ladder_8k;
+
+  memcpy(private, private_key, sizeof(private));
+
+  /* clampC function */
+ private
+  [0] = private[0] & (~(uint8_t)0x7);
+ private
+  [X25519_KEYSIZE_BYTES - 1] =
+      (uint8_t)64 | (private[X25519_KEYSIZE_BYTES - 1] & (uint8_t)0x7F);
+
+  setzero_EltFp25519_1w_x64(Ur1);
+  setzero_EltFp25519_1w_x64(Zr1);
+  setzero_EltFp25519_1w_x64(Zr2);
+  Ur1[0] = 1;
+  Zr1[0] = 1;
+  Zr2[0] = 1;
+
+  /* G-S */
+  Ur2[3] = 0x1eaecdeee27cab34;
+  Ur2[2] = 0xadc7a0b9235d48e2;
+  Ur2[1] = 0xbbf095ae14b2edf8;
+  Ur2[0] = 0x7e94e1fec82faabd;
+
+  /* main-loop */
+  const int ite[4] = {64, 64, 64, 63};
+  const int q = 3;
+  uint64_t swap = 1;
+
+  j = q;
+  for (i = 0; i < NUM_WORDS_ELTFP25519_X64; i++) {
+    while (j < ite[i]) {
+      k = (64 * i + j - q);
+      uint64_t bit = (key[i] >> j) & 0x1;
+      swap = swap ^ bit;
+      cswap_x64(swap, Ur1, Ur2);
+      cswap_x64(swap, Zr1, Zr2);
+      swap = bit;
+      /** Addition */
+      sub_EltFp25519_1w_x64(B, Ur1, Zr1);     /* B = Ur1-Zr1                 */
+      add_EltFp25519_1w_x64(A, Ur1, Zr1);     /* A = Ur1+Zr1                 */
+      mul_EltFp25519_1w_x64(C, &P[4 * k], B); /* C = M0-B                    */
+      sub_EltFp25519_1w_x64(B, A, C);         /* B = (Ur1+Zr1) - M*(Ur1-Zr1) */
+      add_EltFp25519_1w_x64(A, A, C);         /* A = (Ur1+Zr1) + M*(Ur1-Zr1) */
+      sqr_EltFp25519_2w_x64(AB);              /* A = A^2      |  B = B^2     */
+      mul_EltFp25519_2w_x64(UZr1, ZUr2, AB);  /* Ur1 = Zr2*A  |  Zr1 = Ur2*B */
+      j++;
+    }
+    j = 0;
+  }
+
+  /** Doubling */
+  for (i = 0; i < q; i++) {
+    add_EltFp25519_1w_x64(A, Ur1, Zr1);  /*  A = Ur1+Zr1   */
+    sub_EltFp25519_1w_x64(B, Ur1, Zr1);  /*  B = Ur1-Zr1   */
+    sqr_EltFp25519_2w_x64(AB);           /*  A = A**2     B = B**2   */
+    copy_EltFp25519_1w_x64(C, B);        /*  C = B         */
+    sub_EltFp25519_1w_x64(B, A, B);      /*  B = A-B       */
+    mul_a24_EltFp25519_1w_x64(D, B);     /*  D = my_a24*B  */
+    add_EltFp25519_1w_x64(D, D, C);      /*  D = D+C       */
+    mul_EltFp25519_2w_x64(UZr1, AB, CD); /*  Ur1 = A*B   Zr1 = Zr1*A */
+  }
+
+  /* Convert to affine coordinates */
+  inv_EltFp25519_1w_x64(A, Zr1);
+  mul_EltFp25519_1w_x64((uint64_t *)session_key, Ur1, A);
+  fred_EltFp25519_1w_x64((uint64_t *)session_key);
+}

data/ext/x25519_ref10/api.h

@@ -0,0 +1,2 @@
+#define CRYPTO_BYTES 32
+#define CRYPTO_SCALARBYTES 32

data/ext/x25519_ref10/base.c

@@ -0,0 +1,9 @@
+#include "fe.h"
+#include "x25519_ref10.h"
+
+static const uint8_t x25519_basepoint[32] = {9};
+
+int x25519_ref10_scalarmult_base(uint8_t *q, const uint8_t *n)
+{
+  return x25519_ref10_scalarmult(q,n,x25519_basepoint);
+}

data/ext/x25519_ref10/extconf.rb

@@ -0,0 +1,13 @@
+# frozen_string_literal: true
+
+# rubocop:disable Style/GlobalVars
+
+require "mkmf"
+
+require_relative '../extconf_helpers'
+
+add_cflags '-Wall -O3 -pedantic -std=c99'
+
+create_makefile("x25519_ref10")
+
+# rubocop:enable Style/GlobalVars

data/ext/x25519_ref10/fe.c

@@ -0,0 +1,912 @@
+#include "fe.h"
+
+/*
+h = 0
+*/
+
+void fe_0(fe h)
+{
+  h[0] = 0;
+  h[1] = 0;
+  h[2] = 0;
+  h[3] = 0;
+  h[4] = 0;
+  h[5] = 0;
+  h[6] = 0;
+  h[7] = 0;
+  h[8] = 0;
+  h[9] = 0;
+}
+
+/*
+h = 1
+*/
+
+void fe_1(fe h)
+{
+  h[0] = 1;
+  h[1] = 0;
+  h[2] = 0;
+  h[3] = 0;
+  h[4] = 0;
+  h[5] = 0;
+  h[6] = 0;
+  h[7] = 0;
+  h[8] = 0;
+  h[9] = 0;
+}
+
+/*
+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.
+*/
+
+void fe_add(fe h,fe f,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 h0 = f0 + g0;
+  int32_t h1 = f1 + g1;
+  int32_t h2 = f2 + g2;
+  int32_t h3 = f3 + g3;
+  int32_t h4 = f4 + g4;
+  int32_t h5 = f5 + g5;
+  int32_t h6 = f6 + g6;
+  int32_t h7 = f7 + g7;
+  int32_t h8 = f8 + g8;
+  int32_t h9 = f9 + g9;
+  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
+*/
+
+void fe_copy(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];
+  h[0] = f0;
+  h[1] = f1;
+  h[2] = f2;
+  h[3] = f3;
+  h[4] = f4;
+  h[5] = f5;
+  h[6] = f6;
+  h[7] = f7;
+  h[8] = f8;
+  h[9] = f9;
+}
+
+/*
+Replace (f,g) with (g,f) if b == 1;
+replace (f,g) with (f,g) if b == 0.
+
+Preconditions: b in {0,1}.
+*/
+
+void fe_cswap(fe f,fe g,unsigned int b)
+{
+  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 x0 = f0 ^ g0;
+  int32_t x1 = f1 ^ g1;
+  int32_t x2 = f2 ^ g2;
+  int32_t x3 = f3 ^ g3;
+  int32_t x4 = f4 ^ g4;
+  int32_t x5 = f5 ^ g5;
+  int32_t x6 = f6 ^ g6;
+  int32_t x7 = f7 ^ g7;
+  int32_t x8 = f8 ^ g8;
+  int32_t x9 = f9 ^ g9;
+  b = -b;
+  x0 &= b;
+  x1 &= b;
+  x2 &= b;
+  x3 &= b;
+  x4 &= b;
+  x5 &= b;
+  x6 &= b;
+  x7 &= b;
+  x8 &= b;
+  x9 &= b;
+  f[0] = f0 ^ x0;
+  f[1] = f1 ^ x1;
+  f[2] = f2 ^ x2;
+  f[3] = f3 ^ x3;
+  f[4] = f4 ^ x4;
+  f[5] = f5 ^ x5;
+  f[6] = f6 ^ x6;
+  f[7] = f7 ^ x7;
+  f[8] = f8 ^ x8;
+  f[9] = f9 ^ x9;
+  g[0] = g0 ^ x0;
+  g[1] = g1 ^ x1;
+  g[2] = g2 ^ x2;
+  g[3] = g3 ^ x3;
+  g[4] = g4 ^ x4;
+  g[5] = g5 ^ x5;
+  g[6] = g6 ^ x6;
+  g[7] = g7 ^ x7;
+  g[8] = g8 ^ x8;
+  g[9] = g9 ^ x9;
+}
+
+static uint64_t load_3(const unsigned char *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 unsigned char *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;
+}
+
+void fe_frombytes(fe h,const unsigned char *s)
+{
+  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 + (int64_t) (1<<24)) >> 25; h0 += carry9 * 19; h9 -= carry9 << 25;
+  carry1 = (h1 + (int64_t) (1<<24)) >> 25; h2 += carry1; h1 -= carry1 << 25;
+  carry3 = (h3 + (int64_t) (1<<24)) >> 25; h4 += carry3; h3 -= carry3 << 25;
+  carry5 = (h5 + (int64_t) (1<<24)) >> 25; h6 += carry5; h5 -= carry5 << 25;
+  carry7 = (h7 + (int64_t) (1<<24)) >> 25; h8 += carry7; h7 -= carry7 << 25;
+
+  carry0 = (h0 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+  carry2 = (h2 + (int64_t) (1<<25)) >> 26; h3 += carry2; h2 -= carry2 << 26;
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+  carry6 = (h6 + (int64_t) (1<<25)) >> 26; h7 += carry6; h6 -= carry6 << 26;
+  carry8 = (h8 + (int64_t) (1<<25)) >> 26; h9 += carry8; h8 -= carry8 << 26;
+
+  h[0] = (int32_t)h0;
+  h[1] = (int32_t)h1;
+  h[2] = (int32_t)h2;
+  h[3] = (int32_t)h3;
+  h[4] = (int32_t)h4;
+  h[5] = (int32_t)h5;
+  h[6] = (int32_t)h6;
+  h[7] = (int32_t)h7;
+  h[8] = (int32_t)h8;
+  h[9] = (int32_t)h9;
+}
+
+void fe_invert(fe out,fe z)
+{
+  fe t0;
+  fe t1;
+  fe t2;
+  fe t3;
+  int i;
+
+#include "pow225521.h"
+
+  return;
+}
+
+/*
+h = f * g
+Can overlap h with f or g.
+
+Preconditions:
+   |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
+   |g| 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.
+*/
+
+/*
+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.
+*/
+
+void fe_mul(fe h,fe f,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; /* 1.4*2^29 */
+  int32_t g2_19 = 19 * g2; /* 1.4*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| <= (1.1*1.1*2^52*(1+19+19+19+19)+1.1*1.1*2^50*(38+38+38+38+38))
+    i.e. |h0| <= 1.2*2^59; narrower ranges for h2, h4, h6, h8
+  |h1| <= (1.1*1.1*2^51*(1+1+19+19+19+19+19+19+19+19))
+    i.e. |h1| <= 1.5*2^58; narrower ranges for h3, h5, h7, h9
+  */
+
+  carry0 = (h0 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+  /* |h0| <= 2^25 */
+  /* |h4| <= 2^25 */
+  /* |h1| <= 1.51*2^58 */
+  /* |h5| <= 1.51*2^58 */
+
+  carry1 = (h1 + (int64_t) (1<<24)) >> 25; h2 += carry1; h1 -= carry1 << 25;
+  carry5 = (h5 + (int64_t) (1<<24)) >> 25; h6 += carry5; h5 -= carry5 << 25;
+  /* |h1| <= 2^24; from now on fits into int32 */
+  /* |h5| <= 2^24; from now on fits into int32 */
+  /* |h2| <= 1.21*2^59 */
+  /* |h6| <= 1.21*2^59 */
+
+  carry2 = (h2 + (int64_t) (1<<25)) >> 26; h3 += carry2; h2 -= carry2 << 26;
+  carry6 = (h6 + (int64_t) (1<<25)) >> 26; h7 += carry6; h6 -= carry6 << 26;
+  /* |h2| <= 2^25; from now on fits into int32 unchanged */
+  /* |h6| <= 2^25; from now on fits into int32 unchanged */
+  /* |h3| <= 1.51*2^58 */
+  /* |h7| <= 1.51*2^58 */
+
+  carry3 = (h3 + (int64_t) (1<<24)) >> 25; h4 += carry3; h3 -= carry3 << 25;
+  carry7 = (h7 + (int64_t) (1<<24)) >> 25; h8 += carry7; h7 -= carry7 << 25;
+  /* |h3| <= 2^24; from now on fits into int32 unchanged */
+  /* |h7| <= 2^24; from now on fits into int32 unchanged */
+  /* |h4| <= 1.52*2^33 */
+  /* |h8| <= 1.52*2^33 */
+
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+  carry8 = (h8 + (int64_t) (1<<25)) >> 26; h9 += carry8; h8 -= carry8 << 26;
+  /* |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| <= 1.51*2^58 */
+
+  carry9 = (h9 + (int64_t) (1<<24)) >> 25; h0 += carry9 * 19; h9 -= carry9 << 25;
+  /* |h9| <= 2^24; from now on fits into int32 unchanged */
+  /* |h0| <= 1.8*2^37 */
+
+  carry0 = (h0 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+  /* |h0| <= 2^25; from now on fits into int32 unchanged */
+  /* |h1| <= 1.01*2^24 */
+
+  h[0] = (int32_t)h0;
+  h[1] = (int32_t)h1;
+  h[2] = (int32_t)h2;
+  h[3] = (int32_t)h3;
+  h[4] = (int32_t)h4;
+  h[5] = (int32_t)h5;
+  h[6] = (int32_t)h6;
+  h[7] = (int32_t)h7;
+  h[8] = (int32_t)h8;
+  h[9] = (int32_t)h9;
+}
+
+/*
+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.
+*/
+
+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 + (int64_t) (1<<24)) >> 25; h0 += carry9 * 19; h9 -= carry9 << 25;
+  carry1 = (h1 + (int64_t) (1<<24)) >> 25; h2 += carry1; h1 -= carry1 << 25;
+  carry3 = (h3 + (int64_t) (1<<24)) >> 25; h4 += carry3; h3 -= carry3 << 25;
+  carry5 = (h5 + (int64_t) (1<<24)) >> 25; h6 += carry5; h5 -= carry5 << 25;
+  carry7 = (h7 + (int64_t) (1<<24)) >> 25; h8 += carry7; h7 -= carry7 << 25;
+
+  carry0 = (h0 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+  carry2 = (h2 + (int64_t) (1<<25)) >> 26; h3 += carry2; h2 -= carry2 << 26;
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+  carry6 = (h6 + (int64_t) (1<<25)) >> 26; h7 += carry6; h6 -= carry6 << 26;
+  carry8 = (h8 + (int64_t) (1<<25)) >> 26; h9 += carry8; h8 -= carry8 << 26;
+
+  h[0] = (int32_t)h0;
+  h[1] = (int32_t)h1;
+  h[2] = (int32_t)h2;
+  h[3] = (int32_t)h3;
+  h[4] = (int32_t)h4;
+  h[5] = (int32_t)h5;
+  h[6] = (int32_t)h6;
+  h[7] = (int32_t)h7;
+  h[8] = (int32_t)h8;
+  h[9] = (int32_t)h9;
+}
+
+/*
+h = f * f
+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.
+*/
+
+/*
+See fe_mul.c for discussion of implementation strategy.
+*/
+
+void fe_sq(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];
+  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; /* 1.31*2^30 */
+  int32_t f6_19 = 19 * f6; /* 1.31*2^30 */
+  int32_t f7_38 = 38 * f7; /* 1.31*2^30 */
+  int32_t f8_19 = 19 * f8; /* 1.31*2^30 */
+  int32_t f9_38 = 38 * f9; /* 1.31*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 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+
+  carry1 = (h1 + (int64_t) (1<<24)) >> 25; h2 += carry1; h1 -= carry1 << 25;
+  carry5 = (h5 + (int64_t) (1<<24)) >> 25; h6 += carry5; h5 -= carry5 << 25;
+
+  carry2 = (h2 + (int64_t) (1<<25)) >> 26; h3 += carry2; h2 -= carry2 << 26;
+  carry6 = (h6 + (int64_t) (1<<25)) >> 26; h7 += carry6; h6 -= carry6 << 26;
+
+  carry3 = (h3 + (int64_t) (1<<24)) >> 25; h4 += carry3; h3 -= carry3 << 25;
+  carry7 = (h7 + (int64_t) (1<<24)) >> 25; h8 += carry7; h7 -= carry7 << 25;
+
+  carry4 = (h4 + (int64_t) (1<<25)) >> 26; h5 += carry4; h4 -= carry4 << 26;
+  carry8 = (h8 + (int64_t) (1<<25)) >> 26; h9 += carry8; h8 -= carry8 << 26;
+
+  carry9 = (h9 + (int64_t) (1<<24)) >> 25; h0 += carry9 * 19; h9 -= carry9 << 25;
+
+  carry0 = (h0 + (int64_t) (1<<25)) >> 26; h1 += carry0; h0 -= carry0 << 26;
+
+  h[0] = (int32_t)h0;
+  h[1] = (int32_t)h1;
+  h[2] = (int32_t)h2;
+  h[3] = (int32_t)h3;
+  h[4] = (int32_t)h4;
+  h[5] = (int32_t)h5;
+  h[6] = (int32_t)h6;
+  h[7] = (int32_t)h7;
+  h[8] = (int32_t)h8;
+  h[9] = (int32_t)h9;
+}
+
+/*
+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.
+*/
+
+void fe_sub(fe h,fe f,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 h0 = f0 - g0;
+  int32_t h1 = f1 - g1;
+  int32_t h2 = f2 - g2;
+  int32_t h3 = f3 - g3;
+  int32_t h4 = f4 - g4;
+  int32_t h5 = f5 - g5;
+  int32_t h6 = f6 - g6;
+  int32_t h7 = f7 - g7;
+  int32_t h8 = f8 - g8;
+  int32_t h9 = f9 - g9;
+  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*2^25,1.1*2^24,1.1*2^25,1.1*2^24,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|<2^230 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.
+*/
+
+void fe_tobytes(unsigned char *s,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;
+  int32_t carry0;
+  int32_t carry1;
+  int32_t carry2;
+  int32_t carry3;
+  int32_t carry4;
+  int32_t carry5;
+  int32_t carry6;
+  int32_t carry7;
+  int32_t carry8;
+  int32_t carry9;
+
+  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. */
+
+  carry0 = h0 >> 26; h1 += carry0; h0 -= carry0 << 26;
+  carry1 = h1 >> 25; h2 += carry1; h1 -= carry1 << 25;
+  carry2 = h2 >> 26; h3 += carry2; h2 -= carry2 << 26;
+  carry3 = h3 >> 25; h4 += carry3; h3 -= carry3 << 25;
+  carry4 = h4 >> 26; h5 += carry4; h4 -= carry4 << 26;
+  carry5 = h5 >> 25; h6 += carry5; h5 -= carry5 << 25;
+  carry6 = h6 >> 26; h7 += carry6; h6 -= carry6 << 26;
+  carry7 = h7 >> 25; h8 += carry7; h7 -= carry7 << 25;
+  carry8 = h8 >> 26; h9 += carry8; h8 -= carry8 << 26;
+  carry9 = h9 >> 25;               h9 -= carry9 << 25;
+                  /* 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) | (h1 << 2);
+  s[4] = h1 >> 6;
+  s[5] = h1 >> 14;
+  s[6] = (h1 >> 22) | (h2 << 3);
+  s[7] = h2 >> 5;
+  s[8] = h2 >> 13;
+  s[9] = (h2 >> 21) | (h3 << 5);
+  s[10] = h3 >> 3;
+  s[11] = h3 >> 11;
+  s[12] = (h3 >> 19) | (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) | (h6 << 1);
+  s[20] = h6 >> 7;
+  s[21] = h6 >> 15;
+  s[22] = (h6 >> 23) | (h7 << 3);
+  s[23] = h7 >> 5;
+  s[24] = h7 >> 13;
+  s[25] = (h7 >> 21) | (h8 << 4);
+  s[26] = h8 >> 4;
+  s[27] = h8 >> 12;
+  s[28] = (h8 >> 20) | (h9 << 6);
+  s[29] = h9 >> 2;
+  s[30] = h9 >> 10;
+  s[31] = h9 >> 18;
+}