diff --git a/c/common/shared_dictionary.c b/c/common/shared_dictionary.c
new file mode 100644
index 0000000..c7d05c5
--- /dev/null
+++ b/c/common/shared_dictionary.c
@@ -0,0 +1,515 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* Shared Dictionary definition and utilities. */
+
+#include <brotli/shared_dictionary.h>
+
+#include <memory.h>
+#include <stdlib.h>  /* malloc, free */
+#include <stdio.h>
+
+#include "./dictionary.h"
+#include "./platform.h"
+#include "./shared_dictionary_internal.h"
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+#define BROTLI_NUM_ENCODED_LENGTHS (SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH \
+    - SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH + 1)
+
+/* Max allowed by spec */
+#define BROTLI_MAX_SIZE_BITS 15u
+
+/* Returns BROTLI_TRUE on success, BROTLI_FALSE on failure. */
+static BROTLI_BOOL ReadBool(const uint8_t* encoded, size_t size, size_t* pos,
+    BROTLI_BOOL* result) {
+  uint8_t value;
+  size_t position = *pos;
+  if (position >= size) return BROTLI_FALSE;  /* past file end */
+  value = encoded[position++];
+  if (value > 1) return BROTLI_FALSE;  /* invalid bool */
+  *result = TO_BROTLI_BOOL(value);
+  *pos = position;
+  return BROTLI_TRUE;  /* success */
+}
+
+/* Returns BROTLI_TRUE on success, BROTLI_FALSE on failure. */
+static BROTLI_BOOL ReadUint8(const uint8_t* encoded, size_t size, size_t* pos,
+    uint8_t* result) {
+  size_t position = *pos;
+  if (position + sizeof(uint8_t) > size) return BROTLI_FALSE;
+  *result = encoded[position++];
+  *pos = position;
+  return BROTLI_TRUE;
+}
+
+/* Returns BROTLI_TRUE on success, BROTLI_FALSE on failure. */
+static BROTLI_BOOL ReadUint16(const uint8_t* encoded, size_t size, size_t* pos,
+    uint16_t* result) {
+  size_t position = *pos;
+  if (position + sizeof(uint16_t) > size) return BROTLI_FALSE;
+  *result = BROTLI_UNALIGNED_LOAD16LE(&encoded[position]);
+  position += 2;
+  *pos = position;
+  return BROTLI_TRUE;
+}
+
+/* Reads a varint into a uint32_t, and returns error if it's too large */
+/* Returns BROTLI_TRUE on success, BROTLI_FALSE on failure. */
+static BROTLI_BOOL ReadVarint32(const uint8_t* encoded, size_t size,
+    size_t* pos, uint32_t* result) {
+  int num = 0;
+  uint8_t byte;
+  *result = 0;
+  for (;;) {
+    if (*pos >= size) return BROTLI_FALSE;
+    byte = encoded[(*pos)++];
+    if (num == 4 && byte > 15) return BROTLI_FALSE;
+    *result |= (uint32_t)(byte & 127) << (num * 7);
+    if (byte < 128) return BROTLI_TRUE;
+    num++;
+  }
+}
+
+/* Returns the total length of word list. */
+static size_t BrotliSizeBitsToOffsets(const uint8_t* size_bits_by_length,
+    uint32_t* offsets_by_length) {
+  uint32_t pos = 0;
+  uint32_t i;
+  for (i = 0; i <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; i++) {
+    offsets_by_length[i] = pos;
+    if (size_bits_by_length[i] != 0) {
+      pos += i << size_bits_by_length[i];
+    }
+  }
+  return pos;
+}
+
+static BROTLI_BOOL ParseWordList(size_t size, const uint8_t* encoded,
+    size_t* pos, BrotliDictionary* out) {
+  size_t offset;
+  size_t i;
+  size_t position = *pos;
+  if (position + BROTLI_NUM_ENCODED_LENGTHS > size) {
+    return BROTLI_FALSE;
+  }
+
+  memset(out->size_bits_by_length, 0, SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH);
+  memcpy(out->size_bits_by_length + SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH,
+      &encoded[position], BROTLI_NUM_ENCODED_LENGTHS);
+  for (i = SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH;
+      i <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; i++) {
+    if (out->size_bits_by_length[i] > BROTLI_MAX_SIZE_BITS) {
+      return BROTLI_FALSE;
+    }
+  }
+  position += BROTLI_NUM_ENCODED_LENGTHS;
+  offset = BrotliSizeBitsToOffsets(
+      out->size_bits_by_length, out->offsets_by_length);
+
+  out->data = &encoded[position];
+  out->data_size = offset;
+  position += offset;
+  if (position > size) return BROTLI_FALSE;
+  *pos = position;
+  return BROTLI_TRUE;
+}
+
+/* Computes the cutOffTransforms of a BrotliTransforms which already has the
+   transforms data correctly filled in. */
+static void ComputeCutoffTransforms(BrotliTransforms* transforms) {
+  uint32_t i;
+  for (i = 0; i < BROTLI_TRANSFORMS_MAX_CUT_OFF + 1; i++) {
+    transforms->cutOffTransforms[i] = -1;
+  }
+  for (i = 0; i < transforms->num_transforms; i++) {
+    const uint8_t* prefix = BROTLI_TRANSFORM_PREFIX(transforms, i);
+    uint8_t type = BROTLI_TRANSFORM_TYPE(transforms, i);
+    const uint8_t* suffix = BROTLI_TRANSFORM_SUFFIX(transforms, i);
+    if (type <= BROTLI_TRANSFORM_OMIT_LAST_9 && *prefix == 0 && *suffix == 0 &&
+        transforms->cutOffTransforms[type] == -1) {
+      transforms->cutOffTransforms[type] = (int16_t)i;
+    }
+  }
+}
+
+static BROTLI_BOOL ParsePrefixSuffixTable(size_t size, const uint8_t* encoded,
+    size_t* pos, BrotliTransforms* out, uint16_t* out_table,
+    size_t* out_table_size) {
+  size_t position = *pos;
+  size_t offset = 0;
+  size_t stringlet_count = 0;  /* NUM_PREFIX_SUFFIX */
+  size_t data_length = 0;
+
+  /* PREFIX_SUFFIX_LENGTH */
+  if (!ReadUint16(encoded, size, &position, &out->prefix_suffix_size)) {
+    return BROTLI_FALSE;
+  }
+  data_length = out->prefix_suffix_size;
+
+  /* Must at least have space for null terminator. */
+  if (data_length < 1) return BROTLI_FALSE;
+  out->prefix_suffix = &encoded[position];
+  if (position + data_length >= size) return BROTLI_FALSE;
+  while (BROTLI_TRUE) {
+    /* STRING_LENGTH */
+    size_t stringlet_len = encoded[position + offset];
+    out_table[stringlet_count] = (uint16_t)offset;
+    stringlet_count++;
+    offset++;
+    if (stringlet_len == 0) {
+      if (offset == data_length) {
+        break;
+      } else {
+        return BROTLI_FALSE;
+      }
+    }
+    if (stringlet_count > 255) return BROTLI_FALSE;
+    offset += stringlet_len;
+    if (offset >= data_length) return BROTLI_FALSE;
+  }
+
+  position += data_length;
+  *pos = position;
+  *out_table_size = (uint16_t)stringlet_count;
+  return BROTLI_TRUE;
+}
+
+static BROTLI_BOOL ParseTransformsList(size_t size, const uint8_t* encoded,
+    size_t* pos, BrotliTransforms* out, uint16_t* prefix_suffix_table,
+    size_t* prefix_suffix_count) {
+  uint32_t i;
+  BROTLI_BOOL has_params = BROTLI_FALSE;
+  BROTLI_BOOL prefix_suffix_ok = BROTLI_FALSE;
+  size_t position = *pos;
+  size_t stringlet_cnt = 0;
+  if (position >= size) return BROTLI_FALSE;
+
+  prefix_suffix_ok = ParsePrefixSuffixTable(
+      size, encoded, &position, out, prefix_suffix_table, &stringlet_cnt);
+  if (!prefix_suffix_ok) return BROTLI_FALSE;
+  out->prefix_suffix_map = prefix_suffix_table;
+  *prefix_suffix_count = stringlet_cnt;
+
+  out->num_transforms = encoded[position++];
+  out->transforms = &encoded[position];
+  position += (size_t)out->num_transforms * 3;
+  if (position > size) return BROTLI_FALSE;
+  /* Check for errors and read extra parameters. */
+  for (i = 0; i < out->num_transforms; i++) {
+    uint8_t prefix_id = BROTLI_TRANSFORM_PREFIX_ID(out, i);
+    uint8_t type = BROTLI_TRANSFORM_TYPE(out, i);
+    uint8_t suffix_id = BROTLI_TRANSFORM_SUFFIX_ID(out, i);
+    if (prefix_id >= stringlet_cnt) return BROTLI_FALSE;
+    if (type >= BROTLI_NUM_TRANSFORM_TYPES) return BROTLI_FALSE;
+    if (suffix_id >= stringlet_cnt) return BROTLI_FALSE;
+    if (type == BROTLI_TRANSFORM_SHIFT_FIRST ||
+        type == BROTLI_TRANSFORM_SHIFT_ALL) {
+      has_params = BROTLI_TRUE;
+    }
+  }
+  if (has_params) {
+    out->params = &encoded[position];
+    position += (size_t)out->num_transforms * 2;
+    if (position > size) return BROTLI_FALSE;
+    for (i = 0; i < out->num_transforms; i++) {
+      uint8_t type = BROTLI_TRANSFORM_TYPE(out, i);
+      if (type != BROTLI_TRANSFORM_SHIFT_FIRST &&
+          type != BROTLI_TRANSFORM_SHIFT_ALL) {
+        if (out->params[i * 2] != 0 || out->params[i * 2 + 1] != 0) {
+          return BROTLI_FALSE;
+        }
+      }
+    }
+  } else {
+    out->params = NULL;
+  }
+  ComputeCutoffTransforms(out);
+  *pos = position;
+  return BROTLI_TRUE;
+}
+
+static BROTLI_BOOL DryParseDictionary(const uint8_t* encoded,
+    size_t size, uint32_t* num_prefix, BROTLI_BOOL* is_custom_static_dict) {
+  size_t pos = 0;
+  uint32_t chunk_size = 0;
+  uint8_t num_word_lists;
+  uint8_t num_transform_lists;
+  *is_custom_static_dict = BROTLI_FALSE;
+  *num_prefix = 0;
+
+  /* Skip magic header bytes. */
+  pos += 2;
+
+  /* LZ77_DICTIONARY_LENGTH */
+  if (!ReadVarint32(encoded, size, &pos, &chunk_size)) return BROTLI_FALSE;
+  if (chunk_size != 0) {
+    /* This limitation is not specified but the 32-bit Brotli decoder for now */
+    if (chunk_size > 1073741823) return BROTLI_FALSE;
+    *num_prefix = 1;
+    if (pos + chunk_size > size) return BROTLI_FALSE;
+    pos += chunk_size;
+  }
+
+  if (!ReadUint8(encoded, size, &pos, &num_word_lists)) {
+    return BROTLI_FALSE;
+  }
+  if (!ReadUint8(encoded, size, &pos, &num_transform_lists)) {
+    return BROTLI_FALSE;
+  }
+
+  if (num_word_lists > 0 || num_transform_lists > 0) {
+    *is_custom_static_dict = BROTLI_TRUE;
+  }
+
+  return BROTLI_TRUE;
+}
+
+static BROTLI_BOOL ParseDictionary(const uint8_t* encoded, size_t size,
+    BrotliSharedDictionary* dict) {
+  uint32_t i;
+  size_t pos = 0;
+  uint32_t chunk_size = 0;
+  size_t total_prefix_suffix_count = 0;
+  size_t trasform_list_start[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+  uint16_t temporary_prefix_suffix_table[256];
+
+  /* Skip magic header bytes. */
+  pos += 2;
+
+  /* LZ77_DICTIONARY_LENGTH */
+  if (!ReadVarint32(encoded, size, &pos, &chunk_size)) return BROTLI_FALSE;
+  if (chunk_size != 0) {
+    if (pos + chunk_size > size) return BROTLI_FALSE;
+    dict->prefix_size[dict->num_prefix] = chunk_size;
+    dict->prefix[dict->num_prefix] = &encoded[pos];
+    dict->num_prefix++;
+    /* LZ77_DICTIONARY_LENGTH bytes. */
+    pos += chunk_size;
+  }
+
+  /* NUM_WORD_LISTS */
+  if (!ReadUint8(encoded, size, &pos, &dict->num_word_lists)) {
+    return BROTLI_FALSE;
+  }
+  if (dict->num_word_lists > SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS) {
+    return BROTLI_FALSE;
+  }
+
+  if (dict->num_word_lists != 0) {
+    dict->words_instances = (BrotliDictionary*)dict->alloc_func(
+        dict->memory_manager_opaque,
+        dict->num_word_lists * sizeof(*dict->words_instances));
+    if (!dict->words_instances) return BROTLI_FALSE;  /* OOM */
+  }
+  for (i = 0; i < dict->num_word_lists; i++) {
+    if (!ParseWordList(size, encoded, &pos, &dict->words_instances[i])) {
+      return BROTLI_FALSE;
+    }
+  }
+
+  /* NUM_TRANSFORM_LISTS */
+  if (!ReadUint8(encoded, size, &pos, &dict->num_transform_lists)) {
+    return BROTLI_FALSE;
+  }
+  if (dict->num_transform_lists > SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS) {
+    return BROTLI_FALSE;
+  }
+
+  if (dict->num_transform_lists != 0) {
+    dict->transforms_instances = (BrotliTransforms*)dict->alloc_func(
+        dict->memory_manager_opaque,
+        dict->num_transform_lists * sizeof(*dict->transforms_instances));
+    if (!dict->transforms_instances) return BROTLI_FALSE;  /* OOM */
+  }
+  for (i = 0; i < dict->num_transform_lists; i++) {
+    BROTLI_BOOL ok = BROTLI_FALSE;
+    size_t prefix_suffix_count = 0;
+    trasform_list_start[i] = pos;
+    dict->transforms_instances[i].prefix_suffix_map =
+        temporary_prefix_suffix_table;
+    ok = ParseTransformsList(
+        size, encoded, &pos, &dict->transforms_instances[i],
+        temporary_prefix_suffix_table, &prefix_suffix_count);
+    if (!ok) return BROTLI_FALSE;
+    total_prefix_suffix_count += prefix_suffix_count;
+  }
+  if (total_prefix_suffix_count != 0) {
+    dict->prefix_suffix_maps = (uint16_t*)dict->alloc_func(
+        dict->memory_manager_opaque,
+        total_prefix_suffix_count * sizeof(*dict->prefix_suffix_maps));
+    if (!dict->prefix_suffix_maps) return BROTLI_FALSE;  /* OOM */
+  }
+  total_prefix_suffix_count = 0;
+  for (i = 0; i < dict->num_transform_lists; i++) {
+    size_t prefix_suffix_count = 0;
+    size_t position = trasform_list_start[i];
+    uint16_t* prefix_suffix_map =
+      &dict->prefix_suffix_maps[total_prefix_suffix_count];
+    BROTLI_BOOL ok = ParsePrefixSuffixTable(
+        size, encoded, &position, &dict->transforms_instances[i],
+        prefix_suffix_map, &prefix_suffix_count);
+    if (!ok) return BROTLI_FALSE;
+    dict->transforms_instances[i].prefix_suffix_map = prefix_suffix_map;
+    total_prefix_suffix_count += prefix_suffix_count;
+  }
+
+  if (dict->num_word_lists != 0 || dict->num_transform_lists != 0) {
+    if (!ReadUint8(encoded, size, &pos, &dict->num_dictionaries)) {
+      return BROTLI_FALSE;
+    }
+    if (dict->num_dictionaries == 0 ||
+        dict->num_dictionaries > SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS) {
+      return BROTLI_FALSE;
+    }
+    for (i = 0; i < dict->num_dictionaries; i++) {
+      uint8_t words_index;
+      uint8_t transforms_index;
+      if (!ReadUint8(encoded, size, &pos, &words_index)) {
+        return BROTLI_FALSE;
+      }
+      if (words_index > dict->num_word_lists) return BROTLI_FALSE;
+      if (!ReadUint8(encoded, size, &pos, &transforms_index)) {
+        return BROTLI_FALSE;
+      }
+      if (transforms_index > dict->num_transform_lists) return BROTLI_FALSE;
+      dict->words[i] = words_index == dict->num_word_lists ?
+          BrotliGetDictionary() : &dict->words_instances[words_index];
+      dict->transforms[i] = transforms_index == dict->num_transform_lists ?
+          BrotliGetTransforms(): &dict->transforms_instances[transforms_index];
+    }
+    /* CONTEXT_ENABLED */
+    if (!ReadBool(encoded, size, &pos, &dict->context_based)) {
+      return BROTLI_FALSE;
+    }
+
+    /* CONTEXT_MAP */
+    if (dict->context_based) {
+      for (i = 0; i < SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS; i++) {
+        if (!ReadUint8(encoded, size, &pos, &dict->context_map[i])) {
+          return BROTLI_FALSE;
+        }
+        if (dict->context_map[i] >= dict->num_dictionaries) {
+          return BROTLI_FALSE;
+        }
+      }
+    }
+  } else {
+    dict->context_based = BROTLI_FALSE;
+    dict->num_dictionaries = 1;
+    dict->words[0] = BrotliGetDictionary();
+    dict->transforms[0] = BrotliGetTransforms();
+  }
+
+  return BROTLI_TRUE;
+}
+
+/* Decodes shared dictionary and verifies correctness.
+   Returns BROTLI_TRUE if dictionary is valid, BROTLI_FALSE otherwise.
+   The BrotliSharedDictionary must already have been initialized. If the
+   BrotliSharedDictionary already contains data, compound dictionaries
+   will be appended, but an error will be returned if it already has
+   custom words or transforms.
+   TODO: link to RFC for shared brotli once published. */
+static BROTLI_BOOL DecodeSharedDictionary(
+    const uint8_t* encoded, size_t size, BrotliSharedDictionary* dict) {
+  uint32_t num_prefix = 0;
+  BROTLI_BOOL is_custom_static_dict = BROTLI_FALSE;
+  BROTLI_BOOL has_custom_static_dict =
+      dict->num_word_lists > 0 || dict->num_transform_lists > 0;
+
+  /* Check magic header bytes. */
+  if (size < 2) return BROTLI_FALSE;
+  if (encoded[0] != 0x91 || encoded[1] != 0) return BROTLI_FALSE;
+
+  if (!DryParseDictionary(encoded, size, &num_prefix, &is_custom_static_dict)) {
+    return BROTLI_FALSE;
+  }
+
+  if (num_prefix + dict->num_prefix > SHARED_BROTLI_MAX_COMPOUND_DICTS) {
+    return BROTLI_FALSE;
+  }
+
+  /* Cannot combine different static dictionaries, only prefix dictionaries */
+  if (has_custom_static_dict && is_custom_static_dict) return BROTLI_FALSE;
+
+  return ParseDictionary(encoded, size, dict);
+}
+
+void BrotliSharedDictionaryDestroyInstance(
+    BrotliSharedDictionary* dict) {
+  if (!dict) {
+    return;
+  } else {
+    brotli_free_func free_func = dict->free_func;
+    void* opaque = dict->memory_manager_opaque;
+    /* Cleanup. */
+    free_func(opaque, dict->words_instances);
+    free_func(opaque, dict->transforms_instances);
+    free_func(opaque, dict->prefix_suffix_maps);
+    /* Self-destruction. */
+    free_func(opaque, dict);
+  }
+}
+
+BROTLI_BOOL BrotliSharedDictionaryAttach(
+    BrotliSharedDictionary* dict, BrotliSharedDictionaryType type,
+    size_t data_size, const uint8_t* data) {
+  if (!dict) {
+    return BROTLI_FALSE;
+  }
+  if (type == BROTLI_SHARED_DICTIONARY_SERIALIZED) {
+    return DecodeSharedDictionary(data, data_size, dict);
+  } else if (type == BROTLI_SHARED_DICTIONARY_RAW) {
+    if (dict->num_prefix >= SHARED_BROTLI_MAX_COMPOUND_DICTS) {
+      return BROTLI_FALSE;
+    }
+    dict->prefix_size[dict->num_prefix] = data_size;
+    dict->prefix[dict->num_prefix] = data;
+    dict->num_prefix++;
+    return BROTLI_TRUE;
+  } else {
+    return BROTLI_FALSE;
+  }
+}
+
+BrotliSharedDictionary* BrotliSharedDictionaryCreateInstance(
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+  BrotliSharedDictionary* dict = 0;
+  if (!alloc_func && !free_func) {
+    dict = (BrotliSharedDictionary*)malloc(sizeof(BrotliSharedDictionary));
+  } else if (alloc_func && free_func) {
+    dict = (BrotliSharedDictionary*)alloc_func(
+        opaque, sizeof(BrotliSharedDictionary));
+  }
+  if (dict == 0) {
+    return 0;
+  }
+
+  /* TODO: explicitly initialize all the fields? */
+  memset(dict, 0, sizeof(BrotliSharedDictionary));
+
+  dict->context_based = BROTLI_FALSE;
+  dict->num_dictionaries = 1;
+  dict->num_word_lists = 0;
+  dict->num_transform_lists = 0;
+
+  dict->words[0] = BrotliGetDictionary();
+  dict->transforms[0] = BrotliGetTransforms();
+
+  dict->alloc_func = alloc_func ? alloc_func : BrotliDefaultAllocFunc;
+  dict->free_func = free_func ? free_func : BrotliDefaultFreeFunc;
+  dict->memory_manager_opaque = opaque;
+
+  return dict;
+}
+
+#if defined(__cplusplus) || defined(c_plusplus)
+}  /* extern "C" */
+#endif
diff --git a/c/common/shared_dictionary_internal.h b/c/common/shared_dictionary_internal.h
new file mode 100644
index 0000000..a2b76ee
--- /dev/null
+++ b/c/common/shared_dictionary_internal.h
@@ -0,0 +1,74 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* (Transparent) Shared Dictionary definition. */
+
+#ifndef BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_
+#define BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_
+
+#include "./dictionary.h"
+#include <brotli/shared_dictionary.h>
+#include "./transform.h"
+#include <brotli/types.h>
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+struct BrotliSharedDictionaryStruct {
+  /* LZ77 prefixes (compound dictionary). */
+  uint32_t num_prefix;  /* max SHARED_BROTLI_MAX_COMPOUND_DICTS */
+  size_t prefix_size[SHARED_BROTLI_MAX_COMPOUND_DICTS];
+  const uint8_t* prefix[SHARED_BROTLI_MAX_COMPOUND_DICTS];
+
+  /* If set, the context map is used to select word and transform list from 64
+     contexts, if not set, the context map is not used and only words[0] and
+     transforms[0] are to be used. */
+  BROTLI_BOOL context_based;
+
+  uint8_t context_map[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+  /* Amount of word_list+transform_list combinations. */
+  uint8_t num_dictionaries;
+
+  /* Must use num_dictionaries values. */
+  const BrotliDictionary* words[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+  /* Must use num_dictionaries values. */
+  const BrotliTransforms* transforms[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+  /* Amount of custom word lists. May be 0 if only Brotli's built-in is used */
+  uint8_t num_word_lists;
+
+  /* Contents of the custom words lists. Must be NULL if num_word_lists is 0. */
+  BrotliDictionary* words_instances;
+
+  /* Amount of custom transform lists. May be 0 if only Brotli's built-in is
+     used */
+  uint8_t num_transform_lists;
+
+  /* Contents of the custom transform lists. Must be NULL if num_transform_lists
+     is 0. */
+  BrotliTransforms* transforms_instances;
+
+  /* Concatenated prefix_suffix_maps of the custom transform lists. Must be NULL
+     if num_transform_lists is 0. */
+  uint16_t* prefix_suffix_maps;
+
+  /* Memory management */
+  brotli_alloc_func alloc_func;
+  brotli_free_func free_func;
+  void* memory_manager_opaque;
+};
+
+typedef struct BrotliSharedDictionaryStruct BrotliSharedDictionaryInternal;
+#define BrotliSharedDictionary BrotliSharedDictionaryInternal
+
+#if defined(__cplusplus) || defined(c_plusplus)
+}  /* extern "C" */
+#endif
+
+#endif  /* BROTLI_COMMON_SHARED_DICTIONARY_INTERNAL_H_ */
diff --git a/c/dec/decode.c b/c/dec/decode.c
index 7eee968..12cd502 100644
--- a/c/dec/decode.c
+++ b/c/dec/decode.c
@@ -13,6 +13,7 @@
 #include "../common/context.h"
 #include "../common/dictionary.h"
 #include "../common/platform.h"
+#include "../common/shared_dictionary_internal.h"
 #include "../common/transform.h"
 #include "../common/version.h"
 #include "./bit_reader.h"
@@ -42,8 +43,8 @@
 /* We need the slack region for the following reasons:
     - doing up to two 16-byte copies for fast backward copying
     - inserting transformed dictionary word:
-        5 prefix + 24 base + 8 suffix */
-static const uint32_t kRingBufferWriteAheadSlack = 42;
+        255 prefix + 32 base + 255 suffix */
+static const uint32_t kRingBufferWriteAheadSlack = 542;
 
 static const uint8_t kCodeLengthCodeOrder[BROTLI_CODE_LENGTH_CODES] = {
   1, 2, 3, 4, 0, 5, 17, 6, 16, 7, 8, 9, 10, 11, 12, 13, 14, 15,
@@ -1403,6 +1404,114 @@
   BROTLI_DCHECK(0);  /* Unreachable */
 }
 
+static BROTLI_BOOL AttachCompoundDictionary(
+    BrotliDecoderState* state, const uint8_t* data, size_t size) {
+  BrotliDecoderCompoundDictionary* addon = state->compound_dictionary;
+  if (state->state != BROTLI_STATE_UNINITED) return BROTLI_FALSE;
+  if (!addon) {
+    addon = (BrotliDecoderCompoundDictionary*)BROTLI_DECODER_ALLOC(
+        state, sizeof(BrotliDecoderCompoundDictionary));
+    if (!addon) return BROTLI_FALSE;
+    addon->num_chunks = 0;
+    addon->total_size = 0;
+    addon->br_length = 0;
+    addon->br_copied = 0;
+    addon->block_bits = -1;
+    addon->chunk_offsets[0] = 0;
+    state->compound_dictionary = addon;
+  }
+  if (addon->num_chunks == 15) return BROTLI_FALSE;
+  addon->chunks[addon->num_chunks] = data;
+  addon->num_chunks++;
+  addon->total_size += (int)size;
+  addon->chunk_offsets[addon->num_chunks] = addon->total_size;
+  return BROTLI_TRUE;
+}
+
+static void EnsureCoumpoundDictionaryInitialized(BrotliDecoderState* state) {
+  BrotliDecoderCompoundDictionary* addon = state->compound_dictionary;
+  /* 256 = (1 << 8) slots in block map. */
+  int block_bits = 8;
+  int cursor = 0;
+  int index = 0;
+  if (addon->block_bits != -1) return;
+  while (((addon->total_size - 1) >> block_bits) != 0) block_bits++;
+  block_bits -= 8;
+  addon->block_bits = block_bits;
+  while (cursor < addon->total_size) {
+    while (addon->chunk_offsets[index + 1] < cursor) index++;
+    addon->block_map[cursor >> block_bits] = (uint8_t)index;
+    cursor += 1 << block_bits;
+  }
+}
+
+static BROTLI_BOOL InitializeCompoundDictionaryCopy(BrotliDecoderState* s,
+    int address, int length) {
+  BrotliDecoderCompoundDictionary* addon = s->compound_dictionary;
+  int index;
+  EnsureCoumpoundDictionaryInitialized(s);
+  index = addon->block_map[address >> addon->block_bits];
+  while (address >= addon->chunk_offsets[index + 1]) index++;
+  if (addon->total_size < address + length) return BROTLI_FALSE;
+  /* Update the recent distances cache. */
+  s->dist_rb[s->dist_rb_idx & 3] = s->distance_code;
+  ++s->dist_rb_idx;
+  s->meta_block_remaining_len -= length;
+  addon->br_index = index;
+  addon->br_offset = address - addon->chunk_offsets[index];
+  addon->br_length = length;
+  addon->br_copied = 0;
+  return BROTLI_TRUE;
+}
+
+static int GetCompoundDictionarySize(BrotliDecoderState* s) {
+  return s->compound_dictionary ? s->compound_dictionary->total_size : 0;
+}
+
+static int CopyFromCompoundDictionary(BrotliDecoderState* s, int pos) {
+  BrotliDecoderCompoundDictionary* addon = s->compound_dictionary;
+  int orig_pos = pos;
+  while (addon->br_length != addon->br_copied) {
+    uint8_t* copy_dst = &s->ringbuffer[pos];
+    const uint8_t* copy_src =
+        addon->chunks[addon->br_index] + addon->br_offset;
+    int space = s->ringbuffer_size - pos;
+    int rem_chunk_length = (addon->chunk_offsets[addon->br_index + 1] -
+        addon->chunk_offsets[addon->br_index]) - addon->br_offset;
+    int length = addon->br_length - addon->br_copied;
+    if (length > rem_chunk_length) length = rem_chunk_length;
+    if (length > space) length = space;
+    memcpy(copy_dst, copy_src, (size_t)length);
+    pos += length;
+    addon->br_offset += length;
+    addon->br_copied += length;
+    if (length == rem_chunk_length) {
+      addon->br_index++;
+      addon->br_offset = 0;
+    }
+    if (pos == s->ringbuffer_size) break;
+  }
+  return pos - orig_pos;
+}
+
+BROTLI_BOOL BrotliDecoderAttachDictionary(BrotliDecoderState* state,
+    BrotliSharedDictionaryType type, size_t data_size, const uint8_t* data) {
+  uint32_t i;
+  uint32_t num_prefix_before = state->dictionary->num_prefix;
+  if (state->state != BROTLI_STATE_UNINITED) return BROTLI_FALSE;
+  if (!BrotliSharedDictionaryAttach(state->dictionary, type, data_size, data)) {
+    return BROTLI_FALSE;
+  }
+  for (i = num_prefix_before; i < state->dictionary->num_prefix; i++) {
+    if (!AttachCompoundDictionary(
+        state, state->dictionary->prefix[i],
+        state->dictionary->prefix_size[i])) {
+      return BROTLI_FALSE;
+    }
+  }
+  return BROTLI_TRUE;
+}
+
 /* Calculates the smallest feasible ring buffer.
 
    If we know the data size is small, do not allocate more ring buffer
@@ -1737,6 +1846,7 @@
   int i = s->loop_counter;
   BrotliDecoderErrorCode result = BROTLI_DECODER_SUCCESS;
   BrotliBitReader* br = &s->br;
+  int compound_dictionary_size = GetCompoundDictionarySize(s);
 
   if (!CheckInputAmount(safe, br, 28)) {
     result = BROTLI_DECODER_NEEDS_MORE_INPUT;
@@ -1903,20 +2013,75 @@
           pos, s->distance_code, i, s->meta_block_remaining_len));
       return BROTLI_FAILURE(BROTLI_DECODER_ERROR_FORMAT_DISTANCE);
     }
-    if (i >= BROTLI_MIN_DICTIONARY_WORD_LENGTH &&
-        i <= BROTLI_MAX_DICTIONARY_WORD_LENGTH) {
-      int address = s->distance_code - s->max_distance - 1;
-      const BrotliDictionary* words = s->dictionary;
-      const BrotliTransforms* transforms = s->transforms;
-      int offset = (int)s->dictionary->offsets_by_length[i];
-      uint32_t shift = s->dictionary->size_bits_by_length[i];
-
+    if (s->distance_code - s->max_distance - 1 < compound_dictionary_size) {
+      int address = compound_dictionary_size -
+          (s->distance_code - s->max_distance);
+      if (!InitializeCompoundDictionaryCopy(s, address, i)) {
+        return BROTLI_FAILURE(BROTLI_DECODER_ERROR_COMPOUND_DICTIONARY);
+      }
+      pos += CopyFromCompoundDictionary(s, pos);
+      if (pos >= s->ringbuffer_size) {
+        s->state = BROTLI_STATE_COMMAND_POST_WRITE_1;
+        goto saveStateAndReturn;
+      }
+    } else if (i >= SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH &&
+               i <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH) {
+      uint8_t p1 = s->ringbuffer[(pos - 1) & s->ringbuffer_mask];
+      uint8_t p2 = s->ringbuffer[(pos - 2) & s->ringbuffer_mask];
+      uint8_t dict_id = s->dictionary->context_based ?
+          s->dictionary->context_map[BROTLI_CONTEXT(p1, p2, s->context_lookup)]
+          : 0;
+      const BrotliDictionary* words = s->dictionary->words[dict_id];
+      const BrotliTransforms* transforms = s->dictionary->transforms[dict_id];
+      int offset = (int)words->offsets_by_length[i];
+      uint32_t shift = words->size_bits_by_length[i];
+      int address =
+          s->distance_code - s->max_distance - 1 - compound_dictionary_size;
       int mask = (int)BitMask(shift);
       int word_idx = address & mask;
       int transform_idx = address >> shift;
       /* Compensate double distance-ring-buffer roll. */
       s->dist_rb_idx += s->distance_context;
       offset += word_idx * i;
+      /* If the distance is out of bound, select a next static dictionary if
+         there exist multiple. */
+      if ((transform_idx >= (int)transforms->num_transforms ||
+          words->size_bits_by_length[i] == 0) &&
+          s->dictionary->num_dictionaries > 1) {
+        uint8_t dict_id2;
+        int dist_remaining = address -
+            (int)(((1u << shift) & ~1u)) * (int)transforms->num_transforms;
+        for (dict_id2 = 0; dict_id2 < s->dictionary->num_dictionaries;
+            dict_id2++) {
+          const BrotliDictionary* words2 = s->dictionary->words[dict_id2];
+          if (dict_id2 != dict_id && words2->size_bits_by_length[i] != 0) {
+            const BrotliTransforms* transforms2 =
+                s->dictionary->transforms[dict_id2];
+            uint32_t shift2 = words2->size_bits_by_length[i];
+            int num = (int)((1u << shift2) & ~1u) *
+                (int)transforms2->num_transforms;
+            if (dist_remaining < num) {
+              dict_id = dict_id2;
+              words = words2;
+              transforms = transforms2;
+              address = dist_remaining;
+              shift = shift2;
+              mask = (int)BitMask(shift);
+              word_idx = address & mask;
+              transform_idx = address >> shift;
+              offset = (int)words->offsets_by_length[i] + word_idx * i;
+              break;
+            }
+            dist_remaining -= num;
+          }
+        }
+      }
+      if (BROTLI_PREDICT_FALSE(words->size_bits_by_length[i] == 0)) {
+        BROTLI_LOG(("Invalid backward reference. pos: %d distance: %d "
+            "len: %d bytes left: %d\n",
+            pos, s->distance_code, i, s->meta_block_remaining_len));
+        return BROTLI_FAILURE(BROTLI_DECODER_ERROR_FORMAT_DICTIONARY);
+      }
       if (BROTLI_PREDICT_FALSE(!words->data)) {
         return BROTLI_FAILURE(BROTLI_DECODER_ERROR_DICTIONARY_NOT_SET);
       }
@@ -1933,6 +2098,10 @@
           BROTLI_LOG(("[ProcessCommandsInternal] dictionary word: [%.*s],"
                       " transform_idx = %d, transformed: [%.*s]\n",
                       i, word, transform_idx, len, &s->ringbuffer[pos]));
+          if (len == 0 && s->distance_code <= 120) {
+            BROTLI_LOG(("Invalid length-0 dictionary word after transform\n"));
+            return BROTLI_FAILURE(BROTLI_DECODER_ERROR_FORMAT_TRANSFORM);
+          }
         }
         pos += len;
         s->meta_block_remaining_len -= len;
@@ -2483,6 +2652,11 @@
           s->max_distance = s->max_backward_distance;
         }
         if (s->state == BROTLI_STATE_COMMAND_POST_WRITE_1) {
+          BrotliDecoderCompoundDictionary* addon = s->compound_dictionary;
+          if (addon && (addon->br_length != addon->br_copied)) {
+            s->pos += CopyFromCompoundDictionary(s, s->pos);
+            if (s->pos >= s->ringbuffer_size) continue;
+          }
           if (s->meta_block_remaining_len == 0) {
             /* Next metablock, if any. */
             s->state = BROTLI_STATE_METABLOCK_DONE;
diff --git a/c/dec/state.c b/c/dec/state.c
index f847836..66d6820 100644
--- a/c/dec/state.c
+++ b/c/dec/state.c
@@ -8,6 +8,7 @@
 
 #include <stdlib.h>  /* free, malloc */
 
+#include "../common/dictionary.h"
 #include <brotli/types.h>
 #include "./huffman.h"
 
@@ -81,8 +82,10 @@
 
   s->mtf_upper_bound = 63;
 
-  s->dictionary = BrotliGetDictionary();
-  s->transforms = BrotliGetTransforms();
+  s->compound_dictionary = NULL;
+  s->dictionary =
+      BrotliSharedDictionaryCreateInstance(alloc_func, free_func, opaque);
+  if (!s->dictionary) return BROTLI_FALSE;
 
   return BROTLI_TRUE;
 }
@@ -129,6 +132,9 @@
 void BrotliDecoderStateCleanup(BrotliDecoderState* s) {
   BrotliDecoderStateCleanupAfterMetablock(s);
 
+  BROTLI_DECODER_FREE(s, s->compound_dictionary);
+  BrotliSharedDictionaryDestroyInstance(s->dictionary);
+  s->dictionary = NULL;
   BROTLI_DECODER_FREE(s, s->ringbuffer);
   BROTLI_DECODER_FREE(s, s->block_type_trees);
 }
diff --git a/c/dec/state.h b/c/dec/state.h
index 54dab69..7127581 100644
--- a/c/dec/state.h
+++ b/c/dec/state.h
@@ -12,6 +12,7 @@
 #include "../common/constants.h"
 #include "../common/dictionary.h"
 #include "../common/platform.h"
+#include <brotli/shared_dictionary.h>
 #include "../common/transform.h"
 #include <brotli/types.h>
 #include "./bit_reader.h"
@@ -189,6 +190,20 @@
   BROTLI_STATE_READ_BLOCK_LENGTH_SUFFIX
 } BrotliRunningReadBlockLengthState;
 
+/* BrotliDecoderState addon, used for Compound Dictionary functionality. */
+typedef struct BrotliDecoderCompoundDictionary {
+  int num_chunks;
+  int total_size;
+  int br_index;
+  int br_offset;
+  int br_length;
+  int br_copied;
+  const uint8_t* chunks[16];
+  int chunk_offsets[16];
+  int block_bits;
+  uint8_t block_map[256];
+} BrotliDecoderCompoundDictionary;
+
 typedef struct BrotliMetablockHeaderArena {
   BrotliRunningTreeGroupState substate_tree_group;
   BrotliRunningContextMapState substate_context_map;
@@ -327,8 +342,8 @@
   uint8_t* context_map;
   uint8_t* context_modes;
 
-  const BrotliDictionary* dictionary;
-  const BrotliTransforms* transforms;
+  BrotliSharedDictionary* dictionary;
+  BrotliDecoderCompoundDictionary* compound_dictionary;
 
   uint32_t trivial_literal_contexts[8];  /* 256 bits */
 
diff --git a/c/enc/backward_references.c b/c/enc/backward_references.c
index a07a617..1ae1146 100644
--- a/c/enc/backward_references.c
+++ b/c/enc/backward_references.c
@@ -9,12 +9,13 @@
 #include "./backward_references.h"
 
 #include "../common/constants.h"
-#include "../common/context.h"
 #include "../common/dictionary.h"
 #include "../common/platform.h"
 #include <brotli/types.h>
 #include "./command.h"
+#include "./compound_dictionary.h"
 #include "./dictionary_hash.h"
+#include "./encoder_dict.h"
 #include "./memory.h"
 #include "./quality.h"
 
@@ -52,6 +53,7 @@
 #define EXPORT_FN(X) EXPAND_CAT(X, EXPAND_CAT(PREFIX(), HASHER()))
 
 #define PREFIX() N
+#define ENABLE_COMPOUND_DICTIONARY 0
 
 #define HASHER() H2
 /* NOLINTNEXTLINE(build/include) */
@@ -113,6 +115,41 @@
 #include "./backward_references_inc.h"
 #undef HASHER
 
+#undef ENABLE_COMPOUND_DICTIONARY
+#undef PREFIX
+#define PREFIX() D
+#define ENABLE_COMPOUND_DICTIONARY 1
+
+#define HASHER() H5
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H6
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H40
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H41
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H42
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H55
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+#define HASHER() H65
+/* NOLINTNEXTLINE(build/include) */
+#include "./backward_references_inc.h"
+#undef HASHER
+
+#undef ENABLE_COMPOUND_DICTIONARY
 #undef PREFIX
 
 #undef EXPORT_FN
@@ -125,6 +162,28 @@
     ContextLut literal_context_lut, const BrotliEncoderParams* params,
     Hasher* hasher, int* dist_cache, size_t* last_insert_len,
     Command* commands, size_t* num_commands, size_t* num_literals) {
+  if (params->dictionary.compound.num_chunks != 0) {
+    switch (params->hasher.type) {
+#define CASE_(N)                                                    \
+      case N:                                                       \
+        CreateBackwardReferencesDH ## N(num_bytes,                  \
+            position, ringbuffer, ringbuffer_mask,                  \
+            literal_context_lut, params, hasher, dist_cache,        \
+            last_insert_len, commands, num_commands, num_literals); \
+        return;
+      CASE_(5)
+      CASE_(6)
+      CASE_(40)
+      CASE_(41)
+      CASE_(42)
+      CASE_(55)
+      CASE_(65)
+#undef CASE_
+      default:
+        break;
+    }
+  }
+
   switch (params->hasher.type) {
 #define CASE_(N)                                                  \
     case N:                                                       \
diff --git a/c/enc/backward_references_hq.c b/c/enc/backward_references_hq.c
index 61b7ff1..5e17c84 100644
--- a/c/enc/backward_references_hq.c
+++ b/c/enc/backward_references_hq.c
@@ -11,10 +11,11 @@
 #include <string.h>  /* memcpy, memset */
 
 #include "../common/constants.h"
-#include "../common/context.h"
 #include "../common/platform.h"
 #include <brotli/types.h>
 #include "./command.h"
+#include "./compound_dictionary.h"
+#include "./encoder_dict.h"
 #include "./fast_log.h"
 #include "./find_match_length.h"
 #include "./literal_cost.h"
@@ -430,7 +431,8 @@
   size_t min_len;
   size_t result = 0;
   size_t k;
-  size_t gap = 0;
+  const CompoundDictionary* addon = &params->dictionary.compound;
+  size_t gap = addon->total_size;
 
   EvaluateNode(block_start + stream_offset, pos, max_backward_limit, gap,
       starting_dist_cache, model, queue, nodes);
@@ -484,6 +486,24 @@
         len = FindMatchLengthWithLimit(&ringbuffer[prev_ix],
                                        &ringbuffer[cur_ix_masked],
                                        max_len);
+      } else if (backward > dictionary_start) {
+        size_t d = 0;
+        size_t offset;
+        size_t limit;
+        const uint8_t* source;
+        offset = dictionary_start + 1 + addon->total_size - 1;
+        while (offset >= backward + addon->chunk_offsets[d + 1]) d++;
+        source = addon->chunk_source[d];
+        offset = offset - addon->chunk_offsets[d] - backward;
+        limit = addon->chunk_offsets[d + 1] - addon->chunk_offsets[d] - offset;
+        limit = limit > max_len ? max_len : limit;
+        if (best_len >= limit ||
+            continuation != source[offset + best_len]) {
+          continue;
+        }
+        len = FindMatchLengthWithLimit(&source[offset],
+                                       &ringbuffer[cur_ix_masked],
+                                       limit);
       } else {
         /* "Gray" area. It is addressable by decoder, but this encoder
            instance does not have that data -> should not touch it. */
@@ -589,7 +609,7 @@
   size_t pos = 0;
   uint32_t offset = nodes[0].u.next;
   size_t i;
-  size_t gap = 0;
+  size_t gap = params->dictionary.compound.total_size;
   for (i = 0; offset != BROTLI_UINT32_MAX; i++) {
     const ZopfliNode* next = &nodes[pos + offset];
     size_t copy_length = ZopfliNodeCopyLength(next);
@@ -665,6 +685,23 @@
   return ComputeShortestPathFromNodes(num_bytes, nodes);
 }
 
+static void MergeMatches(BackwardMatch* dst,
+    BackwardMatch* src1, size_t len1, BackwardMatch* src2, size_t len2) {
+  while (len1 > 0 && len2 > 0) {
+    size_t l1 = BackwardMatchLength(src1);
+    size_t l2 = BackwardMatchLength(src2);
+    if (l1 < l2 || ((l1 == l2) && (src1->distance < src2->distance))) {
+      *dst++ = *src1++;
+      len1--;
+    } else {
+      *dst++ = *src2++;
+      len2--;
+    }
+  }
+  while (len1-- > 0) *dst++ = *src1++;
+  while (len2-- > 0) *dst++ = *src2++;
+}
+
 /* REQUIRES: nodes != NULL and len(nodes) >= num_bytes + 1 */
 size_t BrotliZopfliComputeShortestPath(MemoryManager* m, size_t num_bytes,
     size_t position, const uint8_t* ringbuffer, size_t ringbuffer_mask,
@@ -679,10 +716,11 @@
   const size_t store_end = num_bytes >= StoreLookaheadH10() ?
       position + num_bytes - StoreLookaheadH10() + 1 : position;
   size_t i;
-  size_t gap = 0;
-  size_t lz_matches_offset = 0;
+  const CompoundDictionary* addon = &params->dictionary.compound;
+  size_t gap = addon->total_size;
+  size_t lz_matches_offset =
+      (addon->num_chunks != 0) ? (MAX_NUM_MATCHES_H10 + 128) : 0;
   ZopfliCostModel* model = BROTLI_ALLOC(m, ZopfliCostModel, 1);
-  BROTLI_UNUSED(literal_context_lut);
   if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(model) || BROTLI_IS_NULL(matches)) {
     return 0;
   }
@@ -700,10 +738,28 @@
         pos + stream_offset, max_backward_limit);
     size_t skip;
     size_t num_matches;
+    int dict_id = 0;
+    if (params->dictionary.contextual.context_based) {
+      uint8_t p1 = pos >= 1 ?
+          ringbuffer[(size_t)(pos - 1) & ringbuffer_mask] : 0;
+      uint8_t p2 = pos >= 2 ?
+          ringbuffer[(size_t)(pos - 2) & ringbuffer_mask] : 0;
+      dict_id = params->dictionary.contextual.context_map[
+          BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+    }
     num_matches = FindAllMatchesH10(&hasher->privat._H10,
-        &params->dictionary,
+        params->dictionary.contextual.dict[dict_id],
         ringbuffer, ringbuffer_mask, pos, num_bytes - i, max_distance,
         dictionary_start + gap, params, &matches[lz_matches_offset]);
+    if (addon->num_chunks != 0) {
+      size_t cd_matches = LookupAllCompoundDictionaryMatches(addon,
+          ringbuffer, ringbuffer_mask, pos, 3, num_bytes - i,
+          dictionary_start, params->dist.max_distance,
+          &matches[lz_matches_offset - 64], 64);
+      MergeMatches(matches, &matches[lz_matches_offset - 64], cd_matches,
+          &matches[lz_matches_offset], num_matches);
+      num_matches += cd_matches;
+    }
     if (num_matches > 0 &&
         BackwardMatchLength(&matches[num_matches - 1]) > max_zopfli_len) {
       matches[0] = matches[num_matches - 1];
@@ -774,9 +830,10 @@
   ZopfliCostModel* model = BROTLI_ALLOC(m, ZopfliCostModel, 1);
   ZopfliNode* nodes;
   BackwardMatch* matches = BROTLI_ALLOC(m, BackwardMatch, matches_size);
-  size_t gap = 0;
-  size_t shadow_matches = 0;
-  BROTLI_UNUSED(literal_context_lut);
+  const CompoundDictionary* addon = &params->dictionary.compound;
+  size_t gap = addon->total_size;
+  size_t shadow_matches =
+      (addon->num_chunks != 0) ? (MAX_NUM_MATCHES_H10 + 128) : 0;
   if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(model) ||
       BROTLI_IS_NULL(num_matches) || BROTLI_IS_NULL(matches)) {
     return;
@@ -790,15 +847,34 @@
     size_t num_found_matches;
     size_t cur_match_end;
     size_t j;
+    int dict_id = 0;
+    if (params->dictionary.contextual.context_based) {
+      uint8_t p1 = pos >= 1 ?
+          ringbuffer[(size_t)(pos - 1) & ringbuffer_mask] : 0;
+      uint8_t p2 = pos >= 2 ?
+          ringbuffer[(size_t)(pos - 2) & ringbuffer_mask] : 0;
+      dict_id = params->dictionary.contextual.context_map[
+          BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+    }
     /* Ensure that we have enough free slots. */
     BROTLI_ENSURE_CAPACITY(m, BackwardMatch, matches, matches_size,
         cur_match_pos + MAX_NUM_MATCHES_H10 + shadow_matches);
     if (BROTLI_IS_OOM(m)) return;
     num_found_matches = FindAllMatchesH10(&hasher->privat._H10,
-        &params->dictionary,
+        params->dictionary.contextual.dict[dict_id],
         ringbuffer, ringbuffer_mask, pos, max_length,
         max_distance, dictionary_start + gap, params,
         &matches[cur_match_pos + shadow_matches]);
+    if (addon->num_chunks != 0) {
+      size_t cd_matches = LookupAllCompoundDictionaryMatches(addon,
+          ringbuffer, ringbuffer_mask, pos, 3, max_length,
+          dictionary_start, params->dist.max_distance,
+          &matches[cur_match_pos + shadow_matches - 64], 64);
+      MergeMatches(&matches[cur_match_pos],
+          &matches[cur_match_pos + shadow_matches - 64], cd_matches,
+          &matches[cur_match_pos + shadow_matches], num_found_matches);
+      num_found_matches += cd_matches;
+    }
     cur_match_end = cur_match_pos + num_found_matches;
     for (j = cur_match_pos; j + 1 < cur_match_end; ++j) {
       BROTLI_DCHECK(BackwardMatchLength(&matches[j]) <=
diff --git a/c/enc/backward_references_inc.h b/c/enc/backward_references_inc.h
index 766bf91..752c12e 100644
--- a/c/enc/backward_references_inc.h
+++ b/c/enc/backward_references_inc.h
@@ -28,13 +28,11 @@
   const size_t random_heuristics_window_size =
       LiteralSpreeLengthForSparseSearch(params);
   size_t apply_random_heuristics = position + random_heuristics_window_size;
-  const size_t gap = 0;
+  const size_t gap = params->dictionary.compound.total_size;
 
   /* Minimum score to accept a backward reference. */
   const score_t kMinScore = BROTLI_SCORE_BASE + 100;
 
-  BROTLI_UNUSED(literal_context_lut);
-
   FN(PrepareDistanceCache)(privat, dist_cache);
 
   while (position + FN(HashTypeLength)() < pos_end) {
@@ -43,13 +41,29 @@
     size_t dictionary_start = BROTLI_MIN(size_t,
         position + position_offset, max_backward_limit);
     HasherSearchResult sr;
+    int dict_id = 0;
+    uint8_t p1 = 0;
+    uint8_t p2 = 0;
+    if (params->dictionary.contextual.context_based) {
+      p1 = position >= 1 ?
+          ringbuffer[(size_t)(position - 1) & ringbuffer_mask] : 0;
+      p2 = position >= 2 ?
+          ringbuffer[(size_t)(position - 2) & ringbuffer_mask] : 0;
+      dict_id = params->dictionary.contextual.context_map[
+          BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+    }
     sr.len = 0;
     sr.len_code_delta = 0;
     sr.distance = 0;
     sr.score = kMinScore;
-    FN(FindLongestMatch)(privat, &params->dictionary,
+    FN(FindLongestMatch)(privat, params->dictionary.contextual.dict[dict_id],
         ringbuffer, ringbuffer_mask, dist_cache, position, max_length,
         max_distance, dictionary_start + gap, params->dist.max_distance, &sr);
+    if (ENABLE_COMPOUND_DICTIONARY) {
+      LookupCompoundDictionaryMatch(&params->dictionary.compound, ringbuffer,
+          ringbuffer_mask, dist_cache, position, max_length,
+          dictionary_start, params->dist.max_distance, &sr);
+    }
     if (sr.score > kMinScore) {
       /* Found a match. Let's look for something even better ahead. */
       int delayed_backward_references_in_row = 0;
@@ -65,11 +79,23 @@
         max_distance = BROTLI_MIN(size_t, position + 1, max_backward_limit);
         dictionary_start = BROTLI_MIN(size_t,
             position + 1 + position_offset, max_backward_limit);
+        if (params->dictionary.contextual.context_based) {
+          p2 = p1;
+          p1 = ringbuffer[position & ringbuffer_mask];
+          dict_id = params->dictionary.contextual.context_map[
+              BROTLI_CONTEXT(p1, p2, literal_context_lut)];
+        }
         FN(FindLongestMatch)(privat,
-            &params->dictionary,
+            params->dictionary.contextual.dict[dict_id],
             ringbuffer, ringbuffer_mask, dist_cache, position + 1, max_length,
             max_distance, dictionary_start + gap, params->dist.max_distance,
             &sr2);
+        if (ENABLE_COMPOUND_DICTIONARY) {
+          LookupCompoundDictionaryMatch(
+              &params->dictionary.compound, ringbuffer,
+              ringbuffer_mask, dist_cache, position + 1, max_length,
+              dictionary_start, params->dist.max_distance, &sr2);
+        }
         if (sr2.score >= sr.score + cost_diff_lazy) {
           /* Ok, let's just write one byte for now and start a match from the
              next byte. */
diff --git a/c/enc/compound_dictionary.c b/c/enc/compound_dictionary.c
new file mode 100644
index 0000000..f2a3f37
--- /dev/null
+++ b/c/enc/compound_dictionary.c
@@ -0,0 +1,200 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+#include "./compound_dictionary.h"
+
+#include "../common/platform.h"
+#include <brotli/types.h>
+#include "./memory.h"
+#include "./quality.h"
+
+static PreparedDictionary* CreatePreparedDictionaryWithParams(MemoryManager* m,
+    const uint8_t* source, size_t source_size, uint32_t bucket_bits,
+    uint32_t slot_bits, uint32_t hash_bits, uint16_t bucket_limit) {
+  /* Step 1: create "bloated" hasher. */
+  uint32_t num_slots = 1u << slot_bits;
+  uint32_t num_buckets = 1u << bucket_bits;
+  uint32_t hash_shift = 64u - bucket_bits;
+  uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+  uint32_t slot_mask = num_slots - 1;
+  size_t alloc_size = (sizeof(uint32_t) << slot_bits) +
+      (sizeof(uint32_t) << slot_bits) +
+      (sizeof(uint16_t) << bucket_bits) +
+      (sizeof(uint32_t) << bucket_bits) +
+      (sizeof(uint32_t) * source_size);
+  uint8_t* flat = NULL;
+  PreparedDictionary* result = NULL;
+  uint16_t* num = NULL;
+  uint32_t* bucket_heads = NULL;
+  uint32_t* next_bucket = NULL;
+  uint32_t* slot_offsets = NULL;
+  uint16_t* heads = NULL;
+  uint32_t* items = NULL;
+  uint8_t* source_copy = NULL;
+  uint32_t i;
+  uint32_t* slot_size = NULL;
+  uint32_t* slot_limit = NULL;
+  uint32_t total_items = 0;
+  if (slot_bits > 16) return NULL;
+  if (slot_bits > bucket_bits) return NULL;
+  if (bucket_bits - slot_bits >= 16) return NULL;
+
+  flat = BROTLI_ALLOC(m, uint8_t, alloc_size);
+  if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(flat)) return NULL;
+
+  slot_size = (uint32_t*)flat;
+  slot_limit = (uint32_t*)(&slot_size[num_slots]);
+  num = (uint16_t*)(&slot_limit[num_slots]);
+  bucket_heads = (uint32_t*)(&num[num_buckets]);
+  next_bucket = (uint32_t*)(&bucket_heads[num_buckets]);
+  memset(num, 0, num_buckets * sizeof(num[0]));
+
+  /* TODO: apply custom "store" order. */
+  for (i = 0; i + 7 < source_size; ++i) {
+    const uint64_t h = (BROTLI_UNALIGNED_LOAD64LE(&source[i]) & hash_mask) *
+        kPreparedDictionaryHashMul64Long;
+    const uint32_t key = (uint32_t)(h >> hash_shift);
+    uint16_t count = num[key];
+    next_bucket[i] = (count == 0) ? ((uint32_t)(-1)) : bucket_heads[key];
+    bucket_heads[key] = i;
+    count++;
+    if (count > bucket_limit) count = bucket_limit;
+    num[key] = count;
+  }
+
+  /* Step 2: find slot limits. */
+  for (i = 0; i < num_slots; ++i) {
+    BROTLI_BOOL overflow = BROTLI_FALSE;
+    slot_limit[i] = bucket_limit;
+    while (BROTLI_TRUE) {
+      uint32_t limit = slot_limit[i];
+      size_t j;
+      uint32_t count = 0;
+      overflow = BROTLI_FALSE;
+      for (j = i; j < num_buckets; j += num_slots) {
+        uint32_t size = num[j];
+        /* Last chain may span behind 64K limit; overflow happens only if
+           we are about to use 0xFFFF+ as item offset. */
+        if (count >= 0xFFFF) {
+          overflow = BROTLI_TRUE;
+          break;
+        }
+        if (size > limit) size = limit;
+        count += size;
+      }
+      if (!overflow) {
+        slot_size[i] = count;
+        total_items += count;
+        break;
+      }
+      slot_limit[i]--;
+    }
+  }
+
+  /* Step 3: transfer data to "slim" hasher. */
+  alloc_size = sizeof(PreparedDictionary) + (sizeof(uint32_t) << slot_bits) +
+      (sizeof(uint16_t) << bucket_bits) + (sizeof(uint32_t) * total_items) +
+      source_size;
+
+  result = (PreparedDictionary*)BROTLI_ALLOC(m, uint8_t, alloc_size);
+  if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(result)) {
+    BROTLI_FREE(m, flat);
+    return NULL;
+  }
+  slot_offsets = (uint32_t*)(&result[1]);
+  heads = (uint16_t*)(&slot_offsets[num_slots]);
+  items = (uint32_t*)(&heads[num_buckets]);
+  source_copy = (uint8_t*)(&items[total_items]);
+
+  result->magic = kPreparedDictionaryMagic;
+  result->source_offset = total_items;
+  result->source_size = (uint32_t)source_size;
+  result->hash_bits = hash_bits;
+  result->bucket_bits = bucket_bits;
+  result->slot_bits = slot_bits;
+
+  total_items = 0;
+  for (i = 0; i < num_slots; ++i) {
+    slot_offsets[i] = total_items;
+    total_items += slot_size[i];
+    slot_size[i] = 0;
+  }
+  for (i = 0; i < num_buckets; ++i) {
+    uint32_t slot = i & slot_mask;
+    uint32_t count = num[i];
+    uint32_t pos;
+    size_t j;
+    size_t cursor = slot_size[slot];
+    if (count > slot_limit[slot]) count = slot_limit[slot];
+    if (count == 0) {
+      heads[i] = 0xFFFF;
+      continue;
+    }
+    heads[i] = (uint16_t)cursor;
+    cursor += slot_offsets[slot];
+    slot_size[slot] += count;
+    pos = bucket_heads[i];
+    for (j = 0; j < count; j++) {
+      items[cursor++] = pos;
+      pos = next_bucket[pos];
+    }
+    items[cursor - 1] |= 0x80000000;
+  }
+
+  BROTLI_FREE(m, flat);
+  memcpy(source_copy, source, source_size);
+  return result;
+}
+
+PreparedDictionary* CreatePreparedDictionary(MemoryManager* m,
+    const uint8_t* source, size_t source_size) {
+  uint32_t bucket_bits = 17;
+  uint32_t slot_bits = 7;
+  uint32_t hash_bits = 40;
+  uint16_t bucket_limit = 32;
+  size_t volume = 16u << bucket_bits;
+  /* Tune parameters to fit dictionary size. */
+  while (volume < source_size && bucket_bits < 22) {
+    bucket_bits++;
+    slot_bits++;
+    volume <<= 1;
+  }
+  return CreatePreparedDictionaryWithParams(m,
+      source, source_size, bucket_bits, slot_bits, hash_bits, bucket_limit);
+}
+
+void DestroyPreparedDictionary(MemoryManager* m,
+    PreparedDictionary* dictionary) {
+  if (!dictionary) return;
+  BROTLI_FREE(m, dictionary);
+}
+
+BROTLI_BOOL AttachPreparedDictionary(
+    CompoundDictionary* compound, const PreparedDictionary* dictionary) {
+  size_t length = 0;
+  size_t index = 0;
+
+  if (compound->num_chunks == SHARED_BROTLI_MAX_COMPOUND_DICTS) {
+    return BROTLI_FALSE;
+  }
+
+  if (!dictionary) return BROTLI_FALSE;
+
+  length = dictionary->source_size;
+  index = compound->num_chunks;
+  compound->total_size += length;
+  compound->chunks[index] = dictionary;
+  compound->chunk_offsets[index + 1] = compound->total_size;
+  {
+    uint32_t* slot_offsets = (uint32_t*)(&dictionary[1]);
+    uint16_t* heads = (uint16_t*)(&slot_offsets[1u << dictionary->slot_bits]);
+    uint32_t* items = (uint32_t*)(&heads[1u << dictionary->bucket_bits]);
+    compound->chunk_source[index] =
+        (const uint8_t*)(&items[dictionary->source_offset]);
+  }
+  compound->num_chunks++;
+  return BROTLI_TRUE;
+}
diff --git a/c/enc/compound_dictionary.h b/c/enc/compound_dictionary.h
new file mode 100644
index 0000000..5e1dfe2
--- /dev/null
+++ b/c/enc/compound_dictionary.h
@@ -0,0 +1,60 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+#ifndef BROTLI_ENC_PREPARED_DICTIONARY_H_
+#define BROTLI_ENC_PREPARED_DICTIONARY_H_
+
+#include "../common/platform.h"
+#include "../common/constants.h"
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
+#include "./memory.h"
+
+static const uint32_t kPreparedDictionaryMagic = 0xDEBCEDE0;
+static const uint64_t kPreparedDictionaryHashMul64Long =
+    BROTLI_MAKE_UINT64_T(0x1FE35A7Bu, 0xD3579BD3u);
+
+typedef struct PreparedDictionary {
+  uint32_t magic;
+  uint32_t source_offset;
+  uint32_t source_size;
+  uint32_t hash_bits;
+  uint32_t bucket_bits;
+  uint32_t slot_bits;
+
+  /* --- Dynamic size members --- */
+
+  /* uint32_t slot_offsets[1 << slot_bits]; */
+  /* uint16_t heads[1 << bucket_bits]; */
+  /* uint32_t items[variable]; */
+
+  /* uint8_t source[source_size] */
+} PreparedDictionary;
+
+BROTLI_INTERNAL PreparedDictionary* CreatePreparedDictionary(MemoryManager* m,
+    const uint8_t* source, size_t source_size);
+
+BROTLI_INTERNAL void DestroyPreparedDictionary(MemoryManager* m,
+    PreparedDictionary* dictionary);
+
+typedef struct CompoundDictionary {
+  /* LZ77 prefix, compound dictionary */
+  size_t num_chunks;
+  size_t total_size;
+  /* Client instances. */
+  const PreparedDictionary* chunks[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+  const uint8_t* chunk_source[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+  size_t chunk_offsets[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+
+  size_t num_prepared_instances_;
+  /* Owned instances. */
+  PreparedDictionary* prepared_instances_[SHARED_BROTLI_MAX_COMPOUND_DICTS + 1];
+} CompoundDictionary;
+
+BROTLI_INTERNAL BROTLI_BOOL AttachPreparedDictionary(
+    CompoundDictionary* compound, const PreparedDictionary* dictionary);
+
+#endif /* BROTLI_ENC_PREPARED_DICTIONARY */
diff --git a/c/enc/encode.c b/c/enc/encode.c
index 23b1d3b..4a68c3c 100644
--- a/c/enc/encode.c
+++ b/c/enc/encode.c
@@ -21,6 +21,7 @@
 #include "./brotli_bit_stream.h"
 #include "./compress_fragment.h"
 #include "./compress_fragment_two_pass.h"
+#include "./dictionary_hash.h"
 #include "./encoder_dict.h"
 #include "./entropy_encode.h"
 #include "./fast_log.h"
@@ -746,7 +747,7 @@
   params->stream_offset = 0;
   params->size_hint = 0;
   params->disable_literal_context_modeling = BROTLI_FALSE;
-  BrotliInitEncoderDictionary(&params->dictionary);
+  BrotliInitSharedEncoderDictionary(&params->dictionary);
   params->dist.distance_postfix_bits = 0;
   params->dist.num_direct_distance_codes = 0;
   params->dist.alphabet_size_max =
@@ -755,6 +756,11 @@
   params->dist.max_distance = BROTLI_MAX_DISTANCE;
 }
 
+static void BrotliEncoderCleanupParams(MemoryManager* m,
+    BrotliEncoderParams* params) {
+  BrotliCleanupSharedEncoderDictionary(m, &params->dictionary);
+}
+
 static void BrotliEncoderInitState(BrotliEncoderState* s) {
   BrotliEncoderInitParams(&s->params);
   s->input_pos_ = 0;
@@ -825,6 +831,7 @@
   BROTLI_FREE(m, s->two_pass_arena_);
   BROTLI_FREE(m, s->command_buf_);
   BROTLI_FREE(m, s->literal_buf_);
+  BrotliEncoderCleanupParams(m, &s->params);
 }
 
 /* Deinitializes and frees BrotliEncoderState instance. */
@@ -922,6 +929,8 @@
   uint64_t cmd_dist = (uint64_t)s->dist_cache_[0];
   uint32_t distance_code = CommandRestoreDistanceCode(last_command,
                                                       &s->params.dist);
+  const CompoundDictionary* dict = &s->params.dictionary.compound;
+  size_t compound_dictionary_size = dict->total_size;
   if (distance_code < BROTLI_NUM_DISTANCE_SHORT_CODES ||
       distance_code - (BROTLI_NUM_DISTANCE_SHORT_CODES - 1) == cmd_dist) {
     if (cmd_dist <= max_distance) {
@@ -932,6 +941,38 @@
         (*wrapped_last_processed_pos)++;
       }
     } else {
+      if ((cmd_dist - max_distance - 1) < compound_dictionary_size &&
+          last_copy_len < cmd_dist - max_distance) {
+        size_t address =
+            compound_dictionary_size - (size_t)(cmd_dist - max_distance) +
+            (size_t)last_copy_len;
+        size_t br_index = 0;
+        size_t br_offset;
+        const uint8_t* chunk;
+        size_t chunk_length;
+        while (address >= dict->chunk_offsets[br_index + 1]) br_index++;
+        br_offset = address - dict->chunk_offsets[br_index];
+        chunk = dict->chunk_source[br_index];
+        chunk_length =
+            dict->chunk_offsets[br_index + 1] - dict->chunk_offsets[br_index];
+        while (*bytes != 0 && data[*wrapped_last_processed_pos & mask] ==
+               chunk[br_offset]) {
+          last_command->copy_len_++;
+          (*bytes)--;
+          (*wrapped_last_processed_pos)++;
+          if (++br_offset == chunk_length) {
+            br_index++;
+            br_offset = 0;
+            if (br_index != dict->num_chunks) {
+              chunk = dict->chunk_source[br_index];
+              chunk_length = dict->chunk_offsets[br_index + 1] -
+                  dict->chunk_offsets[br_index];
+            } else {
+              break;
+            }
+          }
+        }
+      }
     }
     /* The copy length is at most the metablock size, and thus expressible. */
     GetLengthCode(last_command->insert_len_,
@@ -969,6 +1010,7 @@
   data = s->ringbuffer_.buffer_;
   mask = s->ringbuffer_.mask_;
 
+  if (s->params.quality > s->params.dictionary.max_quality) return BROTLI_FALSE;
   /* Adding more blocks after "last" block is forbidden. */
   if (s->is_last_block_emitted_) return BROTLI_FALSE;
   if (is_last) s->is_last_block_emitted_ = BROTLI_TRUE;
@@ -1421,6 +1463,7 @@
   *encoded_size = total_out_size;
   DestroyHasher(m, hasher);
   BROTLI_FREE(m, hasher);
+  BrotliEncoderCleanupParams(m, params);
   BROTLI_FREE(m, params);
   BrotliBootstrapFree(m, m);
   return ok;
@@ -1930,6 +1973,124 @@
   return BROTLI_VERSION;
 }
 
+BrotliEncoderPreparedDictionary* BrotliEncoderPrepareDictionary(
+    BrotliSharedDictionaryType type, size_t size, const uint8_t* data,
+    int quality,
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+  ManagedDictionary* managed_dictionary = NULL;
+  if (type != BROTLI_SHARED_DICTIONARY_RAW &&
+      type != BROTLI_SHARED_DICTIONARY_SERIALIZED) {
+    return NULL;
+  }
+  managed_dictionary =
+      BrotliCreateManagedDictionary(alloc_func, free_func, opaque);
+  if (managed_dictionary == NULL) {
+    return NULL;
+  }
+  if (type == BROTLI_SHARED_DICTIONARY_RAW) {
+    managed_dictionary->dictionary = (uint32_t*)CreatePreparedDictionary(
+        &managed_dictionary->memory_manager_, data, size);
+  } else {
+    SharedEncoderDictionary* dict = (SharedEncoderDictionary*)BrotliAllocate(
+        &managed_dictionary->memory_manager_, sizeof(SharedEncoderDictionary));
+    managed_dictionary->dictionary = (uint32_t*)dict;
+    if (dict != NULL) {
+      BROTLI_BOOL ok = BrotliInitCustomSharedEncoderDictionary(
+          &managed_dictionary->memory_manager_, data, size, quality, dict);
+      if (!ok) {
+        BrotliFree(&managed_dictionary->memory_manager_, dict);
+        managed_dictionary->dictionary = NULL;
+      }
+    }
+  }
+  if (managed_dictionary->dictionary == NULL) {
+    BrotliDestroyManagedDictionary(managed_dictionary);
+    return NULL;
+  }
+  return (BrotliEncoderPreparedDictionary*)managed_dictionary;
+}
+
+void BrotliEncoderDestroyPreparedDictionary(
+    BrotliEncoderPreparedDictionary* dictionary) {
+  ManagedDictionary* dict = (ManagedDictionary*)dictionary;
+  if (!dictionary) return;
+  /* First field of dictionary structs. */
+  /* Only managed dictionaries are eligible for destruction by this method. */
+  if (dict->magic != kManagedDictionaryMagic) {
+    return;
+  }
+  if (dict->dictionary == NULL) {
+    /* This should never ever happen. */
+  } else if (*dict->dictionary == kPreparedDictionaryMagic) {
+    DestroyPreparedDictionary(
+        &dict->memory_manager_, (PreparedDictionary*)dict->dictionary);
+  } else if (*dict->dictionary == kSharedDictionaryMagic) {
+    BrotliCleanupSharedEncoderDictionary(&dict->memory_manager_,
+        (SharedEncoderDictionary*)dict->dictionary);
+    BrotliFree(&dict->memory_manager_, dict->dictionary);
+  } else {
+    /* This should never ever happen. */
+  }
+  dict->dictionary = NULL;
+  BrotliDestroyManagedDictionary(dict);
+}
+
+BROTLI_BOOL BrotliEncoderAttachPreparedDictionary(BrotliEncoderState* state,
+    const BrotliEncoderPreparedDictionary* dictionary) {
+  /* First field of dictionary structs */
+  const BrotliEncoderPreparedDictionary* dict = dictionary;
+  uint32_t magic = *((const uint32_t*)dict);
+  SharedEncoderDictionary* current = NULL;
+  if (magic == kManagedDictionaryMagic) {
+    /* Unwrap managed dictionary. */
+    ManagedDictionary* managed_dictionary = (ManagedDictionary*)dict;
+    magic = *managed_dictionary->dictionary;
+    dict = (BrotliEncoderPreparedDictionary*)managed_dictionary->dictionary;
+  }
+  current = &state->params.dictionary;
+  if (magic == kPreparedDictionaryMagic) {
+    const PreparedDictionary* prepared = (const PreparedDictionary*)dict;
+    if (!AttachPreparedDictionary(&current->compound, prepared)) {
+      return BROTLI_FALSE;
+    }
+  } else if (magic == kSharedDictionaryMagic) {
+    const SharedEncoderDictionary* attached =
+        (const SharedEncoderDictionary*)dict;
+    BROTLI_BOOL was_default = !current->contextual.context_based &&
+        current->contextual.num_dictionaries == 1 &&
+        current->contextual.dict[0]->hash_table_words ==
+        kStaticDictionaryHashWords &&
+        current->contextual.dict[0]->hash_table_lengths ==
+        kStaticDictionaryHashLengths;
+    BROTLI_BOOL new_default = !attached->contextual.context_based &&
+        attached->contextual.num_dictionaries == 1 &&
+        attached->contextual.dict[0]->hash_table_words ==
+        kStaticDictionaryHashWords &&
+        attached->contextual.dict[0]->hash_table_lengths ==
+        kStaticDictionaryHashLengths;
+    size_t i;
+    if (state->is_initialized_) return BROTLI_FALSE;
+    current->max_quality =
+        BROTLI_MIN(int, current->max_quality, attached->max_quality);
+    for (i = 0; i < attached->compound.num_chunks; i++) {
+      if (!AttachPreparedDictionary(&current->compound,
+          attached->compound.chunks[i])) {
+        return BROTLI_FALSE;
+      }
+    }
+    if (!new_default) {
+      if (!was_default) return BROTLI_FALSE;
+      /* Copy by value, but then set num_instances_ to 0 because their memory
+      is managed by attached, not by current */
+      current->contextual = attached->contextual;
+      current->contextual.num_instances_ = 0;
+    }
+  } else {
+    return BROTLI_FALSE;
+  }
+  return BROTLI_TRUE;
+}
+
 #if defined(__cplusplus) || defined(c_plusplus)
 }  /* extern "C" */
 #endif
diff --git a/c/enc/encoder_dict.c b/c/enc/encoder_dict.c
index c9e963b..098727c 100644
--- a/c/enc/encoder_dict.c
+++ b/c/enc/encoder_dict.c
@@ -6,16 +6,43 @@
 
 #include "./encoder_dict.h"
 
+#include <stdlib.h>  /* malloc, free */
+
 #include "../common/dictionary.h"
+#include "../common/platform.h"
+#include "../common/shared_dictionary_internal.h"
 #include "../common/transform.h"
+#include "./compound_dictionary.h"
 #include "./dictionary_hash.h"
+#include "./memory.h"
+#include "./quality.h"
 #include "./hash.h"
 
 #if defined(__cplusplus) || defined(c_plusplus)
 extern "C" {
 #endif
 
-void BrotliInitEncoderDictionary(BrotliEncoderDictionary* dict) {
+#define NUM_HASH_BITS 15u
+#define NUM_HASH_BUCKETS (1u << NUM_HASH_BITS)
+
+static void BrotliTrieInit(BrotliTrie* trie) {
+  trie->pool_capacity = 0;
+  trie->pool_size = 0;
+  trie->pool = 0;
+
+  /* Set up the root node */
+  trie->root.single = 0;
+  trie->root.len_ = 0;
+  trie->root.idx_ = 0;
+  trie->root.sub = 0;
+}
+
+static void BrotliTrieFree(MemoryManager* m, BrotliTrie* trie) {
+  BrotliFree(m, trie->pool);
+}
+
+/* Initializes to RFC 7932 static dictionary / transforms. */
+static void InitEncoderDictionary(BrotliEncoderDictionary* dict) {
   dict->words = BrotliGetDictionary();
   dict->num_transforms = (uint32_t)BrotliGetTransforms()->num_transforms;
 
@@ -26,6 +53,574 @@
 
   dict->cutoffTransformsCount = kCutoffTransformsCount;
   dict->cutoffTransforms = kCutoffTransforms;
+
+  dict->parent = 0;
+
+  dict->hash_table_data_words_ = 0;
+  dict->hash_table_data_lengths_ = 0;
+  dict->buckets_alloc_size_ = 0;
+  dict->buckets_data_ = 0;
+  dict->dict_words_alloc_size_ = 0;
+  dict->dict_words_data_ = 0;
+  dict->words_instance_ = 0;
+  dict->has_words_heavy = BROTLI_FALSE;
+  BrotliTrieInit(&dict->trie);
+}
+
+static void BrotliDestroyEncoderDictionary(MemoryManager* m,
+    BrotliEncoderDictionary* dict) {
+  BrotliFree(m, dict->hash_table_data_words_);
+  BrotliFree(m, dict->hash_table_data_lengths_);
+  BrotliFree(m, dict->buckets_data_);
+  BrotliFree(m, dict->dict_words_data_);
+  BrotliFree(m, dict->words_instance_);
+  BrotliTrieFree(m, &dict->trie);
+}
+
+/* Word length must be at least 4 bytes */
+static uint32_t Hash(const uint8_t* data, int bits) {
+  uint32_t h = BROTLI_UNALIGNED_LOAD32LE(data) * kHashMul32;
+  /* The higher bits contain more mixture from the multiplication,
+     so we take our results from there. */
+  return h >> (32 - bits);
+}
+
+/* Theoretical max possible word size after transform */
+#define kTransformedBufferSize \
+    (256 + 256 + SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH)
+
+/* To be safe buffer must have at least kTransformedBufferSize */
+static void TransformedDictionaryWord(uint32_t word_idx, int len, int transform,
+    const BrotliTransforms* transforms,
+    const BrotliEncoderDictionary* dict,
+    uint8_t* buffer, size_t* size) {
+  const uint8_t* dict_word = &dict->words->data[
+      dict->words->offsets_by_length[len] + (uint32_t)len * word_idx];
+  *size = (size_t)BrotliTransformDictionaryWord(buffer, dict_word, len,
+      transforms, transform);
+}
+
+static DictWord MakeDictWord(uint8_t len, uint8_t transform, uint16_t idx) {
+  DictWord result;
+  result.len = len;
+  result.transform = transform;
+  result.idx = idx;
+  return result;
+}
+
+static uint32_t BrotliTrieAlloc(MemoryManager* m, size_t num, BrotliTrie* trie,
+                                BrotliTrieNode** keep) {
+  uint32_t result;
+  uint32_t keep_index = 0;
+  if (keep && *keep != &trie->root) {
+    /* Optional node to keep, since address may change after re-allocating */
+    keep_index = (uint32_t)(*keep - trie->pool);
+  }
+  if (trie->pool_size == 0) {
+    /* Have a dummy node in the front. We do not want the result to be 0, it
+    must be at least 1, 0 represents "null pointer" */
+    trie->pool_size = 1;
+  }
+  BROTLI_ENSURE_CAPACITY(m, BrotliTrieNode, trie->pool, trie->pool_capacity,
+                         trie->pool_size + num);
+  if (BROTLI_IS_OOM(m)) return 0;
+  /* Init the new nodes to empty */
+  memset(trie->pool + trie->pool_size, 0, sizeof(*trie->pool) * num);
+  result = (uint32_t)trie->pool_size;
+  trie->pool_size += num;
+  if (keep && *keep != &trie->root) {
+    *keep = trie->pool + keep_index;
+  }
+  return result;
+}
+
+/**
+ * len and idx: payload for last node
+ * word, size: the string
+ * index: position in the string
+ */
+static BROTLI_BOOL BrotliTrieNodeAdd(MemoryManager* m, uint8_t len,
+    uint32_t idx, const uint8_t* word, size_t size, int index,
+    BrotliTrieNode* node, BrotliTrie* trie) {
+  BrotliTrieNode* child = 0;
+  uint8_t c;
+  if ((size_t)index == size) {
+    if (!node->len_ || idx < node->idx_) {
+      node->len_ = len;
+      node->idx_ = idx;
+    }
+    return BROTLI_TRUE;
+  }
+  c = word[index];
+  if (node->single && c != node->c) {
+    BrotliTrieNode old = trie->pool[node->sub];
+    uint32_t new_nodes = BrotliTrieAlloc(m, 32, trie, &node);
+    if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+    node->single = 0;
+    node->sub = new_nodes;
+    trie->pool[node->sub + (node->c >> 4)].sub = new_nodes + 16;
+    trie->pool[trie->pool[node->sub + (node->c >> 4)].sub + (node->c & 15)] =
+        old;
+  }
+  if (!node->sub) {
+    uint32_t new_node = BrotliTrieAlloc(m, 1, trie, &node);
+    if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+    node->single = 1;
+    node->c = c;
+    node->sub = new_node;
+  }
+  if (node->single) {
+    child = &trie->pool[node->sub];
+  } else {
+    if (!trie->pool[node->sub + (c >> 4)].sub) {
+      uint32_t new_nodes = BrotliTrieAlloc(m, 16, trie, &node);
+      if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+      trie->pool[node->sub + (c >> 4)].sub = new_nodes;
+    }
+    child = &trie->pool[trie->pool[node->sub + (c >> 4)].sub + (c & 15)];
+  }
+  return BrotliTrieNodeAdd(m, len, idx, word, size, index + 1, child, trie);
+}
+
+static BROTLI_BOOL BrotliTrieAdd(MemoryManager* m, uint8_t len, uint32_t idx,
+                          const uint8_t* word, size_t size, BrotliTrie* trie) {
+  return BrotliTrieNodeAdd(m, len, idx, word, size, 0, &trie->root, trie);
+}
+
+const BrotliTrieNode* BrotliTrieSub(const BrotliTrie* trie,
+                                    const BrotliTrieNode* node, uint8_t c) {
+  BrotliTrieNode* temp_node;
+  if (node->single) {
+    if (node->c == c) return &trie->pool[node->sub];
+    return 0;
+  }
+  if (!node->sub) return 0;
+  temp_node = &trie->pool[node->sub + (c >> 4)];
+  if (!temp_node->sub) return 0;
+  return &trie->pool[temp_node->sub + (c & 15)];
+}
+
+static const BrotliTrieNode* BrotliTrieFind(const BrotliTrie* trie,
+                                            const uint8_t* word, size_t size) {
+  const BrotliTrieNode* node = &trie->root;
+  size_t i;
+  for (i = 0; i < size; i++) {
+    node = BrotliTrieSub(trie, node, word[i]);
+    if (!node) return 0;
+  }
+  return node;
+}
+
+static BROTLI_BOOL BuildDictionaryLut(MemoryManager* m,
+    const BrotliTransforms* transforms,
+    BrotliEncoderDictionary* dict) {
+  uint32_t i;
+  DictWord* dict_words;
+  uint16_t* buckets;
+  DictWord** words_by_hash;
+  size_t* words_by_hash_size;
+  size_t* words_by_hash_capacity;
+  BrotliTrie dedup;
+  uint8_t word[kTransformedBufferSize];
+  size_t word_size;
+  size_t total = 0;
+  uint8_t l;
+  uint16_t idx;
+
+  BrotliTrieInit(&dedup);
+
+  words_by_hash = (DictWord**)BrotliAllocate(m,
+      sizeof(*words_by_hash) * NUM_HASH_BUCKETS);
+  words_by_hash_size = (size_t*)BrotliAllocate(m,
+      sizeof(*words_by_hash_size) * NUM_HASH_BUCKETS);
+  words_by_hash_capacity = (size_t*)BrotliAllocate(m,
+      sizeof(*words_by_hash_capacity) * NUM_HASH_BUCKETS);
+  if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+  memset(words_by_hash, 0, sizeof(*words_by_hash) * NUM_HASH_BUCKETS);
+  memset(words_by_hash_size, 0, sizeof(*words_by_hash_size) * NUM_HASH_BUCKETS);
+  memset(words_by_hash_capacity, 0,
+         sizeof(*words_by_hash_capacity) * NUM_HASH_BUCKETS);
+
+  if (transforms->num_transforms > 0) {
+    for (l = SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH;
+        l <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; ++l) {
+      uint16_t n = dict->words->size_bits_by_length[l] ?
+          (uint16_t)(1 << dict->words->size_bits_by_length[l]) : 0u;
+      for (idx = 0; idx < n; ++idx) {
+        uint32_t key;
+        /* First transform (usually identity) */
+        TransformedDictionaryWord(idx, l, 0, transforms, dict, word,
+                                  &word_size);
+        /* Cannot hash words smaller than 4 bytes */
+        if (word_size < 4) {
+          /* Break instead of continue, all next words of this length will have
+             same length after transform */
+          break;
+        }
+        if (!BrotliTrieAdd(m, 0, idx, word, word_size, &dedup)) {
+          return BROTLI_FALSE;
+        }
+        key = Hash(word, NUM_HASH_BITS);
+        BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+            words_by_hash_capacity[key], words_by_hash_size[key],
+            MakeDictWord(l, 0, idx));
+        ++total;
+      }
+    }
+  }
+
+  /* These LUT transforms only supported if no custom transforms. This is
+     ok, we will use the heavy trie instead. */
+  if (transforms == BrotliGetTransforms()) {
+    for (l = SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH;
+        l <= SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH; ++l) {
+      uint16_t n = dict->words->size_bits_by_length[l] ?
+          (uint16_t)(1 << dict->words->size_bits_by_length[l]) : 0u;
+      for (idx = 0; idx < n; ++idx) {
+        int k;
+        BROTLI_BOOL is_ascii = BROTLI_TRUE;
+        size_t offset = dict->words->offsets_by_length[l] + (size_t)l * idx;
+        const uint8_t* data = &dict->words->data[offset];
+        for (k = 0; k < l; ++k) {
+          if (data[k] >= 128) is_ascii = BROTLI_FALSE;
+        }
+        if (data[0] < 128) {
+          int transform = 9;  /* {empty, uppercase first, empty} */
+          uint32_t ix = idx + (uint32_t)transform * n;
+          const BrotliTrieNode* it;
+          TransformedDictionaryWord(idx, l, transform, transforms,
+                                   dict, word, &word_size);
+          it = BrotliTrieFind(&dedup, word, word_size);
+          if (!it || it->idx_ > ix) {
+            uint32_t key = Hash(word, NUM_HASH_BITS);
+            if (!BrotliTrieAdd(m, 0, ix, word, word_size, &dedup)) {
+              return BROTLI_FALSE;
+            }
+            BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+                words_by_hash_capacity[key], words_by_hash_size[key],
+                MakeDictWord(l, BROTLI_TRANSFORM_UPPERCASE_FIRST, idx));
+            ++total;
+          }
+        }
+        if (is_ascii) {
+          int transform = 44;  /* {empty, uppercase all, empty} */
+          uint32_t ix = idx + (uint32_t)transform * n;
+          const BrotliTrieNode* it;
+          TransformedDictionaryWord(idx, l, transform, transforms,
+                                    dict, word, &word_size);
+          it = BrotliTrieFind(&dedup, word, word_size);
+          if (!it || it->idx_ > ix) {
+            uint32_t key = Hash(word, NUM_HASH_BITS);
+            if (!BrotliTrieAdd(m, 0, ix, word, word_size, &dedup)) {
+              return BROTLI_FALSE;
+            }
+            BROTLI_ENSURE_CAPACITY_APPEND(m, DictWord, words_by_hash[key],
+                words_by_hash_capacity[key], words_by_hash_size[key],
+                MakeDictWord(l, BROTLI_TRANSFORM_UPPERCASE_ALL, idx));
+            ++total;
+          }
+        }
+      }
+    }
+  }
+
+  dict_words = (DictWord*)BrotliAllocate(m,
+      sizeof(*dict->dict_words) * (total + 1));
+  buckets = (uint16_t*)BrotliAllocate(m,
+      sizeof(*dict->buckets) * NUM_HASH_BUCKETS);
+  if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+  dict->dict_words_alloc_size_ = total + 1;
+  dict->dict_words = dict->dict_words_data_ = dict_words;
+  dict->buckets_alloc_size_ = NUM_HASH_BUCKETS;
+  dict->buckets = dict->buckets_data_ = buckets;
+
+  /* Unused; makes offsets start from 1. */
+  dict_words[0] = MakeDictWord(0, 0, 0);
+  total = 1;
+  for (i = 0; i < NUM_HASH_BUCKETS; ++i) {
+    size_t num_words = words_by_hash_size[i];
+    if (num_words > 0) {
+      buckets[i] = (uint16_t)(total);
+      memcpy(&dict_words[total], &words_by_hash[i][0],
+          sizeof(dict_words[0]) * num_words);
+      total += num_words;
+      dict_words[total - 1].len |= 0x80;
+    } else {
+      buckets[i] = 0;
+    }
+  }
+
+  for (i = 0; i < NUM_HASH_BUCKETS; ++i) {
+    BrotliFree(m, words_by_hash[i]);
+  }
+  BrotliFree(m, words_by_hash);
+  BrotliFree(m, words_by_hash_size);
+  BrotliFree(m, words_by_hash_capacity);
+  BrotliTrieFree(m, &dedup);
+
+  return BROTLI_TRUE;
+}
+
+static void BuildDictionaryHashTable(uint16_t* hash_table_words,
+    uint8_t* hash_table_lengths, const BrotliDictionary* dict) {
+  int j, len;
+  /* The order of the loops is such that in case of collision, words with
+     shorter length are preferred, and in case of same length, words with
+     smaller index. There is only a single word per bucket. */
+  /* TODO: consider adding optional user-supplied frequency_map to use
+     for preferred words instead, this can make the encoder better for
+     quality 9 and below without affecting the decoder */
+  memset(hash_table_words, 0, sizeof(kStaticDictionaryHashWords));
+  memset(hash_table_lengths, 0, sizeof(kStaticDictionaryHashLengths));
+  for (len = SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH;
+      len >= SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH; --len) {
+    const size_t num_words = dict->size_bits_by_length[len] ?
+        (1u << dict->size_bits_by_length[len]) : 0;
+    for (j = (int)num_words - 1; j >= 0; --j) {
+      size_t offset = dict->offsets_by_length[len] +
+          (size_t)len * (size_t)j;
+      const uint8_t* word = &dict->data[offset];
+      const uint32_t key = Hash(word, 14);
+      int idx = (int)(key << 1) + (len < 8 ? 1 : 0);
+      BROTLI_DCHECK(idx < (int)NUM_HASH_BUCKETS);
+      hash_table_words[idx] = (uint16_t)j;
+      hash_table_lengths[idx] = (uint8_t)len;
+    }
+  }
+}
+
+static BROTLI_BOOL GenerateWordsHeavy(MemoryManager* m,
+    const BrotliTransforms* transforms,
+    BrotliEncoderDictionary* dict) {
+  int i, j, l;
+  for (j = (int)transforms->num_transforms - 1; j >= 0 ; --j) {
+    for (l = 0; l < 32; l++) {
+      int num = (int)((1u << dict->words->size_bits_by_length[l]) & ~1u);
+      for (i = 0; i < num; i++) {
+        uint8_t transformed[kTransformedBufferSize];
+        size_t size;
+        TransformedDictionaryWord(
+            (uint32_t)i, l, j, transforms, dict, transformed, &size);
+        if (size < 4) continue;
+        if (!BrotliTrieAdd(m, (uint8_t)l, (uint32_t)(i + num * j),
+            transformed, size, &dict->trie)) {
+          return BROTLI_FALSE;
+        }
+      }
+    }
+  }
+  return BROTLI_TRUE;
+}
+
+/* Computes cutoffTransformsCount (in count) and cutoffTransforms (in data) for
+   the custom transforms, where possible within the limits of the
+   cutoffTransforms encoding. The fast encoder uses this to do fast lookup for
+   transforms that remove the N last characters (OmitLast). */
+static void ComputeCutoffTransforms(
+    const BrotliTransforms* transforms,
+    uint32_t* count, uint64_t* data) {
+  int i;
+  /* The encoding in a 64-bit integer of transform N in the data is: (N << 2) +
+     ((cutoffTransforms >> (N * 6)) & 0x3F), so for example the identity
+     transform code must be 0-63, for N=1 the transform code must be 4-67, ...,
+     for N=9 it must be 36-99.
+     TODO: consider a simple flexible uint8_t[10] instead of the uint64_t
+     for the cutoff transforms, so that shared dictionaries can have the
+     OmitLast transforms anywhere without loss. */
+  *count = 0;
+  *data = 0;
+  for (i = 0; i < BROTLI_TRANSFORMS_MAX_CUT_OFF + 1; i++) {
+    int idx = transforms->cutOffTransforms[i];
+    if (idx == -1) break;  /* Not found */
+    if (idx < (i << 2)) break;  /* Too small for the encoding */
+    if (idx >= (i << 2) + 64) break;  /* Too large for the encoding */
+    (*count)++;
+    *data |= (uint64_t)(((uint64_t)idx -
+        ((uint64_t)i << 2u)) << ((uint64_t)i * 6u));
+  }
+}
+
+static BROTLI_BOOL ComputeDictionary(MemoryManager* m, int quality,
+    const BrotliTransforms* transforms,
+    BrotliEncoderDictionary* current) {
+  int default_words = current->words == BrotliGetDictionary();
+  int default_transforms = transforms == BrotliGetTransforms();
+
+  if (default_words && default_transforms) {
+    /* hashes are already set to Brotli defaults */
+    return BROTLI_TRUE;
+  }
+
+  current->hash_table_data_words_ = (uint16_t*)BrotliAllocate(
+      m, sizeof(kStaticDictionaryHashWords));
+  current->hash_table_data_lengths_ = (uint8_t*)BrotliAllocate(
+      m, sizeof(kStaticDictionaryHashLengths));
+  if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+  current->hash_table_words = current->hash_table_data_words_;
+  current->hash_table_lengths = current->hash_table_data_lengths_;
+
+  BuildDictionaryHashTable(current->hash_table_data_words_,
+      current->hash_table_data_lengths_, current->words);
+
+  ComputeCutoffTransforms(transforms,
+      &current->cutoffTransformsCount, &current->cutoffTransforms);
+
+  /* Only compute the data for slow encoder if the requested quality is high
+     enough to need it */
+  if (quality >= ZOPFLIFICATION_QUALITY) {
+    if (!BuildDictionaryLut(m, transforms, current)) return BROTLI_FALSE;
+
+    /* For the built-in Brotli transforms, there is a hard-coded function to
+       handle all transforms, but for custom transforms, we use the following
+       large hammer instead */
+    current->has_words_heavy = !default_transforms;
+    if (current->has_words_heavy) {
+      if (!GenerateWordsHeavy(m, transforms, current)) return BROTLI_FALSE;
+    }
+  }
+
+  return BROTLI_TRUE;
+}
+
+void BrotliInitSharedEncoderDictionary(SharedEncoderDictionary* dict) {
+  dict->magic = kSharedDictionaryMagic;
+
+  dict->compound.num_chunks = 0;
+  dict->compound.total_size = 0;
+  dict->compound.chunk_offsets[0] = 0;
+  dict->compound.num_prepared_instances_ = 0;
+
+  dict->contextual.context_based = 0;
+  dict->contextual.num_dictionaries = 1;
+  dict->contextual.instances_ = 0;
+  dict->contextual.num_instances_ = 1;  /* The instance_ field */
+  dict->contextual.dict[0] = &dict->contextual.instance_;
+  InitEncoderDictionary(&dict->contextual.instance_);
+  dict->contextual.instance_.parent = &dict->contextual;
+
+  dict->max_quality = BROTLI_MAX_QUALITY;
+}
+
+/* TODO: make sure that tooling will warn user if not all the cutoff
+   transforms are available (for low-quality encoder). */
+static BROTLI_BOOL InitCustomSharedEncoderDictionary(
+    MemoryManager* m, const BrotliSharedDictionary* decoded_dict,
+    int quality, SharedEncoderDictionary* dict) {
+  ContextualEncoderDictionary* contextual;
+  CompoundDictionary* compound;
+  BrotliEncoderDictionary* instances;
+  int i;
+  BrotliInitSharedEncoderDictionary(dict);
+
+  contextual = &dict->contextual;
+  compound = &dict->compound;
+
+  for (i = 0; i < (int)decoded_dict->num_prefix; i++) {
+    PreparedDictionary* prepared = CreatePreparedDictionary(m,
+        decoded_dict->prefix[i], decoded_dict->prefix_size[i]);
+    AttachPreparedDictionary(compound, prepared);
+    /* remember for cleanup */
+    compound->prepared_instances_[
+        compound->num_prepared_instances_++] = prepared;
+  }
+
+  dict->max_quality = quality;
+  contextual->context_based = decoded_dict->context_based;
+  if (decoded_dict->context_based) {
+    memcpy(contextual->context_map, decoded_dict->context_map,
+        SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS);
+  }
+
+  contextual->num_dictionaries = decoded_dict->num_dictionaries;
+  contextual->num_instances_ = decoded_dict->num_dictionaries;
+  if (contextual->num_instances_ == 1) {
+    instances = &contextual->instance_;
+  } else {
+    contextual->instances_ = (BrotliEncoderDictionary*)
+        BrotliAllocate(m, sizeof(*contextual->instances_) *
+        contextual->num_instances_);
+    if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+    instances = contextual->instances_;
+  }
+  for (i = 0; i < (int)contextual->num_instances_; i++) {
+    BrotliEncoderDictionary* current = &instances[i];
+    InitEncoderDictionary(current);
+    current->parent = &dict->contextual;
+    if (decoded_dict->words[i] == BrotliGetDictionary()) {
+      current->words = BrotliGetDictionary();
+    } else {
+      current->words_instance_ = (BrotliDictionary*)BrotliAllocate(
+          m, sizeof(BrotliDictionary));
+      if (BROTLI_IS_OOM(m)) return BROTLI_FALSE;
+      *current->words_instance_ = *decoded_dict->words[i];
+      current->words = current->words_instance_;
+    }
+    current->num_transforms =
+        (uint32_t)decoded_dict->transforms[i]->num_transforms;
+    if (!ComputeDictionary(
+        m, quality, decoded_dict->transforms[i], current)) {
+      return BROTLI_FALSE;
+    }
+
+    contextual->dict[i] = current;
+  }
+
+  return BROTLI_TRUE;  /* success */
+}
+
+BROTLI_BOOL BrotliInitCustomSharedEncoderDictionary(
+    MemoryManager* m, const uint8_t* encoded_dict, size_t size,
+    int quality, SharedEncoderDictionary* dict) {
+  BROTLI_BOOL success = BROTLI_FALSE;
+  BrotliSharedDictionary* decoded_dict = BrotliSharedDictionaryCreateInstance(
+      m->alloc_func, m->free_func, m->opaque);
+  if (!decoded_dict) {  /* OOM */
+    return BROTLI_FALSE;
+  }
+  success = BrotliSharedDictionaryAttach(
+      decoded_dict, BROTLI_SHARED_DICTIONARY_SERIALIZED, size, encoded_dict);
+  if (success) {
+    success = InitCustomSharedEncoderDictionary(m,
+        decoded_dict, quality, dict);
+  }
+  BrotliSharedDictionaryDestroyInstance(decoded_dict);
+  return success;
+}
+
+void BrotliCleanupSharedEncoderDictionary(MemoryManager* m,
+                                          SharedEncoderDictionary* dict) {
+  size_t i;
+  for (i = 0; i < dict->compound.num_prepared_instances_; i++) {
+    DestroyPreparedDictionary(m,
+        (PreparedDictionary*)dict->compound.prepared_instances_[i]);
+  }
+  if (dict->contextual.num_instances_ == 1) {
+    BrotliDestroyEncoderDictionary(m, &dict->contextual.instance_);
+  } else if (dict->contextual.num_instances_ > 1) {
+    for (i = 0; i < dict->contextual.num_instances_; i++) {
+      BrotliDestroyEncoderDictionary(m, &dict->contextual.instances_[i]);
+    }
+    BrotliFree(m, dict->contextual.instances_);
+  }
+}
+
+ManagedDictionary* BrotliCreateManagedDictionary(
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque) {
+  ManagedDictionary* result = (ManagedDictionary*)BrotliBootstrapAlloc(
+      sizeof(ManagedDictionary), alloc_func, free_func, opaque);
+  if (result == NULL) return NULL;
+
+  result->magic = kManagedDictionaryMagic;
+  BrotliInitMemoryManager(
+      &result->memory_manager_, alloc_func, free_func, opaque);
+  result->dictionary = NULL;
+
+  return result;
+}
+
+void BrotliDestroyManagedDictionary(ManagedDictionary* dictionary) {
+  if (!dictionary) return;
+  BrotliBootstrapFree(dictionary, &dictionary->memory_manager_);
 }
 
 #if defined(__cplusplus) || defined(c_plusplus)
diff --git a/c/enc/encoder_dict.h b/c/enc/encoder_dict.h
index a1c329f..dfeb796 100644
--- a/c/enc/encoder_dict.h
+++ b/c/enc/encoder_dict.h
@@ -9,13 +9,50 @@
 
 #include "../common/dictionary.h"
 #include "../common/platform.h"
+#include <brotli/shared_dictionary.h>
 #include <brotli/types.h>
+#include "./compound_dictionary.h"
+#include "./memory.h"
 #include "./static_dict_lut.h"
 
 #if defined(__cplusplus) || defined(c_plusplus)
 extern "C" {
 #endif
 
+/*
+Dictionary hierarchy for Encoder:
+-SharedEncoderDictionary
+--CompoundDictionary
+---PreparedDictionary [up to 15x]
+   = prefix dictionary with precomputed hashes
+--ContextualEncoderDictionary
+---BrotliEncoderDictionary [up to 64x]
+   = for each context, precomputed static dictionary with words + transforms
+
+Dictionary hiearchy from common: similar, but without precomputed hashes
+-BrotliSharedDictionary
+--BrotliDictionary [up to 64x]
+--BrotliTransforms [up to 64x]
+--const uint8_t* prefix [up to 15x]: compound dictionaries
+*/
+
+typedef struct BrotliTrieNode {
+  uint8_t single;  /* if 1, sub is a single node for c instead of 256 */
+  uint8_t c;
+  uint8_t len_;  /* untransformed length */
+  uint32_t idx_;  /* word index + num words * transform index */
+  uint32_t sub;  /* index of sub node(s) in the pool */
+} BrotliTrieNode;
+
+typedef struct BrotliTrie {
+  BrotliTrieNode* pool;
+  size_t pool_capacity;
+  size_t pool_size;
+  BrotliTrieNode root;
+} BrotliTrie;
+
+BROTLI_INTERNAL const BrotliTrieNode* BrotliTrieSub(const BrotliTrie* trie,
+    const BrotliTrieNode* node, uint8_t c);
 /* Dictionary data (words and transforms) for 1 possible context */
 typedef struct BrotliEncoderDictionary {
   const BrotliDictionary* words;
@@ -32,9 +69,83 @@
   /* from static_dict_lut.h, for slow encoder */
   const uint16_t* buckets;
   const DictWord* dict_words;
+  /* Heavy version, for use by slow encoder when there are custom transforms.
+     Contains every possible transformed dictionary word in a trie. It encodes
+     about as fast as the non-heavy encoder but consumes a lot of memory and
+     takes time to build. */
+  BrotliTrie trie;
+  BROTLI_BOOL has_words_heavy;
+
+  /* Reference to other dictionaries. */
+  const struct ContextualEncoderDictionary* parent;
+
+  /* Allocated memory, used only when not using the Brotli defaults */
+  uint16_t* hash_table_data_words_;
+  uint8_t* hash_table_data_lengths_;
+  size_t buckets_alloc_size_;
+  uint16_t* buckets_data_;
+  size_t dict_words_alloc_size_;
+  DictWord* dict_words_data_;
+  BrotliDictionary* words_instance_;
 } BrotliEncoderDictionary;
 
-BROTLI_INTERNAL void BrotliInitEncoderDictionary(BrotliEncoderDictionary* dict);
+/* Dictionary data for all 64 contexts */
+typedef struct ContextualEncoderDictionary {
+  BROTLI_BOOL context_based;
+  uint8_t num_dictionaries;
+  uint8_t context_map[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+  const BrotliEncoderDictionary* dict[SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS];
+
+  /* If num_instances_ is 1, instance_ is used, else dynamic allocation with
+     instances_ is used. */
+  size_t num_instances_;
+  BrotliEncoderDictionary instance_;
+  BrotliEncoderDictionary* instances_;
+} ContextualEncoderDictionary;
+
+static const uint32_t kSharedDictionaryMagic = 0xDEBCEDE1;
+static const uint32_t kManagedDictionaryMagic = 0xDEBCEDE2;
+
+typedef struct SharedEncoderDictionary {
+  /* Magic value to distinguish this struct from PreparedDictionary for
+     certain external usages. */
+  uint32_t magic;
+
+  /* LZ77 prefix, compound dictionary */
+  CompoundDictionary compound;
+
+  /* Custom static dictionary (optionally context-based) */
+  ContextualEncoderDictionary contextual;
+
+  /* The maximum quality the dictionary was computed for */
+  int max_quality;
+} SharedEncoderDictionary;
+
+typedef struct ManagedDictionary {
+  uint32_t magic;
+  MemoryManager memory_manager_;
+  uint32_t* dictionary;
+} ManagedDictionary;
+
+/* Initializes to the brotli built-in dictionary */
+BROTLI_INTERNAL void BrotliInitSharedEncoderDictionary(
+    SharedEncoderDictionary* dict);
+
+/* Initializes to shared dictionary that will be parsed from
+   encoded_dict. Requires that you keep the encoded_dict buffer
+   around, parts of data will point to it. */
+BROTLI_INTERNAL BROTLI_BOOL BrotliInitCustomSharedEncoderDictionary(
+    MemoryManager* m, const uint8_t* encoded_dict, size_t size,
+    int quality, SharedEncoderDictionary* dict);
+
+BROTLI_INTERNAL void BrotliCleanupSharedEncoderDictionary(
+    MemoryManager* m, SharedEncoderDictionary* dict);
+
+BROTLI_INTERNAL ManagedDictionary* BrotliCreateManagedDictionary(
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+BROTLI_INTERNAL void BrotliDestroyManagedDictionary(
+    ManagedDictionary* dictionary);
 
 #if defined(__cplusplus) || defined(c_plusplus)
 }  /* extern "C" */
diff --git a/c/enc/hash.h b/c/enc/hash.h
index 114b621..7a54be2 100644
--- a/c/enc/hash.h
+++ b/c/enc/hash.h
@@ -504,6 +504,206 @@
   }
 }
 
+/* NB: when seamless dictionary-ring-buffer copies are implemented, don't forget
+       to add proper guards for non-zero-BROTLI_PARAM_STREAM_OFFSET. */
+static BROTLI_INLINE void FindCompoundDictionaryMatch(
+    const PreparedDictionary* self, const uint8_t* BROTLI_RESTRICT data,
+    const size_t ring_buffer_mask, const int* BROTLI_RESTRICT distance_cache,
+    const size_t cur_ix, const size_t max_length, const size_t distance_offset,
+    const size_t max_distance, HasherSearchResult* BROTLI_RESTRICT out) {
+  const uint32_t source_offset = self->source_offset;
+  const uint32_t source_size = self->source_size;
+  const size_t boundary = distance_offset - source_size;
+  const uint32_t hash_bits = self->hash_bits;
+  const uint32_t bucket_bits = self->bucket_bits;
+  const uint32_t slot_bits = self->slot_bits;
+
+  const uint32_t hash_shift = 64u - bucket_bits;
+  const uint32_t slot_mask = (~((uint32_t)0U)) >> (32 - slot_bits);
+  const uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+
+  const uint32_t* slot_offsets = (uint32_t*)(&self[1]);
+  const uint16_t* heads = (uint16_t*)(&slot_offsets[1u << slot_bits]);
+  const uint32_t* items = (uint32_t*)(&heads[1u << bucket_bits]);
+  const uint8_t* source = (uint8_t*)(&items[source_offset]);
+
+  const size_t cur_ix_masked = cur_ix & ring_buffer_mask;
+  score_t best_score = out->score;
+  size_t best_len = out->len;
+  size_t i;
+  const uint64_t h =
+      (BROTLI_UNALIGNED_LOAD64LE(&data[cur_ix_masked]) & hash_mask) *
+      kPreparedDictionaryHashMul64Long;
+  const uint32_t key = (uint32_t)(h >> hash_shift);
+  const uint32_t slot = key & slot_mask;
+  const uint32_t head = heads[key];
+  const uint32_t* BROTLI_RESTRICT chain = &items[slot_offsets[slot] + head];
+  uint32_t item = (head == 0xFFFF) ? 1 : 0;
+  for (i = 0; i < 4; ++i) {
+    const size_t distance = (size_t)distance_cache[i];
+    size_t offset;
+    size_t limit;
+    size_t len;
+    if (distance <= boundary || distance > distance_offset) continue;
+    offset = distance_offset - distance;
+    limit = source_size - offset;
+    limit = limit > max_length ? max_length : limit;
+    len = FindMatchLengthWithLimit(&source[offset], &data[cur_ix_masked],
+                                   limit);
+    if (len >= 2) {
+      score_t score = BackwardReferenceScoreUsingLastDistance(len);
+      if (best_score < score) {
+        if (i != 0) score -= BackwardReferencePenaltyUsingLastDistance(i);
+        if (best_score < score) {
+          best_score = score;
+          if (len > best_len) best_len = len;
+          out->len = len;
+          out->len_code_delta = 0;
+          out->distance = distance;
+          out->score = best_score;
+        }
+      }
+    }
+  }
+  while (item == 0) {
+    size_t offset;
+    size_t distance;
+    size_t limit;
+    item = *chain;
+    chain++;
+    offset = item & 0x7FFFFFFF;
+    item &= 0x80000000;
+    distance = distance_offset - offset;
+    limit = source_size - offset;
+    limit = (limit > max_length) ? max_length : limit;
+    if (distance > max_distance) continue;
+    if (cur_ix_masked + best_len > ring_buffer_mask ||
+        best_len >= limit ||
+        data[cur_ix_masked + best_len] != source[offset + best_len]) {
+      continue;
+    }
+    {
+      const size_t len = FindMatchLengthWithLimit(&source[offset],
+                                                  &data[cur_ix_masked],
+                                                  limit);
+      if (len >= 4) {
+        score_t score = BackwardReferenceScore(len, distance);
+        if (best_score < score) {
+          best_score = score;
+          best_len = len;
+          out->len = best_len;
+          out->len_code_delta = 0;
+          out->distance = distance;
+          out->score = best_score;
+        }
+      }
+    }
+  }
+}
+
+/* NB: when seamless dictionary-ring-buffer copies are implemented, don't forget
+       to add proper guards for non-zero-BROTLI_PARAM_STREAM_OFFSET. */
+static BROTLI_INLINE size_t FindAllCompoundDictionaryMatches(
+    const PreparedDictionary* self, const uint8_t* BROTLI_RESTRICT data,
+    const size_t ring_buffer_mask, const size_t cur_ix, const size_t min_length,
+    const size_t max_length, const size_t distance_offset,
+    const size_t max_distance, BackwardMatch* matches, size_t match_limit) {
+  const uint32_t source_offset = self->source_offset;
+  const uint32_t source_size = self->source_size;
+  const uint32_t hash_bits = self->hash_bits;
+  const uint32_t bucket_bits = self->bucket_bits;
+  const uint32_t slot_bits = self->slot_bits;
+
+  const uint32_t hash_shift = 64u - bucket_bits;
+  const uint32_t slot_mask = (~((uint32_t)0U)) >> (32 - slot_bits);
+  const uint64_t hash_mask = (~((uint64_t)0U)) >> (64 - hash_bits);
+
+  const uint32_t* slot_offsets = (uint32_t*)(&self[1]);
+  const uint16_t* heads = (uint16_t*)(&slot_offsets[1u << slot_bits]);
+  const uint32_t* items = (uint32_t*)(&heads[1u << bucket_bits]);
+  const uint8_t* source = (uint8_t*)(&items[source_offset]);
+
+  const size_t cur_ix_masked = cur_ix & ring_buffer_mask;
+  size_t best_len = min_length;
+  const uint64_t h =
+      (BROTLI_UNALIGNED_LOAD64LE(&data[cur_ix_masked]) & hash_mask) *
+      kPreparedDictionaryHashMul64Long;
+  const uint32_t key = (uint32_t)(h >> hash_shift);
+  const uint32_t slot = key & slot_mask;
+  const uint32_t head = heads[key];
+  const uint32_t* BROTLI_RESTRICT chain = &items[slot_offsets[slot] + head];
+  uint32_t item = (head == 0xFFFF) ? 1 : 0;
+  size_t found = 0;
+  while (item == 0) {
+    size_t offset;
+    size_t distance;
+    size_t limit;
+    size_t len;
+    item = *chain;
+    chain++;
+    offset = item & 0x7FFFFFFF;
+    item &= 0x80000000;
+    distance = distance_offset - offset;
+    limit = source_size - offset;
+    limit = (limit > max_length) ? max_length : limit;
+    if (distance > max_distance) continue;
+    if (cur_ix_masked + best_len > ring_buffer_mask ||
+        best_len >= limit ||
+        data[cur_ix_masked + best_len] != source[offset + best_len]) {
+      continue;
+    }
+    len = FindMatchLengthWithLimit(
+        &source[offset], &data[cur_ix_masked], limit);
+    if (len > best_len) {
+      best_len = len;
+      InitBackwardMatch(matches++, distance, len);
+      found++;
+      if (found == match_limit) break;
+    }
+  }
+  return found;
+}
+
+static BROTLI_INLINE void LookupCompoundDictionaryMatch(
+    const CompoundDictionary* addon, const uint8_t* BROTLI_RESTRICT data,
+    const size_t ring_buffer_mask, const int* BROTLI_RESTRICT distance_cache,
+    const size_t cur_ix, const size_t max_length,
+    const size_t max_ring_buffer_distance, const size_t max_distance,
+    HasherSearchResult* sr) {
+  size_t base_offset = max_ring_buffer_distance + 1 + addon->total_size - 1;
+  size_t d;
+  for (d = 0; d < addon->num_chunks; ++d) {
+    /* Only one prepared dictionary type is currently supported. */
+    FindCompoundDictionaryMatch(
+        (const PreparedDictionary*)addon->chunks[d], data, ring_buffer_mask,
+        distance_cache, cur_ix, max_length,
+        base_offset - addon->chunk_offsets[d], max_distance, sr);
+  }
+}
+
+static BROTLI_INLINE size_t LookupAllCompoundDictionaryMatches(
+    const CompoundDictionary* addon, const uint8_t* BROTLI_RESTRICT data,
+    const size_t ring_buffer_mask, const size_t cur_ix, size_t min_length,
+    const size_t max_length, const size_t max_ring_buffer_distance,
+    const size_t max_distance, BackwardMatch* matches,
+    size_t match_limit) {
+  size_t base_offset = max_ring_buffer_distance + 1 + addon->total_size - 1;
+  size_t d;
+  size_t total_found = 0;
+  for (d = 0; d < addon->num_chunks; ++d) {
+    /* Only one prepared dictionary type is currently supported. */
+    total_found += FindAllCompoundDictionaryMatches(
+        (const PreparedDictionary*)addon->chunks[d], data, ring_buffer_mask,
+        cur_ix, min_length, max_length, base_offset - addon->chunk_offsets[d],
+        max_distance, matches + total_found, match_limit - total_found);
+    if (total_found == match_limit) break;
+    if (total_found > 0) {
+      min_length = BackwardMatchLength(&matches[total_found - 1]);
+    }
+  }
+  return total_found;
+}
+
 #if defined(__cplusplus) || defined(c_plusplus)
 }  /* extern "C" */
 #endif
diff --git a/c/enc/params.h b/c/enc/params.h
index 54a7f00..c4c0642 100644
--- a/c/enc/params.h
+++ b/c/enc/params.h
@@ -40,7 +40,8 @@
   BROTLI_BOOL large_window;
   BrotliHasherParams hasher;
   BrotliDistanceParams dist;
-  BrotliEncoderDictionary dictionary;
+  /* TODO(eustas): rename to BrotliShared... */
+  SharedEncoderDictionary dictionary;
 } BrotliEncoderParams;
 
 #endif  /* BROTLI_ENC_PARAMS_H_ */
diff --git a/c/enc/static_dict.c b/c/enc/static_dict.c
index 7299ab7..c0e0ec4 100644
--- a/c/enc/static_dict.c
+++ b/c/enc/static_dict.c
@@ -74,10 +74,25 @@
   }
 }
 
-BROTLI_BOOL BrotliFindAllStaticDictionaryMatches(
+/* Finds matches for a single static dictionary */
+static BROTLI_BOOL BrotliFindAllStaticDictionaryMatchesFor(
     const BrotliEncoderDictionary* dictionary, const uint8_t* data,
     size_t min_length, size_t max_length, uint32_t* matches) {
   BROTLI_BOOL has_found_match = BROTLI_FALSE;
+  if (dictionary->has_words_heavy) {
+    const BrotliTrieNode* node = &dictionary->trie.root;
+    size_t l = 0;
+    while (node && l < max_length) {
+      uint8_t c;
+      if (l >= min_length && node->len_) {
+        AddMatch(node->idx_, l, node->len_, matches);
+        has_found_match = BROTLI_TRUE;
+      }
+      c = data[l++];
+      node = BrotliTrieSub(&dictionary->trie, node, c);
+    }
+    return has_found_match;
+  }
   {
     size_t offset = dictionary->buckets[Hash(data)];
     BROTLI_BOOL end = !offset;
@@ -481,6 +496,45 @@
   return has_found_match;
 }
 
+/* Finds matches for one or more dictionaries, if multiple are present
+   in the contextual dictionary */
+BROTLI_BOOL BrotliFindAllStaticDictionaryMatches(
+    const BrotliEncoderDictionary* dictionary, const uint8_t* data,
+    size_t min_length, size_t max_length, uint32_t* matches) {
+  BROTLI_BOOL has_found_match =
+      BrotliFindAllStaticDictionaryMatchesFor(
+          dictionary, data, min_length, max_length, matches);
+
+  if (!!dictionary->parent && dictionary->parent->num_dictionaries > 1) {
+    uint32_t matches2[BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1];
+    int l;
+    const BrotliEncoderDictionary* dictionary2 = dictionary->parent->dict[0];
+    if (dictionary2 == dictionary) {
+      dictionary2 = dictionary->parent->dict[1];
+    }
+
+    for (l = 0; l < BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1; l++) {
+      matches2[l] = kInvalidMatch;
+    }
+
+    has_found_match |= BrotliFindAllStaticDictionaryMatchesFor(
+        dictionary2, data, min_length, max_length, matches2);
+
+    for (l = 0; l < BROTLI_MAX_STATIC_DICTIONARY_MATCH_LEN + 1; l++) {
+      if (matches2[l] != kInvalidMatch) {
+        uint32_t dist = (uint32_t)(matches2[l] >> 5);
+        uint32_t len_code = matches2[l] & 31;
+        uint32_t skipdist = (uint32_t)((uint32_t)(1 << dictionary->words->
+            size_bits_by_length[len_code]) & ~1u) *
+            (uint32_t)dictionary->num_transforms;
+        /* TODO: check for dist overflow */
+        dist += skipdist;
+        AddMatch(dist, (size_t)l, len_code, matches);
+      }
+    }
+  }
+  return has_found_match;
+}
 #if defined(__cplusplus) || defined(c_plusplus)
 }  /* extern "C" */
 #endif
diff --git a/c/include/brotli/decode.h b/c/include/brotli/decode.h
index 0f5c8f9..98b3c7b 100644
--- a/c/include/brotli/decode.h
+++ b/c/include/brotli/decode.h
@@ -13,6 +13,7 @@
 #define BROTLI_DEC_DECODE_H_
 
 #include <brotli/port.h>
+#include <brotli/shared_dictionary.h>
 #include <brotli/types.h>
 
 #if defined(__cplusplus) || defined(c_plusplus)
@@ -85,8 +86,9 @@
   BROTLI_ERROR_CODE(_ERROR_FORMAT_, PADDING_2, -15) SEPARATOR              \
   BROTLI_ERROR_CODE(_ERROR_FORMAT_, DISTANCE, -16) SEPARATOR               \
                                                                            \
-  /* -17..-18 codes are reserved */                                        \
+  /* -17 code is reserved */                                               \
                                                                            \
+  BROTLI_ERROR_CODE(_ERROR_, COMPOUND_DICTIONARY, -18) SEPARATOR           \
   BROTLI_ERROR_CODE(_ERROR_, DICTIONARY_NOT_SET, -19) SEPARATOR            \
   BROTLI_ERROR_CODE(_ERROR_, INVALID_ARGUMENTS, -20) SEPARATOR             \
                                                                            \
@@ -155,6 +157,28 @@
     BrotliDecoderState* state, BrotliDecoderParameter param, uint32_t value);
 
 /**
+ * Adds LZ77 prefix dictionary, adds or replaces built-in static dictionary and
+ * transforms.
+ *
+ * Attached dictionary ownership is not transferred.
+ * Data provided to this method should be kept accessible until
+ * decoding is finished and decoder instance is destroyed.
+ *
+ * @note Dictionaries could NOT be attached after actual decoding is started.
+ *
+ * @param state decoder instance
+ * @param type dictionary data format
+ * @param data_size length of memory region pointed by @p data
+ * @param data dictionary data in format corresponding to @p type
+ * @returns ::BROTLI_FALSE if dictionary is corrupted,
+ *          or dictionary count limit is reached
+ * @returns ::BROTLI_TRUE if dictionary is accepted / attached
+ */
+BROTLI_DEC_API BROTLI_BOOL BrotliDecoderAttachDictionary(
+    BrotliDecoderState* state, BrotliSharedDictionaryType type,
+    size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)]);
+
+/**
  * Creates an instance of ::BrotliDecoderState and initializes it.
  *
  * The instance can be used once for decoding and should then be destroyed with
diff --git a/c/include/brotli/encode.h b/c/include/brotli/encode.h
index b2774cb..72668e1 100644
--- a/c/include/brotli/encode.h
+++ b/c/include/brotli/encode.h
@@ -13,6 +13,7 @@
 #define BROTLI_ENC_ENCODE_H_
 
 #include <brotli/port.h>
+#include <brotli/shared_dictionary.h>
 #include <brotli/types.h>
 
 #if defined(__cplusplus) || defined(c_plusplus)
@@ -269,6 +270,51 @@
  */
 BROTLI_ENC_API void BrotliEncoderDestroyInstance(BrotliEncoderState* state);
 
+/* Opaque type for pointer to different possible internal structures containing
+   dictionary prepared for the encoder */
+typedef struct BrotliEncoderPreparedDictionaryStruct
+    BrotliEncoderPreparedDictionary;
+
+/**
+ * Prepares a shared dictionary from the given file format for the encoder.
+ *
+ * @p alloc_func and @p free_func @b MUST be both zero or both non-zero. In the
+ * case they are both zero, default memory allocators are used. @p opaque is
+ * passed to @p alloc_func and @p free_func when they are called. @p free_func
+ * has to return without doing anything when asked to free a NULL pointer.
+ *
+ * @param type type of dictionary stored in data
+ * @param data_size size of @p data buffer
+ * @param data pointer to the dictionary data
+ * @param quality the maximum Brotli quality to prepare the dictionary for,
+ *        use BROTLI_MAX_QUALITY by default
+ * @param alloc_func custom memory allocation function
+ * @param free_func custom memory free function
+ * @param opaque custom memory manager handle
+ */
+BROTLI_ENC_API BrotliEncoderPreparedDictionary*
+BrotliEncoderPrepareDictionary(BrotliSharedDictionaryType type,
+    size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)],
+    int quality,
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+BROTLI_ENC_API void BrotliEncoderDestroyPreparedDictionary(
+    BrotliEncoderPreparedDictionary* dictionary);
+
+/**
+ * Attaches a prepared dictionary of any type to the encoder. Can be used
+ * multiple times to attach multiple dictionaries. The dictionary type was
+ * determined by BrotliEncoderPrepareDictionary. Multiple raw prefix
+ * dictionaries and/or max 1 serialized dictionary with custom words can be
+ * attached.
+ *
+ * @returns ::BROTLI_FALSE in case of error
+ * @returns ::BROTLI_TRUE otherwise
+ */
+BROTLI_ENC_API BROTLI_BOOL BrotliEncoderAttachPreparedDictionary(
+    BrotliEncoderState* state,
+    const BrotliEncoderPreparedDictionary* dictionary);
+
 /**
  * Calculates the output size bound for the given @p input_size.
  *
diff --git a/c/include/brotli/shared_dictionary.h b/c/include/brotli/shared_dictionary.h
new file mode 100644
index 0000000..ceb6cf1
--- /dev/null
+++ b/c/include/brotli/shared_dictionary.h
@@ -0,0 +1,97 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+/* (Opaque) Shared Dictionary definition and utilities. */
+
+#ifndef BROTLI_COMMON_SHARED_DICTIONARY_H_
+#define BROTLI_COMMON_SHARED_DICTIONARY_H_
+
+#include <brotli/port.h>
+#include <brotli/types.h>
+
+#if defined(__cplusplus) || defined(c_plusplus)
+extern "C" {
+#endif
+
+#define SHARED_BROTLI_MIN_DICTIONARY_WORD_LENGTH 4
+#define SHARED_BROTLI_MAX_DICTIONARY_WORD_LENGTH 31
+#define SHARED_BROTLI_NUM_DICTIONARY_CONTEXTS 64
+#define SHARED_BROTLI_MAX_COMPOUND_DICTS 15
+
+/**
+ * Opaque structure that holds shared dictionary data.
+ *
+ * Allocated and initialized with ::BrotliSharedDictionaryCreateInstance.
+ * Cleaned up and deallocated with ::BrotliSharedDictionaryDestroyInstance.
+ */
+typedef struct BrotliSharedDictionaryStruct BrotliSharedDictionary;
+
+/**
+ * Input data type for ::BrotliSharedDictionaryAttach.
+ */
+typedef enum BrotliSharedDictionaryType {
+  /** Raw LZ77 prefix dictionary. */
+  BROTLI_SHARED_DICTIONARY_RAW = 0,
+  /** Serialized shared dictionary. */
+  BROTLI_SHARED_DICTIONARY_SERIALIZED = 1
+} BrotliSharedDictionaryType;
+
+/**
+ * Creates an instance of ::BrotliSharedDictionary.
+ *
+ * Fresh instance has default word dictionary and transforms
+ * and no LZ77 prefix dictionary.
+ *
+ * @p alloc_func and @p free_func @b MUST be both zero or both non-zero. In the
+ * case they are both zero, default memory allocators are used. @p opaque is
+ * passed to @p alloc_func and @p free_func when they are called. @p free_func
+ * has to return without doing anything when asked to free a NULL pointer.
+ *
+ * @param alloc_func custom memory allocation function
+ * @param free_func custom memory free function
+ * @param opaque custom memory manager handle
+ * @returns @c 0 if instance can not be allocated or initialized
+ * @returns pointer to initialized ::BrotliSharedDictionary otherwise
+ */
+BROTLI_COMMON_API BrotliSharedDictionary* BrotliSharedDictionaryCreateInstance(
+    brotli_alloc_func alloc_func, brotli_free_func free_func, void* opaque);
+
+/**
+ * Deinitializes and frees ::BrotliSharedDictionary instance.
+ *
+ * @param dict shared dictionary instance to be cleaned up and deallocated
+ */
+BROTLI_COMMON_API void BrotliSharedDictionaryDestroyInstance(
+    BrotliSharedDictionary* dict);
+
+/**
+ * Attaches dictionary to a given instance of ::BrotliSharedDictionary.
+ *
+ * Dictionary to be attached is represented in a serialized format as a region
+ * of memory.
+ *
+ * Provided data it partially referenced by a resulting (compound) dictionary,
+ * and should be kept untouched, while at least one compound dictionary uses it.
+ * This way memory overhead is kept minimal by the cost of additional resource
+ * management.
+ *
+ * @param dict dictionary to extend
+ * @param type type of dictionary to attach
+ * @param data_size size of @p data
+ * @param data serialized dictionary of type @p type, with at least @p data_size
+ *        addressable bytes
+ * @returns ::BROTLI_TRUE if provided dictionary is successfully attached
+ * @returns ::BROTLI_FALSE otherwise
+ */
+BROTLI_COMMON_API BROTLI_BOOL BrotliSharedDictionaryAttach(
+    BrotliSharedDictionary* dict, BrotliSharedDictionaryType type,
+    size_t data_size, const uint8_t data[BROTLI_ARRAY_PARAM(data_size)]);
+
+#if defined(__cplusplus) || defined(c_plusplus)
+}  /* extern "C" */
+#endif
+
+#endif  /* BROTLI_COMMON_SHARED_DICTIONARY_H_ */
diff --git a/c/tools/brotli.c b/c/tools/brotli.c
index 77752b2..049371c 100644
--- a/c/tools/brotli.c
+++ b/c/tools/brotli.c
@@ -100,6 +100,7 @@
   BROTLI_BOOL decompress;
   BROTLI_BOOL large_window;
   const char* output_path;
+  const char* dictionary_path;
   const char* suffix;
   int not_input_indices[MAX_OPTIONS];
   size_t longest_path_len;
@@ -108,6 +109,9 @@
   /* Inner state */
   int argc;
   char** argv;
+  uint8_t* dictionary;
+  size_t dictionary_size;
+  BrotliEncoderPreparedDictionary* prepared_dictionary;
   char* modified_path;  /* Storage for path with appended / cut suffix */
   int iterator;
   int ignore;
@@ -359,6 +363,12 @@
                     params->lgwin, BROTLI_MIN_WINDOW_BITS);
             return COMMAND_INVALID;
           }
+        } else if (c == 'D') {
+          if (params->dictionary_path) {
+            fprintf(stderr, "dictionary path already set\n");
+            return COMMAND_INVALID;
+          }
+          params->dictionary_path = argv[i];
         } else if (c == 'S') {
           if (suffix_set) {
             fprintf(stderr, "suffix already set\n");
@@ -446,7 +456,13 @@
         }
         key_len = (size_t)(value - arg);
         value++;
-        if (strncmp("lgwin", arg, key_len) == 0) {
+        if (strncmp("dictionary", arg, key_len) == 0) {
+          if (params->dictionary_path) {
+            fprintf(stderr, "dictionary path already set\n");
+            return COMMAND_INVALID;
+          }
+          params->dictionary_path = value;
+        } else if (strncmp("lgwin", arg, key_len) == 0) {
           if (lgwin_set) {
             fprintf(stderr, "lgwin parameter already set\n");
             return COMMAND_INVALID;
@@ -575,6 +591,8 @@
 "                              decodable with regular brotli decoders\n",
           BROTLI_MIN_WINDOW_BITS, BROTLI_LARGE_MAX_WINDOW_BITS);
   fprintf(media,
+"  -D FILE, --dictionary=FILE  use FILE as raw (LZ77) dictionary\n");
+  fprintf(media,
 "  -S SUF, --suffix=SUF        output file suffix (default:'%s')\n",
           DEFAULT_SUFFIX);
   fprintf(media,
@@ -678,6 +696,69 @@
   }
 }
 
+/* Result ownership is passed to caller.
+   |*dictionary_size| is set to resulting buffer size. */
+static BROTLI_BOOL ReadDictionary(Context* context, Command command) {
+  static const int kMaxDictionarySize =
+      BROTLI_MAX_DISTANCE - BROTLI_MAX_BACKWARD_LIMIT(24);
+  FILE* f;
+  int64_t file_size_64;
+  uint8_t* buffer;
+  size_t bytes_read;
+
+  if (context->dictionary_path == NULL) return BROTLI_TRUE;
+  f = fopen(context->dictionary_path, "rb");
+  if (f == NULL) {
+    fprintf(stderr, "failed to open dictionary file [%s]: %s\n",
+            PrintablePath(context->dictionary_path), strerror(errno));
+    return BROTLI_FALSE;
+  }
+
+  file_size_64 = FileSize(context->dictionary_path);
+  if (file_size_64 == -1) {
+    fprintf(stderr, "could not get size of dictionary file [%s]",
+            PrintablePath(context->dictionary_path));
+    fclose(f);
+    return BROTLI_FALSE;
+  }
+
+  if (file_size_64 > kMaxDictionarySize) {
+    fprintf(stderr, "dictionary [%s] is larger than maximum allowed: %d\n",
+            PrintablePath(context->dictionary_path), kMaxDictionarySize);
+    fclose(f);
+    return BROTLI_FALSE;
+  }
+  context->dictionary_size = (size_t)file_size_64;
+
+  buffer = (uint8_t*)malloc(context->dictionary_size);
+  if (!buffer) {
+    fprintf(stderr, "could not read dictionary: out of memory\n");
+    fclose(f);
+    return BROTLI_FALSE;
+  }
+  bytes_read = fread(buffer, sizeof(uint8_t), context->dictionary_size, f);
+  if (bytes_read != context->dictionary_size) {
+    free(buffer);
+    fprintf(stderr, "failed to read dictionary [%s]: %s\n",
+            PrintablePath(context->dictionary_path), strerror(errno));
+    fclose(f);
+    return BROTLI_FALSE;
+  }
+  fclose(f);
+  context->dictionary = buffer;
+  if (command == COMMAND_COMPRESS) {
+    context->prepared_dictionary = BrotliEncoderPrepareDictionary(
+        BROTLI_SHARED_DICTIONARY_RAW, context->dictionary_size,
+        context->dictionary, BROTLI_MAX_QUALITY, NULL, NULL, NULL);
+    if (context->prepared_dictionary == NULL) {
+      fprintf(stderr, "failed to prepare dictionary [%s]\n",
+              PrintablePath(context->dictionary_path));
+      return BROTLI_FALSE;
+    }
+  }
+  return BROTLI_TRUE;
+}
+
 static BROTLI_BOOL NextFile(Context* context) {
   const char* arg;
   size_t arg_len;
@@ -933,6 +1014,10 @@
        fragmentation (new builds decode streams that old builds don't),
        it is better from used experience perspective. */
     BrotliDecoderSetParameter(s, BROTLI_DECODER_PARAM_LARGE_WINDOW, 1u);
+    if (context->dictionary) {
+      BrotliDecoderAttachDictionary(s, BROTLI_SHARED_DICTIONARY_RAW,
+          context->dictionary_size, context->dictionary);
+    }
     is_ok = OpenFiles(context);
     if (is_ok && !context->current_input_path &&
         !context->force_overwrite && isatty(STDIN_FILENO)) {
@@ -1020,6 +1105,9 @@
           (uint32_t)context->input_file_length : (1u << 30);
       BrotliEncoderSetParameter(s, BROTLI_PARAM_SIZE_HINT, size_hint);
     }
+    if (context->dictionary) {
+      BrotliEncoderAttachPreparedDictionary(s, context->prepared_dictionary);
+    }
     is_ok = OpenFiles(context);
     if (is_ok && !context->current_output_path &&
         !context->force_overwrite && isatty(STDOUT_FILENO)) {
@@ -1051,6 +1139,7 @@
   context.decompress = BROTLI_FALSE;
   context.large_window = BROTLI_FALSE;
   context.output_path = NULL;
+  context.dictionary_path = NULL;
   context.suffix = DEFAULT_SUFFIX;
   for (i = 0; i < MAX_OPTIONS; ++i) context.not_input_indices[i] = 0;
   context.longest_path_len = 1;
@@ -1058,6 +1147,9 @@
 
   context.argc = argc;
   context.argv = argv;
+  context.dictionary = NULL;
+  context.dictionary_size = 0;
+  context.prepared_dictionary = NULL;
   context.modified_path = NULL;
   context.iterator = 0;
   context.ignore = 0;
@@ -1072,6 +1164,7 @@
 
   if (command == COMMAND_COMPRESS || command == COMMAND_DECOMPRESS ||
       command == COMMAND_TEST_INTEGRITY) {
+    if (!ReadDictionary(&context, command)) is_ok = BROTLI_FALSE;
     if (is_ok) {
       size_t modified_path_len =
           context.longest_path_len + strlen(context.suffix) + 1;
@@ -1116,6 +1209,8 @@
 
   if (context.iterator_error) is_ok = BROTLI_FALSE;
 
+  BrotliEncoderDestroyPreparedDictionary(context.prepared_dictionary);
+  free(context.dictionary);
   free(context.modified_path);
   free(context.buffer);
 
diff --git a/c/tools/brotli.md b/c/tools/brotli.md
index c029869..86253b8 100644
--- a/c/tools/brotli.md
+++ b/c/tools/brotli.md
@@ -84,6 +84,9 @@
     `(2**NUM - 16)`; 0 lets compressor decide over the optimal value; bigger
     windows size improve density; decoder might require up to window size
     memory to operate
+* `-D FILE`, `--dictionary=FILE`:
+    use FILE as raw (LZ77) dictionary; same dictionary MUST be used both for
+    compression and decompression
 * `-S SUF`, `--suffix=SUF`:
     output file suffix (default: `.br`)
 * `-V`, `--version`:
diff --git a/java/org/brotli/wrapper/dec/BUILD b/java/org/brotli/wrapper/dec/BUILD
index 4fd51d0..6190504 100644
--- a/java/org/brotli/wrapper/dec/BUILD
+++ b/java/org/brotli/wrapper/dec/BUILD
@@ -24,6 +24,7 @@
         ":dec",
         "//org/brotli/integration:brotli_jni_test_base",
         "//org/brotli/integration:bundle_helper",
+        "//org/brotli/wrapper/enc",
         "@maven//:junit_junit",
     ],
 )
@@ -82,3 +83,16 @@
     test_class = "org.brotli.wrapper.dec.DecoderTest",
     runtime_deps = [":test_lib"],
 )
+
+java_test(
+    name = "CornerCasesTest",
+    size = "large",
+    data = [
+        ":brotli_jni",  # Bazel JNI workaround
+    ],
+    jvm_flags = [
+        "-DBROTLI_JNI_LIBRARY=$(location :brotli_jni)",
+    ],
+    test_class = "org.brotli.wrapper.dec.CornerCasesTest",
+    runtime_deps = [":test_lib"],
+)
diff --git a/java/org/brotli/wrapper/dec/BrotliDecoderChannel.java b/java/org/brotli/wrapper/dec/BrotliDecoderChannel.java
index 6dfe8f2..30f6f1d 100644
--- a/java/org/brotli/wrapper/dec/BrotliDecoderChannel.java
+++ b/java/org/brotli/wrapper/dec/BrotliDecoderChannel.java
@@ -36,6 +36,11 @@
   }
 
   @Override
+  public void attachDictionary(ByteBuffer dictionary) throws IOException {
+    super.attachDictionary(dictionary);
+  }
+
+  @Override
   public boolean isOpen() {
     synchronized (mutex) {
       return !closed;
diff --git a/java/org/brotli/wrapper/dec/BrotliInputStream.java b/java/org/brotli/wrapper/dec/BrotliInputStream.java
index 6e2e6e5..13a8880 100644
--- a/java/org/brotli/wrapper/dec/BrotliInputStream.java
+++ b/java/org/brotli/wrapper/dec/BrotliInputStream.java
@@ -8,6 +8,7 @@
 
 import java.io.IOException;
 import java.io.InputStream;
+import java.nio.ByteBuffer;
 import java.nio.channels.Channels;
 
 /**
@@ -34,6 +35,10 @@
     this(source, DEFAULT_BUFFER_SIZE);
   }
 
+  public void attachDictionary(ByteBuffer dictionary) throws IOException {
+    decoder.attachDictionary(dictionary);
+  }
+
   public void enableEagerOutput() {
     decoder.enableEagerOutput();
   }
diff --git a/java/org/brotli/wrapper/dec/CornerCasesTest.java b/java/org/brotli/wrapper/dec/CornerCasesTest.java
new file mode 100644
index 0000000..ab02f23
--- /dev/null
+++ b/java/org/brotli/wrapper/dec/CornerCasesTest.java
@@ -0,0 +1,33 @@
+/* Copyright 2021 Google Inc. All Rights Reserved.
+
+   Distributed under MIT license.
+   See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
+*/
+
+package org.brotli.wrapper.dec;
+
+import static org.junit.Assert.assertEquals;
+
+import org.brotli.integration.BrotliJniTestBase;
+import org.brotli.wrapper.enc.Encoder;
+import java.io.IOException;
+import org.junit.Test;
+import org.junit.runner.RunWith;
+import org.junit.runners.JUnit4;
+
+/** Tests for {@link org.brotli.wrapper.enc.Encoder}. */
+@RunWith(JUnit4.class)
+public class CornerCasesTest extends BrotliJniTestBase {
+  @Test
+  public void testPowerOfTwoInput() throws IOException {
+    // 24 == max window bits to ensure ring-buffer size is not greater than input.
+    int len = 1 << 24;
+    byte[] data = new byte[len];
+    for (int i = 0; i < len; ++i) {
+      data[i] = (byte) Integer.bitCount(i);
+    }
+    byte[] encoded = Encoder.compress(data);
+    byte[] decoded = Decoder.decompress(encoded);
+    assertEquals(len, decoded.length);
+  }
+}
diff --git a/java/org/brotli/wrapper/dec/Decoder.java b/java/org/brotli/wrapper/dec/Decoder.java
index 018317d..de69d3a 100644
--- a/java/org/brotli/wrapper/dec/Decoder.java
+++ b/java/org/brotli/wrapper/dec/Decoder.java
@@ -50,6 +50,12 @@
     throw new IOException(message);
   }
 
+  void attachDictionary(ByteBuffer dictionary) throws IOException {
+    if (!decoder.attachDictionary(dictionary)) {
+      fail("failed to attach dictionary");
+    }
+  }
+
   public void enableEagerOutput() {
     this.eager = true;
   }
diff --git a/java/org/brotli/wrapper/dec/DecoderJNI.java b/java/org/brotli/wrapper/dec/DecoderJNI.java
index 2319b1e..fc5225d 100644
--- a/java/org/brotli/wrapper/dec/DecoderJNI.java
+++ b/java/org/brotli/wrapper/dec/DecoderJNI.java
@@ -17,6 +17,7 @@
   private static native void nativePush(long[] context, int length);
   private static native ByteBuffer nativePull(long[] context);
   private static native void nativeDestroy(long[] context);
+  private static native boolean nativeAttachDictionary(long[] context, ByteBuffer dictionary);
 
   public enum Status {
     ERROR,
@@ -40,6 +41,19 @@
       }
     }
 
+    public boolean attachDictionary(ByteBuffer dictionary) {
+      if (!dictionary.isDirect()) {
+        throw new IllegalArgumentException("only direct buffers allowed");
+      }
+      if (context[0] == 0) {
+        throw new IllegalStateException("brotli decoder is already destroyed");
+      }
+      if (!fresh) {
+        throw new IllegalStateException("decoding is already started");
+      }
+      return nativeAttachDictionary(context, dictionary);
+    }
+
     public void push(int length) {
       if (length < 0) {
         throw new IllegalArgumentException("negative block length");
diff --git a/java/org/brotli/wrapper/dec/decoder_jni.cc b/java/org/brotli/wrapper/dec/decoder_jni.cc
index 268a10b..e10ffe3 100644
--- a/java/org/brotli/wrapper/dec/decoder_jni.cc
+++ b/java/org/brotli/wrapper/dec/decoder_jni.cc
@@ -15,6 +15,9 @@
 typedef struct DecoderHandle {
   BrotliDecoderState* state;
 
+  jobject dictionary_refs[15];
+  size_t dictionary_count;
+
   uint8_t* input_start;
   size_t input_offset;
   size_t input_length;
@@ -54,6 +57,10 @@
   ok = !!handle;
 
   if (ok) {
+    for (int i = 0; i < 15; ++i) {
+      handle->dictionary_refs[i] = nullptr;
+    }
+    handle->dictionary_count = 0;
     handle->input_offset = 0;
     handle->input_length = 0;
     handle->input_start = nullptr;
@@ -193,10 +200,53 @@
   env->GetLongArrayRegion(ctx, 0, 3, context);
   DecoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0]));
   BrotliDecoderDestroyInstance(handle->state);
+  for (size_t i = 0; i < handle->dictionary_count; ++i) {
+    env->DeleteGlobalRef(handle->dictionary_refs[i]);
+  }
   delete[] handle->input_start;
   delete handle;
 }
 
+JNIEXPORT jboolean JNICALL
+Java_org_brotli_wrapper_dec_DecoderJNI_nativeAttachDictionary(
+    JNIEnv* env, jobject /*jobj*/, jlongArray ctx, jobject dictionary) {
+  jlong context[3];
+  env->GetLongArrayRegion(ctx, 0, 3, context);
+  DecoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0]));
+  jobject ref = nullptr;
+  uint8_t* address = nullptr;
+  jlong capacity = 0;
+
+  bool ok = true;
+  if (ok && !dictionary) {
+    ok = false;
+  }
+  if (ok && handle->dictionary_count >= 15) {
+    ok = false;
+  }
+  if (ok) {
+    ref = env->NewGlobalRef(dictionary);
+    ok = !!ref;
+  }
+  if (ok) {
+    handle->dictionary_refs[handle->dictionary_count] = ref;
+    handle->dictionary_count++;
+    address = static_cast<uint8_t*>(env->GetDirectBufferAddress(ref));
+    ok = !!address;
+  }
+  if (ok) {
+    capacity = env->GetDirectBufferCapacity(ref);
+    ok = (capacity > 0) && (capacity < (1 << 30));
+  }
+  if (ok) {
+    size_t size = static_cast<size_t>(capacity);
+    ok = !!BrotliDecoderAttachDictionary(handle->state,
+        BROTLI_SHARED_DICTIONARY_RAW, size, address);
+  }
+
+  return static_cast<jboolean>(ok);
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/java/org/brotli/wrapper/enc/BUILD b/java/org/brotli/wrapper/enc/BUILD
index 22154d2..0f8d2e1 100644
--- a/java/org/brotli/wrapper/enc/BUILD
+++ b/java/org/brotli/wrapper/enc/BUILD
@@ -18,6 +18,10 @@
         ["*.java"],
         exclude = ["*Test*.java"],
     ),
+    deps = [
+        "//org/brotli/common:shared_dictionary",
+        "//org/brotli/enc:prepared_dictionary",
+    ],
     resource_jars = ["//:license"],
 )
 
@@ -27,8 +31,11 @@
     srcs = glob(["*Test*.java"]),
     deps = [
         ":enc",
+        "//org/brotli/common:shared_dictionary",
+        "//org/brotli/enc:prepared_dictionary",
         "//org/brotli/integration:brotli_jni_test_base",
         "//org/brotli/integration:bundle_helper",
+        "//org/brotli/wrapper/common",
         "//org/brotli/wrapper/dec",
         "@maven//:junit_junit",
     ],
diff --git a/java/org/brotli/wrapper/enc/BrotliEncoderChannel.java b/java/org/brotli/wrapper/enc/BrotliEncoderChannel.java
index 047c356..2ee2ba6 100644
--- a/java/org/brotli/wrapper/enc/BrotliEncoderChannel.java
+++ b/java/org/brotli/wrapper/enc/BrotliEncoderChannel.java
@@ -6,6 +6,7 @@
 
 package org.brotli.wrapper.enc;
 
+import org.brotli.enc.PreparedDictionary;
 import java.io.IOException;
 import java.nio.Buffer;
 import java.nio.ByteBuffer;
@@ -43,6 +44,11 @@
   }
 
   @Override
+  public void attachDictionary(PreparedDictionary dictionary) throws IOException {
+    super.attachDictionary(dictionary);
+  }
+
+  @Override
   public boolean isOpen() {
     synchronized (mutex) {
       return !closed;
diff --git a/java/org/brotli/wrapper/enc/BrotliOutputStream.java b/java/org/brotli/wrapper/enc/BrotliOutputStream.java
index 5bd3957..09cbd5a 100644
--- a/java/org/brotli/wrapper/enc/BrotliOutputStream.java
+++ b/java/org/brotli/wrapper/enc/BrotliOutputStream.java
@@ -6,6 +6,7 @@
 
 package org.brotli.wrapper.enc;
 
+import org.brotli.enc.PreparedDictionary;
 import java.io.IOException;
 import java.io.OutputStream;
 import java.nio.channels.Channels;
@@ -40,6 +41,10 @@
     this(destination, new Encoder.Parameters());
   }
 
+  public void attachDictionary(PreparedDictionary dictionary) throws IOException {
+    encoder.attachDictionary(dictionary);
+  }
+
   @Override
   public void close() throws IOException {
     encoder.close();
diff --git a/java/org/brotli/wrapper/enc/Encoder.java b/java/org/brotli/wrapper/enc/Encoder.java
index af579a3..2aa9890 100644
--- a/java/org/brotli/wrapper/enc/Encoder.java
+++ b/java/org/brotli/wrapper/enc/Encoder.java
@@ -6,17 +6,20 @@
 
 package org.brotli.wrapper.enc;
 
+import org.brotli.enc.PreparedDictionary;
 import java.io.IOException;
 import java.nio.Buffer;
 import java.nio.ByteBuffer;
 import java.nio.channels.WritableByteChannel;
 import java.util.ArrayList;
+import java.util.List;
 
 /**
  * Base class for OutputStream / Channel implementations.
  */
 public class Encoder {
   private final WritableByteChannel destination;
+  private final List<PreparedDictionary> dictionaries;
   private final EncoderJNI.Wrapper encoder;
   private ByteBuffer buffer;
   final ByteBuffer inputBuffer;
@@ -111,6 +114,7 @@
     if (destination == null) {
       throw new NullPointerException("destination can not be null");
     }
+    this.dictionaries = new ArrayList<PreparedDictionary>();
     this.destination = destination;
     this.encoder = new EncoderJNI.Wrapper(inputBufferSize, params.quality, params.lgwin, params.mode);
     this.inputBuffer = this.encoder.getInputBuffer();
@@ -125,6 +129,14 @@
     throw new IOException(message);
   }
 
+  public void attachDictionary(PreparedDictionary dictionary) throws IOException {
+    if (!encoder.attachDictionary(dictionary.getData())) {
+      fail("failed to attach dictionary");
+    }
+    // Reference to native prepared dictionary wrapper should be held till the end of encoding.
+    dictionaries.add(dictionary);
+  }
+
   /**
    * @param force repeat pushing until all output is consumed
    * @return true if all encoder output is consumed
@@ -239,4 +251,15 @@
   public static byte[] compress(byte[] data) throws IOException {
     return compress(data, new Parameters());
   }
+
+  /**
+   * Prepares raw or serialized dictionary for being used by encoder.
+   *
+   * @param dictionary raw / serialized dictionary data; MUST be direct
+   * @param sharedDictionaryType dictionary data type
+   */
+  public static PreparedDictionary prepareDictionary(ByteBuffer dictionary,
+      int sharedDictionaryType) {
+    return EncoderJNI.prepareDictionary(dictionary, sharedDictionaryType);
+  }
 }
diff --git a/java/org/brotli/wrapper/enc/EncoderJNI.java b/java/org/brotli/wrapper/enc/EncoderJNI.java
index d0c5fac..1cbd7c1 100644
--- a/java/org/brotli/wrapper/enc/EncoderJNI.java
+++ b/java/org/brotli/wrapper/enc/EncoderJNI.java
@@ -6,6 +6,7 @@
 
 package org.brotli.wrapper.enc;
 
+import org.brotli.enc.PreparedDictionary;
 import java.io.IOException;
 import java.nio.ByteBuffer;
 
@@ -17,6 +18,9 @@
   private static native void nativePush(long[] context, int length);
   private static native ByteBuffer nativePull(long[] context);
   private static native void nativeDestroy(long[] context);
+  private static native boolean nativeAttachDictionary(long[] context, ByteBuffer dictionary);
+  private static native ByteBuffer nativePrepareDictionary(ByteBuffer dictionary, long type);
+  private static native void nativeDestroyDictionary(ByteBuffer dictionary);
 
   enum Operation {
     PROCESS,
@@ -24,6 +28,47 @@
     FINISH
   }
 
+  private static class PreparedDictionaryImpl implements PreparedDictionary {
+    private ByteBuffer data;
+
+    private PreparedDictionaryImpl(ByteBuffer data) {
+      this.data = data;
+    }
+
+    @Override
+    public ByteBuffer getData() {
+      return data;
+    }
+
+    @Override
+    protected void finalize() throws Throwable {
+      try {
+        ByteBuffer data = this.data;
+        this.data = null;
+        nativeDestroyDictionary(data);
+      } finally {
+        super.finalize();
+      }
+    }
+  }
+
+  /**
+   * Prepares raw or serialized dictionary for being used by encoder.
+   *
+   * @param dictionary raw / serialized dictionary data; MUST be direct
+   * @param sharedDictionaryType dictionary data type
+   */
+  static PreparedDictionary prepareDictionary(ByteBuffer dictionary, int sharedDictionaryType) {
+    if (!dictionary.isDirect()) {
+      throw new IllegalArgumentException("only direct buffers allowed");
+    }
+    ByteBuffer dictionaryData = nativePrepareDictionary(dictionary, sharedDictionaryType);
+    if (dictionaryData == null) {
+      throw new IllegalStateException("OOM");
+    }
+    return new PreparedDictionaryImpl(dictionaryData);
+  }
+
   static class Wrapper {
     protected final long[] context = new long[5];
     private final ByteBuffer inputBuffer;
@@ -48,6 +93,19 @@
       this.context[4] = 0;
     }
 
+    boolean attachDictionary(ByteBuffer dictionary) {
+      if (!dictionary.isDirect()) {
+        throw new IllegalArgumentException("only direct buffers allowed");
+      }
+      if (context[0] == 0) {
+        throw new IllegalStateException("brotli decoder is already destroyed");
+      }
+      if (!fresh) {
+        throw new IllegalStateException("decoding is already started");
+      }
+      return nativeAttachDictionary(context, dictionary);
+    }
+
     void push(Operation op, int length) {
       if (length < 0) {
         throw new IllegalArgumentException("negative block length");
diff --git a/java/org/brotli/wrapper/enc/UseCompoundDictionaryTest.java b/java/org/brotli/wrapper/enc/UseCompoundDictionaryTest.java
new file mode 100644
index 0000000..cbfd0d9
--- /dev/null
+++ b/java/org/brotli/wrapper/enc/UseCompoundDictionaryTest.java
@@ -0,0 +1,114 @@
+package org.brotli.wrapper.enc;
+
+import static org.junit.Assert.assertEquals;
+import static org.junit.Assert.assertTrue;
+
+import org.brotli.common.SharedDictionaryType;
+import org.brotli.enc.PreparedDictionary;
+import org.brotli.enc.PreparedDictionaryGenerator;
+import org.brotli.integration.BrotliJniTestBase;
+import org.brotli.integration.BundleHelper;
+import org.brotli.wrapper.common.BrotliCommon;
+import org.brotli.wrapper.dec.BrotliInputStream;
+import java.io.ByteArrayInputStream;
+import java.io.ByteArrayOutputStream;
+import java.io.FileInputStream;
+import java.io.IOException;
+import java.io.InputStream;
+import java.nio.ByteBuffer;
+import java.util.List;
+import junit.framework.TestCase;
+import junit.framework.TestSuite;
+import org.junit.runner.RunWith;
+import org.junit.runners.AllTests;
+
+/** Tests for compression / decompression aided with LZ77 dictionary. */
+@RunWith(AllTests.class)
+public class UseCompoundDictionaryTest extends BrotliJniTestBase {
+
+  static InputStream getBundle() throws IOException {
+    return new FileInputStream(System.getProperty("TEST_BUNDLE"));
+  }
+
+  /** Creates a test suite. */
+  public static TestSuite suite() throws IOException {
+    TestSuite suite = new TestSuite();
+    InputStream bundle = getBundle();
+    try {
+      List<String> entries = BundleHelper.listEntries(bundle);
+      for (String entry : entries) {
+        suite.addTest(new UseCompoundDictionaryTestCase(entry));
+      }
+    } finally {
+      bundle.close();
+    }
+    return suite;
+  }
+
+  /** Test case with a unique name. */
+  static class UseCompoundDictionaryTestCase extends TestCase {
+    final String entryName;
+    UseCompoundDictionaryTestCase(String entryName) {
+      super("UseCompoundDictionaryTest." + entryName);
+      this.entryName = entryName;
+    }
+
+    @Override
+    protected void runTest() throws Throwable {
+      UseCompoundDictionaryTest.run(entryName);
+    }
+  }
+
+  private static PreparedDictionary prepareRawDictionary(String entryName, ByteBuffer data) {
+    if (entryName.endsWith("E.coli.txt")) {
+      // Default prepared dictionary parameters doesn't work well for DNA data.
+      // Should work well with 8-byte hash.
+      return PreparedDictionaryGenerator.generate(data, 17, 3, 64, 5);
+    } else {
+       return Encoder.prepareDictionary(data, SharedDictionaryType.RAW);
+    }
+  }
+
+  private static void run(String entryName) throws Throwable {
+    InputStream bundle = getBundle();
+    byte[] original;
+    try {
+      original = BundleHelper.readEntry(bundle, entryName);
+    } finally {
+      bundle.close();
+    }
+
+    if (original == null) {
+      throw new RuntimeException("Can't read bundle entry: " + entryName);
+    }
+
+    ByteBuffer dictionary = BrotliCommon.makeNative(original);
+    PreparedDictionary preparedDictionary = prepareRawDictionary(entryName, dictionary);
+
+    ByteArrayOutputStream dst = new ByteArrayOutputStream();
+    BrotliOutputStream encoder =
+        new BrotliOutputStream(dst, new Encoder.Parameters().setQuality(9).setWindow(23), 1 << 23);
+    encoder.attachDictionary(preparedDictionary);
+    try {
+      encoder.write(original);
+    } finally {
+      encoder.close();
+    }
+
+    byte[] compressed = dst.toByteArray();
+
+    // Just copy self from LZ77 dictionary -> ultimate compression ratio.
+    assertTrue(compressed.length < 80 + original.length / 65536);
+
+    BrotliInputStream decoder =
+        new BrotliInputStream(new ByteArrayInputStream(compressed), 1 << 23);
+    decoder.attachDictionary(dictionary);
+    try {
+      long originalCrc = BundleHelper.fingerprintStream(new ByteArrayInputStream(original));
+      long crc = BundleHelper.fingerprintStream(decoder);
+      assertEquals(originalCrc, crc);
+    } finally {
+      decoder.close();
+    }
+  }
+}
\ No newline at end of file
diff --git a/java/org/brotli/wrapper/enc/encoder_jni.cc b/java/org/brotli/wrapper/enc/encoder_jni.cc
index e69f719..00fbfd8 100644
--- a/java/org/brotli/wrapper/enc/encoder_jni.cc
+++ b/java/org/brotli/wrapper/enc/encoder_jni.cc
@@ -15,6 +15,9 @@
 typedef struct EncoderHandle {
   BrotliEncoderState* state;
 
+  jobject dictionary_refs[15];
+  size_t dictionary_count;
+
   uint8_t* input_start;
   size_t input_offset;
   size_t input_last;
@@ -53,6 +56,10 @@
   ok = !!handle;
 
   if (ok) {
+    for (int i = 0; i < 15; ++i) {
+      handle->dictionary_refs[i] = nullptr;
+    }
+    handle->dictionary_count = 0;
     handle->input_offset = 0;
     handle->input_last = 0;
     handle->input_start = nullptr;
@@ -190,10 +197,90 @@
   env->GetLongArrayRegion(ctx, 0, 2, context);
   EncoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0]));
   BrotliEncoderDestroyInstance(handle->state);
+  for (size_t i = 0; i < handle->dictionary_count; ++i) {
+    env->DeleteGlobalRef(handle->dictionary_refs[i]);
+  }
   delete[] handle->input_start;
   delete handle;
 }
 
+JNIEXPORT jboolean JNICALL
+Java_org_brotli_wrapper_enc_EncoderJNI_nativeAttachDictionary(
+    JNIEnv* env, jobject /*jobj*/, jlongArray ctx, jobject dictionary) {
+  jlong context[2];
+  env->GetLongArrayRegion(ctx, 0, 2, context);
+  EncoderHandle* handle = getHandle(reinterpret_cast<void*>(context[0]));
+  jobject ref = nullptr;
+  uint8_t* address = nullptr;
+
+  bool ok = true;
+  if (ok && !dictionary) {
+    ok = false;
+  }
+  if (ok && handle->dictionary_count >= 15) {
+    ok = false;
+  }
+  if (ok) {
+    ref = env->NewGlobalRef(dictionary);
+    ok = !!ref;
+  }
+  if (ok) {
+    handle->dictionary_refs[handle->dictionary_count] = ref;
+    handle->dictionary_count++;
+    address = static_cast<uint8_t*>(env->GetDirectBufferAddress(ref));
+    ok = !!address;
+  }
+  if (ok) {
+    ok = !!BrotliEncoderAttachPreparedDictionary(handle->state,
+        reinterpret_cast<BrotliEncoderPreparedDictionary*>(address));
+  }
+
+  return static_cast<jboolean>(ok);
+}
+
+JNIEXPORT void JNICALL
+Java_org_brotli_wrapper_enc_EncoderJNI_nativeDestroyDictionary(
+    JNIEnv* env, jobject /*jobj*/, jobject dictionary) {
+  if (!dictionary) {
+    return;
+  }
+  uint8_t* address =
+      static_cast<uint8_t*>(env->GetDirectBufferAddress(dictionary));
+  if (!address) {
+    return;
+  }
+  BrotliEncoderDestroyPreparedDictionary(
+      reinterpret_cast<BrotliEncoderPreparedDictionary*>(address));
+}
+
+JNIEXPORT jobject JNICALL
+Java_org_brotli_wrapper_enc_EncoderJNI_nativePrepareDictionary(
+    JNIEnv* env, jobject /*jobj*/, jobject dictionary, jlong type) {
+  if (!dictionary) {
+    return nullptr;
+  }
+  uint8_t* address =
+      static_cast<uint8_t*>(env->GetDirectBufferAddress(dictionary));
+  if (!address) {
+    return nullptr;
+  }
+  jlong capacity = env->GetDirectBufferCapacity(dictionary);
+  if ((capacity <= 0) || (capacity >= (1 << 30))) {
+    return nullptr;
+  }
+  BrotliSharedDictionaryType dictionary_type =
+      static_cast<BrotliSharedDictionaryType>(type);
+  size_t size = static_cast<size_t>(capacity);
+  BrotliEncoderPreparedDictionary* prepared_dictionary =
+      BrotliEncoderPrepareDictionary(dictionary_type, size, address,
+        BROTLI_MAX_QUALITY, nullptr, nullptr, nullptr);
+  if (!prepared_dictionary) {
+    return nullptr;
+  }
+  /* Size is 4 - just enough to check magic bytes. */
+  return env->NewDirectByteBuffer(prepared_dictionary, 4);
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/scripts/sources.lst b/scripts/sources.lst
index 19a6d00..18d3867 100644
--- a/scripts/sources.lst
+++ b/scripts/sources.lst
@@ -9,6 +9,7 @@
   c/common/context.c \
   c/common/dictionary.c \
   c/common/platform.c \
+  c/common/shared_dictionary.c \
   c/common/transform.c
 
 BROTLI_COMMON_H = \
@@ -16,6 +17,7 @@
   c/common/context.h \
   c/common/dictionary.h \
   c/common/platform.h \
+  c/common/shared_dictionary_internal.h \
   c/common/transform.h \
   c/common/version.h
 
@@ -39,6 +41,7 @@
   c/enc/brotli_bit_stream.c \
   c/enc/cluster.c \
   c/enc/command.c \
+  c/enc/compound_dictionary.c \
   c/enc/compress_fragment.c \
   c/enc/compress_fragment_two_pass.c \
   c/enc/dictionary_hash.c \
@@ -66,6 +69,7 @@
   c/enc/cluster.h \
   c/enc/cluster_inc.h \
   c/enc/command.h \
+  c/enc/compound_dictionary.h \
   c/enc/compress_fragment.h \
   c/enc/compress_fragment_two_pass.h \
   c/enc/dictionary_hash.h \
@@ -101,4 +105,5 @@
   c/include/brotli/decode.h \
   c/include/brotli/encode.h \
   c/include/brotli/port.h \
+  c/include/brotli/shared_dictionary.h \
   c/include/brotli/types.h
diff --git a/setup.py b/setup.py
index 25626ec..3afb035 100644
--- a/setup.py
+++ b/setup.py
@@ -185,6 +185,7 @@
             'c/common/context.c',
             'c/common/dictionary.c',
             'c/common/platform.c',
+            'c/common/shared_dictionary.c',
             'c/common/transform.c',
             'c/dec/bit_reader.c',
             'c/dec/decode.c',
@@ -197,6 +198,7 @@
             'c/enc/brotli_bit_stream.c',
             'c/enc/cluster.c',
             'c/enc/command.c',
+            'c/enc/compound_dictionary.c',
             'c/enc/compress_fragment.c',
             'c/enc/compress_fragment_two_pass.c',
             'c/enc/dictionary_hash.c',
@@ -216,6 +218,7 @@
             'c/common/context.h',
             'c/common/dictionary.h',
             'c/common/platform.h',
+            'c/common/shared_dictionary_internal.h',
             'c/common/transform.h',
             'c/common/version.h',
             'c/dec/bit_reader.h',
@@ -234,6 +237,7 @@
             'c/enc/cluster.h',
             'c/enc/cluster_inc.h',
             'c/enc/command.h',
+            'c/enc/compound_dictionary.h',
             'c/enc/compress_fragment.h',
             'c/enc/compress_fragment_two_pass.h',
             'c/enc/dictionary_hash.h',