brotli 0.4.0 → 0.5.0

data/vendor/brotli/c/enc/cluster_inc.h

@@ -12,8 +12,8 @@
 /* Computes the bit cost reduction by combining out[idx1] and out[idx2] and if
    it is below a threshold, stores the pair (idx1, idx2) in the *pairs queue. */
 BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
-    const HistogramType* out, const uint32_t* cluster_size, uint32_t idx1,
-    uint32_t idx2, size_t max_num_pairs, HistogramPair* pairs,
+    const HistogramType* out, HistogramType* tmp, const uint32_t* cluster_size,
+    uint32_t idx1, uint32_t idx2, size_t max_num_pairs, HistogramPair* pairs,
     size_t* num_pairs) CODE({
   BROTLI_BOOL is_good_pair = BROTLI_FALSE;
   HistogramPair p;
@@ -42,10 +42,10 @@ BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
   } else {
     double threshold = *num_pairs == 0 ? 1e99 :
         BROTLI_MAX(double, 0.0, pairs[0].cost_diff);
-    HistogramType combo = out[idx1];
     double cost_combo;
-    FN(HistogramAddHistogram)(&combo, &out[idx2]);
-    cost_combo = FN(BrotliPopulationCost)(&combo);
+    *tmp = out[idx1];
+    FN(HistogramAddHistogram)(tmp, &out[idx2]);
+    cost_combo = FN(BrotliPopulationCost)(tmp);
     if (cost_combo < threshold - p.cost_diff) {
       p.cost_combo = cost_combo;
       is_good_pair = BROTLI_TRUE;
@@ -68,6 +68,7 @@ BROTLI_INTERNAL void FN(BrotliCompareAndPushToQueue)(
 })
 
 BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
+                                                  HistogramType* tmp,
                                                   uint32_t* cluster_size,
                                                   uint32_t* symbols,
                                                   uint32_t* clusters,
@@ -87,7 +88,7 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
     for (idx1 = 0; idx1 < num_clusters; ++idx1) {
       size_t idx2;
       for (idx2 = idx1 + 1; idx2 < num_clusters; ++idx2) {
-        FN(BrotliCompareAndPushToQueue)(out, cluster_size, clusters[idx1],
+        FN(BrotliCompareAndPushToQueue)(out, tmp, cluster_size, clusters[idx1],
             clusters[idx2], max_num_pairs, &pairs[0], &num_pairs);
       }
     }
@@ -146,8 +147,8 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
 
     /* Push new pairs formed with the combined histogram to the heap. */
     for (i = 0; i < num_clusters; ++i) {
-      FN(BrotliCompareAndPushToQueue)(out, cluster_size, best_idx1, clusters[i],
-                                      max_num_pairs, &pairs[0], &num_pairs);
+      FN(BrotliCompareAndPushToQueue)(out, tmp, cluster_size, best_idx1,
+          clusters[i], max_num_pairs, &pairs[0], &num_pairs);
     }
   }
   return num_clusters;
@@ -155,13 +156,14 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramCombine)(HistogramType* out,
 
 /* What is the bit cost of moving histogram from cur_symbol to candidate. */
 BROTLI_INTERNAL double FN(BrotliHistogramBitCostDistance)(
-    const HistogramType* histogram, const HistogramType* candidate) CODE({
+    const HistogramType* histogram, const HistogramType* candidate,
+    HistogramType* tmp) CODE({
   if (histogram->total_count_ == 0) {
     return 0.0;
   } else {
-    HistogramType tmp = *histogram;
-    FN(HistogramAddHistogram)(&tmp, candidate);
-    return FN(BrotliPopulationCost)(&tmp) - candidate->bit_cost_;
+    *tmp = *histogram;
+    FN(HistogramAddHistogram)(tmp, candidate);
+    return FN(BrotliPopulationCost)(tmp) - candidate->bit_cost_;
   }
 })
 
@@ -171,16 +173,16 @@ BROTLI_INTERNAL double FN(BrotliHistogramBitCostDistance)(
    Note: we assume that out[]->bit_cost_ is already up-to-date. */
 BROTLI_INTERNAL void FN(BrotliHistogramRemap)(const HistogramType* in,
     size_t in_size, const uint32_t* clusters, size_t num_clusters,
-    HistogramType* out, uint32_t* symbols) CODE({
+    HistogramType* out, HistogramType* tmp, uint32_t* symbols) CODE({
   size_t i;
   for (i = 0; i < in_size; ++i) {
     uint32_t best_out = i == 0 ? symbols[0] : symbols[i - 1];
     double best_bits =
-        FN(BrotliHistogramBitCostDistance)(&in[i], &out[best_out]);
+        FN(BrotliHistogramBitCostDistance)(&in[i], &out[best_out], tmp);
     size_t j;
     for (j = 0; j < num_clusters; ++j) {
       const double cur_bits =
-          FN(BrotliHistogramBitCostDistance)(&in[i], &out[clusters[j]]);
+          FN(BrotliHistogramBitCostDistance)(&in[i], &out[clusters[j]], tmp);
       if (cur_bits < best_bits) {
         best_bits = cur_bits;
         best_out = clusters[j];
@@ -226,7 +228,7 @@ BROTLI_INTERNAL size_t FN(BrotliHistogramReindex)(MemoryManager* m,
       ++next_index;
     }
   }
-  /* TODO: by using idea of "cycle-sort" we can avoid allocation of
+  /* TODO(eustas): by using idea of "cycle-sort" we can avoid allocation of
      tmp and reduce the number of copying by the factor of 2. */
   tmp = BROTLI_ALLOC(m, HistogramType, next_index);
   if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(tmp)) return 0;
@@ -257,10 +259,12 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
   size_t pairs_capacity = max_input_histograms * max_input_histograms / 2;
   /* For the first pass of clustering, we allow all pairs. */
   HistogramPair* pairs = BROTLI_ALLOC(m, HistogramPair, pairs_capacity + 1);
+  /* TODO(eustas): move to "persistent" arena? */
+  HistogramType* tmp = BROTLI_ALLOC(m, HistogramType, 1);
   size_t i;
 
   if (BROTLI_IS_OOM(m) || BROTLI_IS_NULL(cluster_size) ||
-      BROTLI_IS_NULL(clusters) || BROTLI_IS_NULL(pairs)) {
+      BROTLI_IS_NULL(clusters) || BROTLI_IS_NULL(pairs)|| BROTLI_IS_NULL(tmp)) {
     return;
   }
 
@@ -283,7 +287,7 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
       clusters[num_clusters + j] = (uint32_t)(i + j);
     }
     num_new_clusters =
-        FN(BrotliHistogramCombine)(out, cluster_size,
+        FN(BrotliHistogramCombine)(out, tmp, cluster_size,
                                    &histogram_symbols[i],
                                    &clusters[num_clusters], pairs,
                                    num_to_combine, num_to_combine,
@@ -301,7 +305,7 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
     if (BROTLI_IS_OOM(m)) return;
 
     /* Collapse similar histograms. */
-    num_clusters = FN(BrotliHistogramCombine)(out, cluster_size,
+    num_clusters = FN(BrotliHistogramCombine)(out, tmp, cluster_size,
                                               histogram_symbols, clusters,
                                               pairs, num_clusters, in_size,
                                               max_histograms, max_num_pairs);
@@ -310,7 +314,8 @@ BROTLI_INTERNAL void FN(BrotliClusterHistograms)(
   BROTLI_FREE(m, cluster_size);
   /* Find the optimal map from original histograms to the final ones. */
   FN(BrotliHistogramRemap)(in, in_size, clusters, num_clusters,
-                           out, histogram_symbols);
+                           out, tmp, histogram_symbols);
+  BROTLI_FREE(m, tmp);
   BROTLI_FREE(m, clusters);
   /* Convert the context map to a canonical form. */
   *out_size = FN(BrotliHistogramReindex)(m, out, histogram_symbols, in_size);

data/vendor/brotli/c/enc/command.c

@@ -4,7 +4,7 @@
    See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
 */
 
-#include "./command.h"
+#include "command.h"
 
 #include <brotli/types.h>
 

data/vendor/brotli/c/enc/command.h

@@ -9,12 +9,13 @@
 #ifndef BROTLI_ENC_COMMAND_H_
 #define BROTLI_ENC_COMMAND_H_
 
+#include <brotli/types.h>
+
 #include "../common/constants.h"
 #include "../common/platform.h"
-#include <brotli/types.h>
-#include "./fast_log.h"
-#include "./params.h"
-#include "./prefix.h"
+#include "fast_log.h"
+#include "params.h"
+#include "prefix.h"
 
 #if defined(__cplusplus) || defined(c_plusplus)
 extern "C" {

data/vendor/brotli/c/enc/compound_dictionary.c

@@ -0,0 +1,207 @@
+/* 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 <brotli/types.h>
+
+#include "../common/platform.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_ref = 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(eustas): 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) +
+      sizeof(uint8_t*);
+
+  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_ref = (uint8_t**)(&items[total_items]);
+
+  result->magic = kLeanPreparedDictionaryMagic;
+  result->num_items = total_items;
+  result->source_size = (uint32_t)source_size;
+  result->hash_bits = hash_bits;
+  result->bucket_bits = bucket_bits;
+  result->slot_bits = slot_bits;
+  BROTLI_UNALIGNED_STORE_PTR(source_ref, source);
+
+  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);
+  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]);
+    const void* tail = (void*)&items[dictionary->num_items];
+    if (dictionary->magic == kPreparedDictionaryMagic) {
+      compound->chunk_source[index] = (const uint8_t*)tail;
+    } else {
+      /* dictionary->magic == kLeanPreparedDictionaryMagic */
+      compound->chunk_source[index] =
+          (const uint8_t*)BROTLI_UNALIGNED_LOAD_PTR((const uint8_t**)tail);
+    }
+  }
+  compound->num_chunks++;
+  return BROTLI_TRUE;
+}

data/vendor/brotli/c/enc/compound_dictionary.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
+*/
+
+#ifndef BROTLI_ENC_PREPARED_DICTIONARY_H_
+#define BROTLI_ENC_PREPARED_DICTIONARY_H_
+
+#include <brotli/shared_dictionary.h>
+#include <brotli/types.h>
+
+#include "../common/platform.h"
+#include "../common/constants.h"
+#include "memory.h"
+
+/* "Fat" prepared dictionary, could be cooked outside of C implementation,
+ * e.g. on Java side. LZ77 data is copied inside PreparedDictionary struct. */
+static const uint32_t kPreparedDictionaryMagic = 0xDEBCEDE0;
+
+static const uint32_t kSharedDictionaryMagic = 0xDEBCEDE1;
+
+static const uint32_t kManagedDictionaryMagic = 0xDEBCEDE2;
+
+/* "Lean" prepared dictionary. LZ77 data is referenced. It is the responsibility
+ * of caller of "prepare dictionary" to keep the LZ77 data while prepared
+ * dictionary is in use. */
+static const uint32_t kLeanPreparedDictionaryMagic = 0xDEBCEDE3;
+
+static const uint64_t kPreparedDictionaryHashMul64Long =
+    BROTLI_MAKE_UINT64_T(0x1FE35A7Bu, 0xD3579BD3u);
+
+typedef struct PreparedDictionary {
+  uint32_t magic;
+  uint32_t num_items;
+  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]; */
+
+  /* [maybe] uint8_t* source_ref, depending on magic. */
+  /* [maybe] uint8_t source[source_size], depending on magic. */
+} 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 */