RetroArch/input/bsv/uint32s_index.c

/**
 *  RetroArch - A frontend for libretro.
 *
 *  RetroArch is free software: you can redistribute it and/or modify it under
 *  the terms of the GNU General Public License as published by the Free
 *  Software Foundation, either version 3 of the License, or (at your option)
 *  any later version.
 *
 *  RetroArch is distributed in the hope that it will be useful, but WITHOUT
 *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 *  more details.
 *
 *  You should have received a copy of the GNU General Public License along
 *  with RetroArch. If not, see <http://www.gnu.org/licenses/>.
 **/

#ifdef HAVE_STATESTREAM
#include "uint32s_index.h"
#include <string.h>
#include <array/rhmap.h>
#include <array/rbuf.h>
#include <retro_common_api.h>
#include "../../verbosity.h"

#include <encodings/xxh3.h>

#define HASHMAP_CAP 65536
/* A test build can define its own hash to force collisions. */
#ifndef uint32s_hash_bytes
#define uint32s_hash_bytes(bytes, len) ((uint32_t)xxh3_64(bytes, len, 0))
#endif

/* Room for one more object in every per-object list, so the pushes
 * that record it cannot fail half way. */
static bool uint32s_index_reserve(uint32s_index_t *index)
{
   return RBUF_TRYFIT(index->objects,   RBUF_LEN(index->objects)   + 1)
       && RBUF_TRYFIT(index->counts,    RBUF_LEN(index->counts)    + 1)
       && RBUF_TRYFIT(index->hashes,    RBUF_LEN(index->hashes)    + 1)
       && RBUF_TRYFIT(index->additions, RBUF_LEN(index->additions) + 1);
}

uint32s_index_t *uint32s_index_new(size_t object_size,
      uint8_t commit_interval, uint8_t commit_threshold)
{
   uint32_t *zeros         = (uint32_t*)calloc(object_size ? object_size : 1, sizeof(uint32_t));
   uint32s_index_t *index  = (uint32s_index_t *)malloc(sizeof(uint32s_index_t));
   if (!zeros || !index)
   {
      free(zeros);
      free(index);
      return NULL;
   }
   index->object_size      = object_size;
   index->index            = NULL;
   RHMAP_FIT(index->index, HASHMAP_CAP);
   index->objects          = NULL;
   index->counts           = NULL;
   index->hashes           = NULL;
   index->additions        = NULL;
   index->commit_interval  = commit_interval;
   index->commit_threshold = commit_threshold;
   /* transfers ownership of zero buffer */
   if (!index->index || !uint32s_index_insert_exact(index, 0, zeros, 0))
   {
      free(zeros);
      uint32s_index_free(index);
      return NULL;
   }
   RBUF_CLEAR(index->additions); /* scrap first addition, we never want to delete 0s during rewind */
   return index;
}

void uint32s_bucket_free(struct uint32s_bucket *bucket)
{
   if (bucket->len > 3)
   {
      free(bucket->contents.vec.idxs);
      bucket->contents.vec.idxs = NULL;
      bucket->contents.vec.cap  = 0;
   }
   /* RHMAP_CLEAR leaves value-array entries untouched; subsequent
    * RHMAP_PTR (which auto-creates) on the cleared slot would re-read
    * the stale len/vec.idxs and double-free.  Reset len to 0 so the
    * bucket is unambiguously inline-empty after free. */
   bucket->len = 0;
}

bool uint32s_bucket_get(uint32s_index_t *index, struct uint32s_bucket *bucket, uint32_t *object, size_t size_bytes, uint32_t *out_idx)
{
   uint32_t i;
   uint32_t *coll = bucket->len < 4 ? bucket->contents.idxs : bucket->contents.vec.idxs;
   for (i = 0; i < bucket->len; i++)
   {
      uint32_t idx = coll[i];
      if (memcmp(index->objects[idx], object, size_bytes) == 0)
      {
         *out_idx = idx;
         return true;
      }
   }
   return false;
}

/* Adds idx to the bucket; false, with the bucket unchanged, when it
 * cannot grow. */
static bool uint32s_bucket_expand(struct uint32s_bucket *bucket, uint32_t idx)
{
   if (bucket->len < 3)
      bucket->contents.idxs[bucket->len] = idx;
   else if (bucket->len == 3)
   {
      uint32_t *idxs = (uint32_t*)calloc(8, sizeof(uint32_t));
      if (!idxs)
         return false;
      memcpy(idxs, bucket->contents.idxs, 3*sizeof(uint32_t));
      bucket->contents.vec.cap  = 8;
      bucket->contents.vec.idxs = idxs;
      bucket->contents.vec.idxs[bucket->len] = idx;
   }
   else if (bucket->len < bucket->contents.vec.cap)
      bucket->contents.vec.idxs[bucket->len] = idx;
   else /* bucket->len == bucket->contents.vec.cap */
   {
      uint32_t  new_cap = bucket->contents.vec.cap * 2;
      uint32_t *tmp     = (uint32_t*)realloc(bucket->contents.vec.idxs,
            new_cap * sizeof(uint32_t));
      if (!tmp)
         return false;
      bucket->contents.vec.cap  = new_cap;
      bucket->contents.vec.idxs = tmp;
      bucket->contents.vec.idxs[bucket->len] = idx;
   }
   bucket->len++;
   return true;
}

bool uint32s_bucket_remove(struct uint32s_bucket *bucket, uint32_t idx)
{
   int i;
   bool small     = bucket->len < 4;
   uint32_t *coll = small ? bucket->contents.idxs : bucket->contents.vec.idxs;
   if (idx == 0) /* never remove 0s pattern */
      return false;

   for (i = 0; i < (int)bucket->len; i++)
   {
      if (coll[i] == idx)
      {
         memmove((uint8_t*)(coll+i), (uint8_t*)(coll+i+1), (bucket->len-(i+1))*sizeof(uint32_t));
         bucket->len--;
         if (bucket->len == 3)
         {
            memcpy(bucket->contents.idxs, coll, 3*sizeof(uint32_t));
            free(coll);
         }
         return true;
      }
   }
   RARCH_ERR("[STATESTREAM] didn't find index %d during remove\n",idx);
   return false;
}

uint32s_insert_result_t uint32s_index_insert(uint32s_index_t *index, uint32_t *object, uint64_t frame)
{
   uint32_t idx;
   uint32_t *copy;
   struct uint32s_bucket *bucket;
   uint32s_insert_result_t result;
   size_t size_bytes      = index->object_size * sizeof(uint32_t);
   uint32_t hash          = uint32s_hash_bytes((uint8_t *)object, size_bytes);
   uint32_t additions_len = RBUF_LEN(index->additions);
   result.index  = 0;
   result.is_new = false;
   if (RHMAP_HAS(index->index, hash))
   {
      bucket = RHMAP_PTR(index->index, hash);
      if (uint32s_bucket_get(index, bucket, object, size_bytes, &result.index))
      {
         if (index->objects[result.index])
         {
            result.is_new = false;
            index->counts[result.index]++;
            return result;
         }

         RARCH_LOG("[STATESTREAM] accessed collected index %u\n", result.index);
      }
      idx  = RBUF_LEN(index->objects);
      if (     !uint32s_index_reserve(index)
            || !(copy = (uint32_t*)malloc(size_bytes)))
      {
         result.index = UINT32S_INDEX_NONE;
         return result;
      }
      memcpy(copy, object, size_bytes);
      /* The bucket takes the index first, so an object is never stored
       * where a lookup cannot find it. */
      if (!uint32s_bucket_expand(bucket, idx))
      {
         free(copy);
         result.index = UINT32S_INDEX_NONE;
         return result;
      }
      RBUF_PUSH(index->objects, copy);
      RBUF_PUSH(index->counts, 1);
      RBUF_PUSH(index->hashes, hash);
      result.index = idx;
      result.is_new = true;
   }
   else
   {
      struct uint32s_bucket new_bucket;
      idx  = RBUF_LEN(index->objects);
      if (     !uint32s_index_reserve(index)
            || !(copy = (uint32_t*)malloc(size_bytes)))
      {
         result.index = UINT32S_INDEX_NONE;
         return result;
      }
      memcpy(copy, object, size_bytes);
      RBUF_PUSH(index->objects, copy);
      RBUF_PUSH(index->counts, 1);
      RBUF_PUSH(index->hashes, hash);
      new_bucket.len = 1;
      new_bucket.contents.idxs[0] = idx;
      new_bucket.contents.idxs[1] = 0;
      new_bucket.contents.idxs[2] = 0;
      RHMAP_SET(index->index, hash, new_bucket);
      result.index = idx;
      result.is_new = true;
   }
   if (additions_len == 0 || index->additions[additions_len-1].frame_counter < frame)
   {
      struct uint32s_frame_addition addition;
      addition.frame_counter = frame;
      addition.first_index = result.index;
      RBUF_PUSH(index->additions, addition);
   }
   return result;
}

bool uint32s_index_insert_exact(uint32s_index_t *index, uint32_t idx, uint32_t *object, uint64_t frame)
{
   struct uint32s_bucket *bucket;
   uint32_t hash;
   size_t size_bytes;
   uint32_t additions_len;
   if (idx != RBUF_LEN(index->objects) || !uint32s_index_reserve(index))
      return false;
   size_bytes = index->object_size * sizeof(uint32_t);
   hash = uint32s_hash_bytes((uint8_t *)object, size_bytes);
   additions_len = RBUF_LEN(index->additions);
   if (RHMAP_HAS(index->index, hash))
   {
      bucket = RHMAP_PTR(index->index, hash);
      if (!uint32s_bucket_expand(bucket, idx))
         return false;
   }
   else
   {
      struct uint32s_bucket new_bucket;
      new_bucket.len = 1;
      new_bucket.contents.idxs[0] = idx;
      new_bucket.contents.idxs[1] = 0;
      new_bucket.contents.idxs[2] = 0;
      RHMAP_SET(index->index, hash, new_bucket);
   }
   /* RARCH_LOG("[STATESTREAM] insert index %d\n",idx); */
   RBUF_PUSH(index->objects, object);
   RBUF_PUSH(index->counts, 1);
   RBUF_PUSH(index->hashes, hash);
   if (   additions_len == 0
       || index->additions[additions_len-1].frame_counter < frame)
   {
      struct uint32s_frame_addition addition;
      addition.frame_counter = frame;
      addition.first_index = idx;
      RBUF_PUSH(index->additions, addition);
   }
   return true;
}

void uint32s_index_commit(uint32s_index_t *index)
{
   uint32_t i, interval=index->commit_interval,threshold=index->commit_threshold;
   struct uint32s_frame_addition prev,cur;
   uint32_t additions_len = RBUF_LEN(index->additions), limit;
   /* The interval comes from the replay header.  Below 2 there is no
    * addition after the one being collected, so cur would be read one
    * past the end of the log; collect nothing instead. */
   if (additions_len < interval || interval < 2)
      return;
   prev  = index->additions[additions_len-interval];
   cur   = index->additions[additions_len-(interval-1)];
   limit = cur.first_index;
   if (limit > RBUF_LEN(index->objects))
      limit = (uint32_t)RBUF_LEN(index->objects);
   for (i = prev.first_index; i < limit; i++)
   {
      struct uint32s_bucket *bucket;
      if (index->counts[i] >= threshold || index->objects[i] == NULL)
         continue;
      free(index->objects[i]);
      index->objects[i] = NULL;
      /* RHMAP_PTR adds on a miss; see uint32s_index_pop. */
      if (!RHMAP_HAS(index->index, index->hashes[i]))
         continue;
      bucket = RHMAP_PTR(index->index, index->hashes[i]);
      uint32s_bucket_remove(bucket, i);
      if (bucket->len == 0)
      {
         uint32s_bucket_free(bucket);
         if (!RHMAP_DEL(index->index, index->hashes[i]))
            RARCH_ERR("[STATESTREAM] Trying to remove absent hash %x\n",index->hashes[i]);
      }
   }
}

void uint32s_index_bump_count(uint32s_index_t *index, uint32_t which)
{
   if (which >= RBUF_LEN(index->counts))
      return;
   index->counts[which]++;
}

uint32_t *uint32s_index_get(uint32s_index_t *index, uint32_t which)
{
   if (which >= RBUF_LEN(index->objects))
      return NULL;
   if (!index->objects[which])
   {
      int i;
      RARCH_LOG("[STATESTREAM] accessed garbage collected block %u\n", which);
      for (i = RBUF_LEN(index->additions); i != 0; i--)
      {
         if (which >= index->additions[i - 1].first_index)
         {
            RARCH_LOG("[STATESTREAM] originally allocated on frame "
                  STRING_REP_UINT64 "\n", index->additions[i - 1].frame_counter);
            break;
         }
      }
      return NULL;
   }
   return index->objects[which];
}

void uint32s_index_pop(uint32s_index_t *index)
{
   uint32_t idx;
   uint32_t hash;
   /* Trust boundary: BSV replay file controls index state.  Bail before
    * RBUF_LEN(...)-1 underflows to UINT32_MAX (turning the next line
    * into a wild read of index->hashes[]). */
   if (RBUF_LEN(index->objects) == 0)
   {
      RARCH_ERR("[STATESTREAM] trying to pop from empty index\n");
      return;
   }
   idx  = RBUF_LEN(index->objects) - 1;
   hash = index->hashes[idx];
   /* RHMAP_PTR has add-on-miss semantics; gate on RHMAP_HAS so a
    * corrupted hashes[] entry that points to a no-longer-present hash
    * can't auto-create a stale-content bucket and fall through into
    * uint32s_bucket_remove with garbage len.  The counts are still
    * adjusted below, so uint32s_index_remove_after keeps making
    * progress. */
   if (index->objects[idx] != NULL && !RHMAP_HAS(index->index, hash))
      RARCH_ERR("[STATESTREAM] pop: missing hash %x for idx %u\n",
            hash, idx);
   else if (index->objects[idx] != NULL) {
      struct uint32s_bucket *bucket = RHMAP_PTR(index->index, hash);
      uint32_t old_len = bucket->len;
      if (old_len == 0)
         RARCH_ERR("[STATESTREAM] Trying to remove from empty bucket, this should never happen");
      if (!uint32s_bucket_remove(bucket, idx))
         RARCH_WARN("[STATESTREAM] Failed to remove idx from bucket");
      if (old_len == 1) {
         uint32s_bucket_free(bucket);
         if (!RHMAP_DEL(index->index, hash))
            RARCH_ERR("[STATESTREAM] Trying to remove absent hash %x\n",hash);
      }
   }
   /* else: already garbage collected, just adjust counts */
   /* The resize below drops the last reference to the object. */
   free(index->objects[idx]);
   RBUF_RESIZE(index->objects, idx);
   RBUF_RESIZE(index->counts, idx);
   RBUF_RESIZE(index->hashes, idx);
}

/* goes backwards from end of additions */
void uint32s_index_remove_after(uint32s_index_t *index, uint64_t frame)
{
   int i;
   for(i = RBUF_LEN(index->additions)-1; i >= 0; i--)
   {
      struct uint32s_frame_addition add = index->additions[i];
      if (add.frame_counter <= frame)
         break;
      while(add.first_index < RBUF_LEN(index->objects))
         uint32s_index_pop(index);
   }
   RBUF_RESIZE(index->additions, i+1);
}

/* removes all data from index */
void uint32s_index_clear(uint32s_index_t *index)
{
   size_t i, cap;
   uint32_t *zeros = index->objects[0];
   for(i = 0, cap = RHMAP_CAP(index->index); i != cap; i++)
      if (RHMAP_KEY(index->index, i))
         uint32s_bucket_free(&index->index[i]);
   RHMAP_CLEAR(index->index);
   /* don't dealloc all-zeros pattern */
   for(i = 1; i < RBUF_LEN(index->objects); i++)
      free(index->objects[i]);
   RBUF_CLEAR(index->objects);
   RBUF_CLEAR(index->counts);
   RBUF_CLEAR(index->hashes);
   uint32s_index_insert_exact(index, 0, zeros, 0);
   /* wipe additions */
   RBUF_CLEAR(index->additions);
}

void uint32s_index_free(uint32s_index_t *index)
{
   size_t i, cap;
   if (!index)
      return;
   for(i = 0, cap = RHMAP_CAP(index->index); i != cap; i++)
      if (RHMAP_KEY(index->index, i))
         uint32s_bucket_free(&index->index[i]);
   RHMAP_FREE(index->index);
   for(i = 0; i < RBUF_LEN(index->objects); i++)
      free(index->objects[i]);
   RBUF_FREE(index->objects);
   RBUF_FREE(index->counts);
   RBUF_FREE(index->hashes);
   RBUF_FREE(index->additions);
   free(index);
}

uint32_t uint32s_index_count(uint32s_index_t *index)
{
   if (!index || !index->objects)
      return 0;
   return RBUF_LEN(index->objects);
}

#if DEBUG
#define BIN_COUNT 1000
uint32_t bins[BIN_COUNT];
void uint32s_index_print_count_data(uint32s_index_t *index)
{
   uint32_t i;
   /* TODO/FIXME: Don't count or differently count NULL objects entries */
   uint32_t max=1;
   if (!index)
      return;
   for(i = 0; i < BIN_COUNT; i++)
      bins[i] = 0;
   for(i = 1; i < RBUF_LEN(index->counts); i++)
      max = MAX(max, index->counts[i]);
   max = max + 1;
   for(i = 1; i < RBUF_LEN(index->counts); i++)
      bins[(int)((((float)index->counts[i]) / (float)max)*BIN_COUNT)]++;
   for(i = 0; i < BIN_COUNT; i++)
   {
      uint32_t bin_start = MAX(1,(i*(float)max/(float)BIN_COUNT));
      uint32_t bin_end = ((i+1)*(float)max/(float)BIN_COUNT);
      if (bins[i] == 0)
         continue;
      RARCH_DBG("%d--%d: %d\n", bin_start, bin_end, bins[i]);
   }
}
#endif
#endif