view audioDB-internals.h @ 442:16c5c51a4c32 api-inversion

Add more information to adb_t

In this case, add a vector to store track offsets.

WARNING WARNING WARNING: unlike trackOffsetTable, this is a raw off_t in
bytes of the feature data: in other words, it already has sizeof(double)
and dbH->dim in it.  (I think this is probably minimally unsurprising in
the long run, but as I convert stuff there might well be errors creeping
in...)

typedef struct adb_datum_internal {
  uint32_t nvectors;
  uint32_t dim;
  const char *key;
  void *data;
  void *times;
  void *power;
} adb_datum_internal_t;

struct adb {
  char *path;
  int fd;
  int flags;
  adb_header_t *header;
  std::map<std::string,uint32_t> *keys;
  std::vector<uint32_t> *track_lengths;
  std::vector<off_t> *track_offsets;
};

typedef struct {
  bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
    return strcmp(r1.key, r2.key) < 0;
  }
} adb_result_key_lt;

typedef struct {
  bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
    return r1.qpos < r2.qpos;
  }
} adb_result_qpos_lt;

typedef struct {
  bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
    return r1.dist < r2.dist;
  }
} adb_result_dist_lt;

typedef struct {
  bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
    return r1.dist > r2.dist;
  }
} adb_result_dist_gt;

typedef struct {
  bool operator() (const adb_result_t &r1, const adb_result_t &r2) {
    return ((r1.ipos < r2.ipos) ||
            ((r1.ipos == r2.ipos) && 
             ((r1.qpos < r2.qpos) ||
              ((r1.qpos == r2.qpos) && (strcmp(r1.key, r2.key) < 0)))));
  }
} adb_result_triple_lt;

/* We could go gcc-specific here and use typeof() instead of passing
 * in an explicit type.  Answers on a postcard as to whether that's a
 * good plan or not. */
#define mmap_or_goto_error(type, var, start, length) \
  { void *tmp = mmap(0, length, PROT_READ, MAP_SHARED, adb->fd, (start)); \
    if(tmp == (void *) -1) { \
      goto error; \
    } \
    var = (type) tmp; \
  }

#define maybe_munmap(table, length) \
  { if(table) { \
      munmap(table, length); \
    } \
  }

#define write_or_goto_error(fd, buffer, size) \
  { ssize_t tmp = size; \
    if(write(fd, buffer, size) != tmp) { \
      goto error; \
    } \
  }

#define read_or_goto_error(fd, buffer, size) \
  { ssize_t tmp = size; \
    if(read(fd, buffer, size) != tmp) { \
      goto error; \
    } \
  }

static inline int audiodb_sync_header(adb_t *adb) {
  off_t pos;
  pos = lseek(adb->fd, (off_t) 0, SEEK_CUR);
  if(pos == (off_t) -1) {
    goto error;
  }
  if(lseek(adb->fd, (off_t) 0, SEEK_SET) == (off_t) -1) {
    goto error;
  }
  if(write(adb->fd, adb->header, O2_HEADERSIZE) != O2_HEADERSIZE) {
    goto error;
  }

  /* can be fsync() if fdatasync() is racily exciting and new */
  fdatasync(adb->fd);
  if(lseek(adb->fd, pos, SEEK_SET) == (off_t) -1) {
    goto error;
  }
  return 0;

 error:
  return 1;
}

static inline double audiodb_dot_product(double *p, double *q, size_t count) {
  double result = 0;
  while(count--) {
    result += *p++ * *q++;
  }
  return result;
}

static inline void audiodb_l2norm_buffer(double *d, size_t dim, size_t nvectors, double *l) {
  while(nvectors--) {
    double *d1 = d;
    double *d2 = d;
    *l++ = audiodb_dot_product(d1, d2, dim);
    d += dim;
  }
}

// This is a common pattern in sequence queries: what we are doing is
// taking a window of length seqlen over a buffer of length length,
// and placing the sum of the elements in that window in the first
// element of the window: thus replacing all but the last seqlen
// elements in the buffer with the corresponding windowed sum.
static inline void audiodb_sequence_sum(double *buffer, int length, int seqlen) {
  double tmp1, tmp2, *ps;
  int j, w;

  tmp1 = *buffer;
  j = 1;
  w = seqlen - 1;
  while(w--) {
    *buffer += buffer[j++];
  }
  ps = buffer + 1;
  w = length - seqlen; // +1 - 1
  while(w--) {
    tmp2 = *ps;
    if(isfinite(tmp1)) {
      *ps = *(ps - 1) - tmp1 + *(ps + seqlen - 1);
    } else {
      for(int i = 1; i < seqlen; i++) {
        *ps += *(ps + i);
      }
    }
    tmp1 = tmp2;
    ps++;
  }
}

// In contrast to audiodb_sequence_sum() above,
// audiodb_sequence_sqrt() and audiodb_sequence_average() below are
// simple mappers across the sequence.
static inline void audiodb_sequence_sqrt(double *buffer, int length, int seqlen) {
  int w = length - seqlen + 1;
  while(w--) {
    *buffer = sqrt(*buffer);
    buffer++;
  }
}

static inline void audiodb_sequence_average(double *buffer, int length, int seqlen) {
  int w = length - seqlen + 1;
  while(w--) {
    *buffer /= seqlen;
    buffer++;
  }
}

static inline uint32_t audiodb_key_index(adb_t *adb, const char *key) {
  std::map<std::string,uint32_t>::iterator it;
  it = adb->keys->find(key);
  if(it == adb->keys->end()) {
    return (uint32_t) -1;
  } else {
    return (*it).second;
  }
}

int audiodb_read_data(adb_t *, int, int, double **, size_t *);