Merging upstream version 1.6~pre2.

Signed-off-by: Daniel Baumann <mail@daniel-baumann.ch>

1 files changed, 87 insertions, 47 deletions

diff --git a/encoder.h b/encoder.h
index 4f14075..54b15e7 100644
--- a/encoder.h
+++ b/encoder.h
@@ -15,7 +15,7 @@
     along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-enum { max_num_trials = 1 << 12,
+enum { max_num_trials = 1 << 13,
        price_shift_bits = 6,
        price_step_bits = 2,
        price_step = 1 << price_step_bits };
@@ -53,19 +53,18 @@ extern Prob_prices prob_prices;
 static inline void Prob_prices_init( void )
   {
   int i, j;
-  for( i = price_step / 2; i < bit_model_total; i += price_step )
+  for( i = 0; i < bit_model_total >> price_step_bits; ++i )
     {
-    unsigned val = i;
-    int bits = 0;			/* base 2 logarithm of val */
+    unsigned val = ( i * price_step ) + ( price_step / 2 );
+    int bits = 0;				/* base 2 logarithm of val */
     for( j = 0; j < price_shift_bits; ++j )
       {
       val = val * val;
       bits <<= 1;
       while( val >= 1 << 16 ) { val >>= 1; ++bits; }
       }
-    bits += 15;				/* remaining bits in val */
-    prob_prices[i >> price_step_bits] =
-      ( bit_model_total_bits << price_shift_bits ) - bits;
+    bits += 15;					/* remaining bits in val */
+    prob_prices[i] = ( bit_model_total_bits << price_shift_bits ) - bits;
     }
   }
 
@@ -374,52 +373,93 @@ static inline void Re_encode_matched( struct Range_encoder * const renc,
   while( symbol < 0x10000 );
   }
 
+static inline void Re_encode_len( struct Range_encoder * const renc,
+                                  struct Len_model * const lm,
+                                  int symbol, const int pos_state )
+  {
+  bool bit = ( ( symbol -= min_match_len ) >= len_low_symbols );
+  Re_encode_bit( renc, &lm->choice1, bit );
+  if( !bit )
+    Re_encode_tree( renc, lm->bm_low[pos_state], symbol, len_low_bits );
+  else
+    {
+    bit = ( symbol >= len_low_symbols + len_mid_symbols );
+    Re_encode_bit( renc, &lm->choice2, bit );
+    if( !bit )
+      Re_encode_tree( renc, lm->bm_mid[pos_state],
+                      symbol - len_low_symbols, len_mid_bits );
+    else
+      Re_encode_tree( renc, lm->bm_high,
+                      symbol - len_low_symbols - len_mid_symbols, len_high_bits );
+    }
+  }
+
 
-struct Len_encoder
+struct Len_prices
   {
-  struct Len_model lm;
-  int prices[pos_states][max_len_symbols];
+  const struct Len_model * lm;
   int len_symbols;
+  int count;
+  int prices[pos_states][max_len_symbols];
   int counters[pos_states];
   };
 
-static inline void Lee_update_prices( struct Len_encoder * const le,
-                                      const int pos_state )
+static inline void Lp_update_low_mid_prices( struct Len_prices * const lp,
+                                             const int pos_state )
   {
-  int * const pps = le->prices[pos_state];
-  int tmp = price0( le->lm.choice1 );
+  int * const pps = lp->prices[pos_state];
+  int tmp = price0( lp->lm->choice1 );
   int len = 0;
-  for( ; len < len_low_symbols && len < le->len_symbols; ++len )
-    pps[len] = tmp + price_symbol( le->lm.bm_low[pos_state], len, len_low_bits );
-  tmp = price1( le->lm.choice1 );
-  for( ; len < len_low_symbols + len_mid_symbols && len < le->len_symbols; ++len )
-    pps[len] = tmp + price0( le->lm.choice2 ) +
-               price_symbol( le->lm.bm_mid[pos_state], len - len_low_symbols, len_mid_bits );
-  for( ; len < le->len_symbols; ++len )
-    /* using 4 slots per value makes "Lee_price" faster */
-    le->prices[3][len] = le->prices[2][len] =
-    le->prices[1][len] = le->prices[0][len] =
-      tmp + price1( le->lm.choice2 ) +
-      price_symbol( le->lm.bm_high, len - len_low_symbols - len_mid_symbols, len_high_bits );
-  le->counters[pos_state] = le->len_symbols;
-  }
-
-static inline void Lee_init( struct Len_encoder * const le,
-                             const int match_len_limit )
+  lp->counters[pos_state] = lp->count;
+  for( ; len < len_low_symbols && len < lp->len_symbols; ++len )
+    pps[len] = tmp + price_symbol( lp->lm->bm_low[pos_state], len, len_low_bits );
+  if( len >= lp->len_symbols ) return;
+  tmp = price1( lp->lm->choice1 ) + price0( lp->lm->choice2 );
+  for( ; len < len_low_symbols + len_mid_symbols && len < lp->len_symbols; ++len )
+    pps[len] = tmp +
+               price_symbol( lp->lm->bm_mid[pos_state], len - len_low_symbols, len_mid_bits );
+    }
+
+static inline void Lp_update_high_prices( struct Len_prices * const lp )
+  {
+  const int tmp = price1( lp->lm->choice1 ) + price1( lp->lm->choice2 );
+  int len;
+  for( len = len_low_symbols + len_mid_symbols; len < lp->len_symbols; ++len )
+    /* using 4 slots per value makes "Lp_price" faster */
+    lp->prices[3][len] = lp->prices[2][len] =
+    lp->prices[1][len] = lp->prices[0][len] = tmp +
+      price_symbol( lp->lm->bm_high, len - len_low_symbols - len_mid_symbols, len_high_bits );
+  }
+
+static inline void Lp_init( struct Len_prices * const lp,
+                            const struct Len_model * const lm,
+                            const int match_len_limit )
   {
   int i;
-  Lm_init( &le->lm );
-  le->len_symbols = match_len_limit + 1 - min_match_len;
-  for( i = 0; i < pos_states; ++i ) Lee_update_prices( le, i );
+  lp->lm = lm;
+  lp->len_symbols = match_len_limit + 1 - min_match_len;
+  lp->count = ( match_len_limit > 12 ) ? 1 : lp->len_symbols;
+  for( i = 0; i < pos_states; ++i ) lp->counters[i] = 0;
   }
 
-void Lee_encode( struct Len_encoder * const le,
-                 struct Range_encoder * const renc,
-                 int symbol, const int pos_state );
+static inline void Lp_decrement_counter( struct Len_prices * const lp,
+                                         const int pos_state )
+  { --lp->counters[pos_state]; }
+
+static inline void Lp_update_prices( struct Len_prices * const lp )
+  {
+  int pos_state;
+  bool high_pending = false;
+  for( pos_state = 0; pos_state < pos_states; ++pos_state )
+    if( lp->counters[pos_state] <= 0 )
+      { Lp_update_low_mid_prices( lp, pos_state ); high_pending = true; }
+  if( high_pending && lp->len_symbols > len_low_symbols + len_mid_symbols )
+    Lp_update_high_prices( lp );
+  }
 
-static inline int Lee_price( const struct Len_encoder * const le,
-                             const int symbol, const int pos_state )
-  { return le->prices[pos_state][symbol - min_match_len]; }
+static inline int Lp_price( const struct Len_prices * const lp,
+                            const int symbol, const int pos_state )
+  { return lp->prices[pos_state][symbol - min_match_len]; }
 
 
 enum { infinite_price = 0x0FFFFFFF,
@@ -490,8 +530,10 @@ struct LZ_encoder
 
   struct Matchfinder * matchfinder;
   struct Range_encoder renc;
-  struct Len_encoder match_len_encoder;
-  struct Len_encoder rep_len_encoder;
+  struct Len_model match_len_model;
+  struct Len_model rep_len_model;
+  struct Len_prices match_len_prices;
+  struct Len_prices rep_len_prices;
 
   struct Pair pairs[max_match_len+1];
   struct Trial trials[max_num_trials];
@@ -499,7 +541,6 @@ struct LZ_encoder
   int dis_slot_prices[len_states][2*max_dictionary_bits];
   int dis_prices[len_states][modeled_distances];
   int align_prices[dis_align_size];
-  int align_price_count;
   int num_dis_slots;
   };
 
@@ -558,14 +599,14 @@ static inline int LZe_price_rep0_len( const struct LZ_encoder * const e,
                                       const State state, const int pos_state )
   {
   return LZe_price_rep( e, 0, state, pos_state ) +
-         Lee_price( &e->rep_len_encoder, len, pos_state );
+         Lp_price( &e->rep_len_prices, len, pos_state );
   }
 
 static inline int LZe_price_pair( const struct LZ_encoder * const e,
                                   const int dis, const int len,
                                   const int pos_state )
   {
-  const int price = Lee_price( &e->match_len_encoder, len, pos_state );
+  const int price = Lp_price( &e->match_len_prices, len, pos_state );
   const int len_state = get_len_state( len );
   if( dis < modeled_distances )
     return price + e->dis_prices[len_state][dis];
@@ -600,7 +641,7 @@ static inline void LZe_encode_pair( struct LZ_encoder * const e,
                                     const int pos_state )
   {
   const int dis_slot = get_slot( dis );
-  Lee_encode( &e->match_len_encoder, &e->renc, len, pos_state );
+  Re_encode_len( &e->renc, &e->match_len_model, len, pos_state );
   Re_encode_tree( &e->renc, e->bm_dis_slot[get_len_state(len)], dis_slot,
                   dis_slot_bits );
 
@@ -618,7 +659,6 @@ static inline void LZe_encode_pair( struct LZ_encoder * const e,
       Re_encode( &e->renc, direct_dis >> dis_align_bits,
                  direct_bits - dis_align_bits );
       Re_encode_tree_reversed( &e->renc, e->bm_align, direct_dis, dis_align_bits );
-      --e->align_price_count;
       }
     }
   }