bitstream reader optimize patch by (BERO <bero at geocities dot co dot jp>)

Originally committed as revision 1871 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/common.c

@@ -140,9 +140,19 @@ void init_get_bits(GetBitContext *s,
 #ifdef ALT_BITSTREAM_READER
     s->index=0;
 #elif defined LIBMPEG2_BITSTREAM_READER
+#ifdef LIBMPEG2_BITSTREAM_HACK
+  if ((int)buffer&1) {
+     /* word alignment */
+    s->cache = (*buffer++)<<24;
+    s->buffer_ptr = buffer;
+    s->bit_count = 16-8;
+  } else
+#endif
+  {
     s->buffer_ptr = buffer;
     s->bit_count = 16;
     s->cache = 0;
+  }
 #elif defined A32_BITSTREAM_READER
     s->buffer_ptr = (uint32_t*)buffer;
     s->bit_count = 32;

libavcodec/common.h

@@ -16,6 +16,7 @@
 #define ALT_BITSTREAM_READER
 //#define LIBMPEG2_BITSTREAM_READER
 //#define A32_BITSTREAM_READER
+#define LIBMPEG2_BITSTREAM_READER_HACK //add BERO
 
 #ifdef HAVE_AV_CONFIG_H
 /* only include the following when compiling package */
@@ -472,6 +473,16 @@ LAST_SKIP_BITS(name, gb, num)
 for examples see get_bits, show_bits, skip_bits, get_vlc
 */
 
+static inline int unaligned32_be(const void *v)
+{
+#ifdef CONFIG_ALIGN
+	const uint8_t *p=v;
+	return (((p[0]<<8) | p[1])<<16) | (p[2]<<8) | (p[3]);
+#else
+	return be2me_32( unaligned32(v)); //original
+#endif
+}
+
 #ifdef ALT_BITSTREAM_READER
 #   define MIN_CACHE_BITS 25
 
@@ -483,7 +494,7 @@ for examples see get_bits, show_bits, skip_bits, get_vlc
         (gb)->index= name##_index;\
 
 #   define UPDATE_CACHE(name, gb)\
-        name##_cache= be2me_32( unaligned32( ((uint8_t *)(gb)->buffer)+(name##_index>>3) ) ) << (name##_index&0x07);\
+        name##_cache= unaligned32_be( ((uint8_t *)(gb)->buffer)+(name##_index>>3) ) << (name##_index&0x07);\
 
 #   define SKIP_CACHE(name, gb, num)\
         name##_cache <<= (num);\
@@ -528,6 +539,17 @@ static inline int get_bits_count(GetBitContext *s){
         (gb)->cache= name##_cache;\
         (gb)->buffer_ptr= name##_buffer_ptr;\
 
+#ifdef LIBMPEG2_BITSTREAM_READER_HACK
+
+#   define UPDATE_CACHE(name, gb)\
+    if(name##_bit_count >= 0){\
+        name##_cache+= (int)be2me_16(*(uint16_t*)name##_buffer_ptr++) << name##_bit_count;\
+        name##_buffer_ptr+=2;\
+        name##_bit_count-= 16;\
+    }\
+
+#else
+
 #   define UPDATE_CACHE(name, gb)\
     if(name##_bit_count > 0){\
         name##_cache+= ((name##_buffer_ptr[0]<<8) + name##_buffer_ptr[1]) << name##_bit_count;\
@@ -535,6 +557,8 @@ static inline int get_bits_count(GetBitContext *s){
         name##_bit_count-= 16;\
     }\
 
+#endif
+
 #   define SKIP_CACHE(name, gb, num)\
         name##_cache <<= (num);\
 
@@ -630,6 +654,37 @@ static inline int get_bits_count(GetBitContext *s){
 
 #endif
 
+/* add BERO
+   if MSB not set it is negative 
+*/
+static inline int get_xbits(GetBitContext *s, int n){
+    register int tmp;
+    register int32_t cache;
+    OPEN_READER(re, s)
+    UPDATE_CACHE(re, s)
+    cache = GET_CACHE(re,s);
+    if ((int32_t)cache<0) { //MSB=1
+        tmp = NEG_USR32(cache,n);
+    } else {
+    //   tmp = (-1<<n) | NEG_USR32(cache,n) + 1; mpeg12.c algo
+    //   tmp = - (NEG_USR32(cache,n) ^ ((1 << n) - 1)); h263.c algo
+        tmp = - NEG_USR32(~cache,n);
+    }
+    LAST_SKIP_BITS(re, s, n)
+    CLOSE_READER(re, s)
+    return tmp;
+}
+
+static inline int get_sbits(GetBitContext *s, int n){
+    register int tmp;
+    OPEN_READER(re, s)
+    UPDATE_CACHE(re, s)
+    tmp= SHOW_SBITS(re, s, n);
+    LAST_SKIP_BITS(re, s, n)
+    CLOSE_READER(re, s)
+    return tmp;
+}
+
 static inline unsigned int get_bits(GetBitContext *s, int n){
     register int tmp;
     OPEN_READER(re, s)

libavcodec/h263.c

@@ -3562,12 +3562,10 @@ static int h263_decode_block(MpegEncContext * s, DCTELEM * block,
             if(level == -128){
                 if (s->h263_rv10) {
                     /* XXX: should patch encoder too */
-                    level = get_bits(&s->gb, 12);
-                    level= (level + ((-1)<<11)) ^ ((-1)<<11); //sign extension
+                    level = get_sbits(&s->gb, 12);
                 }else{
                     level = get_bits(&s->gb, 5);
-                    level += get_bits(&s->gb, 6)<<5;
-                    level= (level + ((-1)<<10)) ^ ((-1)<<10); //sign extension
+                    level |= get_sbits(&s->gb, 6)<<5;
                 }
             }
         } else {
@@ -3619,9 +3617,7 @@ static inline int mpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr)
     if (code == 0) {
         level = 0;
     } else {
-        level = get_bits(&s->gb, code);
-        if ((level >> (code - 1)) == 0) /* if MSB not set it is negative*/
-            level = - (level ^ ((1 << code) - 1));
+        level = get_xbits(&s->gb, code);
         if (code > 8){
             if(get_bits1(&s->gb)==0){ /* marker */
                 if(s->error_resilience>=2){
@@ -4083,19 +4079,13 @@ static void mpeg4_decode_sprite_trajectory(MpegEncContext * s)
 
         length= get_vlc(&s->gb, &sprite_trajectory);
         if(length){
-            x= get_bits(&s->gb, length);
-
-            if ((x >> (length - 1)) == 0) /* if MSB not set it is negative*/
-                x = - (x ^ ((1 << length) - 1));
+            x= get_xbits(&s->gb, length);
         }
         if(!(s->divx_version==500 && s->divx_build==413)) skip_bits1(&s->gb); /* marker bit */
         
         length= get_vlc(&s->gb, &sprite_trajectory);
         if(length){
-            y=get_bits(&s->gb, length);
-
-            if ((y >> (length - 1)) == 0) /* if MSB not set it is negative*/
-                y = - (y ^ ((1 << length) - 1));
+            y=get_xbits(&s->gb, length);
         }
         skip_bits1(&s->gb); /* marker bit */
 //printf("%d %d %d %d\n", x, y, i, s->sprite_warping_accuracy);

libavcodec/mpeg12.c

@@ -1223,9 +1223,7 @@ static inline int decode_dc(MpegEncContext *s, int component)
     if (code == 0) {
         diff = 0;
     } else {
-        diff = get_bits(&s->gb, code);
-        if ((diff & (1 << (code - 1))) == 0) 
-            diff = (-1 << code) | (diff + 1);
+        diff = get_xbits(&s->gb, code);
     }
     return diff;
 }